Merge rsync://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / fs / ext3 / namei.c
1 /*
2  *  linux/fs/ext3/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/smp_lock.h>
39
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext3_append(handle_t *handle,
53                                         struct inode *inode,
54                                         u32 *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT3_I(inode)->i_disksize = inode->i_size;
63                 ext3_journal_get_write_access(handle,bh);
64         }
65         return bh;
66 }
67
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command) 
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u32 offs;
144 };
145
146 #ifdef CONFIG_EXT3_INDEX
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned value);
149 static inline unsigned dx_get_hash (struct dx_entry *entry);
150 static void dx_set_hash (struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count (struct dx_entry *entries);
152 static unsigned dx_get_limit (struct dx_entry *entries);
153 static void dx_set_count (struct dx_entry *entries, unsigned value);
154 static void dx_set_limit (struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit (struct inode *dir);
157 static struct dx_frame *dx_probe(struct dentry *dentry,
158                                  struct inode *dir,
159                                  struct dx_hash_info *hinfo,
160                                  struct dx_frame *frame,
161                                  int *err);
162 static void dx_release (struct dx_frame *frames);
163 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167                 struct dx_map_entry *offsets, int count);
168 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171                                  struct dx_frame *frame,
172                                  struct dx_frame *frames, 
173                                  __u32 *start_hash);
174 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175                        struct ext3_dir_entry_2 **res_dir, int *err);
176 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177                              struct inode *inode);
178
179 /*
180  * Future: use high four bits of block for coalesce-on-delete flags
181  * Mask them off for now.
182  */
183
184 static inline unsigned dx_get_block (struct dx_entry *entry)
185 {
186         return le32_to_cpu(entry->block) & 0x00ffffff;
187 }
188
189 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190 {
191         entry->block = cpu_to_le32(value);
192 }
193
194 static inline unsigned dx_get_hash (struct dx_entry *entry)
195 {
196         return le32_to_cpu(entry->hash);
197 }
198
199 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200 {
201         entry->hash = cpu_to_le32(value);
202 }
203
204 static inline unsigned dx_get_count (struct dx_entry *entries)
205 {
206         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207 }
208
209 static inline unsigned dx_get_limit (struct dx_entry *entries)
210 {
211         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212 }
213
214 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215 {
216         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217 }
218
219 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220 {
221         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222 }
223
224 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225 {
226         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
227                 EXT3_DIR_REC_LEN(2) - infosize;
228         return 0? 20: entry_space / sizeof(struct dx_entry);
229 }
230
231 static inline unsigned dx_node_limit (struct inode *dir)
232 {
233         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
234         return 0? 22: entry_space / sizeof(struct dx_entry);
235 }
236
237 /*
238  * Debug
239  */
240 #ifdef DX_DEBUG
241 static void dx_show_index (char * label, struct dx_entry *entries)
242 {
243         int i, n = dx_get_count (entries);
244         printk("%s index ", label);
245         for (i = 0; i < n; i++)
246         {
247                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
248         }
249         printk("\n");
250 }
251
252 struct stats
253
254         unsigned names;
255         unsigned space;
256         unsigned bcount;
257 };
258
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
260                                  int size, int show_names)
261 {
262         unsigned names = 0, space = 0;
263         char *base = (char *) de;
264         struct dx_hash_info h = *hinfo;
265
266         printk("names: ");
267         while ((char *) de < base + size)
268         {
269                 if (de->inode)
270                 {
271                         if (show_names)
272                         {
273                                 int len = de->name_len;
274                                 char *name = de->name;
275                                 while (len--) printk("%c", *name++);
276                                 ext3fs_dirhash(de->name, de->name_len, &h);
277                                 printk(":%x.%u ", h.hash,
278                                        ((char *) de - base));
279                         }
280                         space += EXT3_DIR_REC_LEN(de->name_len);
281                         names++;
282                 }
283                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284         }
285         printk("(%i)\n", names);
286         return (struct stats) { names, space, 1 };
287 }
288
289 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290                              struct dx_entry *entries, int levels)
291 {
292         unsigned blocksize = dir->i_sb->s_blocksize;
293         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294         unsigned bcount = 0;
295         struct buffer_head *bh;
296         int err;
297         printk("%i indexed blocks...\n", count);
298         for (i = 0; i < count; i++, entries++)
299         {
300                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302                 struct stats stats;
303                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
304                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
305                 stats = levels?
