xfs: include reservations in quota reporting
[linux-2.6.git] / fs / xfs / xfs_dir2_leaf.c
1 /*
2  * Copyright (c) 2000-2003,2005 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_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_bmap.h"
33 #include "xfs_dir2_format.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
37
38 /*
39  * Local function declarations.
40  */
41 #ifdef DEBUG
42 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
43 #else
44 #define xfs_dir2_leaf_check(dp, bp)
45 #endif
46 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
47                                     int *indexp, xfs_dabuf_t **dbpp);
48 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
49                                     int first, int last);
50 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
51
52
53 /*
54  * Convert a block form directory to a leaf form directory.
55  */
56 int                                             /* error */
57 xfs_dir2_block_to_leaf(
58         xfs_da_args_t           *args,          /* operation arguments */
59         xfs_dabuf_t             *dbp)           /* input block's buffer */
60 {
61         __be16                  *bestsp;        /* leaf's bestsp entries */
62         xfs_dablk_t             blkno;          /* leaf block's bno */
63         xfs_dir2_data_hdr_t     *hdr;           /* block header */
64         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
65         xfs_dir2_block_tail_t   *btp;           /* block's tail */
66         xfs_inode_t             *dp;            /* incore directory inode */
67         int                     error;          /* error return code */
68         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
69         xfs_dir2_db_t           ldb;            /* leaf block's bno */
70         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
71         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
72         xfs_mount_t             *mp;            /* filesystem mount point */
73         int                     needlog;        /* need to log block header */
74         int                     needscan;       /* need to rescan bestfree */
75         xfs_trans_t             *tp;            /* transaction pointer */
76
77         trace_xfs_dir2_block_to_leaf(args);
78
79         dp = args->dp;
80         mp = dp->i_mount;
81         tp = args->trans;
82         /*
83          * Add the leaf block to the inode.
84          * This interface will only put blocks in the leaf/node range.
85          * Since that's empty now, we'll get the root (block 0 in range).
86          */
87         if ((error = xfs_da_grow_inode(args, &blkno))) {
88                 return error;
89         }
90         ldb = xfs_dir2_da_to_db(mp, blkno);
91         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
92         /*
93          * Initialize the leaf block, get a buffer for it.
94          */
95         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
96                 return error;
97         }
98         ASSERT(lbp != NULL);
99         leaf = lbp->data;
100         hdr = dbp->data;
101         xfs_dir2_data_check(dp, dbp);
102         btp = xfs_dir2_block_tail_p(mp, hdr);
103         blp = xfs_dir2_block_leaf_p(btp);
104         /*
105          * Set the counts in the leaf header.
106          */
107         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
108         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
109         /*
110          * Could compact these but I think we always do the conversion
111          * after squeezing out stale entries.
112          */
113         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
114         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
115         needscan = 0;
116         needlog = 1;
117         /*
118          * Make the space formerly occupied by the leaf entries and block
119          * tail be free.
120          */
121         xfs_dir2_data_make_free(tp, dbp,
122                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
123                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
124                                        (char *)blp),
125                 &needlog, &needscan);
126         /*
127          * Fix up the block header, make it a data block.
128          */
129         hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
130         if (needscan)
131                 xfs_dir2_data_freescan(mp, hdr, &needlog);
132         /*
133          * Set up leaf tail and bests table.
134          */
135         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
136         ltp->bestcount = cpu_to_be32(1);
137         bestsp = xfs_dir2_leaf_bests_p(ltp);
138         bestsp[0] =  hdr->bestfree[0].length;
139         /*
140          * Log the data header and leaf bests table.
141          */
142         if (needlog)
143                 xfs_dir2_data_log_header(tp, dbp);
144         xfs_dir2_leaf_check(dp, lbp);
145         xfs_dir2_data_check(dp, dbp);
146         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
147         xfs_da_buf_done(lbp);
148         return 0;
149 }
150
151 STATIC void
152 xfs_dir2_leaf_find_stale(
153         struct xfs_dir2_leaf    *leaf,
154         int                     index,
155         int                     *lowstale,
156         int                     *highstale)
157 {
158         /*
159          * Find the first stale entry before our index, if any.
160          */
161         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
162                 if (leaf->ents[*lowstale].address ==
163                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
164                         break;
165         }
166
167         /*
168          * Find the first stale entry at or after our index, if any.
169          * Stop if the result would require moving more entries than using
170          * lowstale.
171          */
172         for (*highstale = index;
173              *highstale < be16_to_cpu(leaf->hdr.count);
174              ++*highstale) {
175                 if (leaf->ents[*highstale].address ==
176                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
177                         break;
178                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
179                         break;
180         }
181 }
182
183 struct xfs_dir2_leaf_entry *
184 xfs_dir2_leaf_find_entry(
185         xfs_dir2_leaf_t         *leaf,          /* leaf structure */
186         int                     index,          /* leaf table position */
187         int                     compact,        /* need to compact leaves */
188         int                     lowstale,       /* index of prev stale leaf */
189         int                     highstale,      /* index of next stale leaf */
190         int                     *lfloglow,      /* low leaf logging index */
191         int                     *lfloghigh)     /* high leaf logging index */
192 {
193         if (!leaf->hdr.stale) {
194                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
195
196                 /*
197                  * Now we need to make room to insert the leaf entry.
198                  *
199                  * If there are no stale entries, just insert a hole at index.
200                  */
201                 lep = &leaf->ents[index];
202                 if (index < be16_to_cpu(leaf->hdr.count))
203                         memmove(lep + 1, lep,
204                                 (be16_to_cpu(leaf->hdr.count) - index) *
205                                  sizeof(*lep));
206
207                 /*
208                  * Record low and high logging indices for the leaf.
209                  */
210                 *lfloglow = index;
211                 *lfloghigh = be16_to_cpu(leaf->hdr.count);
212                 be16_add_cpu(&leaf->hdr.count, 1);
213                 return lep;
214         }
215
216         /*
217          * There are stale entries.
218          *
219          * We will use one of them for the new entry.  It's probably not at
220          * the right location, so we'll have to shift some up or down first.
221          *
222          * If we didn't compact before, we need to find the nearest stale
223          * entries before and after our insertion point.
224          */
225         if (compact == 0)
226                 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
227
228         /*
229          * If the low one is better, use it.
230          */
231         if (lowstale >= 0 &&
232             (highstale == be16_to_cpu(leaf->hdr.count) ||
233              index - lowstale - 1 < highstale - index)) {
234                 ASSERT(index - lowstale - 1 >= 0);
235                 ASSERT(leaf->ents[lowstale].address ==
236                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
237
238                 /*
239                  * Copy entries up to cover the stale entry and make room
240                  * for the new entry.
241                  */
242                 if (index - lowstale - 1 > 0) {
243                         memmove(&leaf->ents[lowstale],
244                                 &leaf->ents[lowstale + 1],
245                                 (index - lowstale - 1) *
246                                 sizeof(xfs_dir2_leaf_entry_t));
247                 }
248                 *lfloglow = MIN(lowstale, *lfloglow);
249                 *lfloghigh = MAX(index - 1, *lfloghigh);
250                 be16_add_cpu(&leaf->hdr.stale, -1);
251                 return &leaf->ents[index - 1];
252         }
253
254         /*
255          * The high one is better, so use that one.
256          */
257         ASSERT(highstale - index >= 0);
258         ASSERT(leaf->ents[highstale].address ==
259                cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
260
261         /*
262          * Copy entries down to cover the stale entry and make room for the
263          * new entry.
