]> err.no Git - linux-2.6/blob - fs/gfs2/dir.c
[GFS2] Finally get ref counting correct
[linux-2.6] / fs / gfs2 / dir.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License v.2.
8  */
9
10 /*
11 * Implements Extendible Hashing as described in:
12 *   "Extendible Hashing" by Fagin, et al in
13 *     __ACM Trans. on Database Systems__, Sept 1979.
14 *
15 *
16 * Here's the layout of dirents which is essentially the same as that of ext2
17 * within a single block. The field de_name_len is the number of bytes
18 * actually required for the name (no null terminator). The field de_rec_len
19 * is the number of bytes allocated to the dirent. The offset of the next
20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
21 * deleted, the preceding dirent inherits its allocated space, ie
22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
23 * by adding de_rec_len to the current dirent, this essentially causes the
24 * deleted dirent to get jumped over when iterating through all the dirents.
25 *
26 * When deleting the first dirent in a block, there is no previous dirent so
27 * the field de_ino is set to zero to designate it as deleted. When allocating
28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
30 * dirent is allocated. Otherwise it must go through all the 'used' dirents
31 * searching for one in which the amount of total space minus the amount of
32 * used space will provide enough space for the new dirent.
33 *
34 * There are two types of blocks in which dirents reside. In a stuffed dinode,
35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
36 * the block.  In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
37 * beginning of the leaf block. The dirents reside in leaves when
38 *
39 * dip->i_di.di_flags & GFS2_DIF_EXHASH is true
40 *
41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
42 *
43 * When the dirents are in leaves, the actual contents of the directory file are
44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
45 * dirents are NOT in the directory file itself. There can be more than one block
46 * pointer in the array that points to the same leaf. In fact, when a directory
47 * is first converted from linear to exhash, all of the pointers point to the
48 * same leaf.
49 *
50 * When a leaf is completely full, the size of the hash table can be
51 * doubled unless it is already at the maximum size which is hard coded into
52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
53 * but never before the maximum hash table size has been reached.
54 */
55
56 #include <linux/sched.h>
57 #include <linux/slab.h>
58 #include <linux/spinlock.h>
59 #include <linux/completion.h>
60 #include <linux/buffer_head.h>
61 #include <linux/sort.h>
62 #include <linux/gfs2_ondisk.h>
63 #include <linux/crc32.h>
64 #include <asm/semaphore.h>
65
66 #include "gfs2.h"
67 #include "lm_interface.h"
68 #include "incore.h"
69 #include "dir.h"
70 #include "glock.h"
71 #include "inode.h"
72 #include "meta_io.h"
73 #include "quota.h"
74 #include "rgrp.h"
75 #include "trans.h"
76 #include "bmap.h"
77 #include "util.h"
78
79 #define IS_LEAF     1 /* Hashed (leaf) directory */
80 #define IS_DINODE   2 /* Linear (stuffed dinode block) directory */
81
82 #if 1
83 #define gfs2_disk_hash2offset(h) (((uint64_t)(h)) >> 1)
84 #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)) << 1))
85 #else
86 #define gfs2_disk_hash2offset(h) (((uint64_t)(h)))
87 #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p))))
88 #endif
89
90 typedef int (*leaf_call_t) (struct gfs2_inode *dip,
91                             uint32_t index, uint32_t len, uint64_t leaf_no,
92                             void *data);
93
94 int gfs2_dir_get_buffer(struct gfs2_inode *ip, uint64_t block, int new,
95                          struct buffer_head **bhp)
96 {
97         struct buffer_head *bh;
98         int error = 0;
99
100         if (new) {
101                 bh = gfs2_meta_new(ip->i_gl, block);
102                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
103                 gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
104                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
105         } else {
106                 error = gfs2_meta_read(ip->i_gl, block, DIO_START | DIO_WAIT,
107                                        &bh);
108                 if (error)
109                         return error;
110                 if (gfs2_metatype_check(ip->i_sbd, bh, GFS2_METATYPE_JD)) {
111                         brelse(bh);
112                         return -EIO;
113                 }
114         }
115
116         *bhp = bh;
117         return 0;
118 }
119
120
121
122 static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
123                                   unsigned int offset, unsigned int size)
124                                
125 {
126         struct buffer_head *dibh;
127         int error;
128
129         error = gfs2_meta_inode_buffer(ip, &dibh);
130         if (error)
131                 return error;
132
133         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
134         memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
135         if (ip->i_di.di_size < offset + size)
136                 ip->i_di.di_size = offset + size;
137         ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
138         gfs2_dinode_out(&ip->i_di, dibh->b_data);
139
140         brelse(dibh);
141
142         return size;
143 }
144
145
146
147 /**
148  * gfs2_dir_write_data - Write directory information to the inode
149  * @ip: The GFS2 inode
150  * @buf: The buffer containing information to be written
151  * @offset: The file offset to start writing at
152  * @size: The amount of data to write
153  *
154  * Returns: The number of bytes correctly written or error code
155  */
156 static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
157                                uint64_t offset, unsigned int size)
158 {
159         struct gfs2_sbd *sdp = ip->i_sbd;
160         struct buffer_head *dibh;
161         uint64_t lblock, dblock;
162         uint32_t extlen = 0;
163         unsigned int o;
164         int copied = 0;
165         int error = 0;
166
167         if (!size)
168                 return 0;
169
170         if (gfs2_is_stuffed(ip) &&
171             offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
172                 return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
173                                               size);
174
175         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
176                 return -EINVAL;
177
178         if (gfs2_is_stuffed(ip)) {
179                 error = gfs2_unstuff_dinode(ip, NULL, NULL);
180                 if (error)
181                         return error;
182         }
183
184         lblock = offset;
185         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
186
187         while (copied < size) {
188                 unsigned int amount;
189                 struct buffer_head *bh;
190                 int new;
191
192                 amount = size - copied;
193                 if (amount > sdp->sd_sb.sb_bsize - o)
194                         amount = sdp->sd_sb.sb_bsize - o;
195
196                 if (!extlen) {
197                         new = 1;
198                         error = gfs2_block_map(ip, lblock, &new, &dblock,
199                                                &extlen);
200                         if (error)
201                                 goto fail;
202                         error = -EIO;
203                         if (gfs2_assert_withdraw(sdp, dblock))
204                                 goto fail;
205                 }
206
207                 error = gfs2_dir_get_buffer(ip, dblock,
208                                             (amount == sdp->sd_jbsize) ?
