Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6] / 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_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_bmap.h"
39 #include "xfs_dir2_data.h"
40 #include "xfs_dir2_leaf.h"
41 #include "xfs_dir2_block.h"
42 #include "xfs_dir2_node.h"
43 #include "xfs_dir2_trace.h"
44 #include "xfs_error.h"
45
46 /*
47  * Local function declarations.
48  */
49 #ifdef DEBUG
50 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
51 #else
52 #define xfs_dir2_leaf_check(dp, bp)
53 #endif
54 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
55                                     int *indexp, xfs_dabuf_t **dbpp);
56 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
57                                     int first, int last);
58 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
59
60
61 /*
62  * Convert a block form directory to a leaf form directory.
63  */
64 int                                             /* error */
65 xfs_dir2_block_to_leaf(
66         xfs_da_args_t           *args,          /* operation arguments */
67         xfs_dabuf_t             *dbp)           /* input block's buffer */
68 {
69         __be16                  *bestsp;        /* leaf's bestsp entries */
70         xfs_dablk_t             blkno;          /* leaf block's bno */
71         xfs_dir2_block_t        *block;         /* block structure */
72         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
73         xfs_dir2_block_tail_t   *btp;           /* block's tail */
74         xfs_inode_t             *dp;            /* incore directory inode */
75         int                     error;          /* error return code */
76         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
77         xfs_dir2_db_t           ldb;            /* leaf block's bno */
78         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
79         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
80         xfs_mount_t             *mp;            /* filesystem mount point */
81         int                     needlog;        /* need to log block header */
82         int                     needscan;       /* need to rescan bestfree */
83         xfs_trans_t             *tp;            /* transaction pointer */
84
85         xfs_dir2_trace_args_b("block_to_leaf", args, dbp);
86         dp = args->dp;
87         mp = dp->i_mount;
88         tp = args->trans;
89         /*
90          * Add the leaf block to the inode.
91          * This interface will only put blocks in the leaf/node range.
92          * Since that's empty now, we'll get the root (block 0 in range).
93          */
94         if ((error = xfs_da_grow_inode(args, &blkno))) {
95                 return error;
96         }
97         ldb = XFS_DIR2_DA_TO_DB(mp, blkno);
98         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
99         /*
100          * Initialize the leaf block, get a buffer for it.
101          */
102         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
103                 return error;
104         }
105         ASSERT(lbp != NULL);
106         leaf = lbp->data;
107         block = dbp->data;
108         xfs_dir2_data_check(dp, dbp);
109         btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
110         blp = XFS_DIR2_BLOCK_LEAF_P(btp);
111         /*
112          * Set the counts in the leaf header.
113          */
114         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
115         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
116         /*
117          * Could compact these but I think we always do the conversion
118          * after squeezing out stale entries.
119          */
120         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
121         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
122         needscan = 0;
123         needlog = 1;
124         /*
125          * Make the space formerly occupied by the leaf entries and block
126          * tail be free.
127          */
128         xfs_dir2_data_make_free(tp, dbp,
129                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
130                 (xfs_dir2_data_aoff_t)((char *)block + mp->m_dirblksize -
131                                        (char *)blp),
132                 &needlog, &needscan);
133         /*
134          * Fix up the block header, make it a data block.
135          */
136         block->hdr.magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
137         if (needscan)
138                 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
139                         NULL);
140         /*
141          * Set up leaf tail and bests table.
142          */
143         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
144         ltp->bestcount = cpu_to_be32(1);
145         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
146         bestsp[0] =  block->hdr.bestfree[0].length;
147         /*
148          * Log the data header and leaf bests table.
149          */
150         if (needlog)
151                 xfs_dir2_data_log_header(tp, dbp);
152         xfs_dir2_leaf_check(dp, lbp);
153         xfs_dir2_data_check(dp, dbp);
154         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
155         xfs_da_buf_done(lbp);
156         return 0;
157 }
158
159 /*
160  * Add an entry to a leaf form directory.
161  */
162 int                                             /* error */
163 xfs_dir2_leaf_addname(
164         xfs_da_args_t           *args)          /* operation arguments */
165 {
166         __be16                  *bestsp;        /* freespace table in leaf */
167         int                     compact;        /* need to compact leaves */
168         xfs_dir2_data_t         *data;          /* data block structure */
169         xfs_dabuf_t             *dbp;           /* data block buffer */
170         xfs_dir2_data_entry_t   *dep;           /* data block entry */
171         xfs_inode_t             *dp;            /* incore directory inode */
172         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
173         int                     error;          /* error return value */
174         int                     grown;          /* allocated new data block */
175         int                     highstale;      /* index of next stale leaf */
176         int                     i;              /* temporary, index */
177         int                     index;          /* leaf table position */
178         xfs_dabuf_t             *lbp;           /* leaf's buffer */
179         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
180         int                     length;         /* length of new entry */
181         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
182         int                     lfloglow;       /* low leaf logging index */
183         int                     lfloghigh;      /* high leaf logging index */
184         int                     lowstale;       /* index of prev stale leaf */
185         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
186         xfs_mount_t             *mp;            /* filesystem mount point */
187         int                     needbytes;      /* leaf block bytes needed */
188         int                     needlog;        /* need to log data header */
189         int                     needscan;       /* need to rescan data free */
190         __be16                  *tagp;          /* end of data entry */
191         xfs_trans_t             *tp;            /* transaction pointer */
192         xfs_dir2_db_t           use_block;      /* data block number */
193
194         xfs_dir2_trace_args("leaf_addname", args);
195         dp = args->dp;
196         tp = args->trans;
197         mp = dp->i_mount;
198         /*
199          * Read the leaf block.
200          */
201         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
202                 XFS_DATA_FORK);
203         if (error) {
204                 return error;
205         }
206         ASSERT(lbp != NULL);
207         /*
208          * Look up the entry by hash value and name.
209          * We know it's not there, our caller has already done a lookup.
210          * So the index is of the entry to insert in front of.
211          * But if there are dup hash values the index is of the first of those.
212          */
213         index = xfs_dir2_leaf_search_hash(args, lbp);
214         leaf = lbp->data;
215         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
216         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
217         length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
218         /*
219          * See if there are any entries with the same hash value
220          * and space in their block for the new entry.
221          * This is good because it puts multiple same-hash value entries
222          * in a data block, improving the lookup of those entries.
223          */
224         for (use_block = -1, lep = &leaf->ents[index];
225              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
226              index++, lep++) {
227                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
228                         continue;
229                 i = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
230                 ASSERT(i < be32_to_cpu(ltp->bestcount));
231                 ASSERT(be16_to_cpu(bestsp[i]) != NULLDATAOFF);
232                 if (be16_to_cpu(bestsp[i]) >= length) {
233                         use_block = i;
234                         break;
235                 }
236         }
237         /*
238          * Didn't find a block yet, linear search all the data blocks.
239          */
240         if (use_block == -1) {
241                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
242                         /*
243                          * Remember a block we see that's missing.
244                          */
245                         if (be16_to_cpu(bestsp[i]) == NULLDATAOFF && use_block == -1)
246                                 use_block = i;
247                         else if (be16_to_cpu(bestsp[i]) >= length) {
248                                 use_block = i;
249                                 break;
250                         }
251                 }
252         }
253         /*
254          * How many bytes do we need in the leaf block?
255          */
256         needbytes =
257                 (leaf->hdr.stale ? 0 : (uint)sizeof(leaf->ents[0])) +
258                 (use_block != -1 ? 0 : (uint)sizeof(leaf->bests[0]));
259         /*
260          * Now kill use_block if it refers to a missing block, so we
261          * can use it as an indication of allocation needed.