264          */
265         if (highstale - index > 0) {
266                 memmove(&leaf->ents[index + 1],
267                         &leaf->ents[index],
268                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
269         }
270         *lfloglow = MIN(index, *lfloglow);
271         *lfloghigh = MAX(highstale, *lfloghigh);
272         be16_add_cpu(&leaf->hdr.stale, -1);
273         return &leaf->ents[index];
274 }
275
276 /*
277  * Add an entry to a leaf form directory.
278  */
279 int                                             /* error */
280 xfs_dir2_leaf_addname(
281         xfs_da_args_t           *args)          /* operation arguments */
282 {
283         __be16                  *bestsp;        /* freespace table in leaf */
284         int                     compact;        /* need to compact leaves */
285         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
286         xfs_dabuf_t             *dbp;           /* data block buffer */
287         xfs_dir2_data_entry_t   *dep;           /* data block entry */
288         xfs_inode_t             *dp;            /* incore directory inode */
289         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
290         int                     error;          /* error return value */
291         int                     grown;          /* allocated new data block */
292         int                     highstale;      /* index of next stale leaf */
293         int                     i;              /* temporary, index */
294         int                     index;          /* leaf table position */
295         xfs_dabuf_t             *lbp;           /* leaf's buffer */
296         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
297         int                     length;         /* length of new entry */
298         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
299         int                     lfloglow;       /* low leaf logging index */
300         int                     lfloghigh;      /* high leaf logging index */
301         int                     lowstale;       /* index of prev stale leaf */
302         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
303         xfs_mount_t             *mp;            /* filesystem mount point */
304         int                     needbytes;      /* leaf block bytes needed */
305         int                     needlog;        /* need to log data header */
306         int                     needscan;       /* need to rescan data free */
307         __be16                  *tagp;          /* end of data entry */
308         xfs_trans_t             *tp;            /* transaction pointer */
309         xfs_dir2_db_t           use_block;      /* data block number */
310
311         trace_xfs_dir2_leaf_addname(args);
312
313         dp = args->dp;
314         tp = args->trans;
315         mp = dp->i_mount;
316         /*
317          * Read the leaf block.
318          */
319         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
320                 XFS_DATA_FORK);
321         if (error) {
322                 return error;
323         }
324         ASSERT(lbp != NULL);
325         /*
326          * Look up the entry by hash value and name.
327          * We know it's not there, our caller has already done a lookup.
328          * So the index is of the entry to insert in front of.
329          * But if there are dup hash values the index is of the first of those.
330          */
331         index = xfs_dir2_leaf_search_hash(args, lbp);
332         leaf = lbp->data;
333         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
334         bestsp = xfs_dir2_leaf_bests_p(ltp);
335         length = xfs_dir2_data_entsize(args->namelen);
336         /*
337          * See if there are any entries with the same hash value
338          * and space in their block for the new entry.
339          * This is good because it puts multiple same-hash value entries
340          * in a data block, improving the lookup of those entries.
341          */
342         for (use_block = -1, lep = &leaf->ents[index];
343              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
344              index++, lep++) {
345                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
346                         continue;
347                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
348                 ASSERT(i < be32_to_cpu(ltp->bestcount));
349                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
350                 if (be16_to_cpu(bestsp[i]) >= length) {
351                         use_block = i;
352                         break;
353                 }
354         }
355         /*
356          * Didn't find a block yet, linear search all the data blocks.
357          */
358         if (use_block == -1) {
359                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
360                         /*
361                          * Remember a block we see that's missing.
362                          */
363                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
364                             use_block == -1)
365                                 use_block = i;
366                         else if (be16_to_cpu(bestsp[i]) >= length) {
367                                 use_block = i;
368                                 break;
369                         }
370                 }
371         }
372         /*
373          * How many bytes do we need in the leaf block?
374          */
375         needbytes = 0;
376         if (!leaf->hdr.stale)
377                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
378         if (use_block == -1)
379                 needbytes += sizeof(xfs_dir2_data_off_t);
380
381         /*
382          * Now kill use_block if it refers to a missing block, so we
383          * can use it as an indication of allocation needed.
384          */
385         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
386                 use_block = -1;
387         /*
388          * If we don't have enough free bytes but we can make enough
389          * by compacting out stale entries, we'll do that.
390          */
391         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
392                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
393                 compact = 1;
394         }
395         /*
396          * Otherwise if we don't have enough free bytes we need to
397          * convert to node form.
398          */
399         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
400                                                 leaf->hdr.count)] < needbytes) {
401                 /*
402                  * Just checking or no space reservation, give up.
403                  */
404                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
405                                                         args->total == 0) {
406                         xfs_da_brelse(tp, lbp);
407                         return XFS_ERROR(ENOSPC);
408                 }
409                 /*
410                  * Convert to node form.
411                  */
412                 error = xfs_dir2_leaf_to_node(args, lbp);
413                 xfs_da_buf_done(lbp);
414                 if (error)
415                         return error;
416                 /*
417                  * Then add the new entry.
418                  */
419                 return xfs_dir2_node_addname(args);
420         }
421         /*
422          * Otherwise it will fit without compaction.
423          */
424         else
425                 compact = 0;
426         /*
427          * If just checking, then it will fit unless we needed to allocate
428          * a new data block.
429          */
430         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
431                 xfs_da_brelse(tp, lbp);
432                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
433         }
434         /*
435          * If no allocations are allowed, return now before we've
436          * changed anything.
437          */
438         if (args->total == 0 && use_block == -1) {
439                 xfs_da_brelse(tp, lbp);
440                 return XFS_ERROR(ENOSPC);
441         }
442         /*
443          * Need to compact the leaf entries, removing stale ones.
444          * Leave one stale entry behind - the one closest to our
445          * insertion index - and we'll shift that one to our insertion
446          * point later.
447          */
448         if (compact) {
449                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
450                         &lfloglow, &lfloghigh);
451         }
452         /*
453          * There are stale entries, so we'll need log-low and log-high
454          * impossibly bad values later.
455          */
456         else if (be16_to_cpu(leaf->hdr.stale)) {
457                 lfloglow = be16_to_cpu(leaf->hdr.count);
458                 lfloghigh = -1;
459         }
460         /*
461          * If there was no data block space found, we need to allocate
462          * a new one.
463          */
464         if (use_block == -1) {
465                 /*
466                  * Add the new data block.
467                  */
468                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
469                                 &use_block))) {
470                         xfs_da_brelse(tp, lbp);
471                         return error;
472                 }
473                 /*
474                  * Initialize the block.
475                  */
476                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
477                         xfs_da_brelse(tp, lbp);
478                         return error;
479                 }
480                 /*
481                  * If we're adding a new data block on the end we need to
482                  * extend the bests table.  Copy it up one entry.
483                  */
484                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
485                         bestsp--;
486                         memmove(&bestsp[0], &bestsp[1],
487                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
488                         be32_add_cpu(&ltp->bestcount, 1);
489                         xfs_dir2_leaf_log_tail(tp, lbp);
490                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
491                 }
492                 /*
493                  * If we're filling in a previously empty block just log it.
494                  */
495                 else
496                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
497                 hdr = dbp->data;
498                 bestsp[use_block] = hdr->bestfree[0].length;
499                 grown = 1;
500         }
501         /*
502          * Already had space in some data block.
503          * Just read that one in.