209                                             1 : new, &bh);
210                 if (error)
211                         goto fail;
212
213                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
214                 memcpy(bh->b_data + o, buf, amount);
215                 brelse(bh);
216                 if (error)
217                         goto fail;
218
219                 copied += amount;
220                 lblock++;
221                 dblock++;
222                 extlen--;
223
224                 o = sizeof(struct gfs2_meta_header);
225         }
226
227 out:
228         error = gfs2_meta_inode_buffer(ip, &dibh);
229         if (error)
230                 return error;
231
232         if (ip->i_di.di_size < offset + copied)
233                 ip->i_di.di_size = offset + copied;
234         ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
235
236         gfs2_trans_add_bh(ip->i_gl, dibh, 1);
237         gfs2_dinode_out(&ip->i_di, dibh->b_data);
238         brelse(dibh);
239
240         return copied;
241 fail:
242         if (copied)
243                 goto out;
244         return error;
245 }
246
247 static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf,
248                                  unsigned int offset, unsigned int size)
249 {
250         struct buffer_head *dibh;
251         int error;
252
253         error = gfs2_meta_inode_buffer(ip, &dibh);
254         if (!error) {
255                 offset += sizeof(struct gfs2_dinode);
256                 memcpy(buf, dibh->b_data + offset, size);
257                 brelse(dibh);
258         }
259
260         return (error) ? error : size;
261 }
262
263
264 /**
265  * gfs2_dir_read_data - Read a data from a directory inode
266  * @ip: The GFS2 Inode
267  * @buf: The buffer to place result into
268  * @offset: File offset to begin jdata_readng from
269  * @size: Amount of data to transfer
270  *
271  * Returns: The amount of data actually copied or the error
272  */
273 static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf,
274                               uint64_t offset, unsigned int size)
275 {
276         struct gfs2_sbd *sdp = ip->i_sbd;
277         uint64_t lblock, dblock;
278         uint32_t extlen = 0;
279         unsigned int o;
280         int copied = 0;
281         int error = 0;
282
283         if (offset >= ip->i_di.di_size)
284                 return 0;
285
286         if ((offset + size) > ip->i_di.di_size)
287                 size = ip->i_di.di_size - offset;
288
289         if (!size)
290                 return 0;
291
292         if (gfs2_is_stuffed(ip))
293                 return gfs2_dir_read_stuffed(ip, buf, (unsigned int)offset,
294                                              size);
295
296         if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
297                 return -EINVAL;
298
299         lblock = offset;
300         o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
301
302         while (copied < size) {
303                 unsigned int amount;
304                 struct buffer_head *bh;
305                 int new;
306
307                 amount = size - copied;
308                 if (amount > sdp->sd_sb.sb_bsize - o)
309                         amount = sdp->sd_sb.sb_bsize - o;
310
311                 if (!extlen) {
312                         new = 0;
313                         error = gfs2_block_map(ip, lblock, &new, &dblock,
314                                                &extlen);
315                         if (error)
316                                 goto fail;
317                 }
318
319                 if (extlen > 1)
320                         gfs2_meta_ra(ip->i_gl, dblock, extlen);
321
322                 if (dblock) {
323                         error = gfs2_dir_get_buffer(ip, dblock, new, &bh);
324                         if (error)
325                                 goto fail;
326                         dblock++;
327                         extlen--;
328                 } else
329                         bh = NULL;
330
331                 memcpy(buf, bh->b_data + o, amount);
332                 brelse(bh);
333                 if (error)
334                         goto fail;
335
336                 copied += amount;
337                 lblock++;
338
339                 o = sizeof(struct gfs2_meta_header);
340         }
341
342         return copied;
343 fail:
344         return (copied) ? copied : error;
345 }
346
347 typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
348                             const struct qstr *name,
349                             void *opaque);
350
351 static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
352                                      const struct qstr *name, int ret)
353 {
354         if (dent->de_inum.no_addr != 0 &&
355             be32_to_cpu(dent->de_hash) == name->hash &&
356             be16_to_cpu(dent->de_name_len) == name->len &&
357             memcmp((char *)(dent+1), name->name, name->len) == 0)
358                 return ret;
359         return 0;
360 }
361
362 static int gfs2_dirent_find(const struct gfs2_dirent *dent,
363                             const struct qstr *name,
364                             void *opaque)
365 {
366         return __gfs2_dirent_find(dent, name, 1);
367 }
368
369 static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
370                             const struct qstr *name,
371                             void *opaque)
372 {
373         return __gfs2_dirent_find(dent, name, 2);
374 }
375
376 /*
377  * name->name holds ptr to start of block.
378  * name->len holds size of block.
379  */
380 static int gfs2_dirent_last(const struct gfs2_dirent *dent,
381                             const struct qstr *name,
382                             void *opaque)
383 {
384         const char *start = name->name;
385         const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
386         if (name->len == (end - start))
387                 return 1;
388         return 0;
389 }
390
391 static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
392                                   const struct qstr *name,
393                                   void *opaque)
394 {
395         unsigned required = GFS2_DIRENT_SIZE(name->len);
396         unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
397         unsigned totlen = be16_to_cpu(dent->de_rec_len);
398
399         if (!dent->de_inum.no_addr)
400                 actual = GFS2_DIRENT_SIZE(0);
401         if ((totlen - actual) >= required)
402                 return 1;
403         return 0;
404 }
405
406 struct dirent_gather {
407         const struct gfs2_dirent **pdent;
408         unsigned offset;
409 };
410
411 static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
412                               const struct qstr *name,
413                               void *opaque)
414 {
415         struct dirent_gather *g = opaque;
416         if (dent->de_inum.no_addr) {
417                 g->pdent[g->offset++] = dent;
418         }
419         return 0;
420 }
421
422 /*
423  * Other possible things to check:
424  * - Inode located within filesystem size (and on valid block)
425  * - Valid directory entry type
426  * Not sure how heavy-weight we want to make this... could also check
427  * hash is correct for example, but that would take a lot of extra time.
428  * For now the most important thing is to check that the various sizes
429  * are correct.
430  */
431 static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset,
432                              unsigned int size, unsigned int len, int first)
433 {
434         const char *msg = "gfs2_dirent too small";
435         if (unlikely(size < sizeof(struct gfs2_dirent)))
436                 goto error;
437         msg = "gfs2_dirent misaligned";
438         if (unlikely(offset & 0x7))
439                 goto error;
440         msg = "gfs2_dirent points beyond end of block";
441         if (unlikely(offset + size > len))
442                 goto error;
443         msg = "zero inode number";
444         if (unlikely(!first && !dent->de_inum.no_addr))
445                 goto error;
446         msg = "name length is greater than space in dirent";
447         if (dent->de_inum.no_addr &&
448             unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
449                      size))
450                 goto error;
451         return 0;
452 error:
453         printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg,
454                first ? "first in block" : "not first in block");
455         return -EIO;
456 }
457
458 static int gfs2_dirent_offset(const void *buf)
459 {
460         const struct gfs2_meta_header *h = buf;
461         int offset;
462
463         BUG_ON(buf == NULL);
464
465         switch(be32_to_cpu(h->mh_type)) {
466         case GFS2_METATYPE_LF:
467                 offset = sizeof(struct gfs2_leaf);
468                 break;
469         case GFS2_METATYPE_DI:
470                 offset = sizeof(struct gfs2_dinode);
471                 break;
472         default:
473                 goto wrong_type;
474         }
475         return offset;
476 wrong_type:
477         printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n",
478                be32_to_cpu(h->mh_type));
479         return -1;
480 }
481
482 static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode,
483                                             void *buf,
484                                             unsigned int len, gfs2_dscan_t scan,
485                                             const struct qstr *name,
486                                             void *opaque)
487 {
488         struct gfs2_dirent *dent, *prev;
489         unsigned offset;
490         unsigned size;
491         int ret = 0;
492
493         ret = gfs2_dirent_offset(buf);
494         if (ret < 0)
495                 goto consist_inode;
496
497         offset = ret;
498         prev = NULL;
499         dent = (struct gfs2_dirent *)(buf + offset);
500         size = be16_to_cpu(dent->de_rec_len);
501         if (gfs2_check_dirent(dent, offset, size, len, 1))
502                 goto consist_inode;
503         do {
504                 ret = scan(dent, name, opaque);
505                 if (ret)
506                         break;
507                 offset += size;
508                 if (offset == len)
509                         break;
510                 prev = dent;
511                 dent = (struct gfs2_dirent *)(buf + offset);
512                 size = be16_to_cpu(dent->de_rec_len);
513                 if (gfs2_check_dirent(dent, offset, size, len, 0))
514                         goto consist_inode;
515         } while(1);
516
517         switch(ret) {
518         case 0:
519                 return NULL;
520         case 1:
521                 return dent;
522         case 2:
523                 return prev ? prev : dent;
524         default:
525                 BUG_ON(ret > 0);
526                 return ERR_PTR(ret);
527         }
528
529 consist_inode:
530         gfs2_consist_inode(inode->u.generic_ip);
531         return ERR_PTR(-EIO);
532 }
533
534
535 /**
536  * dirent_first - Return the first dirent
537  * @dip: the directory
538  * @bh: The buffer
539  * @dent: Pointer to list of dirents
540  *
541  * return first dirent whether bh points to leaf or stuffed dinode
542  *
543  * Returns: IS_LEAF, IS_DINODE, or -errno
544  */
545
546 static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh,
547                         struct gfs2_dirent **dent)
548 {
549         struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data;
550
551         if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) {
552                 if (gfs2_meta_check(dip->i_sbd, bh))
553                         return -EIO;
554                 *dent = (struct gfs2_dirent *)(bh->b_data +
555                                                sizeof(struct gfs2_leaf));
556                 return IS_LEAF;
557         } else {
558                 if (gfs2_metatype_check(dip->i_sbd, bh, GFS2_METATYPE_DI))
559                         return -EIO;
560                 *dent = (struct gfs2_dirent *)(bh->b_data +
561                                                sizeof(struct gfs2_dinode));
562                 return IS_DINODE;
563         }
564 }
565
566 /**
567  * dirent_next - Next dirent
568  * @dip: the directory
569  * @bh: The buffer
570  * @dent: Pointer to list of dirents
571  *
572  * Returns: 0 on success, error code otherwise
573  */
574
575 static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
576                        struct gfs2_dirent **dent)
577 {
578         struct gfs2_dirent *tmp, *cur;
579         char *bh_end;
580         uint16_t cur_rec_len;
581
582         cur = *dent;
583         bh_end = bh->b_data + bh->b_size;
584         cur_rec_len = be16_to_cpu(cur->de_rec_len);
585
586         if ((char *)cur + cur_rec_len >= bh_end) {
587                 if ((char *)cur + cur_rec_len > bh_end) {
588                         gfs2_consist_inode(dip);
589                         return -EIO;
590                 }
591                 return -ENOENT;
592         }
593
594         tmp = (struct gfs2_dirent *)((char *)cur + cur_rec_len);
595
596         if ((char *)tmp + be16_to_cpu(tmp->de_rec_len) > bh_end) {
597                 gfs2_consist_inode(dip);
598                 return -EIO;
599         }
600
601         if (cur_rec_len == 0) {
602                 gfs2_consist_inode(dip);
603                 return -EIO;
604         }
605
606         /* Only the first dent could ever have de_inum.no_addr == 0 */
607         if (!tmp->de_inum.no_addr) {
608                 gfs2_consist_inode(dip);
609                 return -EIO;
610         }
611
612         *dent = tmp;
613
614         return 0;
615 }
616
617 /**
618  * dirent_del - Delete a dirent
619  * @dip: The GFS2 inode
620  * @bh: The buffer
621  * @prev: The previous dirent
622  * @cur: The current dirent
623  *
624  */
625
626 static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
627                        struct gfs2_dirent *prev, struct gfs2_dirent *cur)
628 {
629         uint16_t cur_rec_len, prev_rec_len;
630
631         if (!cur->de_inum.no_addr) {
632                 gfs2_consist_inode(dip);
633                 return;
634         }
635
636         gfs2_trans_add_bh(dip->i_gl, bh, 1);
637
638         /* If there is no prev entry, this is the first entry in the block.
639            The de_rec_len is already as big as it needs to be.  Just zero
640            out the inode number and return.  */
641
642         if (!prev) {
643                 cur->de_inum.no_addr = 0;       /* No endianess worries */
644                 return;
645         }
646
647         /*  Combine this dentry with the previous one.  */
648
649         prev_rec_len = be16_to_cpu(prev->de_rec_len);
650         cur_rec_len = be16_to_cpu(cur->de_rec_len);
651
652         if ((char *)prev + prev_rec_len != (char *)cur)
653                 gfs2_consist_inode(dip);
654         if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
655                 gfs2_consist_inode(dip);
656
657         prev_rec_len += cur_rec_len;
658         prev->de_rec_len = cpu_to_be16(prev_rec_len);
659 }
660
661 /*
662  * Takes a dent from which to grab space as an argument. Returns the
663  * newly created dent.
664  */
665 struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
666                                      struct gfs2_dirent *dent,
667                                      const struct qstr *name,
668                                      struct buffer_head *bh)
669 {
670         struct gfs2_inode *ip = inode->u.generic_ip;
671         struct gfs2_dirent *ndent;
672         unsigned offset = 0, totlen;
673
674         if (dent->de_inum.no_addr)
675                 offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
676         totlen = be16_to_cpu(dent->de_rec_len);
677         BUG_ON(offset + name->len > totlen);
678         gfs2_trans_add_bh(ip->i_gl, bh, 1);
679         ndent = (struct gfs2_dirent *)((char *)dent + offset);
680         dent->de_rec_len = cpu_to_be16(offset);
681         gfs2_qstr2dirent(name, totlen - offset, ndent);
682         return ndent;
683 }
684
685 static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode,
686                                              struct buffer_head *bh,
687                                              const struct qstr *name)
688 {
689         struct gfs2_dirent *dent;
690         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, 
691                                 gfs2_dirent_find_space, name, NULL);
692         if (!dent || IS_ERR(dent))
693                 return dent;
694         return gfs2_init_dirent(inode, dent, name, bh);
695 }
696
697 static int get_leaf(struct gfs2_inode *dip, uint64_t leaf_no,
698                     struct buffer_head **bhp)
699 {
700         int error;
701
702         error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_START | DIO_WAIT, bhp);
703         if (!error && gfs2_metatype_check(dip->i_sbd, *bhp, GFS2_METATYPE_LF))
704                 error = -EIO;
705
706         return error;
707 }
708
709 /**
710  * get_leaf_nr - Get a leaf number associated with the index
711  * @dip: The GFS2 inode
712  * @index:
713  * @leaf_out:
714  *
715  * Returns: 0 on success, error code otherwise
716  */
717
718 static int get_leaf_nr(struct gfs2_inode *dip, uint32_t index,
719                        uint64_t *leaf_out)
720 {
721         uint64_t leaf_no;
722         int error;
723
724         error = gfs2_dir_read_data(dip, (char *)&leaf_no,
725                                     index * sizeof(uint64_t),
