2abf945e5844da54537ff36d5a8112c9bd13464c
[muen/linux.git] / fs / xfs / libxfs / xfs_dir2_leaf.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_bmap.h"
17 #include "xfs_dir2.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_cksum.h"
24 #include "xfs_log.h"
25
26 /*
27  * Local function declarations.
28  */
29 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
30                                     int *indexp, struct xfs_buf **dbpp);
31 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
32                                     struct xfs_buf *bp, int first, int last);
33 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
34                                    struct xfs_buf *bp);
35
36 /*
37  * Check the internal consistency of a leaf1 block.
38  * Pop an assert if something is wrong.
39  */
40 #ifdef DEBUG
41 static xfs_failaddr_t
42 xfs_dir3_leaf1_check(
43         struct xfs_inode        *dp,
44         struct xfs_buf          *bp)
45 {
46         struct xfs_dir2_leaf    *leaf = bp->b_addr;
47         struct xfs_dir3_icleaf_hdr leafhdr;
48
49         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
50
51         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
52                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
53                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
54                         return __this_address;
55         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
56                 return __this_address;
57
58         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
59 }
60
61 static inline void
62 xfs_dir3_leaf_check(
63         struct xfs_inode        *dp,
64         struct xfs_buf          *bp)
65 {
66         xfs_failaddr_t          fa;
67
68         fa = xfs_dir3_leaf1_check(dp, bp);
69         if (!fa)
70                 return;
71         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
72                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
73                         fa);
74         ASSERT(0);
75 }
76 #else
77 #define xfs_dir3_leaf_check(dp, bp)
78 #endif
79
80 xfs_failaddr_t
81 xfs_dir3_leaf_check_int(
82         struct xfs_mount        *mp,
83         struct xfs_inode        *dp,
84         struct xfs_dir3_icleaf_hdr *hdr,
85         struct xfs_dir2_leaf    *leaf)
86 {
87         struct xfs_dir2_leaf_entry *ents;
88         xfs_dir2_leaf_tail_t    *ltp;
89         int                     stale;
90         int                     i;
91         const struct xfs_dir_ops *ops;
92         struct xfs_dir3_icleaf_hdr leafhdr;
93         struct xfs_da_geometry  *geo = mp->m_dir_geo;
94
95         /*
96          * we can be passed a null dp here from a verifier, so we need to go the
97          * hard way to get them.
98          */
99         ops = xfs_dir_get_ops(mp, dp);
100
101         if (!hdr) {
102                 ops->leaf_hdr_from_disk(&leafhdr, leaf);
103                 hdr = &leafhdr;
104         }
105
106         ents = ops->leaf_ents_p(leaf);
107         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
108
109         /*
110          * XXX (dgc): This value is not restrictive enough.
111          * Should factor in the size of the bests table as well.
112          * We can deduce a value for that from di_size.
113          */
114         if (hdr->count > ops->leaf_max_ents(geo))
115                 return __this_address;
116
117         /* Leaves and bests don't overlap in leaf format. */
118         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
119              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
120             (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
121                 return __this_address;
122
123         /* Check hash value order, count stale entries.  */
124         for (i = stale = 0; i < hdr->count; i++) {
125                 if (i + 1 < hdr->count) {
126                         if (be32_to_cpu(ents[i].hashval) >
127                                         be32_to_cpu(ents[i + 1].hashval))
128                                 return __this_address;
129                 }
130                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
131                         stale++;
132         }
133         if (hdr->stale != stale)
134                 return __this_address;
135         return NULL;
136 }
137
138 /*
139  * We verify the magic numbers before decoding the leaf header so that on debug
140  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
141  * to incorrect magic numbers.
142  */
143 static xfs_failaddr_t
144 xfs_dir3_leaf_verify(
145         struct xfs_buf          *bp)
146 {
147         struct xfs_mount        *mp = bp->b_target->bt_mount;
148         struct xfs_dir2_leaf    *leaf = bp->b_addr;
149         xfs_failaddr_t          fa;
150
151         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
152         if (fa)
153                 return fa;
154
155         return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
156 }
157
158 static void
159 xfs_dir3_leaf_read_verify(
160         struct xfs_buf  *bp)
161 {
162         struct xfs_mount        *mp = bp->b_target->bt_mount;
163         xfs_failaddr_t          fa;
164
165         if (xfs_sb_version_hascrc(&mp->m_sb) &&
166              !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
167                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
168         else {
169                 fa = xfs_dir3_leaf_verify(bp);
170                 if (fa)
171                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
172         }
173 }
174
175 static void
176 xfs_dir3_leaf_write_verify(
177         struct xfs_buf  *bp)
178 {
179         struct xfs_mount        *mp = bp->b_target->bt_mount;
180         struct xfs_buf_log_item *bip = bp->b_log_item;
181         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
182         xfs_failaddr_t          fa;
183
184         fa = xfs_dir3_leaf_verify(bp);
185         if (fa) {
186                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
187                 return;
188         }
189
190         if (!xfs_sb_version_hascrc(&mp->m_sb))
191                 return;
192
193         if (bip)
194                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
195
196         xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
197 }
198
199 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
200         .name = "xfs_dir3_leaf1",
201         .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
202                      cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
203         .verify_read = xfs_dir3_leaf_read_verify,
204         .verify_write = xfs_dir3_leaf_write_verify,
205         .verify_struct = xfs_dir3_leaf_verify,
206 };
207
208 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
209         .name = "xfs_dir3_leafn",
210         .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
211                      cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
212         .verify_read = xfs_dir3_leaf_read_verify,
213         .verify_write = xfs_dir3_leaf_write_verify,
214         .verify_struct = xfs_dir3_leaf_verify,
215 };
216
217 int
218 xfs_dir3_leaf_read(
219         struct xfs_trans        *tp,
220         struct xfs_inode        *dp,
221         xfs_dablk_t             fbno,
222         xfs_daddr_t             mappedbno,
223         struct xfs_buf          **bpp)
224 {
225         int                     err;
226
227         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
228                                 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
229         if (!err && tp && *bpp)
230                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
231         return err;
232 }
233
234 int
235 xfs_dir3_leafn_read(
236         struct xfs_trans        *tp,
237         struct xfs_inode        *dp,
238         xfs_dablk_t             fbno,
239         xfs_daddr_t             mappedbno,
240         struct xfs_buf          **bpp)
241 {
242         int                     err;
243
244         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
245                                 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
246         if (!err && tp && *bpp)
247                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
248         return err;
249 }
250
251 /*
252  * Initialize a new leaf block, leaf1 or leafn magic accepted.