504          */
505         else {
506                 if ((error =
507                     xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, use_block),
508                             -1, &dbp, XFS_DATA_FORK))) {
509                         xfs_da_brelse(tp, lbp);
510                         return error;
511                 }
512                 hdr = dbp->data;
513                 grown = 0;
514         }
515         xfs_dir2_data_check(dp, dbp);
516         /*
517          * Point to the biggest freespace in our data block.
518          */
519         dup = (xfs_dir2_data_unused_t *)
520               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
521         ASSERT(be16_to_cpu(dup->length) >= length);
522         needscan = needlog = 0;
523         /*
524          * Mark the initial part of our freespace in use for the new entry.
525          */
526         xfs_dir2_data_use_free(tp, dbp, dup,
527                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
528                 &needlog, &needscan);
529         /*
530          * Initialize our new entry (at last).
531          */
532         dep = (xfs_dir2_data_entry_t *)dup;
533         dep->inumber = cpu_to_be64(args->inumber);
534         dep->namelen = args->namelen;
535         memcpy(dep->name, args->name, dep->namelen);
536         tagp = xfs_dir2_data_entry_tag_p(dep);
537         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
538         /*
539          * Need to scan fix up the bestfree table.
540          */
541         if (needscan)
542                 xfs_dir2_data_freescan(mp, hdr, &needlog);
543         /*
544          * Need to log the data block's header.
545          */
546         if (needlog)
547                 xfs_dir2_data_log_header(tp, dbp);
548         xfs_dir2_data_log_entry(tp, dbp, dep);
549         /*
550          * If the bests table needs to be changed, do it.
551          * Log the change unless we've already done that.
552          */
553         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
554                 bestsp[use_block] = hdr->bestfree[0].length;
555                 if (!grown)
556                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
557         }
558
559         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
560                                        highstale, &lfloglow, &lfloghigh);
561
562         /*
563          * Fill in the new leaf entry.
564          */
565         lep->hashval = cpu_to_be32(args->hashval);
566         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
567                                 be16_to_cpu(*tagp)));
568         /*
569          * Log the leaf fields and give up the buffers.
570          */
571         xfs_dir2_leaf_log_header(tp, lbp);
572         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
573         xfs_dir2_leaf_check(dp, lbp);
574         xfs_da_buf_done(lbp);
575         xfs_dir2_data_check(dp, dbp);
576         xfs_da_buf_done(dbp);
577         return 0;
578 }
579
580 #ifdef DEBUG
581 /*
582  * Check the internal consistency of a leaf1 block.
583  * Pop an assert if something is wrong.
584  */
585 STATIC void
586 xfs_dir2_leaf_check(
587         xfs_inode_t             *dp,            /* incore directory inode */
588         xfs_dabuf_t             *bp)            /* leaf's buffer */
589 {
590         int                     i;              /* leaf index */
591         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
592         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
593         xfs_mount_t             *mp;            /* filesystem mount point */
594         int                     stale;          /* count of stale leaves */
595
596         leaf = bp->data;
597         mp = dp->i_mount;
598         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
599         /*
600          * This value is not restrictive enough.
601          * Should factor in the size of the bests table as well.
602          * We can deduce a value for that from di_size.
603          */
604         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
605         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
606         /*
607          * Leaves and bests don't overlap.
608          */
609         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
610                (char *)xfs_dir2_leaf_bests_p(ltp));
611         /*
612          * Check hash value order, count stale entries.
613          */
614         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
615                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
616                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
617                                be32_to_cpu(leaf->ents[i + 1].hashval));
618                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
619                         stale++;
620         }
621         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
622 }
623 #endif  /* DEBUG */
624
625 /*
626  * Compact out any stale entries in the leaf.
627  * Log the header and changed leaf entries, if any.
628  */
629 void
630 xfs_dir2_leaf_compact(
631         xfs_da_args_t   *args,          /* operation arguments */
632         xfs_dabuf_t     *bp)            /* leaf buffer */
633 {
634         int             from;           /* source leaf index */
635         xfs_dir2_leaf_t *leaf;          /* leaf structure */
636         int             loglow;         /* first leaf entry to log */
637         int             to;             /* target leaf index */
638
639         leaf = bp->data;
640         if (!leaf->hdr.stale) {
641                 return;
642         }
643         /*
644          * Compress out the stale entries in place.
645          */
646         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
647                 if (leaf->ents[from].address ==
648                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
649                         continue;
650                 /*
651                  * Only actually copy the entries that are different.
652                  */
653                 if (from > to) {
654                         if (loglow == -1)
655                                 loglow = to;
656                         leaf->ents[to] = leaf->ents[from];
657                 }
658                 to++;
659         }
660         /*
661          * Update and log the header, log the leaf entries.
662          */
663         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
664         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
665         leaf->hdr.stale = 0;
666         xfs_dir2_leaf_log_header(args->trans, bp);
667         if (loglow != -1)
668                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
669 }
670
671 /*
672  * Compact the leaf entries, removing stale ones.
673  * Leave one stale entry behind - the one closest to our
674  * insertion index - and the caller will shift that one to our insertion
675  * point later.
676  * Return new insertion index, where the remaining stale entry is,
677  * and leaf logging indices.
678  */
679 void
680 xfs_dir2_leaf_compact_x1(
681         xfs_dabuf_t     *bp,            /* leaf buffer */
682         int             *indexp,        /* insertion index */
683         int             *lowstalep,     /* out: stale entry before us */
684         int             *highstalep,    /* out: stale entry after us */
685         int             *lowlogp,       /* out: low log index */
686         int             *highlogp)      /* out: high log index */
687 {
688         int             from;           /* source copy index */
689         int             highstale;      /* stale entry at/after index */
690         int             index;          /* insertion index */
691         int             keepstale;      /* source index of kept stale */
692         xfs_dir2_leaf_t *leaf;          /* leaf structure */
693         int             lowstale;       /* stale entry before index */
694         int             newindex=0;     /* new insertion index */
695         int             to;             /* destination copy index */
696
697         leaf = bp->data;
698         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
699         index = *indexp;
700
701         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
702
703         /*
704          * Pick the better of lowstale and highstale.
705          */
706         if (lowstale >= 0 &&
707             (highstale == be16_to_cpu(leaf->hdr.count) ||
708              index - lowstale <= highstale - index))
709                 keepstale = lowstale;
710         else
711                 keepstale = highstale;
712         /*
713          * Copy the entries in place, removing all the stale entries
714          * except keepstale.
715          */
716         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
717                 /*
718                  * Notice the new value of index.
719                  */
720                 if (index == from)
721                         newindex = to;
722                 if (from != keepstale &&
723                     leaf->ents[from].address ==
724                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
725                         if (from == to)
726                                 *lowlogp = to;
727                         continue;
728                 }
729                 /*
730                  * Record the new keepstale value for the insertion.
731                  */
732                 if (from == keepstale)
733                         lowstale = highstale = to;
734                 /*
735                  * Copy only the entries that have moved.
736                  */
737                 if (from > to)
738                         leaf->ents[to] = leaf->ents[from];
739                 to++;
740         }
741         ASSERT(from > to);
742         /*
743          * If the insertion point was past the last entry,
744          * set the new insertion point accordingly.
745          */
746         if (index == from)
747                 newindex = to;
748         *indexp = newindex;
749         /*
750          * Adjust the leaf header values.
751          */
752         be16_add_cpu(&leaf->hdr.count, -(from - to));
753         leaf->hdr.stale = cpu_to_be16(1);
754         /*
755          * Remember the low/high stale value only in the "right"
756          * direction.