306                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
308                 names += stats.names;
309                 space += stats.space;
310                 bcount += stats.bcount;
311                 brelse (bh);
312         }
313         if (bcount)
314                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
315                         names, space/bcount,(space/bcount)*100/blocksize);
316         return (struct stats) { names, space, bcount};
317 }
318 #endif /* DX_DEBUG */
319
320 /*
321  * Probe for a directory leaf block to search.
322  *
323  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324  * error in the directory index, and the caller should fall back to
325  * searching the directory normally.  The callers of dx_probe **MUST**
326  * check for this error code, and make sure it never gets reflected
327  * back to userspace.
328  */
329 static struct dx_frame *
330 dx_probe(struct dentry *dentry, struct inode *dir,
331          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332 {
333         unsigned count, indirect;
334         struct dx_entry *at, *entries, *p, *q, *m;
335         struct dx_root *root;
336         struct buffer_head *bh;
337         struct dx_frame *frame = frame_in;
338         u32 hash;
339
340         frame->bh = NULL;
341         if (dentry)
342                 dir = dentry->d_parent->d_inode;
343         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
344                 goto fail;
345         root = (struct dx_root *) bh->b_data;
346         if (root->info.hash_version != DX_HASH_TEA &&
347             root->info.hash_version != DX_HASH_HALF_MD4 &&
348             root->info.hash_version != DX_HASH_LEGACY) {
349                 ext3_warning(dir->i_sb, __FUNCTION__,
350                              "Unrecognised inode hash code %d",
351                              root->info.hash_version);
352                 brelse(bh);
353                 *err = ERR_BAD_DX_DIR;
354                 goto fail;
355         }
356         hinfo->hash_version = root->info.hash_version;
357         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
358         if (dentry)
359                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
360         hash = hinfo->hash;
361
362         if (root->info.unused_flags & 1) {
363                 ext3_warning(dir->i_sb, __FUNCTION__,
364                              "Unimplemented inode hash flags: %#06x",
365                              root->info.unused_flags);
366                 brelse(bh);
367                 *err = ERR_BAD_DX_DIR;
368                 goto fail;
369         }
370
371         if ((indirect = root->info.indirect_levels) > 1) {
372                 ext3_warning(dir->i_sb, __FUNCTION__,
373                              "Unimplemented inode hash depth: %#06x",
374                              root->info.indirect_levels);
375                 brelse(bh);
376                 *err = ERR_BAD_DX_DIR;
377                 goto fail;
378         }
379
380         entries = (struct dx_entry *) (((char *)&root->info) +
381                                        root->info.info_length);
382         assert(dx_get_limit(entries) == dx_root_limit(dir,
383                                                       root->info.info_length));
384         dxtrace (printk("Look up %x", hash));
385         while (1)
386         {
387                 count = dx_get_count(entries);
388                 assert (count && count <= dx_get_limit(entries));
389                 p = entries + 1;
390                 q = entries + count - 1;
391                 while (p <= q)
392                 {
393                         m = p + (q - p)/2;
394                         dxtrace(printk("."));
395                         if (dx_get_hash(m) > hash)
396                                 q = m - 1;
397                         else
398                                 p = m + 1;
399                 }
400
401                 if (0) // linear search cross check
402                 {
403                         unsigned n = count - 1;
404                         at = entries;
405                         while (n--)
406                         {
407                                 dxtrace(printk(","));
408                                 if (dx_get_hash(++at) > hash)
409                                 {
410                                         at--;
411                                         break;
412                                 }
413                         }
414                         assert (at == p - 1);
415                 }
416
417                 at = p - 1;
418                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
419                 frame->bh = bh;
420                 frame->entries = entries;
421                 frame->at = at;
422                 if (!indirect--) return frame;
423                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
424                         goto fail2;
425                 at = entries = ((struct dx_node *) bh->b_data)->entries;
426                 assert (dx_get_limit(entries) == dx_node_limit (dir));
427                 frame++;
428         }
429 fail2:
430         while (frame >= frame_in) {
431                 brelse(frame->bh);
432                 frame--;
433         }
434 fail:
435         return NULL;
436 }
437
438 static void dx_release (struct dx_frame *frames)
439 {
440         if (frames[0].bh == NULL)
441                 return;
442
443         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
444                 brelse(frames[1].bh);
445         brelse(frames[0].bh);
446 }
447
448 /*
449  * This function increments the frame pointer to search the next leaf
450  * block, and reads in the necessary intervening nodes if the search
451  * should be necessary.  Whether or not the search is necessary is
452  * controlled by the hash parameter.  If the hash value is even, then
453  * the search is only continued if the next block starts with that
454  * hash value.  This is used if we are searching for a specific file.