726                                     sizeof(uint64_t));
727         if (error != sizeof(uint64_t))
728                 return (error < 0) ? error : -EIO;
729
730         *leaf_out = be64_to_cpu(leaf_no);
731
732         return 0;
733 }
734
735 static int get_first_leaf(struct gfs2_inode *dip, uint32_t index,
736                           struct buffer_head **bh_out)
737 {
738         uint64_t leaf_no;
739         int error;
740
741         error = get_leaf_nr(dip, index, &leaf_no);
742         if (!error)
743                 error = get_leaf(dip, leaf_no, bh_out);
744
745         return error;
746 }
747
748 static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
749                                               const struct qstr *name,
750                                               gfs2_dscan_t scan,
751                                               struct buffer_head **pbh)
752 {
753         struct buffer_head *bh;
754         struct gfs2_dirent *dent;
755         struct gfs2_inode *ip = inode->u.generic_ip;
756         int error;
757
758         if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
759                 struct gfs2_leaf *leaf;
760                 unsigned hsize = 1 << ip->i_di.di_depth;
761                 unsigned index;
762                 u64 ln;
763                 if (hsize * sizeof(u64) != ip->i_di.di_size) {
764                         gfs2_consist_inode(ip);
765                         return ERR_PTR(-EIO);
766                 }
767
768                 index = name->hash >> (32 - ip->i_di.di_depth);
769                 error = get_first_leaf(ip, index, &bh);
770                 if (error)
771                         return ERR_PTR(error);
772                 do {
773                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
774                                                 scan, name, NULL);
775                         if (dent)
776                                 goto got_dent;
777                         leaf = (struct gfs2_leaf *)bh->b_data;
778                         ln = be64_to_cpu(leaf->lf_next);
779                         if (!ln)
780                                 break;
781                         brelse(bh);
782                         error = get_leaf(ip, ln, &bh);
783                 } while(!error);
784
785                 return error ? ERR_PTR(error) : NULL;
786         }
787
788         error = gfs2_meta_inode_buffer(ip, &bh);
789         if (error)
790                 return ERR_PTR(error);
791         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
792 got_dent:
793         if (unlikely(IS_ERR(dent))) {
794                 brelse(bh);
795                 bh = NULL;
796         }
797         *pbh = bh;
798         return dent;
799 }
800
801 static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
802 {
803         struct gfs2_inode *ip = inode->u.generic_ip;
804         u64 bn = gfs2_alloc_meta(ip);
805         struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn);
806         struct gfs2_leaf *leaf;
807         struct gfs2_dirent *dent;
808         struct qstr name = { .name = "", .len = 0, .hash = 0 };
809         if (!bh)
810                 return NULL;
811         gfs2_trans_add_bh(ip->i_gl, bh, 1);
812         gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
813         leaf = (struct gfs2_leaf *)bh->b_data;
814         leaf->lf_depth = cpu_to_be16(depth);
815         leaf->lf_entries = cpu_to_be16(0);
816         leaf->lf_dirent_format = cpu_to_be16(GFS2_FORMAT_DE);
817         leaf->lf_next = cpu_to_be64(0);
818         memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
819         dent = (struct gfs2_dirent *)(leaf+1);
820         gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
821         *pbh = bh;
822         return leaf;
823 }
824
825 /**
826  * dir_make_exhash - Convert a stuffed directory into an ExHash directory
827  * @dip: The GFS2 inode
828  *
829  * Returns: 0 on success, error code otherwise
830  */
831
832 static int dir_make_exhash(struct inode *inode)
833 {
834         struct gfs2_inode *dip = inode->u.generic_ip;
835         struct gfs2_sbd *sdp = dip->i_sbd;
836         struct gfs2_dirent *dent;
837         struct qstr args;
838         struct buffer_head *bh, *dibh;
839         struct gfs2_leaf *leaf;
840         int y;
841         uint32_t x;
842         uint64_t *lp, bn;
843         int error;
844
845         error = gfs2_meta_inode_buffer(dip, &dibh);
846         if (error)
847                 return error;
848
849         /*  Turn over a new leaf  */
850
851         leaf = new_leaf(inode, &bh, 0);
852         if (!leaf)
853                 return -ENOSPC;
854         bn = bh->b_blocknr;
855
856         gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16));
857         leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries);
858
859         /*  Copy dirents  */
860
861         gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
862                              sizeof(struct gfs2_dinode));
863
864         /*  Find last entry  */
865
866         x = 0;
867         args.len = bh->b_size - sizeof(struct gfs2_dinode) +
868                    sizeof(struct gfs2_leaf);
869         args.name = bh->b_data;
870         dent = gfs2_dirent_scan(dip->i_vnode, bh->b_data, bh->b_size,
871                                 gfs2_dirent_last, &args, NULL);
872         if (!dent) {
873                 brelse(bh);
874                 brelse(dibh);
875                 return -EIO;
876         }
877         if (IS_ERR(dent)) {
878                 brelse(bh);
879                 brelse(dibh);
880                 return PTR_ERR(dent);
881         }
882
883         /*  Adjust the last dirent's record length
884            (Remember that dent still points to the last entry.)  */
885
886         dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
887                 sizeof(struct gfs2_dinode) -
888                 sizeof(struct gfs2_leaf));
889
890         brelse(bh);
891
892         /*  We're done with the new leaf block, now setup the new
893             hash table.  */
894
895         gfs2_trans_add_bh(dip->i_gl, dibh, 1);
896         gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
897
898         lp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode));
899
900         for (x = sdp->sd_hash_ptrs; x--; lp++)
901                 *lp = cpu_to_be64(bn);
902
903         dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2;
904         dip->i_di.di_blocks++;
905         dip->i_di.di_flags |= GFS2_DIF_EXHASH;
906         dip->i_di.di_payload_format = 0;
907
908         for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ;
909         dip->i_di.di_depth = y;
910
911         gfs2_dinode_out(&dip->i_di, dibh->b_data);
912
913         brelse(dibh);
914
915         return 0;
916 }
917
918 /**
919  * dir_split_leaf - Split a leaf block into two
920  * @dip: The GFS2 inode
921  * @index:
922  * @leaf_no:
923  *
924  * Returns: 0 on success, error code on failure
925  */
926
927 static int dir_split_leaf(struct inode *inode, const struct qstr *name)
928 {
929         struct gfs2_inode *dip = inode->u.generic_ip;
930         struct buffer_head *nbh, *obh, *dibh;
931         struct gfs2_leaf *nleaf, *oleaf;
932         struct gfs2_dirent *dent, *prev = NULL, *next = NULL, *new;
933         uint32_t start, len, half_len, divider;
934         uint64_t bn, *lp, leaf_no;
935         uint32_t index;
936         int x, moved = 0;
937         int error;
938
939         index = name->hash >> (32 - dip->i_di.di_depth);
940         error = get_leaf_nr(dip, index, &leaf_no);
941         if (error)
942                 return error;
943
944         /*  Get the old leaf block  */
945         error = get_leaf(dip, leaf_no, &obh);
946         if (error)
947                 return error;
948
949         oleaf = (struct gfs2_leaf *)obh->b_data;
950         if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) {
951                 brelse(obh);
952                 return 1; /* can't split */
953         }
954
955         gfs2_trans_add_bh(dip->i_gl, obh, 1);
956
957         nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
958         if (!nleaf) {
959                 brelse(obh);
960                 return -ENOSPC;
961         }
962         bn = nbh->b_blocknr;
963
964         /*  Compute the start and len of leaf pointers in the hash table.  */
965         len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth));
966         half_len = len >> 1;
967         if (!half_len) {
968                 printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index);
969                 gfs2_consist_inode(dip);
970                 error = -EIO;
971                 goto fail_brelse;
972         }
973
974         start = (index & ~(len - 1));
975
976         /* Change the pointers.