262          */
263         if (use_block != -1 && be16_to_cpu(bestsp[use_block]) == NULLDATAOFF)
264                 use_block = -1;
265         /*
266          * If we don't have enough free bytes but we can make enough
267          * by compacting out stale entries, we'll do that.
268          */
269         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] < needbytes &&
270             be16_to_cpu(leaf->hdr.stale) > 1) {
271                 compact = 1;
272         }
273         /*
274          * Otherwise if we don't have enough free bytes we need to
275          * convert to node form.
276          */
277         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
278                  needbytes) {
279                 /*
280                  * Just checking or no space reservation, give up.
281                  */
282                 if (args->justcheck || args->total == 0) {
283                         xfs_da_brelse(tp, lbp);
284                         return XFS_ERROR(ENOSPC);
285                 }
286                 /*
287                  * Convert to node form.
288                  */
289                 error = xfs_dir2_leaf_to_node(args, lbp);
290                 xfs_da_buf_done(lbp);
291                 if (error)
292                         return error;
293                 /*
294                  * Then add the new entry.
295                  */
296                 return xfs_dir2_node_addname(args);
297         }
298         /*
299          * Otherwise it will fit without compaction.
300          */
301         else
302                 compact = 0;
303         /*
304          * If just checking, then it will fit unless we needed to allocate
305          * a new data block.
306          */
307         if (args->justcheck) {
308                 xfs_da_brelse(tp, lbp);
309                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
310         }
311         /*
312          * If no allocations are allowed, return now before we've
313          * changed anything.
314          */
315         if (args->total == 0 && use_block == -1) {
316                 xfs_da_brelse(tp, lbp);
317                 return XFS_ERROR(ENOSPC);
318         }
319         /*
320          * Need to compact the leaf entries, removing stale ones.
321          * Leave one stale entry behind - the one closest to our
322          * insertion index - and we'll shift that one to our insertion
323          * point later.
324          */
325         if (compact) {
326                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
327                         &lfloglow, &lfloghigh);
328         }
329         /*
330          * There are stale entries, so we'll need log-low and log-high
331          * impossibly bad values later.
332          */
333         else if (be16_to_cpu(leaf->hdr.stale)) {
334                 lfloglow = be16_to_cpu(leaf->hdr.count);
335                 lfloghigh = -1;
336         }
337         /*
338          * If there was no data block space found, we need to allocate
339          * a new one.
340          */
341         if (use_block == -1) {
342                 /*
343                  * Add the new data block.
344                  */
345                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
346                                 &use_block))) {
347                         xfs_da_brelse(tp, lbp);
348                         return error;
349                 }
350                 /*
351                  * Initialize the block.
352                  */
353                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
354                         xfs_da_brelse(tp, lbp);
355                         return error;
356                 }
357                 /*
358                  * If we're adding a new data block on the end we need to
359                  * extend the bests table.  Copy it up one entry.
360                  */
361                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
362                         bestsp--;
363                         memmove(&bestsp[0], &bestsp[1],
364                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
365                         be32_add(&ltp->bestcount, 1);
366                         xfs_dir2_leaf_log_tail(tp, lbp);
367                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
368                 }
369                 /*
370                  * If we're filling in a previously empty block just log it.
371                  */
372                 else
373                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
374                 data = dbp->data;
375                 bestsp[use_block] = data->hdr.bestfree[0].length;
376                 grown = 1;
377         }
378         /*
379          * Already had space in some data block.
380          * Just read that one in.
381          */
382         else {
383                 if ((error =
384                     xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, use_block),
385                             -1, &dbp, XFS_DATA_FORK))) {
386                         xfs_da_brelse(tp, lbp);
387                         return error;
388                 }
389                 data = dbp->data;
390                 grown = 0;
391         }
392         xfs_dir2_data_check(dp, dbp);
393         /*
394          * Point to the biggest freespace in our data block.
395          */
396         dup = (xfs_dir2_data_unused_t *)
397               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
398         ASSERT(be16_to_cpu(dup->length) >= length);
399         needscan = needlog = 0;
400         /*
401          * Mark the initial part of our freespace in use for the new entry.
402          */
403         xfs_dir2_data_use_free(tp, dbp, dup,
404                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
405                 &needlog, &needscan);
406         /*
407          * Initialize our new entry (at last).
408          */
409         dep = (xfs_dir2_data_entry_t *)dup;
410         INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
411         dep->namelen = args->namelen;
412         memcpy(dep->name, args->name, dep->namelen);
413         tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
414         *tagp = cpu_to_be16((char *)dep - (char *)data);
415         /*
416          * Need to scan fix up the bestfree table.
417          */
418         if (needscan)
419                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
420         /*
421          * Need to log the data block's header.
422          */
423         if (needlog)
424                 xfs_dir2_data_log_header(tp, dbp);
425         xfs_dir2_data_log_entry(tp, dbp, dep);
426         /*
427          * If the bests table needs to be changed, do it.
428          * Log the change unless we've already done that.
429          */
430         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
431                 bestsp[use_block] = data->hdr.bestfree[0].length;
432                 if (!grown)
433                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
434         }
435         /*
436          * Now we need to make room to insert the leaf entry.
437          * If there are no stale entries, we just insert a hole at index.
438          */
439         if (!leaf->hdr.stale) {
440                 /*
441                  * lep is still good as the index leaf entry.
442                  */
443                 if (index < be16_to_cpu(leaf->hdr.count))
444                         memmove(lep + 1, lep,
445                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
446                 /*
447                  * Record low and high logging indices for the leaf.
448                  */
449                 lfloglow = index;
450                 lfloghigh = be16_to_cpu(leaf->hdr.count);
451                 be16_add(&leaf->hdr.count, 1);
452         }
453         /*
454          * There are stale entries.
455          * We will use one of them for the new entry.
456          * It's probably not at the right location, so we'll have to
457          * shift some up or down first.
458          */
459         else {
460                 /*
461                  * If we didn't compact before, we need to find the nearest
462                  * stale entries before and after our insertion point.
463                  */
464                 if (compact == 0) {
465                         /*
466                          * Find the first stale entry before the insertion
467                          * point, if any.
468                          */
469                         for (lowstale = index - 1;
470                              lowstale >= 0 &&
471                                 be32_to_cpu(leaf->ents[lowstale].address) !=
472                                 XFS_DIR2_NULL_DATAPTR;
473                              lowstale--)
474                                 continue;
475                         /*
476                          * Find the next stale entry at or after the insertion
477                          * point, if any.   Stop if we go so far that the
478                          * lowstale entry would be better.
479                          */
480                         for (highstale = index;
481                              highstale < be16_to_cpu(leaf->hdr.count) &&
482                                 be32_to_cpu(leaf->ents[highstale].address) !=
483                                 XFS_DIR2_NULL_DATAPTR &&
484                                 (lowstale < 0 ||
485                                  index - lowstale - 1 >= highstale - index);
486                              highstale++)
487                                 continue;
488                 }
489                 /*
490                  * If the low one is better, use it.
491                  */
492                 if (lowstale >= 0 &&
493                     (highstale == be16_to_cpu(leaf->hdr.count) ||
494                      index - lowstale - 1 < highstale - index)) {
495                         ASSERT(index - lowstale - 1 >= 0);
496                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
497                                XFS_DIR2_NULL_DATAPTR);
498                         /*
499                          * Copy entries up to cover the stale entry
500                          * and make room for the new entry.
501                          */
502                         if (index - lowstale - 1 > 0)
503                                 memmove(&leaf->ents[lowstale],
504                                         &leaf->ents[lowstale + 1],
505                                         (index - lowstale - 1) * sizeof(*lep));
506                         lep = &leaf->ents[index - 1];
507                         lfloglow = MIN(lowstale, lfloglow);
508                         lfloghigh = MAX(index - 1, lfloghigh);
509                 }
510                 /*
511                  * The high one is better, so use that one.