253  */
254 static void
255 xfs_dir3_leaf_init(
256         struct xfs_mount        *mp,
257         struct xfs_trans        *tp,
258         struct xfs_buf          *bp,
259         xfs_ino_t               owner,
260         uint16_t                type)
261 {
262         struct xfs_dir2_leaf    *leaf = bp->b_addr;
263
264         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
265
266         if (xfs_sb_version_hascrc(&mp->m_sb)) {
267                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
268
269                 memset(leaf3, 0, sizeof(*leaf3));
270
271                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
272                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
273                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
274                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
275                 leaf3->info.owner = cpu_to_be64(owner);
276                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
277         } else {
278                 memset(leaf, 0, sizeof(*leaf));
279                 leaf->hdr.info.magic = cpu_to_be16(type);
280         }
281
282         /*
283          * If it's a leaf-format directory initialize the tail.
284          * Caller is responsible for initialising the bests table.
285          */
286         if (type == XFS_DIR2_LEAF1_MAGIC) {
287                 struct xfs_dir2_leaf_tail *ltp;
288
289                 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
290                 ltp->bestcount = 0;
291                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
292                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
293         } else {
294                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
295                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
296         }
297 }
298
299 int
300 xfs_dir3_leaf_get_buf(
301         xfs_da_args_t           *args,
302         xfs_dir2_db_t           bno,
303         struct xfs_buf          **bpp,
304         uint16_t                magic)
305 {
306         struct xfs_inode        *dp = args->dp;
307         struct xfs_trans        *tp = args->trans;
308         struct xfs_mount        *mp = dp->i_mount;
309         struct xfs_buf          *bp;
310         int                     error;
311
312         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
313         ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
314                bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
315
316         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
317                                -1, &bp, XFS_DATA_FORK);
318         if (error)
319                 return error;
320
321         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
322         xfs_dir3_leaf_log_header(args, bp);
323         if (magic == XFS_DIR2_LEAF1_MAGIC)
324                 xfs_dir3_leaf_log_tail(args, bp);
325         *bpp = bp;
326         return 0;
327 }
328
329 /*
330  * Convert a block form directory to a leaf form directory.
331  */
332 int                                             /* error */
333 xfs_dir2_block_to_leaf(
334         xfs_da_args_t           *args,          /* operation arguments */
335         struct xfs_buf          *dbp)           /* input block's buffer */
336 {
337         __be16                  *bestsp;        /* leaf's bestsp entries */
338         xfs_dablk_t             blkno;          /* leaf block's bno */
339         xfs_dir2_data_hdr_t     *hdr;           /* block header */
340         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
341         xfs_dir2_block_tail_t   *btp;           /* block's tail */
342         xfs_inode_t             *dp;            /* incore directory inode */
343         int                     error;          /* error return code */
344         struct xfs_buf          *lbp;           /* leaf block's buffer */
345         xfs_dir2_db_t           ldb;            /* leaf block's bno */
346         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
347         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
348         int                     needlog;        /* need to log block header */
349         int                     needscan;       /* need to rescan bestfree */
350         xfs_trans_t             *tp;            /* transaction pointer */
351         struct xfs_dir2_data_free *bf;
352         struct xfs_dir2_leaf_entry *ents;
353         struct xfs_dir3_icleaf_hdr leafhdr;
354
355         trace_xfs_dir2_block_to_leaf(args);
356
357         dp = args->dp;
358         tp = args->trans;
359         /*
360          * Add the leaf block to the inode.
361          * This interface will only put blocks in the leaf/node range.
362          * Since that's empty now, we'll get the root (block 0 in range).
363          */
364         if ((error = xfs_da_grow_inode(args, &blkno))) {
365                 return error;
366         }
367         ldb = xfs_dir2_da_to_db(args->geo, blkno);
368         ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
369         /*
370          * Initialize the leaf block, get a buffer for it.
371          */
372         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
373         if (error)
374                 return error;
375
376         leaf = lbp->b_addr;
377         hdr = dbp->b_addr;
378         xfs_dir3_data_check(dp, dbp);
379         btp = xfs_dir2_block_tail_p(args->geo, hdr);
380         blp = xfs_dir2_block_leaf_p(btp);
381         bf = dp->d_ops->data_bestfree_p(hdr);
382         ents = dp->d_ops->leaf_ents_p(leaf);
383
384         /*
385          * Set the counts in the leaf header.
386          */
387         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
388         leafhdr.count = be32_to_cpu(btp->count);
389         leafhdr.stale = be32_to_cpu(btp->stale);
390         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
391         xfs_dir3_leaf_log_header(args, lbp);
392
393         /*
394          * Could compact these but I think we always do the conversion
395          * after squeezing out stale entries.
396          */
397         memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
398         xfs_dir3_leaf_log_ents(args, lbp, 0, leafhdr.count - 1);
399         needscan = 0;
400         needlog = 1;
401         /*
402          * Make the space formerly occupied by the leaf entries and block
403          * tail be free.
404          */
405         xfs_dir2_data_make_free(args, dbp,
406                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
407                 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
408                                        (char *)blp),
409                 &needlog, &needscan);
410         /*
411          * Fix up the block header, make it a data block.
412          */
413         dbp->b_ops = &xfs_dir3_data_buf_ops;
414         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
415         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
416                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
417         else
418                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
419
420         if (needscan)
421                 xfs_dir2_data_freescan(dp, hdr, &needlog);
422         /*
423          * Set up leaf tail and bests table.
424          */
425         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
426         ltp->bestcount = cpu_to_be32(1);
427         bestsp = xfs_dir2_leaf_bests_p(ltp);
428         bestsp[0] =  bf[0].length;
429         /*
430          * Log the data header and leaf bests table.
431          */
432         if (needlog)
433                 xfs_dir2_data_log_header(args, dbp);
434         xfs_dir3_leaf_check(dp, lbp);
435         xfs_dir3_data_check(dp, dbp);
436         xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
437         return 0;
438 }
439
440 STATIC void
441 xfs_dir3_leaf_find_stale(
442         struct xfs_dir3_icleaf_hdr *leafhdr,
443         struct xfs_dir2_leaf_entry *ents,
444         int                     index,
445         int                     *lowstale,
446         int                     *highstale)
447 {
448         /*
449          * Find the first stale entry before our index, if any.
450          */
451         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
452                 if (ents[*lowstale].address ==
453                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
454                         break;
455         }
456
457         /*
458          * Find the first stale entry at or after our index, if any.
459          * Stop if the result would require moving more entries than using
460          * lowstale.
461          */
462         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
463                 if (ents[*highstale].address ==
464                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
465                         break;
466                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
467                         break;
468         }
469 }
470
471 struct xfs_dir2_leaf_entry *
472 xfs_dir3_leaf_find_entry(
473         struct xfs_dir3_icleaf_hdr *leafhdr,
474         struct xfs_dir2_leaf_entry *ents,
475         int                     index,          /* leaf table position */
476         int                     compact,        /* need to compact leaves */
477         int                     lowstale,       /* index of prev stale leaf */
478         int                     highstale,      /* index of next stale leaf */
479         int                     *lfloglow,      /* low leaf logging index */
480         int                     *lfloghigh)     /* high leaf logging index */
481 {
482         if (!leafhdr->stale) {
483                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
484
485                 /*
486                  * Now we need to make room to insert the leaf entry.