977            Don't bother distinguishing stuffed from non-stuffed.
978            This code is complicated enough already. */
979         lp = kmalloc(half_len * sizeof(uint64_t), GFP_NOFS | __GFP_NOFAIL);
980         /*  Change the pointers  */
981         for (x = 0; x < half_len; x++)
982                 lp[x] = cpu_to_be64(bn);
983
984         error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(uint64_t),
985                                     half_len * sizeof(uint64_t));
986         if (error != half_len * sizeof(uint64_t)) {
987                 if (error >= 0)
988                         error = -EIO;
989                 goto fail_lpfree;
990         }
991
992         kfree(lp);
993
994         /*  Compute the divider  */
995         divider = (start + half_len) << (32 - dip->i_di.di_depth);
996
997         /*  Copy the entries  */
998         dirent_first(dip, obh, &dent);
999
1000         do {
1001                 next = dent;
1002                 if (dirent_next(dip, obh, &next))
1003                         next = NULL;
1004
1005                 if (dent->de_inum.no_addr &&
1006                     be32_to_cpu(dent->de_hash) < divider) {
1007                         struct qstr str;
1008                         str.name = (char*)(dent+1);
1009                         str.len = be16_to_cpu(dent->de_name_len);
1010                         str.hash = be32_to_cpu(dent->de_hash);
1011                         new = gfs2_dirent_alloc(inode, nbh, &str);
1012                         if (IS_ERR(new)) {
1013                                 error = PTR_ERR(new);
1014                                 break;
1015                         }
1016
1017                         new->de_inum = dent->de_inum; /* No endian worries */
1018                         new->de_type = dent->de_type; /* No endian worries */
1019                         nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1);
1020
1021                         dirent_del(dip, obh, prev, dent);
1022
1023                         if (!oleaf->lf_entries)
1024                                 gfs2_consist_inode(dip);
1025                         oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1);
1026
1027                         if (!prev)
1028                                 prev = dent;
1029
1030                         moved = 1;
1031                 } else {
1032                         prev = dent;
1033                 }
1034                 dent = next;
1035         } while (dent);
1036
1037         oleaf->lf_depth = nleaf->lf_depth;
1038
1039         error = gfs2_meta_inode_buffer(dip, &dibh);
1040         if (!gfs2_assert_withdraw(dip->i_sbd, !error)) {
1041                 dip->i_di.di_blocks++;
1042                 gfs2_dinode_out(&dip->i_di, dibh->b_data);
1043                 brelse(dibh);
1044         }
1045
1046         brelse(obh);
1047         brelse(nbh);
1048
1049         return error;
1050
1051 fail_lpfree:
1052         kfree(lp);
1053
1054 fail_brelse:
1055         brelse(obh);
1056         brelse(nbh);
1057         return error;
1058 }
1059
1060 /**
1061  * dir_double_exhash - Double size of ExHash table
1062  * @dip: The GFS2 dinode
1063  *
1064  * Returns: 0 on success, error code on failure
1065  */
1066
1067 static int dir_double_exhash(struct gfs2_inode *dip)
1068 {
1069         struct gfs2_sbd *sdp = dip->i_sbd;
1070         struct buffer_head *dibh;
1071         uint32_t hsize;
1072         uint64_t *buf;
1073         uint64_t *from, *to;
1074         uint64_t block;
1075         int x;
1076         int error = 0;
1077
1078         hsize = 1 << dip->i_di.di_depth;
1079         if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
1080                 gfs2_consist_inode(dip);
1081                 return -EIO;
1082         }
1083
1084         /*  Allocate both the "from" and "to" buffers in one big chunk  */
1085
1086         buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL);
1087
1088         for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) {
1089                 error = gfs2_dir_read_data(dip, (char *)buf,
1090                                             block * sdp->sd_hash_bsize,
1091                                             sdp->sd_hash_bsize);
1092                 if (error != sdp->sd_hash_bsize) {
1093                         if (error >= 0)
1094                                 error = -EIO;
1095                         goto fail;
1096                 }
1097
1098                 from = buf;
1099                 to = (uint64_t *)((char *)buf + sdp->sd_hash_bsize);
1100
1101                 for (x = sdp->sd_hash_ptrs; x--; from++) {
1102                         *to++ = *from;  /*  No endianess worries  */
1103                         *to++ = *from;
1104                 }
1105
1106                 error = gfs2_dir_write_data(dip,
1107                                              (char *)buf + sdp->sd_hash_bsize,
1108                                              block * sdp->sd_sb.sb_bsize,
1109                                              sdp->sd_sb.sb_bsize);
1110                 if (error != sdp->sd_sb.sb_bsize) {
1111                         if (error >= 0)
1112                                 error = -EIO;
1113                         goto fail;
1114                 }
1115         }
1116
1117         kfree(buf);
1118
1119         error = gfs2_meta_inode_buffer(dip, &dibh);
1120         if (!gfs2_assert_withdraw(sdp, !error)) {
1121                 dip->i_di.di_depth++;
1122                 gfs2_dinode_out(&dip->i_di, dibh->b_data);
1123                 brelse(dibh);
1124         }
1125
1126         return error;
1127
1128  fail:
1129         kfree(buf);
1130
1131         return error;
1132 }
1133
1134 /**
1135  * compare_dents - compare directory entries by hash value
1136  * @a: first dent
1137  * @b: second dent
1138  *
1139  * When comparing the hash entries of @a to @b:
1140  *   gt: returns 1
1141  *   lt: returns -1
1142  *   eq: returns 0
1143  */
1144
1145 static int compare_dents(const void *a, const void *b)
1146 {
1147         struct gfs2_dirent *dent_a, *dent_b;
1148         uint32_t hash_a, hash_b;
1149         int ret = 0;
1150
1151         dent_a = *(struct gfs2_dirent **)a;
1152         hash_a = be32_to_cpu(dent_a->de_hash);
1153
1154         dent_b = *(struct gfs2_dirent **)b;
1155         hash_b = be32_to_cpu(dent_b->de_hash);
1156
1157         if (hash_a > hash_b)
1158                 ret = 1;
1159         else if (hash_a < hash_b)
1160                 ret = -1;
1161         else {
1162                 unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
1163                 unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
1164
1165                 if (len_a > len_b)
1166                         ret = 1;
1167                 else if (len_a < len_b)
1168                         ret = -1;
1169                 else
1170                         ret = memcmp((char *)(dent_a + 1),
1171                                      (char *)(dent_b + 1),
1172                                      len_a);
1173         }
1174
1175         return ret;
1176 }
1177
1178 /**
1179  * do_filldir_main - read out directory entries
1180  * @dip: The GFS2 inode
1181  * @offset: The offset in the file to read from
1182  * @opaque: opaque data to pass to filldir
1183  * @filldir: The function to pass entries to
1184  * @darr: an array of struct gfs2_dirent pointers to read
1185  * @entries: the number of entries in darr
1186  * @copied: pointer to int that's non-zero if a entry has been copied out
1187  *
1188  * Jump through some hoops to make sure that if there are hash collsions,
1189  * they are read out at the beginning of a buffer.  We want to minimize
1190  * the possibility that they will fall into different readdir buffers or
1191  * that someone will want to seek to that location.