512                  */
513                 else {
514                         ASSERT(highstale - index >= 0);
515                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
516                                XFS_DIR2_NULL_DATAPTR);
517                         /*
518                          * Copy entries down to cover the stale entry
519                          * and make room for the new entry.
520                          */
521                         if (highstale - index > 0)
522                                 memmove(&leaf->ents[index + 1],
523                                         &leaf->ents[index],
524                                         (highstale - index) * sizeof(*lep));
525                         lep = &leaf->ents[index];
526                         lfloglow = MIN(index, lfloglow);
527                         lfloghigh = MAX(highstale, lfloghigh);
528                 }
529                 be16_add(&leaf->hdr.stale, -1);
530         }
531         /*
532          * Fill in the new leaf entry.
533          */
534         lep->hashval = cpu_to_be32(args->hashval);
535         lep->address = cpu_to_be32(XFS_DIR2_DB_OFF_TO_DATAPTR(mp, use_block,
536                                 be16_to_cpu(*tagp)));
537         /*
538          * Log the leaf fields and give up the buffers.
539          */
540         xfs_dir2_leaf_log_header(tp, lbp);
541         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
542         xfs_dir2_leaf_check(dp, lbp);
543         xfs_da_buf_done(lbp);
544         xfs_dir2_data_check(dp, dbp);
545         xfs_da_buf_done(dbp);
546         return 0;
547 }
548
549 #ifdef DEBUG
550 /*
551  * Check the internal consistency of a leaf1 block.
552  * Pop an assert if something is wrong.
553  */
554 void
555 xfs_dir2_leaf_check(
556         xfs_inode_t             *dp,            /* incore directory inode */
557         xfs_dabuf_t             *bp)            /* leaf's buffer */
558 {
559         int                     i;              /* leaf index */
560         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
561         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
562         xfs_mount_t             *mp;            /* filesystem mount point */
563         int                     stale;          /* count of stale leaves */
564
565         leaf = bp->data;
566         mp = dp->i_mount;
567         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
568         /*
569          * This value is not restrictive enough.
570          * Should factor in the size of the bests table as well.
571          * We can deduce a value for that from di_size.
572          */
573         ASSERT(be16_to_cpu(leaf->hdr.count) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
574         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
575         /*
576          * Leaves and bests don't overlap.
577          */
578         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
579                (char *)XFS_DIR2_LEAF_BESTS_P(ltp));
580         /*
581          * Check hash value order, count stale entries.
582          */
583         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
584                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
585                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
586                                be32_to_cpu(leaf->ents[i + 1].hashval));
587                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
588                         stale++;
589         }
590         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
591 }
592 #endif  /* DEBUG */
593
594 /*
595  * Compact out any stale entries in the leaf.
596  * Log the header and changed leaf entries, if any.
597  */
598 void
599 xfs_dir2_leaf_compact(
600         xfs_da_args_t   *args,          /* operation arguments */
601         xfs_dabuf_t     *bp)            /* leaf buffer */
602 {
603         int             from;           /* source leaf index */
604         xfs_dir2_leaf_t *leaf;          /* leaf structure */
605         int             loglow;         /* first leaf entry to log */
606         int             to;             /* target leaf index */
607
608         leaf = bp->data;
609         if (!leaf->hdr.stale) {
610                 return;
611         }
612         /*
613          * Compress out the stale entries in place.
614          */
615         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
616                 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
617                         continue;
618                 /*
619                  * Only actually copy the entries that are different.
620                  */
621                 if (from > to) {
622                         if (loglow == -1)
623                                 loglow = to;
624                         leaf->ents[to] = leaf->ents[from];
625                 }
626                 to++;
627         }
628         /*
629          * Update and log the header, log the leaf entries.
630          */
631         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
632         be16_add(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
633         leaf->hdr.stale = 0;
634         xfs_dir2_leaf_log_header(args->trans, bp);
635         if (loglow != -1)
636                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
637 }
638
639 /*
640  * Compact the leaf entries, removing stale ones.
641  * Leave one stale entry behind - the one closest to our
642  * insertion index - and the caller will shift that one to our insertion
643  * point later.
644  * Return new insertion index, where the remaining stale entry is,
645  * and leaf logging indices.
646  */
647 void
648 xfs_dir2_leaf_compact_x1(
649         xfs_dabuf_t     *bp,            /* leaf buffer */
650         int             *indexp,        /* insertion index */
651         int             *lowstalep,     /* out: stale entry before us */
652         int             *highstalep,    /* out: stale entry after us */
653         int             *lowlogp,       /* out: low log index */
654         int             *highlogp)      /* out: high log index */
655 {
656         int             from;           /* source copy index */
657         int             highstale;      /* stale entry at/after index */
658         int             index;          /* insertion index */
659         int             keepstale;      /* source index of kept stale */
660         xfs_dir2_leaf_t *leaf;          /* leaf structure */
661         int             lowstale;       /* stale entry before index */
662         int             newindex=0;     /* new insertion index */
663         int             to;             /* destination copy index */
664
665         leaf = bp->data;
666         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
667         index = *indexp;
668         /*
669          * Find the first stale entry before our index, if any.
670          */
671         for (lowstale = index - 1;
672              lowstale >= 0 &&
673                 be32_to_cpu(leaf->ents[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
674              lowstale--)
675                 continue;
676         /*
677          * Find the first stale entry at or after our index, if any.
678          * Stop if the answer would be worse than lowstale.
679          */
680         for (highstale = index;
681              highstale < be16_to_cpu(leaf->hdr.count) &&
682                 be32_to_cpu(leaf->ents[highstale].address) != XFS_DIR2_NULL_DATAPTR &&
683                 (lowstale < 0 || index - lowstale > highstale - index);
684              highstale++)
685                 continue;
686         /*
687          * Pick the better of lowstale and highstale.
688          */
689         if (lowstale >= 0 &&
690             (highstale == be16_to_cpu(leaf->hdr.count) ||
691              index - lowstale <= highstale - index))
692                 keepstale = lowstale;
693         else
694                 keepstale = highstale;
695         /*
696          * Copy the entries in place, removing all the stale entries
697          * except keepstale.
698          */
699         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
700                 /*
701                  * Notice the new value of index.
702                  */
703                 if (index == from)
704                         newindex = to;
705                 if (from != keepstale &&
706                     be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR) {
707                         if (from == to)
708                                 *lowlogp = to;
709                         continue;
710                 }
711                 /*
712                  * Record the new keepstale value for the insertion.
713                  */
714                 if (from == keepstale)
715                         lowstale = highstale = to;
716                 /*
717                  * Copy only the entries that have moved.
718                  */
719                 if (from > to)
720                         leaf->ents[to] = leaf->ents[from];
721                 to++;
722         }
723         ASSERT(from > to);
724         /*
725          * If the insertion point was past the last entry,
726          * set the new insertion point accordingly.
727          */
728         if (index == from)
729                 newindex = to;
730         *indexp = newindex;
731         /*
732          * Adjust the leaf header values.
733          */
734         be16_add(&leaf->hdr.count, -(from - to));
735         leaf->hdr.stale = cpu_to_be16(1);
736         /*
737          * Remember the low/high stale value only in the "right"
738          * direction.
739          */
740         if (lowstale >= newindex)
741                 lowstale = -1;
742         else
743                 highstale = be16_to_cpu(leaf->hdr.count);
744         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
745         *lowstalep = lowstale;
746         *highstalep = highstale;
747 }
748
749 /*
750  * Getdents (readdir) for leaf and node directories.
751  * This reads the data blocks only, so is the same for both forms.