455  *
456  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
457  *
458  * This function returns 1 if the caller should continue to search,
459  * or 0 if it should not.  If there is an error reading one of the
460  * index blocks, it will a negative error code.
461  *
462  * If start_hash is non-null, it will be filled in with the starting
463  * hash of the next page.
464  */
465 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
466                                  struct dx_frame *frame,
467                                  struct dx_frame *frames, 
468                                  __u32 *start_hash)
469 {
470         struct dx_frame *p;
471         struct buffer_head *bh;
472         int err, num_frames = 0;
473         __u32 bhash;
474
475         p = frame;
476         /*
477          * Find the next leaf page by incrementing the frame pointer.
478          * If we run out of entries in the interior node, loop around and
479          * increment pointer in the parent node.  When we break out of
480          * this loop, num_frames indicates the number of interior
481          * nodes need to be read.
482          */
483         while (1) {
484                 if (++(p->at) < p->entries + dx_get_count(p->entries))
485                         break;
486                 if (p == frames)
487                         return 0;
488                 num_frames++;
489                 p--;
490         }
491
492         /*
493          * If the hash is 1, then continue only if the next page has a
494          * continuation hash of any value.  This is used for readdir
495          * handling.  Otherwise, check to see if the hash matches the
496          * desired contiuation hash.  If it doesn't, return since
497          * there's no point to read in the successive index pages.
498          */
499         bhash = dx_get_hash(p->at);
500         if (start_hash)
501                 *start_hash = bhash;
502         if ((hash & 1) == 0) {
503                 if ((bhash & ~1) != hash)
504                         return 0;
505         }
506         /*
507          * If the hash is HASH_NB_ALWAYS, we always go to the next
508          * block so no check is necessary
509          */
510         while (num_frames--) {
511                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
512                                       0, &err)))
513                         return err; /* Failure */
514                 p++;
515                 brelse (p->bh);
516                 p->bh = bh;
517                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
518         }
519         return 1;
520 }
521
522
523 /*
524  * p is at least 6 bytes before the end of page
525  */
526 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
527 {
528         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
529 }
530
531 /*
532  * This function fills a red-black tree with information from a
533  * directory block.  It returns the number directory entries loaded
534  * into the tree.  If there is an error it is returned in err.
535  */
536 static int htree_dirblock_to_tree(struct file *dir_file,
537                                   struct inode *dir, int block,
538                                   struct dx_hash_info *hinfo,
539                                   __u32 start_hash, __u32 start_minor_hash)
540 {
541         struct buffer_head *bh;
542         struct ext3_dir_entry_2 *de, *top;
543         int err, count = 0;
544
545         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
546         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
547                 return err;
548
549         de = (struct ext3_dir_entry_2 *) bh->b_data;
550         top = (struct ext3_dir_entry_2 *) ((char *) de +
551                                            dir->i_sb->s_blocksize -
552                                            EXT3_DIR_REC_LEN(0));
553         for (; de < top; de = ext3_next_entry(de)) {
554                 ext3fs_dirhash(de->name, de->name_len, hinfo);
555                 if ((hinfo->hash < start_hash) ||
556                     ((hinfo->hash == start_hash) &&
557                      (hinfo->minor_hash < start_minor_hash)))
558                         continue;
559                 if (de->inode == 0)
560                         continue;
561                 if ((err = ext3_htree_store_dirent(dir_file,
562                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
563                         brelse(bh);
564                         return err;
565                 }
566                 count++;
567         }
568         brelse(bh);
569         return count;
570 }
571
572
573 /*
574  * This function fills a red-black tree with information from a
575  * directory.  We start scanning the directory in hash order, starting
576  * at start_hash and start_minor_hash.