757          */
758         if (lowstale >= newindex)
759                 lowstale = -1;
760         else
761                 highstale = be16_to_cpu(leaf->hdr.count);
762         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
763         *lowstalep = lowstale;
764         *highstalep = highstale;
765 }
766
767 /*
768  * Getdents (readdir) for leaf and node directories.
769  * This reads the data blocks only, so is the same for both forms.
770  */
771 int                                             /* error */
772 xfs_dir2_leaf_getdents(
773         xfs_inode_t             *dp,            /* incore directory inode */
774         void                    *dirent,
775         size_t                  bufsize,
776         xfs_off_t               *offset,
777         filldir_t               filldir)
778 {
779         xfs_dabuf_t             *bp;            /* data block buffer */
780         int                     byteoff;        /* offset in current block */
781         xfs_dir2_db_t           curdb;          /* db for current block */
782         xfs_dir2_off_t          curoff;         /* current overall offset */
783         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
784         xfs_dir2_data_entry_t   *dep;           /* data entry */
785         xfs_dir2_data_unused_t  *dup;           /* unused entry */
786         int                     error = 0;      /* error return value */
787         int                     i;              /* temporary loop index */
788         int                     j;              /* temporary loop index */
789         int                     length;         /* temporary length value */
790         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
791         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
792         xfs_dablk_t             map_off;        /* last mapped file offset */
793         int                     map_size;       /* total entries in *map */
794         int                     map_valid;      /* valid entries in *map */
795         xfs_mount_t             *mp;            /* filesystem mount point */
796         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
797         int                     nmap;           /* mappings to ask xfs_bmapi */
798         char                    *ptr = NULL;    /* pointer to current data */
799         int                     ra_current;     /* number of read-ahead blks */
800         int                     ra_index;       /* *map index for read-ahead */
801         int                     ra_offset;      /* map entry offset for ra */
802         int                     ra_want;        /* readahead count wanted */
803
804         /*
805          * If the offset is at or past the largest allowed value,
806          * give up right away.
807          */
808         if (*offset >= XFS_DIR2_MAX_DATAPTR)
809                 return 0;
810
811         mp = dp->i_mount;
812
813         /*
814          * Set up to bmap a number of blocks based on the caller's
815          * buffer size, the directory block size, and the filesystem
816          * block size.
817          */
818         map_size = howmany(bufsize + mp->m_dirblksize, mp->m_sb.sb_blocksize);
819         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
820         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
821         bp = NULL;
822
823         /*
824          * Inside the loop we keep the main offset value as a byte offset
825          * in the directory file.
826          */
827         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
828
829         /*
830          * Force this conversion through db so we truncate the offset
831          * down to get the start of the data block.
832          */
833         map_off = xfs_dir2_db_to_da(mp, xfs_dir2_byte_to_db(mp, curoff));
834         /*
835          * Loop over directory entries until we reach the end offset.
836          * Get more blocks and readahead as necessary.
837          */
838         while (curoff < XFS_DIR2_LEAF_OFFSET) {
839                 /*
840                  * If we have no buffer, or we're off the end of the
841                  * current buffer, need to get another one.
842                  */
843                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
844                         /*
845                          * If we have a buffer, we need to release it and
846                          * take it out of the mapping.
847                          */
848                         if (bp) {
849                                 xfs_da_brelse(NULL, bp);
850                                 bp = NULL;
851                                 map_blocks -= mp->m_dirblkfsbs;
852                                 /*
853                                  * Loop to get rid of the extents for the
854                                  * directory block.
855                                  */
856                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
857                                         j = MIN((int)map->br_blockcount, i);
858                                         map->br_blockcount -= j;
859                                         map->br_startblock += j;
860                                         map->br_startoff += j;
861                                         /*
862                                          * If mapping is done, pitch it from
863                                          * the table.
864                                          */
865                                         if (!map->br_blockcount && --map_valid)
866                                                 memmove(&map[0], &map[1],
867                                                         sizeof(map[0]) *
868                                                         map_valid);
869                                         i -= j;
870                                 }
871                         }
872                         /*
873                          * Recalculate the readahead blocks wanted.
874                          */
875                         ra_want = howmany(bufsize + mp->m_dirblksize,
876                                           mp->m_sb.sb_blocksize) - 1;
877                         ASSERT(ra_want >= 0);
878
879                         /*
880                          * If we don't have as many as we want, and we haven't
881                          * run out of data blocks, get some more mappings.
882                          */
883                         if (1 + ra_want > map_blocks &&
884                             map_off <
885                             xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
886                                 /*
887                                  * Get more bmaps, fill in after the ones
888                                  * we already have in the table.
889                                  */
890                                 nmap = map_size - map_valid;
891                                 error = xfs_bmapi_read(dp, map_off,
892                                         xfs_dir2_byte_to_da(mp,
893                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
894                                         &map[map_valid], &nmap, 0);
895                                 /*
896                                  * Don't know if we should ignore this or
897                                  * try to return an error.
898                                  * The trouble with returning errors
899                                  * is that readdir will just stop without
900                                  * actually passing the error through.
901                                  */
902                                 if (error)
903                                         break;  /* XXX */
904                                 /*
905                                  * If we got all the mappings we asked for,
906                                  * set the final map offset based on the
907                                  * last bmap value received.
908                                  * Otherwise, we've reached the end.
909                                  */
910                                 if (nmap == map_size - map_valid)
911                                         map_off =
912                                         map[map_valid + nmap - 1].br_startoff +
913                                         map[map_valid + nmap - 1].br_blockcount;
914                                 else
915                                         map_off =
916                                                 xfs_dir2_byte_to_da(mp,
917                                                         XFS_DIR2_LEAF_OFFSET);
918                                 /*
919                                  * Look for holes in the mapping, and
920                                  * eliminate them.  Count up the valid blocks.
921                                  */
922                                 for (i = map_valid; i < map_valid + nmap; ) {
923                                         if (map[i].br_startblock ==
924                                             HOLESTARTBLOCK) {
925                                                 nmap--;
926                                                 length = map_valid + nmap - i;
927                                                 if (length)
928                                                         memmove(&map[i],
929                                                                 &map[i + 1],
930                                                                 sizeof(map[i]) *
931                                                                 length);
932                                         } else {
933                                                 map_blocks +=
934                                                         map[i].br_blockcount;
935                                                 i++;
936                                         }
937                                 }
938                                 map_valid += nmap;
939                         }
940                         /*
941                          * No valid mappings, so no more data blocks.
942                          */
943                         if (!map_valid) {
944                                 curoff = xfs_dir2_da_to_byte(mp, map_off);
945                                 break;
946                         }
947                         /*
948                          * Read the directory block starting at the first
949                          * mapping.
950                          */
951                         curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
952                         error = xfs_da_read_buf(NULL, dp, map->br_startoff,
953                                 map->br_blockcount >= mp->m_dirblkfsbs ?
954                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
955                                     -1,
956                                 &bp, XFS_DATA_FORK);
957                         /*
958                          * Should just skip over the data block instead
959                          * of giving up.
960                          */
961                         if (error)
962                                 break;  /* XXX */
963                         /*
964                          * Adjust the current amount of read-ahead: we just
965                          * read a block that was previously ra.
966                          */
967                         if (ra_current)
968                                 ra_current -= mp->m_dirblkfsbs;
969                         /*
970                          * Do we need more readahead?