752  */
753 int                                             /* error */
754 xfs_dir2_leaf_getdents(
755         xfs_trans_t             *tp,            /* transaction pointer */
756         xfs_inode_t             *dp,            /* incore directory inode */
757         uio_t                   *uio,           /* I/O control & vectors */
758         int                     *eofp,          /* out: reached end of dir */
759         xfs_dirent_t            *dbp,           /* caller's buffer */
760         xfs_dir2_put_t          put)            /* ABI formatting routine */
761 {
762         xfs_dabuf_t             *bp;            /* data block buffer */
763         int                     byteoff;        /* offset in current block */
764         xfs_dir2_db_t           curdb;          /* db for current block */
765         xfs_dir2_off_t          curoff;         /* current overall offset */
766         xfs_dir2_data_t         *data;          /* data block structure */
767         xfs_dir2_data_entry_t   *dep;           /* data entry */
768         xfs_dir2_data_unused_t  *dup;           /* unused entry */
769         int                     eof;            /* reached end of directory */
770         int                     error = 0;      /* error return value */
771         int                     i;              /* temporary loop index */
772         int                     j;              /* temporary loop index */
773         int                     length;         /* temporary length value */
774         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
775         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
776         xfs_dablk_t             map_off;        /* last mapped file offset */
777         int                     map_size;       /* total entries in *map */
778         int                     map_valid;      /* valid entries in *map */
779         xfs_mount_t             *mp;            /* filesystem mount point */
780         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
781         int                     nmap;           /* mappings to ask xfs_bmapi */
782         xfs_dir2_put_args_t     *p;             /* formatting arg bundle */
783         char                    *ptr = NULL;    /* pointer to current data */
784         int                     ra_current;     /* number of read-ahead blks */
785         int                     ra_index;       /* *map index for read-ahead */
786         int                     ra_offset;      /* map entry offset for ra */
787         int                     ra_want;        /* readahead count wanted */
788
789         /*
790          * If the offset is at or past the largest allowed value,
791          * give up right away, return eof.
792          */
793         if (uio->uio_offset >= XFS_DIR2_MAX_DATAPTR) {
794                 *eofp = 1;
795                 return 0;
796         }
797         mp = dp->i_mount;
798         /*
799          * Setup formatting arguments.
800          */
801         p = kmem_alloc(sizeof(*p), KM_SLEEP);
802         p->dbp = dbp;
803         p->put = put;
804         p->uio = uio;
805         /*
806          * Set up to bmap a number of blocks based on the caller's
807          * buffer size, the directory block size, and the filesystem
808          * block size.
809          */
810         map_size =
811                 howmany(uio->uio_resid + mp->m_dirblksize,
812                         mp->m_sb.sb_blocksize);
813         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
814         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
815         bp = NULL;
816         eof = 1;
817         /*
818          * Inside the loop we keep the main offset value as a byte offset
819          * in the directory file.
820          */
821         curoff = XFS_DIR2_DATAPTR_TO_BYTE(mp, uio->uio_offset);
822         /*
823          * Force this conversion through db so we truncate the offset
824          * down to get the start of the data block.
825          */
826         map_off = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_BYTE_TO_DB(mp, curoff));
827         /*
828          * Loop over directory entries until we reach the end offset.
829          * Get more blocks and readahead as necessary.
830          */
831         while (curoff < XFS_DIR2_LEAF_OFFSET) {
832                 /*
833                  * If we have no buffer, or we're off the end of the
834                  * current buffer, need to get another one.
835                  */
836                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
837                         /*
838                          * If we have a buffer, we need to release it and
839                          * take it out of the mapping.
840                          */
841                         if (bp) {
842                                 xfs_da_brelse(tp, bp);
843                                 bp = NULL;
844                                 map_blocks -= mp->m_dirblkfsbs;
845                                 /*
846                                  * Loop to get rid of the extents for the
847                                  * directory block.
848                                  */
849                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
850                                         j = MIN((int)map->br_blockcount, i);
851                                         map->br_blockcount -= j;
852                                         map->br_startblock += j;
853                                         map->br_startoff += j;
854                                         /*
855                                          * If mapping is done, pitch it from
856                                          * the table.
857                                          */
858                                         if (!map->br_blockcount && --map_valid)
859                                                 memmove(&map[0], &map[1],
860                                                         sizeof(map[0]) *
861                                                         map_valid);
862                                         i -= j;
863                                 }
864                         }
865                         /*
866                          * Recalculate the readahead blocks wanted.
867                          */
868                         ra_want = howmany(uio->uio_resid + mp->m_dirblksize,
869                                           mp->m_sb.sb_blocksize) - 1;
870                         /*
871                          * If we don't have as many as we want, and we haven't
872                          * run out of data blocks, get some more mappings.
873                          */
874                         if (1 + ra_want > map_blocks &&
875                             map_off <
876                             XFS_DIR2_BYTE_TO_DA(mp, XFS_DIR2_LEAF_OFFSET)) {
877                                 /*
878                                  * Get more bmaps, fill in after the ones
879                                  * we already have in the table.
880                                  */
881                                 nmap = map_size - map_valid;
882                                 error = xfs_bmapi(tp, dp,
883                                         map_off,
884                                         XFS_DIR2_BYTE_TO_DA(mp,
885                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
886                                         XFS_BMAPI_METADATA, NULL, 0,
887                                         &map[map_valid], &nmap, NULL);
888                                 /*
889                                  * Don't know if we should ignore this or
890                                  * try to return an error.
891                                  * The trouble with returning errors
892                                  * is that readdir will just stop without
893                                  * actually passing the error through.
894                                  */
895                                 if (error)
896                                         break;  /* XXX */
897                                 /*
898                                  * If we got all the mappings we asked for,
899                                  * set the final map offset based on the
900                                  * last bmap value received.
901                                  * Otherwise, we've reached the end.
902                                  */
903                                 if (nmap == map_size - map_valid)
904                                         map_off =
905                                         map[map_valid + nmap - 1].br_startoff +
906                                         map[map_valid + nmap - 1].br_blockcount;
907                                 else
908                                         map_off =
909                                                 XFS_DIR2_BYTE_TO_DA(mp,
910                                                         XFS_DIR2_LEAF_OFFSET);
911                                 /*
912                                  * Look for holes in the mapping, and
913                                  * eliminate them.  Count up the valid blocks.
914                                  */
915                                 for (i = map_valid; i < map_valid + nmap; ) {
916                                         if (map[i].br_startblock ==
917                                             HOLESTARTBLOCK) {
918                                                 nmap--;
919                                                 length = map_valid + nmap - i;
920                                                 if (length)
921                                                         memmove(&map[i],
922                                                                 &map[i + 1],
923                                                                 sizeof(map[i]) *
924                                                                 length);
925                                         } else {
926                                                 map_blocks +=
927                                                         map[i].br_blockcount;
928                                                 i++;
929                                         }
930                                 }
931                                 map_valid += nmap;
932                         }
933                         /*
934                          * No valid mappings, so no more data blocks.
935                          */
936                         if (!map_valid) {
937                                 curoff = XFS_DIR2_DA_TO_BYTE(mp, map_off);
938                                 break;
939                         }
940                         /*
941                          * Read the directory block starting at the first
942                          * mapping.
943                          */
944                         curdb = XFS_DIR2_DA_TO_DB(mp, map->br_startoff);
945                         error = xfs_da_read_buf(tp, dp, map->br_startoff,
946                                 map->br_blockcount >= mp->m_dirblkfsbs ?
947                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
948                                     -1,
949                                 &bp, XFS_DATA_FORK);
950                         /*
951                          * Should just skip over the data block instead
952                          * of giving up.
953                          */
954                         if (error)
955                                 break;  /* XXX */
956                         /*
957                          * Adjust the current amount of read-ahead: we just
958                          * read a block that was previously ra.