1192  *
1193  * Returns: errno, >0 on exception from filldir
1194  */
1195
1196 static int do_filldir_main(struct gfs2_inode *dip, uint64_t *offset,
1197                            void *opaque, gfs2_filldir_t filldir,
1198                            const struct gfs2_dirent **darr, uint32_t entries,
1199                            int *copied)
1200 {
1201         const struct gfs2_dirent *dent, *dent_next;
1202         struct gfs2_inum inum;
1203         uint64_t off, off_next;
1204         unsigned int x, y;
1205         int run = 0;
1206         int error = 0;
1207
1208         sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL);
1209
1210         dent_next = darr[0];
1211         off_next = be32_to_cpu(dent_next->de_hash);
1212         off_next = gfs2_disk_hash2offset(off_next);
1213
1214         for (x = 0, y = 1; x < entries; x++, y++) {
1215                 dent = dent_next;
1216                 off = off_next;
1217
1218                 if (y < entries) {
1219                         dent_next = darr[y];
1220                         off_next = be32_to_cpu(dent_next->de_hash);
1221                         off_next = gfs2_disk_hash2offset(off_next);
1222
1223                         if (off < *offset)
1224                                 continue;
1225                         *offset = off;
1226
1227                         if (off_next == off) {
1228                                 if (*copied && !run)
1229                                         return 1;
1230                                 run = 1;
1231                         } else
1232                                 run = 0;
1233                 } else {
1234                         if (off < *offset)
1235                                 continue;
1236                         *offset = off;
1237                 }
1238
1239                 gfs2_inum_in(&inum, (char *)&dent->de_inum);
1240
1241                 error = filldir(opaque, (char *)(dent + 1),
1242                                 be16_to_cpu(dent->de_name_len),
1243                                 off, &inum,
1244                                 be16_to_cpu(dent->de_type));
1245                 if (error)
1246                         return 1;
1247
1248                 *copied = 1;
1249         }
1250
1251         /* Increment the *offset by one, so the next time we come into the
1252            do_filldir fxn, we get the next entry instead of the last one in the
1253            current leaf */
1254
1255         (*offset)++;
1256
1257         return 0;
1258 }
1259
1260 static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque,
1261                               gfs2_filldir_t filldir, int *copied,
1262                               unsigned *depth, u64 leaf_no)
1263 {
1264         struct gfs2_inode *ip = inode->u.generic_ip;
1265         struct buffer_head *bh;
1266         struct gfs2_leaf *lf;
1267         unsigned entries = 0;
1268         unsigned leaves = 0;
1269         const struct gfs2_dirent **darr, *dent;
1270         struct dirent_gather g;
1271         struct buffer_head **larr;
1272         int leaf = 0;
1273         int error, i;
1274         u64 lfn = leaf_no;
1275
1276         do {
1277                 error = get_leaf(ip, lfn, &bh);
1278                 if (error)
1279                         goto out;
1280                 lf = (struct gfs2_leaf *)bh->b_data;
1281                 if (leaves == 0)
1282                         *depth = be16_to_cpu(lf->lf_depth);
1283                 entries += be16_to_cpu(lf->lf_entries);
1284                 leaves++;
1285                 lfn = be64_to_cpu(lf->lf_next);
1286                 brelse(bh);
1287         } while(lfn);
1288
1289         if (!entries)
1290                 return 0;
1291
1292         error = -ENOMEM;
1293         larr = kmalloc((leaves + entries) * sizeof(void*), GFP_KERNEL);
1294         if (!larr)
1295                 goto out;
1296         darr = (const struct gfs2_dirent **)(larr + leaves);
1297         g.pdent = darr;
1298         g.offset = 0;
1299         lfn = leaf_no;
1300
1301         do {
1302                 error = get_leaf(ip, lfn, &bh);
1303                 if (error)
1304                         goto out_kfree;
1305                 lf = (struct gfs2_leaf *)bh->b_data;
1306                 lfn = be64_to_cpu(lf->lf_next);
1307                 if (lf->lf_entries) {
1308                         dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
1309                                                 gfs2_dirent_gather, NULL, &g);
1310                         error = PTR_ERR(dent);
1311                         if (IS_ERR(dent)) {
1312                                 goto out_kfree;
1313                         }
1314                         error = 0;
1315                         larr[leaf++] = bh;
1316                 } else {
1317                         brelse(bh);
1318                 }
1319         } while(lfn);
1320
1321         error = do_filldir_main(ip, offset, opaque, filldir, darr,
1322                                 entries, copied);
1323 out_kfree:
1324         for(i = 0; i < leaf; i++)
1325                 brelse(larr[i]);
1326         kfree(larr);
1327 out:
1328         return error;
1329 }
1330
1331 /**
1332  * dir_e_read - Reads the entries from a directory into a filldir buffer
1333  * @dip: dinode pointer
1334  * @offset: the hash of the last entry read shifted to the right once
1335  * @opaque: buffer for the filldir function to fill
1336  * @filldir: points to the filldir function to use
1337  *
1338  * Returns: errno
1339  */
1340
1341 static int dir_e_read(struct inode *inode, uint64_t *offset, void *opaque,
1342                       gfs2_filldir_t filldir)
1343 {
1344         struct gfs2_inode *dip = inode->u.generic_ip;
1345         struct gfs2_sbd *sdp = dip->i_sbd;
1346         uint32_t hsize, len = 0;
1347         uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
1348         uint32_t hash, index;
1349         uint64_t *lp;
1350         int copied = 0;
1351         int error = 0;
1352         unsigned depth;
1353
1354         hsize = 1 << dip->i_di.di_depth;
1355         if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
1356                 gfs2_consist_inode(dip);
1357                 return -EIO;
1358         }
1359
1360         hash = gfs2_dir_offset2hash(*offset);
1361         index = hash >> (32 - dip->i_di.di_depth);
1362
1363         lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1364         if (!lp)
1365                 return -ENOMEM;
1366
1367         while (index < hsize) {
1368                 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1369                 ht_offset = index - lp_offset;
1370
1371                 if (ht_offset_cur != ht_offset) {
1372                         error = gfs2_dir_read_data(dip, (char *)lp,
1373                                                 ht_offset * sizeof(uint64_t),
1374                                                 sdp->sd_hash_bsize);
1375                         if (error != sdp->sd_hash_bsize) {
1376                                 if (error >= 0)
1377                                         error = -EIO;
1378                                 goto out;
1379                         }
1380                         ht_offset_cur = ht_offset;
1381                 }
1382
1383                 error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
1384                                            &copied, &depth,
1385                                            be64_to_cpu(lp[lp_offset]));
1386                 if (error)
1387                         break;
1388
1389                 len = 1 << (dip->i_di.di_depth - depth);
1390                 index = (index & ~(len - 1)) + len;
1391         }
1392
1393 out:
1394         kfree(lp);
1395         if (error > 0)
1396                 error = 0;
1397         return error;
1398 }
1399
1400 int gfs2_dir_read(struct inode *inode, uint64_t *offset, void *opaque,
1401                   gfs2_filldir_t filldir)
1402 {
1403         struct gfs2_inode *dip = inode->u.generic_ip;
1404         struct dirent_gather g;
1405         const struct gfs2_dirent **darr, *dent;
1406         struct buffer_head *dibh;
1407         int copied = 0;
1408         int error;
1409
1410         if (!dip->i_di.di_entries)
1411                 return 0;
1412
1413         if (dip->i_di.di_flags & GFS2_DIF_EXHASH)
1414                 return dir_e_read(inode, offset, opaque, filldir);
1415
1416         if (!gfs2_is_stuffed(dip)) {
1417                 gfs2_consist_inode(dip);
1418                 return -EIO;
1419         }
1420
1421         error = gfs2_meta_inode_buffer(dip, &dibh);
1422         if (error)
1423                 return error;
1424
1425         error = -ENOMEM;
1426         darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *),
1427                        GFP_KERNEL);
1428         if (darr) {
1429                 g.pdent = darr;
1430                 g.offset = 0;
1431                 dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
1432                                         gfs2_dirent_gather, NULL, &g);
1433                 if (IS_ERR(dent)) {
1434                         error = PTR_ERR(dent);
1435                         goto out;
1436                 }
1437                 error = do_filldir_main(dip, offset, opaque, filldir, darr,
1438                                         dip->i_di.di_entries, &copied);
1439 out:
1440                 kfree(darr);
1441         }
1442
1443         if (error > 0)
1444                 error = 0;
1445
1446         brelse(dibh);
1447
1448         return error;
1449 }
1450
1451 /**
1452  * gfs2_dir_search - Search a directory
1453  * @dip: The GFS2 inode
1454  * @filename:
1455  * @inode:
1456  *
1457  * This routine searches a directory for a file or another directory.
1458  * Assumes a glock is held on dip.