971                          */
972                         for (ra_index = ra_offset = i = 0;
973                              ra_want > ra_current && i < map_blocks;
974                              i += mp->m_dirblkfsbs) {
975                                 ASSERT(ra_index < map_valid);
976                                 /*
977                                  * Read-ahead a contiguous directory block.
978                                  */
979                                 if (i > ra_current &&
980                                     map[ra_index].br_blockcount >=
981                                     mp->m_dirblkfsbs) {
982                                         xfs_buf_readahead(mp->m_ddev_targp,
983                                                 XFS_FSB_TO_DADDR(mp,
984                                                    map[ra_index].br_startblock +
985                                                    ra_offset),
986                                                 (int)BTOBB(mp->m_dirblksize));
987                                         ra_current = i;
988                                 }
989                                 /*
990                                  * Read-ahead a non-contiguous directory block.
991                                  * This doesn't use our mapping, but this
992                                  * is a very rare case.
993                                  */
994                                 else if (i > ra_current) {
995                                         (void)xfs_da_reada_buf(NULL, dp,
996                                                 map[ra_index].br_startoff +
997                                                 ra_offset, XFS_DATA_FORK);
998                                         ra_current = i;
999                                 }
1000                                 /*
1001                                  * Advance offset through the mapping table.
1002                                  */
1003                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1004                                         /*
1005                                          * The rest of this extent but not
1006                                          * more than a dir block.
1007                                          */
1008                                         length = MIN(mp->m_dirblkfsbs,
1009                                                 (int)(map[ra_index].br_blockcount -
1010                                                 ra_offset));
1011                                         j += length;
1012                                         ra_offset += length;
1013                                         /*
1014                                          * Advance to the next mapping if
1015                                          * this one is used up.
1016                                          */
1017                                         if (ra_offset ==
1018                                             map[ra_index].br_blockcount) {
1019                                                 ra_offset = 0;
1020                                                 ra_index++;
1021                                         }
1022                                 }
1023                         }
1024                         /*
1025                          * Having done a read, we need to set a new offset.
1026                          */
1027                         newoff = xfs_dir2_db_off_to_byte(mp, curdb, 0);
1028                         /*
1029                          * Start of the current block.
1030                          */
1031                         if (curoff < newoff)
1032                                 curoff = newoff;
1033                         /*
1034                          * Make sure we're in the right block.
1035                          */
1036                         else if (curoff > newoff)
1037                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1038                                        curdb);
1039                         hdr = bp->data;
1040                         xfs_dir2_data_check(dp, bp);
1041                         /*
1042                          * Find our position in the block.
1043                          */
1044                         ptr = (char *)(hdr + 1);
1045                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1046                         /*
1047                          * Skip past the header.
1048                          */
1049                         if (byteoff == 0)
1050                                 curoff += (uint)sizeof(*hdr);
1051                         /*
1052                          * Skip past entries until we reach our offset.
1053                          */
1054                         else {
1055                                 while ((char *)ptr - (char *)hdr < byteoff) {
1056                                         dup = (xfs_dir2_data_unused_t *)ptr;
1057
1058                                         if (be16_to_cpu(dup->freetag)
1059                                                   == XFS_DIR2_DATA_FREE_TAG) {
1060
1061                                                 length = be16_to_cpu(dup->length);
1062                                                 ptr += length;
1063                                                 continue;
1064                                         }
1065                                         dep = (xfs_dir2_data_entry_t *)ptr;
1066                                         length =
1067                                            xfs_dir2_data_entsize(dep->namelen);
1068                                         ptr += length;
1069                                 }
1070                                 /*
1071                                  * Now set our real offset.
1072                                  */
1073                                 curoff =
1074                                         xfs_dir2_db_off_to_byte(mp,
1075                                             xfs_dir2_byte_to_db(mp, curoff),
1076                                             (char *)ptr - (char *)hdr);
1077                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1078                                         continue;
1079                                 }
1080                         }
1081                 }
1082                 /*
1083                  * We have a pointer to an entry.
1084                  * Is it a live one?
1085                  */
1086                 dup = (xfs_dir2_data_unused_t *)ptr;
1087                 /*
1088                  * No, it's unused, skip over it.
1089                  */
1090                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1091                         length = be16_to_cpu(dup->length);
1092                         ptr += length;
1093                         curoff += length;
1094                         continue;
1095                 }
1096
1097                 dep = (xfs_dir2_data_entry_t *)ptr;
1098                 length = xfs_dir2_data_entsize(dep->namelen);
1099
1100                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1101                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1102                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1103                         break;
1104
1105                 /*
1106                  * Advance to next entry in the block.
1107                  */
1108                 ptr += length;
1109                 curoff += length;
1110                 /* bufsize may have just been a guess; don't go negative */
1111                 bufsize = bufsize > length ? bufsize - length : 0;
1112         }
1113
1114         /*
1115          * All done.  Set output offset value to current offset.
1116          */
1117         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1118                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1119         else
1120                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1121         kmem_free(map);
1122         if (bp)
1123                 xfs_da_brelse(NULL, bp);
1124         return error;
1125 }
1126
1127 /*
1128  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1129  */
1130 int
1131 xfs_dir2_leaf_init(
1132         xfs_da_args_t           *args,          /* operation arguments */
1133         xfs_dir2_db_t           bno,            /* directory block number */
1134         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1135         int                     magic)          /* magic number for block */
1136 {
1137         xfs_dabuf_t             *bp;            /* leaf buffer */
1138         xfs_inode_t             *dp;            /* incore directory inode */
1139         int                     error;          /* error return code */
1140         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1141         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1142         xfs_mount_t             *mp;            /* filesystem mount point */
1143         xfs_trans_t             *tp;            /* transaction pointer */
1144
1145         dp = args->dp;
1146         ASSERT(dp != NULL);
1147         tp = args->trans;
1148         mp = dp->i_mount;
1149         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1150                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1151         /*
1152          * Get the buffer for the block.
1153          */
1154         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1155                 XFS_DATA_FORK);
1156         if (error) {
1157                 return error;
1158         }
1159         ASSERT(bp != NULL);
1160         leaf = bp->data;
1161         /*
1162          * Initialize the header.
1163          */
1164         leaf->hdr.info.magic = cpu_to_be16(magic);
1165         leaf->hdr.info.forw = 0;
1166         leaf->hdr.info.back = 0;
1167         leaf->hdr.count = 0;
1168         leaf->hdr.stale = 0;
1169         xfs_dir2_leaf_log_header(tp, bp);
1170         /*
1171          * If it's a leaf-format directory initialize the tail.
1172          * In this case our caller has the real bests table to copy into
1173          * the block.
1174          */
1175         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1176                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1177                 ltp->bestcount = 0;
1178                 xfs_dir2_leaf_log_tail(tp, bp);
1179         }
1180         *bpp = bp;
1181         return 0;
1182 }
1183
1184 /*
1185  * Log the bests entries indicated from a leaf1 block.
1186  */
1187 static void
1188 xfs_dir2_leaf_log_bests(
1189         xfs_trans_t             *tp,            /* transaction pointer */
1190         xfs_dabuf_t             *bp,            /* leaf buffer */
1191         int                     first,          /* first entry to log */
1192         int                     last)           /* last entry to log */
1193 {
1194         __be16                  *firstb;        /* pointer to first entry */
1195         __be16                  *lastb;         /* pointer to last entry */
1196         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1197         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1198
1199         leaf = bp->data;
1200         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1201         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1202         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1203         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1204         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1205                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1206 }
1207
1208 /*
1209  * Log the leaf entries indicated from a leaf1 or leafn block.