959                          */
960                         if (ra_current)
961                                 ra_current -= mp->m_dirblkfsbs;
962                         /*
963                          * Do we need more readahead?
964                          */
965                         for (ra_index = ra_offset = i = 0;
966                              ra_want > ra_current && i < map_blocks;
967                              i += mp->m_dirblkfsbs) {
968                                 ASSERT(ra_index < map_valid);
969                                 /*
970                                  * Read-ahead a contiguous directory block.
971                                  */
972                                 if (i > ra_current &&
973                                     map[ra_index].br_blockcount >=
974                                     mp->m_dirblkfsbs) {
975                                         xfs_baread(mp->m_ddev_targp,
976                                                 XFS_FSB_TO_DADDR(mp,
977                                                    map[ra_index].br_startblock +
978                                                    ra_offset),
979                                                 (int)BTOBB(mp->m_dirblksize));
980                                         ra_current = i;
981                                 }
982                                 /*
983                                  * Read-ahead a non-contiguous directory block.
984                                  * This doesn't use our mapping, but this
985                                  * is a very rare case.
986                                  */
987                                 else if (i > ra_current) {
988                                         (void)xfs_da_reada_buf(tp, dp,
989                                                 map[ra_index].br_startoff +
990                                                 ra_offset, XFS_DATA_FORK);
991                                         ra_current = i;
992                                 }
993                                 /*
994                                  * Advance offset through the mapping table.
995                                  */
996                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
997                                         /*
998                                          * The rest of this extent but not
999                                          * more than a dir block.
1000                                          */
1001                                         length = MIN(mp->m_dirblkfsbs,
1002                                                 (int)(map[ra_index].br_blockcount -
1003                                                 ra_offset));
1004                                         j += length;
1005                                         ra_offset += length;
1006                                         /*
1007                                          * Advance to the next mapping if
1008                                          * this one is used up.
1009                                          */
1010                                         if (ra_offset ==
1011                                             map[ra_index].br_blockcount) {
1012                                                 ra_offset = 0;
1013                                                 ra_index++;
1014                                         }
1015                                 }
1016                         }
1017                         /*
1018                          * Having done a read, we need to set a new offset.
1019                          */
1020                         newoff = XFS_DIR2_DB_OFF_TO_BYTE(mp, curdb, 0);
1021                         /*
1022                          * Start of the current block.
1023                          */
1024                         if (curoff < newoff)
1025                                 curoff = newoff;
1026                         /*
1027                          * Make sure we're in the right block.
1028                          */
1029                         else if (curoff > newoff)
1030                                 ASSERT(XFS_DIR2_BYTE_TO_DB(mp, curoff) ==
1031                                        curdb);
1032                         data = bp->data;
1033                         xfs_dir2_data_check(dp, bp);
1034                         /*
1035                          * Find our position in the block.
1036                          */
1037                         ptr = (char *)&data->u;
1038                         byteoff = XFS_DIR2_BYTE_TO_OFF(mp, curoff);
1039                         /*
1040                          * Skip past the header.
1041                          */
1042                         if (byteoff == 0)
1043                                 curoff += (uint)sizeof(data->hdr);
1044                         /*
1045                          * Skip past entries until we reach our offset.
1046                          */
1047                         else {
1048                                 while ((char *)ptr - (char *)data < byteoff) {
1049                                         dup = (xfs_dir2_data_unused_t *)ptr;
1050
1051                                         if (be16_to_cpu(dup->freetag)
1052                                                   == XFS_DIR2_DATA_FREE_TAG) {
1053
1054                                                 length = be16_to_cpu(dup->length);
1055                                                 ptr += length;
1056                                                 continue;
1057                                         }
1058                                         dep = (xfs_dir2_data_entry_t *)ptr;
1059                                         length =
1060                                            XFS_DIR2_DATA_ENTSIZE(dep->namelen);
1061                                         ptr += length;
1062                                 }
1063                                 /*
1064                                  * Now set our real offset.
1065                                  */
1066                                 curoff =
1067                                         XFS_DIR2_DB_OFF_TO_BYTE(mp,
1068                                             XFS_DIR2_BYTE_TO_DB(mp, curoff),
1069                                             (char *)ptr - (char *)data);
1070                                 if (ptr >= (char *)data + mp->m_dirblksize) {
1071                                         continue;
1072                                 }
1073                         }
1074                 }
1075                 /*
1076                  * We have a pointer to an entry.
1077                  * Is it a live one?
1078                  */
1079                 dup = (xfs_dir2_data_unused_t *)ptr;
1080                 /*
1081                  * No, it's unused, skip over it.
1082                  */
1083                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1084                         length = be16_to_cpu(dup->length);
1085                         ptr += length;
1086                         curoff += length;
1087                         continue;
1088                 }
1089
1090                 /*
1091                  * Copy the entry into the putargs, and try formatting it.
1092                  */
1093                 dep = (xfs_dir2_data_entry_t *)ptr;
1094
1095                 p->namelen = dep->namelen;
1096
1097                 length = XFS_DIR2_DATA_ENTSIZE(p->namelen);
1098
1099                 p->cook = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff + length);
1100
1101                 p->ino = INT_GET(dep->inumber, ARCH_CONVERT);
1102 #if XFS_BIG_INUMS
1103                 p->ino += mp->m_inoadd;
1104 #endif
1105                 p->name = (char *)dep->name;
1106
1107                 error = p->put(p);
1108
1109                 /*
1110                  * Won't fit.  Return to caller.
1111                  */
1112                 if (!p->done) {
1113                         eof = 0;
1114                         break;
1115                 }
1116                 /*
1117                  * Advance to next entry in the block.
1118                  */
1119                 ptr += length;
1120                 curoff += length;
1121         }
1122
1123         /*
1124          * All done.  Set output offset value to current offset.
1125          */
1126         *eofp = eof;
1127         if (curoff > XFS_DIR2_DATAPTR_TO_BYTE(mp, XFS_DIR2_MAX_DATAPTR))
1128                 uio->uio_offset = XFS_DIR2_MAX_DATAPTR;
1129         else
1130                 uio->uio_offset = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff);
1131         kmem_free(map, map_size * sizeof(*map));
1132         kmem_free(p, sizeof(*p));
1133         if (bp)
1134                 xfs_da_brelse(tp, bp);
1135         return error;
1136 }
1137
1138 /*
1139  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1140  */
1141 int
1142 xfs_dir2_leaf_init(
1143         xfs_da_args_t           *args,          /* operation arguments */
1144         xfs_dir2_db_t           bno,            /* directory block number */
1145         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1146         int                     magic)          /* magic number for block */
1147 {
1148         xfs_dabuf_t             *bp;            /* leaf buffer */
1149         xfs_inode_t             *dp;            /* incore directory inode */
1150         int                     error;          /* error return code */
1151         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1152         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1153         xfs_mount_t             *mp;            /* filesystem mount point */
1154         xfs_trans_t             *tp;            /* transaction pointer */
1155
1156         dp = args->dp;
1157         ASSERT(dp != NULL);
1158         tp = args->trans;
1159         mp = dp->i_mount;
1160         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1161                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1162         /*
1163          * Get the buffer for the block.
1164          */
1165         error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, bno), -1, &bp,
1166                 XFS_DATA_FORK);
1167         if (error) {
1168                 return error;
1169         }
1170         ASSERT(bp != NULL);
1171         leaf = bp->data;
1172         /*
1173          * Initialize the header.
1174          */
1175         leaf->hdr.info.magic = cpu_to_be16(magic);
1176         leaf->hdr.info.forw = 0;
1177         leaf->hdr.info.back = 0;
1178         leaf->hdr.count = 0;
1179         leaf->hdr.stale = 0;
1180         xfs_dir2_leaf_log_header(tp, bp);
1181         /*
1182          * If it's a leaf-format directory initialize the tail.