487                  *
488                  * If there are no stale entries, just insert a hole at index.
489                  */
490                 lep = &ents[index];
491                 if (index < leafhdr->count)
492                         memmove(lep + 1, lep,
493                                 (leafhdr->count - index) * sizeof(*lep));
494
495                 /*
496                  * Record low and high logging indices for the leaf.
497                  */
498                 *lfloglow = index;
499                 *lfloghigh = leafhdr->count++;
500                 return lep;
501         }
502
503         /*
504          * There are stale entries.
505          *
506          * We will use one of them for the new entry.  It's probably not at
507          * the right location, so we'll have to shift some up or down first.
508          *
509          * If we didn't compact before, we need to find the nearest stale
510          * entries before and after our insertion point.
511          */
512         if (compact == 0)
513                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
514                                          &lowstale, &highstale);
515
516         /*
517          * If the low one is better, use it.
518          */
519         if (lowstale >= 0 &&
520             (highstale == leafhdr->count ||
521              index - lowstale - 1 < highstale - index)) {
522                 ASSERT(index - lowstale - 1 >= 0);
523                 ASSERT(ents[lowstale].address ==
524                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
525
526                 /*
527                  * Copy entries up to cover the stale entry and make room
528                  * for the new entry.
529                  */
530                 if (index - lowstale - 1 > 0) {
531                         memmove(&ents[lowstale], &ents[lowstale + 1],
532                                 (index - lowstale - 1) *
533                                         sizeof(xfs_dir2_leaf_entry_t));
534                 }
535                 *lfloglow = min(lowstale, *lfloglow);
536                 *lfloghigh = max(index - 1, *lfloghigh);
537                 leafhdr->stale--;
538                 return &ents[index - 1];
539         }
540
541         /*
542          * The high one is better, so use that one.
543          */
544         ASSERT(highstale - index >= 0);
545         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
546
547         /*
548          * Copy entries down to cover the stale entry and make room for the
549          * new entry.
550          */
551         if (highstale - index > 0) {
552                 memmove(&ents[index + 1], &ents[index],
553                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
554         }
555         *lfloglow = min(index, *lfloglow);
556         *lfloghigh = max(highstale, *lfloghigh);
557         leafhdr->stale--;
558         return &ents[index];
559 }
560
561 /*
562  * Add an entry to a leaf form directory.
563  */
564 int                                             /* error */
565 xfs_dir2_leaf_addname(
566         xfs_da_args_t           *args)          /* operation arguments */
567 {
568         __be16                  *bestsp;        /* freespace table in leaf */
569         int                     compact;        /* need to compact leaves */
570         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
571         struct xfs_buf          *dbp;           /* data block buffer */
572         xfs_dir2_data_entry_t   *dep;           /* data block entry */
573         xfs_inode_t             *dp;            /* incore directory inode */
574         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
575         int                     error;          /* error return value */
576         int                     grown;          /* allocated new data block */
577         int                     highstale = 0;  /* index of next stale leaf */
578         int                     i;              /* temporary, index */
579         int                     index;          /* leaf table position */
580         struct xfs_buf          *lbp;           /* leaf's buffer */
581         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
582         int                     length;         /* length of new entry */
583         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
584         int                     lfloglow;       /* low leaf logging index */
585         int                     lfloghigh;      /* high leaf logging index */
586         int                     lowstale = 0;   /* index of prev stale leaf */
587         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
588         int                     needbytes;      /* leaf block bytes needed */
589         int                     needlog;        /* need to log data header */
590         int                     needscan;       /* need to rescan data free */
591         __be16                  *tagp;          /* end of data entry */
592         xfs_trans_t             *tp;            /* transaction pointer */
593         xfs_dir2_db_t           use_block;      /* data block number */
594         struct xfs_dir2_data_free *bf;          /* bestfree table */
595         struct xfs_dir2_leaf_entry *ents;
596         struct xfs_dir3_icleaf_hdr leafhdr;
597
598         trace_xfs_dir2_leaf_addname(args);
599
600         dp = args->dp;
601         tp = args->trans;
602
603         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
604         if (error)
605                 return error;
606
607         /*
608          * Look up the entry by hash value and name.
609          * We know it's not there, our caller has already done a lookup.
610          * So the index is of the entry to insert in front of.
611          * But if there are dup hash values the index is of the first of those.
612          */
613         index = xfs_dir2_leaf_search_hash(args, lbp);
614         leaf = lbp->b_addr;
615         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
616         ents = dp->d_ops->leaf_ents_p(leaf);
617         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
618         bestsp = xfs_dir2_leaf_bests_p(ltp);
619         length = dp->d_ops->data_entsize(args->namelen);
620
621         /*
622          * See if there are any entries with the same hash value
623          * and space in their block for the new entry.
624          * This is good because it puts multiple same-hash value entries
625          * in a data block, improving the lookup of those entries.
626          */
627         for (use_block = -1, lep = &ents[index];
628              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
629              index++, lep++) {
630                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
631                         continue;
632                 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
633                 ASSERT(i < be32_to_cpu(ltp->bestcount));
634                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
635                 if (be16_to_cpu(bestsp[i]) >= length) {
636                         use_block = i;
637                         break;
638                 }
639         }
640         /*
641          * Didn't find a block yet, linear search all the data blocks.
642          */
643         if (use_block == -1) {
644                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
645                         /*
646                          * Remember a block we see that's missing.
647                          */
648                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
649                             use_block == -1)
650                                 use_block = i;
651                         else if (be16_to_cpu(bestsp[i]) >= length) {
652                                 use_block = i;
653                                 break;
654                         }
655                 }
656         }
657         /*
658          * How many bytes do we need in the leaf block?
659          */
660         needbytes = 0;
661         if (!leafhdr.stale)
662                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
663         if (use_block == -1)
664                 needbytes += sizeof(xfs_dir2_data_off_t);
665
666         /*
667          * Now kill use_block if it refers to a missing block, so we
668          * can use it as an indication of allocation needed.
669          */
670         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
671                 use_block = -1;
672         /*
673          * If we don't have enough free bytes but we can make enough
674          * by compacting out stale entries, we'll do that.
675          */
676         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
677             leafhdr.stale > 1)
678                 compact = 1;
679
680         /*
681          * Otherwise if we don't have enough free bytes we need to
682          * convert to node form.
683          */
684         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
685                 /*
686                  * Just checking or no space reservation, give up.
687                  */
688                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
689                                                         args->total == 0) {
690                         xfs_trans_brelse(tp, lbp);
691                         return -ENOSPC;
692                 }
693                 /*
694                  * Convert to node form.
695                  */
696                 error = xfs_dir2_leaf_to_node(args, lbp);
697                 if (error)
698                         return error;
699                 /*
700                  * Then add the new entry.