1210  */
1211 void
1212 xfs_dir2_leaf_log_ents(
1213         xfs_trans_t             *tp,            /* transaction pointer */
1214         xfs_dabuf_t             *bp,            /* leaf buffer */
1215         int                     first,          /* first entry to log */
1216         int                     last)           /* last entry to log */
1217 {
1218         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1219         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1220         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1221
1222         leaf = bp->data;
1223         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1224                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1225         firstlep = &leaf->ents[first];
1226         lastlep = &leaf->ents[last];
1227         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1228                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1229 }
1230
1231 /*
1232  * Log the header of the leaf1 or leafn block.
1233  */
1234 void
1235 xfs_dir2_leaf_log_header(
1236         xfs_trans_t             *tp,            /* transaction pointer */
1237         xfs_dabuf_t             *bp)            /* leaf buffer */
1238 {
1239         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1240
1241         leaf = bp->data;
1242         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1243                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1244         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1245                 (uint)(sizeof(leaf->hdr) - 1));
1246 }
1247
1248 /*
1249  * Log the tail of the leaf1 block.
1250  */
1251 STATIC void
1252 xfs_dir2_leaf_log_tail(
1253         xfs_trans_t             *tp,            /* transaction pointer */
1254         xfs_dabuf_t             *bp)            /* leaf buffer */
1255 {
1256         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1257         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1258         xfs_mount_t             *mp;            /* filesystem mount point */
1259
1260         mp = tp->t_mountp;
1261         leaf = bp->data;
1262         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1263         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1264         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1265                 (uint)(mp->m_dirblksize - 1));
1266 }
1267
1268 /*
1269  * Look up the entry referred to by args in the leaf format directory.
1270  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1271  * is also used by the node-format code.
1272  */
1273 int
1274 xfs_dir2_leaf_lookup(
1275         xfs_da_args_t           *args)          /* operation arguments */
1276 {
1277         xfs_dabuf_t             *dbp;           /* data block buffer */
1278         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1279         xfs_inode_t             *dp;            /* incore directory inode */
1280         int                     error;          /* error return code */
1281         int                     index;          /* found entry index */
1282         xfs_dabuf_t             *lbp;           /* leaf buffer */
1283         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1284         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1285         xfs_trans_t             *tp;            /* transaction pointer */
1286
1287         trace_xfs_dir2_leaf_lookup(args);
1288
1289         /*
1290          * Look up name in the leaf block, returning both buffers and index.
1291          */
1292         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1293                 return error;
1294         }
1295         tp = args->trans;
1296         dp = args->dp;
1297         xfs_dir2_leaf_check(dp, lbp);
1298         leaf = lbp->data;
1299         /*
1300          * Get to the leaf entry and contained data entry address.
1301          */
1302         lep = &leaf->ents[index];
1303         /*
1304          * Point to the data entry.
1305          */
1306         dep = (xfs_dir2_data_entry_t *)
1307               ((char *)dbp->data +
1308                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1309         /*
1310          * Return the found inode number & CI name if appropriate
1311          */
1312         args->inumber = be64_to_cpu(dep->inumber);
1313         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1314         xfs_da_brelse(tp, dbp);
1315         xfs_da_brelse(tp, lbp);
1316         return XFS_ERROR(error);
1317 }
1318
1319 /*
1320  * Look up name/hash in the leaf block.
1321  * Fill in indexp with the found index, and dbpp with the data buffer.
1322  * If not found dbpp will be NULL, and ENOENT comes back.
1323  * lbpp will always be filled in with the leaf buffer unless there's an error.
1324  */
1325 static int                                      /* error */
1326 xfs_dir2_leaf_lookup_int(
1327         xfs_da_args_t           *args,          /* operation arguments */
1328         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1329         int                     *indexp,        /* out: index in leaf block */
1330         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1331 {
1332         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1333         xfs_dabuf_t             *dbp = NULL;    /* data buffer */
1334         xfs_dir2_data_entry_t   *dep;           /* data entry */
1335         xfs_inode_t             *dp;            /* incore directory inode */
1336         int                     error;          /* error return code */
1337         int                     index;          /* index in leaf block */
1338         xfs_dabuf_t             *lbp;           /* leaf buffer */
1339         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1340         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1341         xfs_mount_t             *mp;            /* filesystem mount point */
1342         xfs_dir2_db_t           newdb;          /* new data block number */
1343         xfs_trans_t             *tp;            /* transaction pointer */
1344         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1345         enum xfs_dacmp          cmp;            /* name compare result */
1346
1347         dp = args->dp;
1348         tp = args->trans;
1349         mp = dp->i_mount;
1350         /*
1351          * Read the leaf block into the buffer.
1352          */
1353         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1354                                                         XFS_DATA_FORK);
1355         if (error)
1356                 return error;
1357         *lbpp = lbp;
1358         leaf = lbp->data;
1359         xfs_dir2_leaf_check(dp, lbp);
1360         /*
1361          * Look for the first leaf entry with our hash value.
1362          */
1363         index = xfs_dir2_leaf_search_hash(args, lbp);
1364         /*
1365          * Loop over all the entries with the right hash value
1366          * looking to match the name.
1367          */
1368         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1369                                 be32_to_cpu(lep->hashval) == args->hashval;
1370                                 lep++, index++) {
1371                 /*
1372                  * Skip over stale leaf entries.
1373                  */
1374                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1375                         continue;
1376                 /*
1377                  * Get the new data block number.
1378                  */
1379                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1380                 /*
1381                  * If it's not the same as the old data block number,
1382                  * need to pitch the old one and read the new one.
1383                  */
1384                 if (newdb != curdb) {
1385                         if (dbp)
1386                                 xfs_da_brelse(tp, dbp);
1387                         error = xfs_da_read_buf(tp, dp,
1388                                                 xfs_dir2_db_to_da(mp, newdb),
1389                                                 -1, &dbp, XFS_DATA_FORK);
1390                         if (error) {
1391                                 xfs_da_brelse(tp, lbp);
1392                                 return error;
1393                         }
1394                         xfs_dir2_data_check(dp, dbp);
1395                         curdb = newdb;
1396                 }
1397                 /*
1398                  * Point to the data entry.
1399                  */
1400                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->data +
1401                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1402                 /*
1403                  * Compare name and if it's an exact match, return the index
1404                  * and buffer. If it's the first case-insensitive match, store
1405                  * the index and buffer and continue looking for an exact match.
1406                  */
1407                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1408                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1409                         args->cmpresult = cmp;
1410                         *indexp = index;
1411                         /* case exact match: return the current buffer. */
1412                         if (cmp == XFS_CMP_EXACT) {
1413                                 *dbpp = dbp;
1414                                 return 0;
1415                         }
1416                         cidb = curdb;
1417                 }
1418         }
1419         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1420         /*
1421          * Here, we can only be doing a lookup (not a rename or remove).
1422          * If a case-insensitive match was found earlier, re-read the
1423          * appropriate data block if required and return it.
1424          */
1425         if (args->cmpresult == XFS_CMP_CASE) {
1426                 ASSERT(cidb != -1);
1427                 if (cidb != curdb) {
1428                         xfs_da_brelse(tp, dbp);
1429                         error = xfs_da_read_buf(tp, dp,
1430                                                 xfs_dir2_db_to_da(mp, cidb),
1431                                                 -1, &dbp, XFS_DATA_FORK);
1432                         if (error) {
1433                                 xfs_da_brelse(tp, lbp);
1434                                 return error;
1435                         }
1436                 }
1437                 *dbpp = dbp;
1438                 return 0;
1439         }
1440         /*
1441          * No match found, return ENOENT.