1183          * In this case our caller has the real bests table to copy into
1184          * the block.
1185          */
1186         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1187                 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1188                 ltp->bestcount = 0;
1189                 xfs_dir2_leaf_log_tail(tp, bp);
1190         }
1191         *bpp = bp;
1192         return 0;
1193 }
1194
1195 /*
1196  * Log the bests entries indicated from a leaf1 block.
1197  */
1198 static void
1199 xfs_dir2_leaf_log_bests(
1200         xfs_trans_t             *tp,            /* transaction pointer */
1201         xfs_dabuf_t             *bp,            /* leaf buffer */
1202         int                     first,          /* first entry to log */
1203         int                     last)           /* last entry to log */
1204 {
1205         __be16                  *firstb;        /* pointer to first entry */
1206         __be16                  *lastb;         /* pointer to last entry */
1207         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1208         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1209
1210         leaf = bp->data;
1211         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1212         ltp = XFS_DIR2_LEAF_TAIL_P(tp->t_mountp, leaf);
1213         firstb = XFS_DIR2_LEAF_BESTS_P(ltp) + first;
1214         lastb = XFS_DIR2_LEAF_BESTS_P(ltp) + last;
1215         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1216                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1217 }
1218
1219 /*
1220  * Log the leaf entries indicated from a leaf1 or leafn block.
1221  */
1222 void
1223 xfs_dir2_leaf_log_ents(
1224         xfs_trans_t             *tp,            /* transaction pointer */
1225         xfs_dabuf_t             *bp,            /* leaf buffer */
1226         int                     first,          /* first entry to log */
1227         int                     last)           /* last entry to log */
1228 {
1229         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1230         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1231         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1232
1233         leaf = bp->data;
1234         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1235                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1236         firstlep = &leaf->ents[first];
1237         lastlep = &leaf->ents[last];
1238         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1239                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1240 }
1241
1242 /*
1243  * Log the header of the leaf1 or leafn block.
1244  */
1245 void
1246 xfs_dir2_leaf_log_header(
1247         xfs_trans_t             *tp,            /* transaction pointer */
1248         xfs_dabuf_t             *bp)            /* leaf buffer */
1249 {
1250         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1251
1252         leaf = bp->data;
1253         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1254                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1255         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1256                 (uint)(sizeof(leaf->hdr) - 1));
1257 }
1258
1259 /*
1260  * Log the tail of the leaf1 block.
1261  */
1262 STATIC void
1263 xfs_dir2_leaf_log_tail(
1264         xfs_trans_t             *tp,            /* transaction pointer */
1265         xfs_dabuf_t             *bp)            /* leaf buffer */
1266 {
1267         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1268         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1269         xfs_mount_t             *mp;            /* filesystem mount point */
1270
1271         mp = tp->t_mountp;
1272         leaf = bp->data;
1273         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1274         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1275         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1276                 (uint)(mp->m_dirblksize - 1));
1277 }
1278
1279 /*
1280  * Look up the entry referred to by args in the leaf format directory.
1281  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1282  * is also used by the node-format code.
1283  */
1284 int
1285 xfs_dir2_leaf_lookup(
1286         xfs_da_args_t           *args)          /* operation arguments */
1287 {
1288         xfs_dabuf_t             *dbp;           /* data block buffer */
1289         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1290         xfs_inode_t             *dp;            /* incore directory inode */
1291         int                     error;          /* error return code */
1292         int                     index;          /* found entry index */
1293         xfs_dabuf_t             *lbp;           /* leaf buffer */
1294         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1295         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1296         xfs_trans_t             *tp;            /* transaction pointer */
1297
1298         xfs_dir2_trace_args("leaf_lookup", args);
1299         /*
1300          * Look up name in the leaf block, returning both buffers and index.
1301          */
1302         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1303                 return error;
1304         }
1305         tp = args->trans;
1306         dp = args->dp;
1307         xfs_dir2_leaf_check(dp, lbp);
1308         leaf = lbp->data;
1309         /*
1310          * Get to the leaf entry and contained data entry address.
1311          */
1312         lep = &leaf->ents[index];
1313         /*
1314          * Point to the data entry.
1315          */
1316         dep = (xfs_dir2_data_entry_t *)
1317               ((char *)dbp->data +
1318                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, be32_to_cpu(lep->address)));
1319         /*
1320          * Return the found inode number.
1321          */
1322         args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
1323         xfs_da_brelse(tp, dbp);
1324         xfs_da_brelse(tp, lbp);
1325         return XFS_ERROR(EEXIST);
1326 }
1327
1328 /*
1329  * Look up name/hash in the leaf block.
1330  * Fill in indexp with the found index, and dbpp with the data buffer.
1331  * If not found dbpp will be NULL, and ENOENT comes back.
1332  * lbpp will always be filled in with the leaf buffer unless there's an error.
1333  */
1334 static int                                      /* error */
1335 xfs_dir2_leaf_lookup_int(
1336         xfs_da_args_t           *args,          /* operation arguments */
1337         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1338         int                     *indexp,        /* out: index in leaf block */
1339         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1340 {
1341         xfs_dir2_db_t           curdb;          /* current data block number */
1342         xfs_dabuf_t             *dbp;           /* data buffer */
1343         xfs_dir2_data_entry_t   *dep;           /* data entry */
1344         xfs_inode_t             *dp;            /* incore directory inode */
1345         int                     error;          /* error return code */
1346         int                     index;          /* index in leaf block */
1347         xfs_dabuf_t             *lbp;           /* leaf buffer */
1348         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1349         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1350         xfs_mount_t             *mp;            /* filesystem mount point */
1351         xfs_dir2_db_t           newdb;          /* new data block number */
1352         xfs_trans_t             *tp;            /* transaction pointer */
1353
1354         dp = args->dp;
1355         tp = args->trans;
1356         mp = dp->i_mount;
1357         /*
1358          * Read the leaf block into the buffer.
1359          */
1360         if ((error =
1361             xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1362                     XFS_DATA_FORK))) {
1363                 return error;
1364         }
1365         *lbpp = lbp;
1366         leaf = lbp->data;
1367         xfs_dir2_leaf_check(dp, lbp);
1368         /*
1369          * Look for the first leaf entry with our hash value.
1370          */
1371         index = xfs_dir2_leaf_search_hash(args, lbp);
1372         /*
1373          * Loop over all the entries with the right hash value
1374          * looking to match the name.
1375          */
1376         for (lep = &leaf->ents[index], dbp = NULL, curdb = -1;
1377              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
1378              lep++, index++) {
1379                 /*
1380                  * Skip over stale leaf entries.
1381                  */
1382                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1383                         continue;
1384                 /*
1385                  * Get the new data block number.
1386                  */
1387                 newdb = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
1388                 /*
1389                  * If it's not the same as the old data block number,
1390                  * need to pitch the old one and read the new one.
1391                  */
1392                 if (newdb != curdb) {
1393                         if (dbp)
1394                                 xfs_da_brelse(tp, dbp);
1395                         if ((error =
1396                             xfs_da_read_buf(tp, dp,
1397                                     XFS_DIR2_DB_TO_DA(mp, newdb), -1, &dbp,
1398                                     XFS_DATA_FORK))) {
1399                                 xfs_da_brelse(tp, lbp);
1400                                 return error;
1401                         }
1402                         xfs_dir2_data_check(dp, dbp);
1403                         curdb = newdb;
1404                 }
1405                 /*
1406                  * Point to the data entry.
1407                  */
1408                 dep = (xfs_dir2_data_entry_t *)
1409                       ((char *)dbp->data +
1410                        XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address)));
1411                 /*
1412                  * If it matches then return it.