701                  */
702                 return xfs_dir2_node_addname(args);
703         }
704         /*
705          * Otherwise it will fit without compaction.
706          */
707         else
708                 compact = 0;
709         /*
710          * If just checking, then it will fit unless we needed to allocate
711          * a new data block.
712          */
713         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
714                 xfs_trans_brelse(tp, lbp);
715                 return use_block == -1 ? -ENOSPC : 0;
716         }
717         /*
718          * If no allocations are allowed, return now before we've
719          * changed anything.
720          */
721         if (args->total == 0 && use_block == -1) {
722                 xfs_trans_brelse(tp, lbp);
723                 return -ENOSPC;
724         }
725         /*
726          * Need to compact the leaf entries, removing stale ones.
727          * Leave one stale entry behind - the one closest to our
728          * insertion index - and we'll shift that one to our insertion
729          * point later.
730          */
731         if (compact) {
732                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
733                         &highstale, &lfloglow, &lfloghigh);
734         }
735         /*
736          * There are stale entries, so we'll need log-low and log-high
737          * impossibly bad values later.
738          */
739         else if (leafhdr.stale) {
740                 lfloglow = leafhdr.count;
741                 lfloghigh = -1;
742         }
743         /*
744          * If there was no data block space found, we need to allocate
745          * a new one.
746          */
747         if (use_block == -1) {
748                 /*
749                  * Add the new data block.
750                  */
751                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
752                                 &use_block))) {
753                         xfs_trans_brelse(tp, lbp);
754                         return error;
755                 }
756                 /*
757                  * Initialize the block.
758                  */
759                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
760                         xfs_trans_brelse(tp, lbp);
761                         return error;
762                 }
763                 /*
764                  * If we're adding a new data block on the end we need to
765                  * extend the bests table.  Copy it up one entry.
766                  */
767                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
768                         bestsp--;
769                         memmove(&bestsp[0], &bestsp[1],
770                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
771                         be32_add_cpu(&ltp->bestcount, 1);
772                         xfs_dir3_leaf_log_tail(args, lbp);
773                         xfs_dir3_leaf_log_bests(args, lbp, 0,
774                                                 be32_to_cpu(ltp->bestcount) - 1);
775                 }
776                 /*
777                  * If we're filling in a previously empty block just log it.
778                  */
779                 else
780                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
781                 hdr = dbp->b_addr;
782                 bf = dp->d_ops->data_bestfree_p(hdr);
783                 bestsp[use_block] = bf[0].length;
784                 grown = 1;
785         } else {
786                 /*
787                  * Already had space in some data block.
788                  * Just read that one in.
789                  */
790                 error = xfs_dir3_data_read(tp, dp,
791                                    xfs_dir2_db_to_da(args->geo, use_block),
792                                    -1, &dbp);
793                 if (error) {
794                         xfs_trans_brelse(tp, lbp);
795                         return error;
796                 }
797                 hdr = dbp->b_addr;
798                 bf = dp->d_ops->data_bestfree_p(hdr);
799                 grown = 0;
800         }
801         /*
802          * Point to the biggest freespace in our data block.
803          */
804         dup = (xfs_dir2_data_unused_t *)
805               ((char *)hdr + be16_to_cpu(bf[0].offset));
806         needscan = needlog = 0;
807         /*
808          * Mark the initial part of our freespace in use for the new entry.
809          */
810         error = xfs_dir2_data_use_free(args, dbp, dup,
811                         (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
812                         length, &needlog, &needscan);
813         if (error) {
814                 xfs_trans_brelse(tp, lbp);
815                 return error;
816         }
817         /*
818          * Initialize our new entry (at last).
819          */
820         dep = (xfs_dir2_data_entry_t *)dup;
821         dep->inumber = cpu_to_be64(args->inumber);
822         dep->namelen = args->namelen;
823         memcpy(dep->name, args->name, dep->namelen);
824         dp->d_ops->data_put_ftype(dep, args->filetype);
825         tagp = dp->d_ops->data_entry_tag_p(dep);
826         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
827         /*
828          * Need to scan fix up the bestfree table.
829          */
830         if (needscan)
831                 xfs_dir2_data_freescan(dp, hdr, &needlog);
832         /*
833          * Need to log the data block's header.
834          */
835         if (needlog)
836                 xfs_dir2_data_log_header(args, dbp);
837         xfs_dir2_data_log_entry(args, dbp, dep);
838         /*
839          * If the bests table needs to be changed, do it.
840          * Log the change unless we've already done that.
841          */
842         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
843                 bestsp[use_block] = bf[0].length;
844                 if (!grown)
845                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
846         }
847
848         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
849                                        highstale, &lfloglow, &lfloghigh);
850
851         /*
852          * Fill in the new leaf entry.
853          */
854         lep->hashval = cpu_to_be32(args->hashval);
855         lep->address = cpu_to_be32(
856                                 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
857                                 be16_to_cpu(*tagp)));
858         /*
859          * Log the leaf fields and give up the buffers.
860          */
861         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
862         xfs_dir3_leaf_log_header(args, lbp);
863         xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
864         xfs_dir3_leaf_check(dp, lbp);
865         xfs_dir3_data_check(dp, dbp);
866         return 0;
867 }
868
869 /*
870  * Compact out any stale entries in the leaf.
871  * Log the header and changed leaf entries, if any.
872  */
873 void
874 xfs_dir3_leaf_compact(
875         xfs_da_args_t   *args,          /* operation arguments */
876         struct xfs_dir3_icleaf_hdr *leafhdr,
877         struct xfs_buf  *bp)            /* leaf buffer */
878 {
879         int             from;           /* source leaf index */
880         xfs_dir2_leaf_t *leaf;          /* leaf structure */
881         int             loglow;         /* first leaf entry to log */
882         int             to;             /* target leaf index */
883         struct xfs_dir2_leaf_entry *ents;
884         struct xfs_inode *dp = args->dp;
885
886         leaf = bp->b_addr;
887         if (!leafhdr->stale)
888                 return;
889
890         /*
891          * Compress out the stale entries in place.
892          */
893         ents = dp->d_ops->leaf_ents_p(leaf);
894         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
895                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
896                         continue;
897                 /*
898                  * Only actually copy the entries that are different.
899                  */
900                 if (from > to) {
901                         if (loglow == -1)
902                                 loglow = to;
903                         ents[to] = ents[from];
904                 }
905                 to++;
906         }
907         /*
908          * Update and log the header, log the leaf entries.
909          */
910         ASSERT(leafhdr->stale == from - to);
911         leafhdr->count -= leafhdr->stale;
912         leafhdr->stale = 0;
913
914         dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
915         xfs_dir3_leaf_log_header(args, bp);
916         if (loglow != -1)
917                 xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
918 }
919
920 /*
921  * Compact the leaf entries, removing stale ones.