577  *
578  * This function returns the number of entries inserted into the tree,
579  * or a negative error code.
580  */
581 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
582                          __u32 start_minor_hash, __u32 *next_hash)
583 {
584         struct dx_hash_info hinfo;
585         struct ext3_dir_entry_2 *de;
586         struct dx_frame frames[2], *frame;
587         struct inode *dir;
588         int block, err;
589         int count = 0;
590         int ret;
591         __u32 hashval;
592
593         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
594                        start_minor_hash));
595         dir = dir_file->f_dentry->d_inode;
596         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
597                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
598                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
599                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
600                                                start_hash, start_minor_hash);
601                 *next_hash = ~0;
602                 return count;
603         }
604         hinfo.hash = start_hash;
605         hinfo.minor_hash = 0;
606         frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
607         if (!frame)
608                 return err;
609
610         /* Add '.' and '..' from the htree header */
611         if (!start_hash && !start_minor_hash) {
612                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
613                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
614                         goto errout;
615                 count++;
616         }
617         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
618                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
619                 de = ext3_next_entry(de);
620                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
621                         goto errout;
622                 count++;
623         }
624
625         while (1) {
626                 block = dx_get_block(frame->at);
627                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
628                                              start_hash, start_minor_hash);
629                 if (ret < 0) {
630                         err = ret;
631                         goto errout;
632                 }
633                 count += ret;
634                 hashval = ~0;
635                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS, 
636                                             frame, frames, &hashval);
637                 *next_hash = hashval;
638                 if (ret < 0) {
639                         err = ret;
640                         goto errout;
641                 }
642                 /*
643                  * Stop if:  (a) there are no more entries, or
644                  * (b) we have inserted at least one entry and the
645                  * next hash value is not a continuation
646                  */
647                 if ((ret == 0) ||
648                     (count && ((hashval & 1) == 0)))
649                         break;
650         }
651         dx_release(frames);
652         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n", 
653                        count, *next_hash));
654         return count;
655 errout:
656         dx_release(frames);
657         return (err);
658 }
659
660
661 /*
662  * Directory block splitting, compacting
663  */
664
665 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
666                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
667 {
668         int count = 0;
669         char *base = (char *) de;
670         struct dx_hash_info h = *hinfo;
671
672         while ((char *) de < base + size)
673         {
674                 if (de->name_len && de->inode) {
675                         ext3fs_dirhash(de->name, de->name_len, &h);
676                         map_tail--;
677                         map_tail->hash = h.hash;
678                         map_tail->offs = (u32) ((char *) de - base);
679                         count++;
680                         cond_resched();
681                 }
682                 /* XXX: do we need to check rec_len == 0 case? -Chris */
683                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
684         }
685         return count;
686 }
687
688 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
689 {
690         struct dx_map_entry *p, *q, *top = map + count - 1;
691         int more;
692         /* Combsort until bubble sort doesn't suck */
693         while (count > 2)
694         {
695                 count = count*10/13;
696                 if (count - 9 < 2) /* 9, 10 -> 11 */
697                         count = 11;
698                 for (p = top, q = p - count; q >= map; p--, q--)
699                         if (p->hash < q->hash)
700                                 swap(*p, *q);
701         }
702         /* Garden variety bubble sort */
703         do {
704                 more = 0;
705                 q = top;
706                 while (q-- > map)
707                 {
708                         if (q[1].hash >= q[0].hash)
709                                 continue;
710                         swap(*(q+1), *q);
711                         more = 1;
712                 }
713         } while(more);
714 }
715
716 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
717 {
718         struct dx_entry *entries = frame->entries;
719         struct dx_entry *old = frame->at, *new = old + 1;
720         int count = dx_get_count(entries);
721
722         assert(count < dx_get_limit(entries));
723         assert(old < entries + count);
724         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
725         dx_set_hash(new, hash);
726         dx_set_block(new, block);
727         dx_set_count(entries, count + 1);
728 }
729 #endif
730
731
732 static void ext3_update_dx_flag(struct inode *inode)
733 {
734         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
735                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
736                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
737 }
738
739 /*
740  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
741  *
742  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
743  * `de != NULL' is guaranteed by caller.