1413                  */
1414                 if (dep->namelen == args->namelen &&
1415                     dep->name[0] == args->name[0] &&
1416                     memcmp(dep->name, args->name, args->namelen) == 0) {
1417                         *dbpp = dbp;
1418                         *indexp = index;
1419                         return 0;
1420                 }
1421         }
1422         /*
1423          * No match found, return ENOENT.
1424          */
1425         ASSERT(args->oknoent);
1426         if (dbp)
1427                 xfs_da_brelse(tp, dbp);
1428         xfs_da_brelse(tp, lbp);
1429         return XFS_ERROR(ENOENT);
1430 }
1431
1432 /*
1433  * Remove an entry from a leaf format directory.
1434  */
1435 int                                             /* error */
1436 xfs_dir2_leaf_removename(
1437         xfs_da_args_t           *args)          /* operation arguments */
1438 {
1439         __be16                  *bestsp;        /* leaf block best freespace */
1440         xfs_dir2_data_t         *data;          /* data block structure */
1441         xfs_dir2_db_t           db;             /* data block number */
1442         xfs_dabuf_t             *dbp;           /* data block buffer */
1443         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1444         xfs_inode_t             *dp;            /* incore directory inode */
1445         int                     error;          /* error return code */
1446         xfs_dir2_db_t           i;              /* temporary data block # */
1447         int                     index;          /* index into leaf entries */
1448         xfs_dabuf_t             *lbp;           /* leaf buffer */
1449         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1450         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1451         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1452         xfs_mount_t             *mp;            /* filesystem mount point */
1453         int                     needlog;        /* need to log data header */
1454         int                     needscan;       /* need to rescan data frees */
1455         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1456         xfs_trans_t             *tp;            /* transaction pointer */
1457
1458         xfs_dir2_trace_args("leaf_removename", args);
1459         /*
1460          * Lookup the leaf entry, get the leaf and data blocks read in.
1461          */
1462         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1463                 return error;
1464         }
1465         dp = args->dp;
1466         tp = args->trans;
1467         mp = dp->i_mount;
1468         leaf = lbp->data;
1469         data = dbp->data;
1470         xfs_dir2_data_check(dp, dbp);
1471         /*
1472          * Point to the leaf entry, use that to point to the data entry.
1473          */
1474         lep = &leaf->ents[index];
1475         db = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
1476         dep = (xfs_dir2_data_entry_t *)
1477               ((char *)data + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address)));
1478         needscan = needlog = 0;
1479         oldbest = be16_to_cpu(data->hdr.bestfree[0].length);
1480         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1481         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1482         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1483         /*
1484          * Mark the former data entry unused.
1485          */
1486         xfs_dir2_data_make_free(tp, dbp,
1487                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)data),
1488                 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
1489         /*
1490          * We just mark the leaf entry stale by putting a null in it.
1491          */
1492         be16_add(&leaf->hdr.stale, 1);
1493         xfs_dir2_leaf_log_header(tp, lbp);
1494         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1495         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1496         /*
1497          * Scan the freespace in the data block again if necessary,
1498          * log the data block header if necessary.
1499          */
1500         if (needscan)
1501                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1502         if (needlog)
1503                 xfs_dir2_data_log_header(tp, dbp);
1504         /*
1505          * If the longest freespace in the data block has changed,
1506          * put the new value in the bests table and log that.
1507          */
1508         if (be16_to_cpu(data->hdr.bestfree[0].length) != oldbest) {
1509                 bestsp[db] = data->hdr.bestfree[0].length;
1510                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1511         }
1512         xfs_dir2_data_check(dp, dbp);
1513         /*
1514          * If the data block is now empty then get rid of the data block.
1515          */
1516         if (be16_to_cpu(data->hdr.bestfree[0].length) ==
1517             mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1518                 ASSERT(db != mp->m_dirdatablk);
1519                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1520                         /*
1521                          * Nope, can't get rid of it because it caused
1522                          * allocation of a bmap btree block to do so.
1523                          * Just go on, returning success, leaving the
1524                          * empty block in place.
1525                          */
1526                         if (error == ENOSPC && args->total == 0) {
1527                                 xfs_da_buf_done(dbp);
1528                                 error = 0;
1529                         }
1530                         xfs_dir2_leaf_check(dp, lbp);
1531                         xfs_da_buf_done(lbp);
1532                         return error;
1533                 }
1534                 dbp = NULL;
1535                 /*
1536                  * If this is the last data block then compact the
1537                  * bests table by getting rid of entries.
1538                  */
1539                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1540                         /*
1541                          * Look for the last active entry (i).
1542                          */
1543                         for (i = db - 1; i > 0; i--) {
1544                                 if (be16_to_cpu(bestsp[i]) != NULLDATAOFF)
1545                                         break;
1546                         }
1547                         /*
1548                          * Copy the table down so inactive entries at the
1549                          * end are removed.
1550                          */
1551                         memmove(&bestsp[db - i], bestsp,
1552                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1553                         be32_add(&ltp->bestcount, -(db - i));
1554                         xfs_dir2_leaf_log_tail(tp, lbp);
1555                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1556                 } else
1557                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1558         }
1559         /*
1560          * If the data block was not the first one, drop it.
1561          */
1562         else if (db != mp->m_dirdatablk && dbp != NULL) {
1563                 xfs_da_buf_done(dbp);
1564                 dbp = NULL;
1565         }
1566         xfs_dir2_leaf_check(dp, lbp);
1567         /*
1568          * See if we can convert to block form.
1569          */
1570         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1571 }
1572
1573 /*
1574  * Replace the inode number in a leaf format directory entry.
1575  */
1576 int                                             /* error */
1577 xfs_dir2_leaf_replace(
1578         xfs_da_args_t           *args)          /* operation arguments */
1579 {
1580         xfs_dabuf_t             *dbp;           /* data block buffer */
1581         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1582         xfs_inode_t             *dp;            /* incore directory inode */
1583         int                     error;          /* error return code */
1584         int                     index;          /* index of leaf entry */
1585         xfs_dabuf_t             *lbp;           /* leaf buffer */
1586         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1587         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1588         xfs_trans_t             *tp;            /* transaction pointer */
1589
1590         xfs_dir2_trace_args("leaf_replace", args);
1591         /*
1592          * Look up the entry.
1593          */
1594         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1595                 return error;
1596         }
1597         dp = args->dp;
1598         leaf = lbp->data;
1599         /*
1600          * Point to the leaf entry, get data address from it.
1601          */
1602         lep = &leaf->ents[index];
1603         /*
1604          * Point to the data entry.
1605          */
1606         dep = (xfs_dir2_data_entry_t *)
1607               ((char *)dbp->data +
1608                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, be32_to_cpu(lep->address)));
1609         ASSERT(args->inumber != INT_GET(dep->inumber, ARCH_CONVERT));
1610         /*
1611          * Put the new inode number in, log it.
1612          */
1613         INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
1614         tp = args->trans;
1615         xfs_dir2_data_log_entry(tp, dbp, dep);
1616         xfs_da_buf_done(dbp);
1617         xfs_dir2_leaf_check(dp, lbp);
1618         xfs_da_brelse(tp, lbp);
1619         return 0;
1620 }
1621
1622 /*
1623  * Return index in the leaf block (lbp) which is either the first
1624  * one with this hash value, or if there are none, the insert point
1625  * for that hash value.
1626  */
1627 int                                             /* index value */
1628 xfs_dir2_leaf_search_hash(
1629         xfs_da_args_t           *args,          /* operation arguments */
1630         xfs_dabuf_t             *lbp)           /* leaf buffer */
1631 {
1632         xfs_dahash_t            hash=0;         /* hash from this entry */
1633         xfs_dahash_t            hashwant;       /* hash value looking for */
1634         int                     high;           /* high leaf index */
1635         int                     low;            /* low leaf index */
1636         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1637         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1638         int                     mid=0;          /* current leaf index */
1639
1640         leaf = lbp->data;
1641 #ifndef __KERNEL__
1642         if (!leaf->hdr.count)
1643                 return 0;
1644 #endif
1645         /*
1646          * Note, the table cannot be empty, so we have to go through the loop.