922  * Leave one stale entry behind - the one closest to our
923  * insertion index - and the caller will shift that one to our insertion
924  * point later.
925  * Return new insertion index, where the remaining stale entry is,
926  * and leaf logging indices.
927  */
928 void
929 xfs_dir3_leaf_compact_x1(
930         struct xfs_dir3_icleaf_hdr *leafhdr,
931         struct xfs_dir2_leaf_entry *ents,
932         int             *indexp,        /* insertion index */
933         int             *lowstalep,     /* out: stale entry before us */
934         int             *highstalep,    /* out: stale entry after us */
935         int             *lowlogp,       /* out: low log index */
936         int             *highlogp)      /* out: high log index */
937 {
938         int             from;           /* source copy index */
939         int             highstale;      /* stale entry at/after index */
940         int             index;          /* insertion index */
941         int             keepstale;      /* source index of kept stale */
942         int             lowstale;       /* stale entry before index */
943         int             newindex=0;     /* new insertion index */
944         int             to;             /* destination copy index */
945
946         ASSERT(leafhdr->stale > 1);
947         index = *indexp;
948
949         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
950
951         /*
952          * Pick the better of lowstale and highstale.
953          */
954         if (lowstale >= 0 &&
955             (highstale == leafhdr->count ||
956              index - lowstale <= highstale - index))
957                 keepstale = lowstale;
958         else
959                 keepstale = highstale;
960         /*
961          * Copy the entries in place, removing all the stale entries
962          * except keepstale.
963          */
964         for (from = to = 0; from < leafhdr->count; from++) {
965                 /*
966                  * Notice the new value of index.
967                  */
968                 if (index == from)
969                         newindex = to;
970                 if (from != keepstale &&
971                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
972                         if (from == to)
973                                 *lowlogp = to;
974                         continue;
975                 }
976                 /*
977                  * Record the new keepstale value for the insertion.
978                  */
979                 if (from == keepstale)
980                         lowstale = highstale = to;
981                 /*
982                  * Copy only the entries that have moved.
983                  */
984                 if (from > to)
985                         ents[to] = ents[from];
986                 to++;
987         }
988         ASSERT(from > to);
989         /*
990          * If the insertion point was past the last entry,
991          * set the new insertion point accordingly.
992          */
993         if (index == from)
994                 newindex = to;
995         *indexp = newindex;
996         /*
997          * Adjust the leaf header values.
998          */
999         leafhdr->count -= from - to;
1000         leafhdr->stale = 1;
1001         /*
1002          * Remember the low/high stale value only in the "right"
1003          * direction.
1004          */
1005         if (lowstale >= newindex)
1006                 lowstale = -1;
1007         else
1008                 highstale = leafhdr->count;
1009         *highlogp = leafhdr->count - 1;
1010         *lowstalep = lowstale;
1011         *highstalep = highstale;
1012 }
1013
1014 /*
1015  * Log the bests entries indicated from a leaf1 block.
1016  */
1017 static void
1018 xfs_dir3_leaf_log_bests(
1019         struct xfs_da_args      *args,
1020         struct xfs_buf          *bp,            /* leaf buffer */
1021         int                     first,          /* first entry to log */
1022         int                     last)           /* last entry to log */
1023 {
1024         __be16                  *firstb;        /* pointer to first entry */
1025         __be16                  *lastb;         /* pointer to last entry */
1026         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1027         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1028
1029         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1030                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1031
1032         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1033         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1034         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1035         xfs_trans_log_buf(args->trans, bp,
1036                 (uint)((char *)firstb - (char *)leaf),
1037                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1038 }
1039
1040 /*
1041  * Log the leaf entries indicated from a leaf1 or leafn block.
1042  */
1043 void
1044 xfs_dir3_leaf_log_ents(
1045         struct xfs_da_args      *args,
1046         struct xfs_buf          *bp,
1047         int                     first,
1048         int                     last)
1049 {
1050         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1051         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1052         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1053         struct xfs_dir2_leaf_entry *ents;
1054
1055         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1056                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1057                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1058                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1059
1060         ents = args->dp->d_ops->leaf_ents_p(leaf);
1061         firstlep = &ents[first];
1062         lastlep = &ents[last];
1063         xfs_trans_log_buf(args->trans, bp,
1064                 (uint)((char *)firstlep - (char *)leaf),
1065                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1066 }
1067
1068 /*
1069  * Log the header of the leaf1 or leafn block.
1070  */
1071 void
1072 xfs_dir3_leaf_log_header(
1073         struct xfs_da_args      *args,
1074         struct xfs_buf          *bp)
1075 {
1076         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1077
1078         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1079                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1080                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1081                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1082
1083         xfs_trans_log_buf(args->trans, bp,
1084                           (uint)((char *)&leaf->hdr - (char *)leaf),
1085                           args->dp->d_ops->leaf_hdr_size - 1);
1086 }
1087
1088 /*
1089  * Log the tail of the leaf1 block.
1090  */
1091 STATIC void
1092 xfs_dir3_leaf_log_tail(
1093         struct xfs_da_args      *args,
1094         struct xfs_buf          *bp)
1095 {
1096         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1097         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1098
1099         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1100                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1101                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1102                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1103
1104         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1105         xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1106                 (uint)(args->geo->blksize - 1));
1107 }
1108
1109 /*
1110  * Look up the entry referred to by args in the leaf format directory.
1111  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1112  * is also used by the node-format code.
1113  */
1114 int
1115 xfs_dir2_leaf_lookup(
1116         xfs_da_args_t           *args)          /* operation arguments */
1117 {
1118         struct xfs_buf          *dbp;           /* data block buffer */
1119         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1120         xfs_inode_t             *dp;            /* incore directory inode */
1121         int                     error;          /* error return code */
1122         int                     index;          /* found entry index */
1123         struct xfs_buf          *lbp;           /* leaf buffer */
1124         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1125         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1126         xfs_trans_t             *tp;            /* transaction pointer */
1127         struct xfs_dir2_leaf_entry *ents;
1128
1129         trace_xfs_dir2_leaf_lookup(args);
1130
1131         /*
1132          * Look up name in the leaf block, returning both buffers and index.
1133          */
1134         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1135                 return error;
1136         }
1137         tp = args->trans;
1138         dp = args->dp;
1139         xfs_dir3_leaf_check(dp, lbp);
1140         leaf = lbp->b_addr;
1141         ents = dp->d_ops->leaf_ents_p(leaf);
1142         /*
1143          * Get to the leaf entry and contained data entry address.
1144          */
1145         lep = &ents[index];
1146
1147         /*
1148          * Point to the data entry.