1459  *
1460  * Returns: errno
1461  */
1462
1463 int gfs2_dir_search(struct inode *dir, const struct qstr *name,
1464                     struct gfs2_inum *inum, unsigned int *type)
1465 {
1466         struct buffer_head *bh;
1467         struct gfs2_dirent *dent;
1468
1469         dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
1470         if (dent) {
1471                 if (IS_ERR(dent))
1472                         return PTR_ERR(dent);
1473                 if (inum)
1474                         gfs2_inum_in(inum, (char *)&dent->de_inum);
1475                 if (type)
1476                         *type = be16_to_cpu(dent->de_type);
1477                 brelse(bh);
1478                 return 0;
1479         }
1480         brelse(bh);
1481         return -ENOENT;
1482 }
1483
1484 static int dir_new_leaf(struct inode *inode, const struct qstr *name)
1485 {
1486         struct buffer_head *bh, *obh;
1487         struct gfs2_inode *ip = inode->u.generic_ip;
1488         struct gfs2_leaf *leaf, *oleaf;
1489         int error;
1490         u32 index;
1491         u64 bn;
1492
1493         index = name->hash >> (32 - ip->i_di.di_depth);
1494         error = get_first_leaf(ip, index, &obh);
1495         if (error)
1496                 return error;
1497         do {
1498                 oleaf = (struct gfs2_leaf *)obh->b_data;
1499                 bn = be64_to_cpu(oleaf->lf_next);
1500                 if (!bn)
1501                         break;
1502                 brelse(obh);
1503                 error = get_leaf(ip, bn, &obh);
1504                 if (error)
1505                         return error;
1506         } while(1);
1507
1508         gfs2_trans_add_bh(ip->i_gl, obh, 1);
1509
1510         leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
1511         if (!leaf) {
1512                 brelse(obh);
1513                 return -ENOSPC;
1514         }
1515         oleaf->lf_next = cpu_to_be64(bn);
1516         brelse(bh);
1517         brelse(obh);
1518
1519         error = gfs2_meta_inode_buffer(ip, &bh);
1520         if (error)
1521                 return error;
1522         gfs2_trans_add_bh(ip->i_gl, bh, 1);
1523         ip->i_di.di_blocks++;
1524         gfs2_dinode_out(&ip->i_di, bh->b_data);
1525         brelse(bh);
1526         return 0;
1527 }
1528
1529 /**
1530  * gfs2_dir_add - Add new filename into directory
1531  * @dip: The GFS2 inode
1532  * @filename: The new name
1533  * @inode: The inode number of the entry
1534  * @type: The type of the entry
1535  *
1536  * Returns: 0 on success, error code on failure
1537  */
1538
1539 int gfs2_dir_add(struct inode *inode, const struct qstr *name,
1540                  const struct gfs2_inum *inum, unsigned type)
1541 {
1542         struct gfs2_inode *ip = inode->u.generic_ip;
1543         struct buffer_head *bh;
1544         struct gfs2_dirent *dent;
1545         struct gfs2_leaf *leaf;
1546         int error;
1547
1548         while(1) {
1549                 dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space,
1550                                           &bh);
1551                 if (dent) {
1552                         if (IS_ERR(dent))
1553                                 return PTR_ERR(dent);
1554                         dent = gfs2_init_dirent(inode, dent, name, bh);
1555                         gfs2_inum_out(inum, (char *)&dent->de_inum);
1556                         dent->de_type = cpu_to_be16(type);
1557                         if (ip->i_di.di_flags & GFS2_DIF_EXHASH) {
1558                                 leaf = (struct gfs2_leaf *)bh->b_data;
1559                                 leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1);
1560                         }
1561                         brelse(bh);
1562                         error = gfs2_meta_inode_buffer(ip, &bh);
1563                         if (error)
1564                                 break;
1565                         gfs2_trans_add_bh(ip->i_gl, bh, 1);
1566                         ip->i_di.di_entries++;
1567                         ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds();
1568                         gfs2_dinode_out(&ip->i_di, bh->b_data);
1569                         brelse(bh);
1570                         error = 0;
1571                         break;
1572                 }
1573                 if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
1574                         error = dir_make_exhash(inode);
1575                         if (error)
1576                                 break;
1577                         continue;
1578                 }
1579                 error = dir_split_leaf(inode, name);
1580                 if (error == 0)
1581                         continue;
1582                 if (error < 0)
1583                         break;
1584                 if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) {
1585                         error = dir_double_exhash(ip);
1586                         if (error)
1587                                 break;
1588                         error = dir_split_leaf(inode, name);
1589                         if (error < 0)
1590                                 break;
1591                         if (error == 0)
1592                                 continue;
1593                 }
1594                 error = dir_new_leaf(inode, name);
1595                 if (!error)
1596                         continue;
1597                 error = -ENOSPC;
1598                 break;
1599         }
1600         return error;
1601 }
1602
1603
1604 /**
1605  * gfs2_dir_del - Delete a directory entry
1606  * @dip: The GFS2 inode
1607  * @filename: The filename
1608  *
1609  * Returns: 0 on success, error code on failure
1610  */
1611
1612 int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name)
1613 {
1614         struct gfs2_dirent *dent, *prev = NULL;
1615         struct buffer_head *bh;
1616         int error;
1617
1618         /* Returns _either_ the entry (if its first in block) or the
1619            previous entry otherwise */
1620         dent = gfs2_dirent_search(dip->i_vnode, name, gfs2_dirent_prev, &bh);
1621         if (!dent) {
1622                 brelse(bh);
1623                 gfs2_consist_inode(dip);
1624                 return -EIO;
1625         }
1626         if (IS_ERR(dent)) {
1627                 gfs2_consist_inode(dip);
1628                 return PTR_ERR(dent);
1629         }
1630         /* If not first in block, adjust pointers accordingly */
1631         if (gfs2_dirent_find(dent, name, NULL) == 0) {
1632                 prev = dent;
1633                 dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
1634         }
1635
1636         dirent_del(dip, bh, prev, dent);
1637         if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1638                 struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
1639                 u16 entries = be16_to_cpu(leaf->lf_entries);
1640                 if (!entries)
1641                         gfs2_consist_inode(dip);
1642                 leaf->lf_entries = cpu_to_be16(--entries);
1643         }
1644         brelse(bh);
1645
1646         error = gfs2_meta_inode_buffer(dip, &bh);
1647         if (error)
1648                 return error;
1649
1650         if (!dip->i_di.di_entries)
1651                 gfs2_consist_inode(dip);
1652         gfs2_trans_add_bh(dip->i_gl, bh, 1);
1653         dip->i_di.di_entries--;
1654         dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
1655         gfs2_dinode_out(&dip->i_di, bh->b_data);
1656         brelse(bh);
1657
1658         return error;
1659 }
1660
1661 /**
1662  * gfs2_dir_mvino - Change inode number of directory entry
1663  * @dip: The GFS2 inode
1664  * @filename:
1665  * @new_inode:
1666  *
1667  * This routine changes the inode number of a directory entry.  It's used
1668  * by rename to change ".." when a directory is moved.
1669  * Assumes a glock is held on dvp.