1647          * Binary search the leaf entries looking for our hash value.
1648          */
1649         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1650                 hashwant = args->hashval;
1651              low <= high; ) {
1652                 mid = (low + high) >> 1;
1653                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1654                         break;
1655                 if (hash < hashwant)
1656                         low = mid + 1;
1657                 else
1658                         high = mid - 1;
1659         }
1660         /*
1661          * Found one, back up through all the equal hash values.
1662          */
1663         if (hash == hashwant) {
1664                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1665                         mid--;
1666                 }
1667         }
1668         /*
1669          * Need to point to an entry higher than ours.
1670          */
1671         else if (hash < hashwant)
1672                 mid++;
1673         return mid;
1674 }
1675
1676 /*
1677  * Trim off a trailing data block.  We know it's empty since the leaf
1678  * freespace table says so.
1679  */
1680 int                                             /* error */
1681 xfs_dir2_leaf_trim_data(
1682         xfs_da_args_t           *args,          /* operation arguments */
1683         xfs_dabuf_t             *lbp,           /* leaf buffer */
1684         xfs_dir2_db_t           db)             /* data block number */
1685 {
1686         __be16                  *bestsp;        /* leaf bests table */
1687 #ifdef DEBUG
1688         xfs_dir2_data_t         *data;          /* data block structure */
1689 #endif
1690         xfs_dabuf_t             *dbp;           /* data block buffer */
1691         xfs_inode_t             *dp;            /* incore directory inode */
1692         int                     error;          /* error return value */
1693         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1694         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1695         xfs_mount_t             *mp;            /* filesystem mount point */
1696         xfs_trans_t             *tp;            /* transaction pointer */
1697
1698         dp = args->dp;
1699         mp = dp->i_mount;
1700         tp = args->trans;
1701         /*
1702          * Read the offending data block.  We need its buffer.
1703          */
1704         if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, db), -1, &dbp,
1705                         XFS_DATA_FORK))) {
1706                 return error;
1707         }
1708 #ifdef DEBUG
1709         data = dbp->data;
1710         ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1711 #endif
1712         /* this seems to be an error
1713          * data is only valid if DEBUG is defined?
1714          * RMC 09/08/1999
1715          */
1716
1717         leaf = lbp->data;
1718         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1719         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) ==
1720                mp->m_dirblksize - (uint)sizeof(data->hdr));
1721         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1722         /*
1723          * Get rid of the data block.
1724          */
1725         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1726                 ASSERT(error != ENOSPC);
1727                 xfs_da_brelse(tp, dbp);
1728                 return error;
1729         }
1730         /*
1731          * Eliminate the last bests entry from the table.
1732          */
1733         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1734         be32_add(&ltp->bestcount, -1);
1735         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1736         xfs_dir2_leaf_log_tail(tp, lbp);
1737         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1738         return 0;
1739 }
1740
1741 /*
1742  * Convert node form directory to leaf form directory.
1743  * The root of the node form dir needs to already be a LEAFN block.
1744  * Just return if we can't do anything.
1745  */
1746 int                                             /* error */
1747 xfs_dir2_node_to_leaf(
1748         xfs_da_state_t          *state)         /* directory operation state */
1749 {
1750         xfs_da_args_t           *args;          /* operation arguments */
1751         xfs_inode_t             *dp;            /* incore directory inode */
1752         int                     error;          /* error return code */
1753         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1754         xfs_fileoff_t           fo;             /* freespace file offset */
1755         xfs_dir2_free_t         *free;          /* freespace structure */
1756         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1757         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1758         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1759         xfs_mount_t             *mp;            /* filesystem mount point */
1760         int                     rval;           /* successful free trim? */
1761         xfs_trans_t             *tp;            /* transaction pointer */
1762
1763         /*
1764          * There's more than a leaf level in the btree, so there must
1765          * be multiple leafn blocks.  Give up.
1766          */
1767         if (state->path.active > 1)
1768                 return 0;
1769         args = state->args;
1770         xfs_dir2_trace_args("node_to_leaf", args);
1771         mp = state->mp;
1772         dp = args->dp;
1773         tp = args->trans;
1774         /*
1775          * Get the last offset in the file.
1776          */
1777         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1778                 return error;
1779         }
1780         fo -= mp->m_dirblkfsbs;
1781         /*
1782          * If there are freespace blocks other than the first one,
1783          * take this opportunity to remove trailing empty freespace blocks
1784          * that may have been left behind during no-space-reservation
1785          * operations.
1786          */
1787         while (fo > mp->m_dirfreeblk) {
1788                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1789                         return error;
1790                 }
1791                 if (rval)
1792                         fo -= mp->m_dirblkfsbs;
1793                 else
1794                         return 0;
1795         }
1796         /*
1797          * Now find the block just before the freespace block.
1798          */
1799         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1800                 return error;
1801         }
1802         /*
1803          * If it's not the single leaf block, give up.
1804          */
1805         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1806                 return 0;
1807         lbp = state->path.blk[0].bp;
1808         leaf = lbp->data;
1809         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1810         /*
1811          * Read the freespace block.
1812          */
1813         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1814                         XFS_DATA_FORK))) {
1815                 return error;
1816         }
1817         free = fbp->data;
1818         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1819         ASSERT(!free->hdr.firstdb);
1820         /*
1821          * Now see if the leafn and free data will fit in a leaf1.
1822          * If not, release the buffer and give up.
1823          */
1824         if ((uint)sizeof(leaf->hdr) +
1825             (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale)) * (uint)sizeof(leaf->ents[0]) +
1826             be32_to_cpu(free->hdr.nvalid) * (uint)sizeof(leaf->bests[0]) +
1827             (uint)sizeof(leaf->tail) >
1828             mp->m_dirblksize) {
1829                 xfs_da_brelse(tp, fbp);
1830                 return 0;
1831         }
1832         /*
1833          * If the leaf has any stale entries in it, compress them out.
1834          * The compact routine will log the header.
1835          */
1836         if (be16_to_cpu(leaf->hdr.stale))
1837                 xfs_dir2_leaf_compact(args, lbp);
1838         else
1839                 xfs_dir2_leaf_log_header(tp, lbp);
1840         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1841         /*
1842          * Set up the leaf tail from the freespace block.
1843          */
1844         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1845         ltp->bestcount = free->hdr.nvalid;
1846         /*
1847          * Set up the leaf bests table.
1848          */
1849         memcpy(XFS_DIR2_LEAF_BESTS_P(ltp), free->bests,
1850                 be32_to_cpu(ltp->bestcount) * sizeof(leaf->bests[0]));
1851         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1852         xfs_dir2_leaf_log_tail(tp, lbp);
1853         xfs_dir2_leaf_check(dp, lbp);
1854         /*
1855          * Get rid of the freespace block.
1856          */
1857         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1858         if (error) {
1859                 /*
1860                  * This can't fail here because it can only happen when
1861                  * punching out the middle of an extent, and this is an
1862                  * isolated block.
1863                  */
1864                 ASSERT(error != ENOSPC);
1865                 return error;
1866         }
1867         fbp = NULL;
1868         /*
1869          * Now see if we can convert the single-leaf directory
1870          * down to a block form directory.
1871          * This routine always kills the dabuf for the leaf, so
1872          * eliminate it from the path.
1873          */
1874         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1875         state->path.blk[0].bp = NULL;
1876         return error;
1877 }