1149          */
1150         dep = (xfs_dir2_data_entry_t *)
1151               ((char *)dbp->b_addr +
1152                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1153         /*
1154          * Return the found inode number & CI name if appropriate
1155          */
1156         args->inumber = be64_to_cpu(dep->inumber);
1157         args->filetype = dp->d_ops->data_get_ftype(dep);
1158         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1159         xfs_trans_brelse(tp, dbp);
1160         xfs_trans_brelse(tp, lbp);
1161         return error;
1162 }
1163
1164 /*
1165  * Look up name/hash in the leaf block.
1166  * Fill in indexp with the found index, and dbpp with the data buffer.
1167  * If not found dbpp will be NULL, and ENOENT comes back.
1168  * lbpp will always be filled in with the leaf buffer unless there's an error.
1169  */
1170 static int                                      /* error */
1171 xfs_dir2_leaf_lookup_int(
1172         xfs_da_args_t           *args,          /* operation arguments */
1173         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1174         int                     *indexp,        /* out: index in leaf block */
1175         struct xfs_buf          **dbpp)         /* out: data buffer */
1176 {
1177         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1178         struct xfs_buf          *dbp = NULL;    /* data buffer */
1179         xfs_dir2_data_entry_t   *dep;           /* data entry */
1180         xfs_inode_t             *dp;            /* incore directory inode */
1181         int                     error;          /* error return code */
1182         int                     index;          /* index in leaf block */
1183         struct xfs_buf          *lbp;           /* leaf buffer */
1184         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1185         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1186         xfs_mount_t             *mp;            /* filesystem mount point */
1187         xfs_dir2_db_t           newdb;          /* new data block number */
1188         xfs_trans_t             *tp;            /* transaction pointer */
1189         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1190         enum xfs_dacmp          cmp;            /* name compare result */
1191         struct xfs_dir2_leaf_entry *ents;
1192         struct xfs_dir3_icleaf_hdr leafhdr;
1193
1194         dp = args->dp;
1195         tp = args->trans;
1196         mp = dp->i_mount;
1197
1198         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1199         if (error)
1200                 return error;
1201
1202         *lbpp = lbp;
1203         leaf = lbp->b_addr;
1204         xfs_dir3_leaf_check(dp, lbp);
1205         ents = dp->d_ops->leaf_ents_p(leaf);
1206         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1207
1208         /*
1209          * Look for the first leaf entry with our hash value.
1210          */
1211         index = xfs_dir2_leaf_search_hash(args, lbp);
1212         /*
1213          * Loop over all the entries with the right hash value
1214          * looking to match the name.
1215          */
1216         for (lep = &ents[index];
1217              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1218              lep++, index++) {
1219                 /*
1220                  * Skip over stale leaf entries.
1221                  */
1222                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1223                         continue;
1224                 /*
1225                  * Get the new data block number.
1226                  */
1227                 newdb = xfs_dir2_dataptr_to_db(args->geo,
1228                                                be32_to_cpu(lep->address));
1229                 /*
1230                  * If it's not the same as the old data block number,
1231                  * need to pitch the old one and read the new one.
1232                  */
1233                 if (newdb != curdb) {
1234                         if (dbp)
1235                                 xfs_trans_brelse(tp, dbp);
1236                         error = xfs_dir3_data_read(tp, dp,
1237                                            xfs_dir2_db_to_da(args->geo, newdb),
1238                                            -1, &dbp);
1239                         if (error) {
1240                                 xfs_trans_brelse(tp, lbp);
1241                                 return error;
1242                         }
1243                         curdb = newdb;
1244                 }
1245                 /*
1246                  * Point to the data entry.
1247                  */
1248                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1249                         xfs_dir2_dataptr_to_off(args->geo,
1250                                                 be32_to_cpu(lep->address)));
1251                 /*
1252                  * Compare name and if it's an exact match, return the index
1253                  * and buffer. If it's the first case-insensitive match, store
1254                  * the index and buffer and continue looking for an exact match.
1255                  */
1256                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1257                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1258                         args->cmpresult = cmp;
1259                         *indexp = index;
1260                         /* case exact match: return the current buffer. */
1261                         if (cmp == XFS_CMP_EXACT) {
1262                                 *dbpp = dbp;
1263                                 return 0;
1264                         }
1265                         cidb = curdb;
1266                 }
1267         }
1268         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1269         /*
1270          * Here, we can only be doing a lookup (not a rename or remove).
1271          * If a case-insensitive match was found earlier, re-read the
1272          * appropriate data block if required and return it.
1273          */
1274         if (args->cmpresult == XFS_CMP_CASE) {
1275                 ASSERT(cidb != -1);
1276                 if (cidb != curdb) {
1277                         xfs_trans_brelse(tp, dbp);
1278                         error = xfs_dir3_data_read(tp, dp,
1279                                            xfs_dir2_db_to_da(args->geo, cidb),
1280                                            -1, &dbp);
1281                         if (error) {
1282                                 xfs_trans_brelse(tp, lbp);
1283                                 return error;
1284                         }
1285                 }
1286                 *dbpp = dbp;
1287                 return 0;
1288         }
1289         /*
1290          * No match found, return -ENOENT.
1291          */
1292         ASSERT(cidb == -1);
1293         if (dbp)
1294                 xfs_trans_brelse(tp, dbp);
1295         xfs_trans_brelse(tp, lbp);
1296         return -ENOENT;
1297 }
1298
1299 /*
1300  * Remove an entry from a leaf format directory.
1301  */
1302 int                                             /* error */
1303 xfs_dir2_leaf_removename(
1304         xfs_da_args_t           *args)          /* operation arguments */
1305 {
1306         __be16                  *bestsp;        /* leaf block best freespace */
1307         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1308         xfs_dir2_db_t           db;             /* data block number */
1309         struct xfs_buf          *dbp;           /* data block buffer */
1310         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1311         xfs_inode_t             *dp;            /* incore directory inode */
1312         int                     error;          /* error return code */
1313         xfs_dir2_db_t           i;              /* temporary data block # */
1314         int                     index;          /* index into leaf entries */
1315         struct xfs_buf          *lbp;           /* leaf buffer */
1316         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1317         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1318         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1319         int                     needlog;        /* need to log data header */
1320         int                     needscan;       /* need to rescan data frees */
1321         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1322         struct xfs_dir2_data_free *bf;          /* bestfree table */
1323         struct xfs_dir2_leaf_entry *ents;
1324         struct xfs_dir3_icleaf_hdr leafhdr;
1325
1326         trace_xfs_dir2_leaf_removename(args);
1327
1328         /*
1329          * Lookup the leaf entry, get the leaf and data blocks read in.
1330          */
1331         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1332                 return error;
1333         }
1334         dp = args->dp;
1335         leaf = lbp->b_addr;
1336         hdr = dbp->b_addr;
1337         xfs_dir3_data_check(dp, dbp);
1338         bf = dp->d_ops->data_bestfree_p(hdr);
1339         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1340         ents = dp->d_ops->leaf_ents_p(leaf);
1341         /*
1342          * Point to the leaf entry, use that to point to the data entry.