1670  *
1671  * Returns: errno
1672  */
1673
1674 int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
1675                    struct gfs2_inum *inum, unsigned int new_type)
1676 {
1677         struct buffer_head *bh;
1678         struct gfs2_dirent *dent;
1679         int error;
1680
1681         dent = gfs2_dirent_search(dip->i_vnode, filename, gfs2_dirent_find, &bh);
1682         if (!dent) {
1683                 brelse(bh);
1684                 gfs2_consist_inode(dip);
1685                 return -EIO;
1686         }
1687         if (IS_ERR(dent))
1688                 return PTR_ERR(dent);
1689
1690         gfs2_trans_add_bh(dip->i_gl, bh, 1);
1691         gfs2_inum_out(inum, (char *)&dent->de_inum);
1692         dent->de_type = cpu_to_be16(new_type);
1693
1694         if (dip->i_di.di_flags & GFS2_DIF_EXHASH) {
1695                 brelse(bh);
1696                 error = gfs2_meta_inode_buffer(dip, &bh);
1697                 if (error)
1698                         return error;
1699                 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1700         }
1701
1702         dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds();
1703         gfs2_dinode_out(&dip->i_di, bh->b_data);
1704         brelse(bh);
1705         return 0;
1706 }
1707
1708 /**
1709  * foreach_leaf - call a function for each leaf in a directory
1710  * @dip: the directory
1711  * @lc: the function to call for each each
1712  * @data: private data to pass to it
1713  *
1714  * Returns: errno
1715  */
1716
1717 static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data)
1718 {
1719         struct gfs2_sbd *sdp = dip->i_sbd;
1720         struct buffer_head *bh;
1721         struct gfs2_leaf *leaf;
1722         uint32_t hsize, len;
1723         uint32_t ht_offset, lp_offset, ht_offset_cur = -1;
1724         uint32_t index = 0;
1725         uint64_t *lp;
1726         uint64_t leaf_no;
1727         int error = 0;
1728
1729         hsize = 1 << dip->i_di.di_depth;
1730         if (hsize * sizeof(uint64_t) != dip->i_di.di_size) {
1731                 gfs2_consist_inode(dip);
1732                 return -EIO;
1733         }
1734
1735         lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL);
1736         if (!lp)
1737                 return -ENOMEM;
1738
1739         while (index < hsize) {
1740                 lp_offset = index & (sdp->sd_hash_ptrs - 1);
1741                 ht_offset = index - lp_offset;
1742
1743                 if (ht_offset_cur != ht_offset) {
1744                         error = gfs2_dir_read_data(dip, (char *)lp,
1745                                                 ht_offset * sizeof(uint64_t),
1746                                                 sdp->sd_hash_bsize);
1747                         if (error != sdp->sd_hash_bsize) {
1748                                 if (error >= 0)
1749                                         error = -EIO;
1750                                 goto out;
1751                         }
1752                         ht_offset_cur = ht_offset;
1753                 }
1754
1755                 leaf_no = be64_to_cpu(lp[lp_offset]);
1756                 if (leaf_no) {
1757                         error = get_leaf(dip, leaf_no, &bh);
1758                         if (error)
1759                                 goto out;
1760                         leaf = (struct gfs2_leaf *)bh->b_data;
1761                         brelse(bh);
1762
1763                         len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth));
1764
1765                         error = lc(dip, index, len, leaf_no, data);
1766                         if (error)
1767                                 goto out;
1768
1769                         index = (index & ~(len - 1)) + len;
1770                 } else
1771                         index++;
1772         }
1773
1774         if (index != hsize) {
1775                 gfs2_consist_inode(dip);
1776                 error = -EIO;
1777         }
1778
1779  out:
1780         kfree(lp);
1781
1782         return error;
1783 }
1784
1785 /**
1786  * leaf_dealloc - Deallocate a directory leaf
1787  * @dip: the directory
1788  * @index: the hash table offset in the directory
1789  * @len: the number of pointers to this leaf
1790  * @leaf_no: the leaf number
1791  * @data: not used
1792  *
1793  * Returns: errno
1794  */
1795
1796 static int leaf_dealloc(struct gfs2_inode *dip, uint32_t index, uint32_t len,
1797                         uint64_t leaf_no, void *data)
1798 {
1799         struct gfs2_sbd *sdp = dip->i_sbd;
1800         struct gfs2_leaf *tmp_leaf;
1801         struct gfs2_rgrp_list rlist;
1802         struct buffer_head *bh, *dibh;
1803         uint64_t blk, nblk;
1804         unsigned int rg_blocks = 0, l_blocks = 0;
1805         char *ht;
1806         unsigned int x, size = len * sizeof(uint64_t);
1807         int error;
1808
1809         memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
1810
1811         ht = kzalloc(size, GFP_KERNEL);
1812         if (!ht)
1813                 return -ENOMEM;
1814
1815         gfs2_alloc_get(dip);
1816
1817         error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1818         if (error)
1819                 goto out;
1820
1821         error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh);
1822         if (error)
1823                 goto out_qs;
1824
1825         /*  Count the number of leaves  */
1826
1827         for (blk = leaf_no; blk; blk = nblk) {
1828                 error = get_leaf(dip, blk, &bh);
1829                 if (error)
1830                         goto out_rlist;
1831                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1832                 nblk = be64_to_cpu(tmp_leaf->lf_next);
1833                 brelse(bh);
1834
1835                 gfs2_rlist_add(sdp, &rlist, blk);
1836                 l_blocks++;
1837         }
1838
1839         gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0);
1840
1841         for (x = 0; x < rlist.rl_rgrps; x++) {
1842                 struct gfs2_rgrpd *rgd;
1843                 rgd = rlist.rl_ghs[x].gh_gl->gl_object;
1844                 rg_blocks += rgd->rd_ri.ri_length;
1845         }
1846
1847         error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
1848         if (error)
1849                 goto out_rlist;
1850
1851         error = gfs2_trans_begin(sdp,
1852                         rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
1853                         RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks);
1854         if (error)
1855                 goto out_rg_gunlock;
1856
1857         for (blk = leaf_no; blk; blk = nblk) {
1858                 error = get_leaf(dip, blk, &bh);
1859                 if (error)
1860                         goto out_end_trans;
1861                 tmp_leaf = (struct gfs2_leaf *)bh->b_data;
1862                 nblk = be64_to_cpu(tmp_leaf->lf_next);
1863                 brelse(bh);
1864
1865                 gfs2_free_meta(dip, blk, 1);
1866
1867                 if (!dip->i_di.di_blocks)
1868                         gfs2_consist_inode(dip);
1869                 dip->i_di.di_blocks--;
1870         }
1871
1872         error = gfs2_dir_write_data(dip, ht, index * sizeof(uint64_t), size);
1873         if (error != size) {
1874                 if (error >= 0)
1875                         error = -EIO;
1876                 goto out_end_trans;
1877         }
1878
1879         error = gfs2_meta_inode_buffer(dip, &dibh);
1880         if (error)
1881                 goto out_end_trans;
1882
1883         gfs2_trans_add_bh(dip->i_gl, dibh, 1);
1884         gfs2_dinode_out(&dip->i_di, dibh->b_data);
1885         brelse(dibh);
1886
1887  out_end_trans:
1888         gfs2_trans_end(sdp);
1889
1890  out_rg_gunlock:
1891         gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
1892
1893  out_rlist:
1894         gfs2_rlist_free(&rlist);
1895         gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh);
1896
1897  out_qs:
1898         gfs2_quota_unhold(dip);
1899
1900  out:
1901         gfs2_alloc_put(dip);
1902         kfree(ht);
1903
1904         return error;
1905 }
1906
1907 /**
1908  * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1909  * @dip: the directory
1910  *
1911  * Dealloc all on-disk directory leaves to FREEMETA state
1912  * Change on-disk inode type to "regular file"
1913  *
1914  * Returns: errno
1915  */
1916
1917 int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
1918 {
1919         struct gfs2_sbd *sdp = dip->i_sbd;
1920         struct buffer_head *bh;
1921         int error;
1922
1923         /* Dealloc on-disk leaves to FREEMETA state */
1924         error = foreach_leaf(dip, leaf_dealloc, NULL);
1925         if (error)
1926                 return error;
1927
1928         /* Make this a regular file in case we crash.
1929            (We don't want to free these blocks a second time.)  */
1930
1931         error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1932         if (error)
1933                 return error;
1934
1935         error = gfs2_meta_inode_buffer(dip, &bh);
1936         if (!error) {
1937                 gfs2_trans_add_bh(dip->i_gl, bh, 1);
1938                 ((struct gfs2_dinode *)bh->b_data)->di_mode =
1939                                                 cpu_to_be32(S_IFREG);
1940                 brelse(bh);
1941         }
1942
1943         gfs2_trans_end(sdp);
1944
1945         return error;
1946 }
1947
1948 /**
1949  * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1950  * @ip: the file being written to
1951  * @filname: the filename that's going to be added
1952  *
1953  * Returns: 1 if alloc required, 0 if not, -ve on error
1954  */
1955
1956 int gfs2_diradd_alloc_required(struct inode *inode,
1957                                const struct qstr *name)
1958 {
1959         struct gfs2_dirent *dent;
1960         struct buffer_head *bh;
1961
1962         dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
1963         if (!dent) {
1964                 brelse(bh);
1965                 return 1;
1966         }
1967         if (IS_ERR(dent))
1968                 return PTR_ERR(dent);
1969         brelse(bh);
1970         return 0;
1971 }
1972