1442          */
1443         ASSERT(cidb == -1);
1444         if (dbp)
1445                 xfs_da_brelse(tp, dbp);
1446         xfs_da_brelse(tp, lbp);
1447         return XFS_ERROR(ENOENT);
1448 }
1449
1450 /*
1451  * Remove an entry from a leaf format directory.
1452  */
1453 int                                             /* error */
1454 xfs_dir2_leaf_removename(
1455         xfs_da_args_t           *args)          /* operation arguments */
1456 {
1457         __be16                  *bestsp;        /* leaf block best freespace */
1458         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1459         xfs_dir2_db_t           db;             /* data block number */
1460         xfs_dabuf_t             *dbp;           /* data block buffer */
1461         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1462         xfs_inode_t             *dp;            /* incore directory inode */
1463         int                     error;          /* error return code */
1464         xfs_dir2_db_t           i;              /* temporary data block # */
1465         int                     index;          /* index into leaf entries */
1466         xfs_dabuf_t             *lbp;           /* leaf buffer */
1467         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1468         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1469         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1470         xfs_mount_t             *mp;            /* filesystem mount point */
1471         int                     needlog;        /* need to log data header */
1472         int                     needscan;       /* need to rescan data frees */
1473         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1474         xfs_trans_t             *tp;            /* transaction pointer */
1475
1476         trace_xfs_dir2_leaf_removename(args);
1477
1478         /*
1479          * Lookup the leaf entry, get the leaf and data blocks read in.
1480          */
1481         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1482                 return error;
1483         }
1484         dp = args->dp;
1485         tp = args->trans;
1486         mp = dp->i_mount;
1487         leaf = lbp->data;
1488         hdr = dbp->data;
1489         xfs_dir2_data_check(dp, dbp);
1490         /*
1491          * Point to the leaf entry, use that to point to the data entry.
1492          */
1493         lep = &leaf->ents[index];
1494         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1495         dep = (xfs_dir2_data_entry_t *)
1496               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1497         needscan = needlog = 0;
1498         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1499         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1500         bestsp = xfs_dir2_leaf_bests_p(ltp);
1501         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1502         /*
1503          * Mark the former data entry unused.
1504          */
1505         xfs_dir2_data_make_free(tp, dbp,
1506                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1507                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1508         /*
1509          * We just mark the leaf entry stale by putting a null in it.
1510          */
1511         be16_add_cpu(&leaf->hdr.stale, 1);
1512         xfs_dir2_leaf_log_header(tp, lbp);
1513         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1514         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1515         /*
1516          * Scan the freespace in the data block again if necessary,
1517          * log the data block header if necessary.
1518          */
1519         if (needscan)
1520                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1521         if (needlog)
1522                 xfs_dir2_data_log_header(tp, dbp);
1523         /*
1524          * If the longest freespace in the data block has changed,
1525          * put the new value in the bests table and log that.
1526          */
1527         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1528                 bestsp[db] = hdr->bestfree[0].length;
1529                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1530         }
1531         xfs_dir2_data_check(dp, dbp);
1532         /*
1533          * If the data block is now empty then get rid of the data block.
1534          */
1535         if (be16_to_cpu(hdr->bestfree[0].length) ==
1536             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1537                 ASSERT(db != mp->m_dirdatablk);
1538                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1539                         /*
1540                          * Nope, can't get rid of it because it caused
1541                          * allocation of a bmap btree block to do so.
1542                          * Just go on, returning success, leaving the
1543                          * empty block in place.
1544                          */
1545                         if (error == ENOSPC && args->total == 0) {
1546                                 xfs_da_buf_done(dbp);
1547                                 error = 0;
1548                         }
1549                         xfs_dir2_leaf_check(dp, lbp);
1550                         xfs_da_buf_done(lbp);
1551                         return error;
1552                 }
1553                 dbp = NULL;
1554                 /*
1555                  * If this is the last data block then compact the
1556                  * bests table by getting rid of entries.
1557                  */
1558                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1559                         /*
1560                          * Look for the last active entry (i).
1561                          */
1562                         for (i = db - 1; i > 0; i--) {
1563                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1564                                         break;
1565                         }
1566                         /*
1567                          * Copy the table down so inactive entries at the
1568                          * end are removed.
1569                          */
1570                         memmove(&bestsp[db - i], bestsp,
1571                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1572                         be32_add_cpu(&ltp->bestcount, -(db - i));
1573                         xfs_dir2_leaf_log_tail(tp, lbp);
1574                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1575                 } else
1576                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1577         }
1578         /*
1579          * If the data block was not the first one, drop it.
1580          */
1581         else if (db != mp->m_dirdatablk && dbp != NULL) {
1582                 xfs_da_buf_done(dbp);
1583                 dbp = NULL;
1584         }
1585         xfs_dir2_leaf_check(dp, lbp);
1586         /*
1587          * See if we can convert to block form.
1588          */
1589         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1590 }
1591
1592 /*
1593  * Replace the inode number in a leaf format directory entry.
1594  */
1595 int                                             /* error */
1596 xfs_dir2_leaf_replace(
1597         xfs_da_args_t           *args)          /* operation arguments */
1598 {
1599         xfs_dabuf_t             *dbp;           /* data block buffer */
1600         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1601         xfs_inode_t             *dp;            /* incore directory inode */
1602         int                     error;          /* error return code */
1603         int                     index;          /* index of leaf entry */
1604         xfs_dabuf_t             *lbp;           /* leaf buffer */
1605         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1606         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1607         xfs_trans_t             *tp;            /* transaction pointer */
1608
1609         trace_xfs_dir2_leaf_replace(args);
1610
1611         /*
1612          * Look up the entry.
1613          */
1614         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1615                 return error;
1616         }
1617         dp = args->dp;
1618         leaf = lbp->data;
1619         /*
1620          * Point to the leaf entry, get data address from it.
1621          */
1622         lep = &leaf->ents[index];
1623         /*
1624          * Point to the data entry.
1625          */
1626         dep = (xfs_dir2_data_entry_t *)
1627               ((char *)dbp->data +
1628                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1629         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1630         /*
1631          * Put the new inode number in, log it.
1632          */
1633         dep->inumber = cpu_to_be64(args->inumber);
1634         tp = args->trans;
1635         xfs_dir2_data_log_entry(tp, dbp, dep);
1636         xfs_da_buf_done(dbp);
1637         xfs_dir2_leaf_check(dp, lbp);
1638         xfs_da_brelse(tp, lbp);
1639         return 0;
1640 }
1641
1642 /*
1643  * Return index in the leaf block (lbp) which is either the first
1644  * one with this hash value, or if there are none, the insert point
1645  * for that hash value.
1646  */
1647 int                                             /* index value */
1648 xfs_dir2_leaf_search_hash(
1649         xfs_da_args_t           *args,          /* operation arguments */
1650         xfs_dabuf_t             *lbp)           /* leaf buffer */
1651 {
1652         xfs_dahash_t            hash=0;         /* hash from this entry */
1653         xfs_dahash_t            hashwant;       /* hash value looking for */
1654         int                     high;           /* high leaf index */
1655         int                     low;            /* low leaf index */
1656         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1657         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1658         int                     mid=0;          /* current leaf index */
1659
1660         leaf = lbp->data;
1661 #ifndef __KERNEL__
1662         if (!leaf->hdr.count)
1663                 return 0;
1664 #endif
1665         /*
1666          * Note, the table cannot be empty, so we have to go through the loop.