1343          */
1344         lep = &ents[index];
1345         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1346         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1347                 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1348         needscan = needlog = 0;
1349         oldbest = be16_to_cpu(bf[0].length);
1350         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1351         bestsp = xfs_dir2_leaf_bests_p(ltp);
1352         if (be16_to_cpu(bestsp[db]) != oldbest)
1353                 return -EFSCORRUPTED;
1354         /*
1355          * Mark the former data entry unused.
1356          */
1357         xfs_dir2_data_make_free(args, dbp,
1358                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1359                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1360         /*
1361          * We just mark the leaf entry stale by putting a null in it.
1362          */
1363         leafhdr.stale++;
1364         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1365         xfs_dir3_leaf_log_header(args, lbp);
1366
1367         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1368         xfs_dir3_leaf_log_ents(args, lbp, index, index);
1369
1370         /*
1371          * Scan the freespace in the data block again if necessary,
1372          * log the data block header if necessary.
1373          */
1374         if (needscan)
1375                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1376         if (needlog)
1377                 xfs_dir2_data_log_header(args, dbp);
1378         /*
1379          * If the longest freespace in the data block has changed,
1380          * put the new value in the bests table and log that.
1381          */
1382         if (be16_to_cpu(bf[0].length) != oldbest) {
1383                 bestsp[db] = bf[0].length;
1384                 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1385         }
1386         xfs_dir3_data_check(dp, dbp);
1387         /*
1388          * If the data block is now empty then get rid of the data block.
1389          */
1390         if (be16_to_cpu(bf[0].length) ==
1391                         args->geo->blksize - dp->d_ops->data_entry_offset) {
1392                 ASSERT(db != args->geo->datablk);
1393                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1394                         /*
1395                          * Nope, can't get rid of it because it caused
1396                          * allocation of a bmap btree block to do so.
1397                          * Just go on, returning success, leaving the
1398                          * empty block in place.
1399                          */
1400                         if (error == -ENOSPC && args->total == 0)
1401                                 error = 0;
1402                         xfs_dir3_leaf_check(dp, lbp);
1403                         return error;
1404                 }
1405                 dbp = NULL;
1406                 /*
1407                  * If this is the last data block then compact the
1408                  * bests table by getting rid of entries.
1409                  */
1410                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1411                         /*
1412                          * Look for the last active entry (i).
1413                          */
1414                         for (i = db - 1; i > 0; i--) {
1415                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1416                                         break;
1417                         }
1418                         /*
1419                          * Copy the table down so inactive entries at the
1420                          * end are removed.
1421                          */
1422                         memmove(&bestsp[db - i], bestsp,
1423                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1424                         be32_add_cpu(&ltp->bestcount, -(db - i));
1425                         xfs_dir3_leaf_log_tail(args, lbp);
1426                         xfs_dir3_leaf_log_bests(args, lbp, 0,
1427                                                 be32_to_cpu(ltp->bestcount) - 1);
1428                 } else
1429                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1430         }
1431         /*
1432          * If the data block was not the first one, drop it.
1433          */
1434         else if (db != args->geo->datablk)
1435                 dbp = NULL;
1436
1437         xfs_dir3_leaf_check(dp, lbp);
1438         /*
1439          * See if we can convert to block form.
1440          */
1441         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1442 }
1443
1444 /*
1445  * Replace the inode number in a leaf format directory entry.
1446  */
1447 int                                             /* error */
1448 xfs_dir2_leaf_replace(
1449         xfs_da_args_t           *args)          /* operation arguments */
1450 {
1451         struct xfs_buf          *dbp;           /* data block buffer */
1452         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1453         xfs_inode_t             *dp;            /* incore directory inode */
1454         int                     error;          /* error return code */
1455         int                     index;          /* index of leaf entry */
1456         struct xfs_buf          *lbp;           /* leaf buffer */
1457         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1458         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1459         xfs_trans_t             *tp;            /* transaction pointer */
1460         struct xfs_dir2_leaf_entry *ents;
1461
1462         trace_xfs_dir2_leaf_replace(args);
1463
1464         /*
1465          * Look up the entry.
1466          */
1467         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1468                 return error;
1469         }
1470         dp = args->dp;
1471         leaf = lbp->b_addr;
1472         ents = dp->d_ops->leaf_ents_p(leaf);
1473         /*
1474          * Point to the leaf entry, get data address from it.
1475          */
1476         lep = &ents[index];
1477         /*
1478          * Point to the data entry.
1479          */
1480         dep = (xfs_dir2_data_entry_t *)
1481               ((char *)dbp->b_addr +
1482                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1483         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1484         /*
1485          * Put the new inode number in, log it.
1486          */
1487         dep->inumber = cpu_to_be64(args->inumber);
1488         dp->d_ops->data_put_ftype(dep, args->filetype);
1489         tp = args->trans;
1490         xfs_dir2_data_log_entry(args, dbp, dep);
1491         xfs_dir3_leaf_check(dp, lbp);
1492         xfs_trans_brelse(tp, lbp);
1493         return 0;
1494 }
1495
1496 /*
1497  * Return index in the leaf block (lbp) which is either the first
1498  * one with this hash value, or if there are none, the insert point
1499  * for that hash value.
1500  */
1501 int                                             /* index value */
1502 xfs_dir2_leaf_search_hash(
1503         xfs_da_args_t           *args,          /* operation arguments */
1504         struct xfs_buf          *lbp)           /* leaf buffer */
1505 {
1506         xfs_dahash_t            hash=0;         /* hash from this entry */
1507         xfs_dahash_t            hashwant;       /* hash value looking for */
1508         int                     high;           /* high leaf index */
1509         int                     low;            /* low leaf index */
1510         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1511         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1512         int                     mid=0;          /* current leaf index */
1513         struct xfs_dir2_leaf_entry *ents;
1514         struct xfs_dir3_icleaf_hdr leafhdr;
1515
1516         leaf = lbp->b_addr;
1517         ents = args->dp->d_ops->leaf_ents_p(leaf);
1518         args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1519
1520         /*
1521          * Note, the table cannot be empty, so we have to go through the loop.
1522          * Binary search the leaf entries looking for our hash value.
1523          */
1524         for (lep = ents, low = 0, high = leafhdr.count - 1,
1525                 hashwant = args->hashval;
1526              low <= high; ) {
1527                 mid = (low + high) >> 1;
1528                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1529                         break;
1530                 if (hash < hashwant)
1531                         low = mid + 1;
1532                 else
1533                         high = mid - 1;
1534         }
1535         /*
1536          * Found one, back up through all the equal hash values.
1537          */
1538         if (hash == hashwant) {
1539                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1540                         mid--;
1541                 }
1542         }
1543         /*
1544          * Need to point to an entry higher than ours.