1667          * Binary search the leaf entries looking for our hash value.
1668          */
1669         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1670                 hashwant = args->hashval;
1671              low <= high; ) {
1672                 mid = (low + high) >> 1;
1673                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1674                         break;
1675                 if (hash < hashwant)
1676                         low = mid + 1;
1677                 else
1678                         high = mid - 1;
1679         }
1680         /*
1681          * Found one, back up through all the equal hash values.
1682          */
1683         if (hash == hashwant) {
1684                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1685                         mid--;
1686                 }
1687         }
1688         /*
1689          * Need to point to an entry higher than ours.
1690          */
1691         else if (hash < hashwant)
1692                 mid++;
1693         return mid;
1694 }
1695
1696 /*
1697  * Trim off a trailing data block.  We know it's empty since the leaf
1698  * freespace table says so.
1699  */
1700 int                                             /* error */
1701 xfs_dir2_leaf_trim_data(
1702         xfs_da_args_t           *args,          /* operation arguments */
1703         xfs_dabuf_t             *lbp,           /* leaf buffer */
1704         xfs_dir2_db_t           db)             /* data block number */
1705 {
1706         __be16                  *bestsp;        /* leaf bests table */
1707         xfs_dabuf_t             *dbp;           /* data block buffer */
1708         xfs_inode_t             *dp;            /* incore directory inode */
1709         int                     error;          /* error return value */
1710         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1711         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1712         xfs_mount_t             *mp;            /* filesystem mount point */
1713         xfs_trans_t             *tp;            /* transaction pointer */
1714
1715         dp = args->dp;
1716         mp = dp->i_mount;
1717         tp = args->trans;
1718         /*
1719          * Read the offending data block.  We need its buffer.
1720          */
1721         if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp,
1722                         XFS_DATA_FORK))) {
1723                 return error;
1724         }
1725
1726         leaf = lbp->data;
1727         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1728
1729 #ifdef DEBUG
1730 {
1731         struct xfs_dir2_data_hdr *hdr = dbp->data;
1732
1733         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1734         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1735                mp->m_dirblksize - (uint)sizeof(*hdr));
1736         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1737 }
1738 #endif
1739
1740         /*
1741          * Get rid of the data block.
1742          */
1743         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1744                 ASSERT(error != ENOSPC);
1745                 xfs_da_brelse(tp, dbp);
1746                 return error;
1747         }
1748         /*
1749          * Eliminate the last bests entry from the table.
1750          */
1751         bestsp = xfs_dir2_leaf_bests_p(ltp);
1752         be32_add_cpu(&ltp->bestcount, -1);
1753         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1754         xfs_dir2_leaf_log_tail(tp, lbp);
1755         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1756         return 0;
1757 }
1758
1759 static inline size_t
1760 xfs_dir2_leaf_size(
1761         struct xfs_dir2_leaf_hdr        *hdr,
1762         int                             counts)
1763 {
1764         int                     entries;
1765
1766         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1767         return sizeof(xfs_dir2_leaf_hdr_t) +
1768             entries * sizeof(xfs_dir2_leaf_entry_t) +
1769             counts * sizeof(xfs_dir2_data_off_t) +
1770             sizeof(xfs_dir2_leaf_tail_t);
1771 }
1772
1773 /*
1774  * Convert node form directory to leaf form directory.
1775  * The root of the node form dir needs to already be a LEAFN block.
1776  * Just return if we can't do anything.
1777  */
1778 int                                             /* error */
1779 xfs_dir2_node_to_leaf(
1780         xfs_da_state_t          *state)         /* directory operation state */
1781 {
1782         xfs_da_args_t           *args;          /* operation arguments */
1783         xfs_inode_t             *dp;            /* incore directory inode */
1784         int                     error;          /* error return code */
1785         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1786         xfs_fileoff_t           fo;             /* freespace file offset */
1787         xfs_dir2_free_t         *free;          /* freespace structure */
1788         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1789         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1790         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1791         xfs_mount_t             *mp;            /* filesystem mount point */
1792         int                     rval;           /* successful free trim? */
1793         xfs_trans_t             *tp;            /* transaction pointer */
1794
1795         /*
1796          * There's more than a leaf level in the btree, so there must
1797          * be multiple leafn blocks.  Give up.
1798          */
1799         if (state->path.active > 1)
1800                 return 0;
1801         args = state->args;
1802
1803         trace_xfs_dir2_node_to_leaf(args);
1804
1805         mp = state->mp;
1806         dp = args->dp;
1807         tp = args->trans;
1808         /*
1809          * Get the last offset in the file.
1810          */
1811         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1812                 return error;
1813         }
1814         fo -= mp->m_dirblkfsbs;
1815         /*
1816          * If there are freespace blocks other than the first one,
1817          * take this opportunity to remove trailing empty freespace blocks
1818          * that may have been left behind during no-space-reservation
1819          * operations.
1820          */
1821         while (fo > mp->m_dirfreeblk) {
1822                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1823                         return error;
1824                 }
1825                 if (rval)
1826                         fo -= mp->m_dirblkfsbs;
1827                 else
1828                         return 0;
1829         }
1830         /*
1831          * Now find the block just before the freespace block.
1832          */
1833         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1834                 return error;
1835         }
1836         /*
1837          * If it's not the single leaf block, give up.
1838          */
1839         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1840                 return 0;
1841         lbp = state->path.blk[0].bp;
1842         leaf = lbp->data;
1843         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1844         /*
1845          * Read the freespace block.
1846          */
1847         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1848                         XFS_DATA_FORK))) {
1849                 return error;
1850         }
1851         free = fbp->data;
1852         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1853         ASSERT(!free->hdr.firstdb);
1854
1855         /*
1856          * Now see if the leafn and free data will fit in a leaf1.
1857          * If not, release the buffer and give up.
1858          */
1859         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1860                         mp->m_dirblksize) {
1861                 xfs_da_brelse(tp, fbp);
1862                 return 0;
1863         }
1864
1865         /*
1866          * If the leaf has any stale entries in it, compress them out.
1867          * The compact routine will log the header.
1868          */
1869         if (be16_to_cpu(leaf->hdr.stale))
1870                 xfs_dir2_leaf_compact(args, lbp);
1871         else
1872                 xfs_dir2_leaf_log_header(tp, lbp);
1873         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1874         /*
1875          * Set up the leaf tail from the freespace block.
1876          */
1877         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1878         ltp->bestcount = free->hdr.nvalid;
1879         /*
1880          * Set up the leaf bests table.
1881          */
1882         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1883                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1884         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1885         xfs_dir2_leaf_log_tail(tp, lbp);
1886         xfs_dir2_leaf_check(dp, lbp);
1887         /*
1888          * Get rid of the freespace block.
1889          */
1890         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1891         if (error) {
1892                 /*
1893                  * This can't fail here because it can only happen when
1894                  * punching out the middle of an extent, and this is an
1895                  * isolated block.
1896                  */
1897                 ASSERT(error != ENOSPC);
1898                 return error;
1899         }
1900         fbp = NULL;
1901         /*
1902          * Now see if we can convert the single-leaf directory
1903          * down to a block form directory.
1904          * This routine always kills the dabuf for the leaf, so
1905          * eliminate it from the path.
1906          */
1907         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1908         state->path.blk[0].bp = NULL;
1909         return error;
1910 }