1545          */
1546         else if (hash < hashwant)
1547                 mid++;
1548         return mid;
1549 }
1550
1551 /*
1552  * Trim off a trailing data block.  We know it's empty since the leaf
1553  * freespace table says so.
1554  */
1555 int                                             /* error */
1556 xfs_dir2_leaf_trim_data(
1557         xfs_da_args_t           *args,          /* operation arguments */
1558         struct xfs_buf          *lbp,           /* leaf buffer */
1559         xfs_dir2_db_t           db)             /* data block number */
1560 {
1561         __be16                  *bestsp;        /* leaf bests table */
1562         struct xfs_buf          *dbp;           /* data block buffer */
1563         xfs_inode_t             *dp;            /* incore directory inode */
1564         int                     error;          /* error return value */
1565         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1566         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1567         xfs_trans_t             *tp;            /* transaction pointer */
1568
1569         dp = args->dp;
1570         tp = args->trans;
1571         /*
1572          * Read the offending data block.  We need its buffer.
1573          */
1574         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1575                                    -1, &dbp);
1576         if (error)
1577                 return error;
1578
1579         leaf = lbp->b_addr;
1580         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1581
1582 #ifdef DEBUG
1583 {
1584         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1585         struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1586
1587         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1588                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1589         ASSERT(be16_to_cpu(bf[0].length) ==
1590                args->geo->blksize - dp->d_ops->data_entry_offset);
1591         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1592 }
1593 #endif
1594
1595         /*
1596          * Get rid of the data block.
1597          */
1598         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1599                 ASSERT(error != -ENOSPC);
1600                 xfs_trans_brelse(tp, dbp);
1601                 return error;
1602         }
1603         /*
1604          * Eliminate the last bests entry from the table.
1605          */
1606         bestsp = xfs_dir2_leaf_bests_p(ltp);
1607         be32_add_cpu(&ltp->bestcount, -1);
1608         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1609         xfs_dir3_leaf_log_tail(args, lbp);
1610         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1611         return 0;
1612 }
1613
1614 static inline size_t
1615 xfs_dir3_leaf_size(
1616         struct xfs_dir3_icleaf_hdr      *hdr,
1617         int                             counts)
1618 {
1619         int     entries;
1620         int     hdrsize;
1621
1622         entries = hdr->count - hdr->stale;
1623         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1624             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1625                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1626         else
1627                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1628
1629         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1630                        + counts * sizeof(xfs_dir2_data_off_t)
1631                        + sizeof(xfs_dir2_leaf_tail_t);
1632 }
1633
1634 /*
1635  * Convert node form directory to leaf form directory.
1636  * The root of the node form dir needs to already be a LEAFN block.
1637  * Just return if we can't do anything.
1638  */
1639 int                                             /* error */
1640 xfs_dir2_node_to_leaf(
1641         xfs_da_state_t          *state)         /* directory operation state */
1642 {
1643         xfs_da_args_t           *args;          /* operation arguments */
1644         xfs_inode_t             *dp;            /* incore directory inode */
1645         int                     error;          /* error return code */
1646         struct xfs_buf          *fbp;           /* buffer for freespace block */
1647         xfs_fileoff_t           fo;             /* freespace file offset */
1648         xfs_dir2_free_t         *free;          /* freespace structure */
1649         struct xfs_buf          *lbp;           /* buffer for leaf block */
1650         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1651         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1652         xfs_mount_t             *mp;            /* filesystem mount point */
1653         int                     rval;           /* successful free trim? */
1654         xfs_trans_t             *tp;            /* transaction pointer */
1655         struct xfs_dir3_icleaf_hdr leafhdr;
1656         struct xfs_dir3_icfree_hdr freehdr;
1657
1658         /*
1659          * There's more than a leaf level in the btree, so there must
1660          * be multiple leafn blocks.  Give up.
1661          */
1662         if (state->path.active > 1)
1663                 return 0;
1664         args = state->args;
1665
1666         trace_xfs_dir2_node_to_leaf(args);
1667
1668         mp = state->mp;
1669         dp = args->dp;
1670         tp = args->trans;
1671         /*
1672          * Get the last offset in the file.
1673          */
1674         if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1675                 return error;
1676         }
1677         fo -= args->geo->fsbcount;
1678         /*
1679          * If there are freespace blocks other than the first one,
1680          * take this opportunity to remove trailing empty freespace blocks
1681          * that may have been left behind during no-space-reservation
1682          * operations.
1683          */
1684         while (fo > args->geo->freeblk) {
1685                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1686                         return error;
1687                 }
1688                 if (rval)
1689                         fo -= args->geo->fsbcount;
1690                 else
1691                         return 0;
1692         }
1693         /*
1694          * Now find the block just before the freespace block.
1695          */
1696         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1697                 return error;
1698         }
1699         /*
1700          * If it's not the single leaf block, give up.
1701          */
1702         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1703                 return 0;
1704         lbp = state->path.blk[0].bp;
1705         leaf = lbp->b_addr;
1706         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1707
1708         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1709                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1710
1711         /*
1712          * Read the freespace block.
1713          */
1714         error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1715         if (error)
1716                 return error;
1717         free = fbp->b_addr;
1718         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1719
1720         ASSERT(!freehdr.firstdb);
1721
1722         /*
1723          * Now see if the leafn and free data will fit in a leaf1.
1724          * If not, release the buffer and give up.
1725          */
1726         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1727                 xfs_trans_brelse(tp, fbp);
1728                 return 0;
1729         }
1730
1731         /*
1732          * If the leaf has any stale entries in it, compress them out.
1733          */
1734         if (leafhdr.stale)
1735                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1736
1737         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1738         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1739         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1740                                         ? XFS_DIR2_LEAF1_MAGIC
1741                                         : XFS_DIR3_LEAF1_MAGIC;
1742
1743         /*
1744          * Set up the leaf tail from the freespace block.
1745          */
1746         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1747         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1748
1749         /*
1750          * Set up the leaf bests table.
1751          */
1752         memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1753                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1754
1755         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1756         xfs_dir3_leaf_log_header(args, lbp);
1757         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1758         xfs_dir3_leaf_log_tail(args, lbp);
1759         xfs_dir3_leaf_check(dp, lbp);
1760
1761         /*
1762          * Get rid of the freespace block.
1763          */
1764         error = xfs_dir2_shrink_inode(args,
1765                         xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1766                         fbp);
1767         if (error) {
1768                 /*
1769                  * This can't fail here because it can only happen when
1770                  * punching out the middle of an extent, and this is an
1771                  * isolated block.
1772                  */
1773                 ASSERT(error != -ENOSPC);
1774                 return error;
1775         }
1776         fbp = NULL;
1777         /*
1778          * Now see if we can convert the single-leaf directory
1779          * down to a block form directory.
1780          * This routine always kills the dabuf for the leaf, so
1781          * eliminate it from the path.
1782          */
1783         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1784         state->path.blk[0].bp = NULL;
1785         return error;
1786 }