744  */
745 static inline int ext3_match (int len, const char * const name,
746                               struct ext3_dir_entry_2 * de)
747 {
748         if (len != de->name_len)
749                 return 0;
750         if (!de->inode)
751                 return 0;
752         return !memcmp(name, de->name, len);
753 }
754
755 /*
756  * Returns 0 if not found, -1 on failure, and 1 on success
757  */
758 static inline int search_dirblock(struct buffer_head * bh,
759                                   struct inode *dir,
760                                   struct dentry *dentry,
761                                   unsigned long offset,
762                                   struct ext3_dir_entry_2 ** res_dir)
763 {
764         struct ext3_dir_entry_2 * de;
765         char * dlimit;
766         int de_len;
767         const char *name = dentry->d_name.name;
768         int namelen = dentry->d_name.len;
769
770         de = (struct ext3_dir_entry_2 *) bh->b_data;
771         dlimit = bh->b_data + dir->i_sb->s_blocksize;
772         while ((char *) de < dlimit) {
773                 /* this code is executed quadratically often */
774                 /* do minimal checking `by hand' */
775
776                 if ((char *) de + namelen <= dlimit &&
777                     ext3_match (namelen, name, de)) {
778                         /* found a match - just to be sure, do a full check */
779                         if (!ext3_check_dir_entry("ext3_find_entry",
780                                                   dir, de, bh, offset))
781                                 return -1;
782                         *res_dir = de;
783                         return 1;
784                 }
785                 /* prevent looping on a bad block */
786                 de_len = le16_to_cpu(de->rec_len);
787                 if (de_len <= 0)
788                         return -1;
789                 offset += de_len;
790                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
791         }
792         return 0;
793 }
794
795
796 /*
797  *      ext3_find_entry()
798  *
799  * finds an entry in the specified directory with the wanted name. It
800  * returns the cache buffer in which the entry was found, and the entry
801  * itself (as a parameter - res_dir). It does NOT read the inode of the
802  * entry - you'll have to do that yourself if you want to.
803  *
804  * The returned buffer_head has ->b_count elevated.  The caller is expected
805  * to brelse() it when appropriate.
806  */
807 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
808                                         struct ext3_dir_entry_2 ** res_dir)
809 {
810         struct super_block * sb;
811         struct buffer_head * bh_use[NAMEI_RA_SIZE];
812         struct buffer_head * bh, *ret = NULL;
813         unsigned long start, block, b;
814         int ra_max = 0;         /* Number of bh's in the readahead
815                                    buffer, bh_use[] */
816         int ra_ptr = 0;         /* Current index into readahead
817                                    buffer */
818         int num = 0;
819         int nblocks, i, err;
820         struct inode *dir = dentry->d_parent->d_inode;
821         int namelen;
822         const u8 *name;
823         unsigned blocksize;
824
825         *res_dir = NULL;
826         sb = dir->i_sb;
827         blocksize = sb->s_blocksize;
828         namelen = dentry->d_name.len;
829         name = dentry->d_name.name;
830         if (namelen > EXT3_NAME_LEN)
831                 return NULL;
832 #ifdef CONFIG_EXT3_INDEX
833         if (is_dx(dir)) {
834                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
835                 /*
836                  * On success, or if the error was file not found,
837                  * return.  Otherwise, fall back to doing a search the
838                  * old fashioned way.
839                  */
840                 if (bh || (err != ERR_BAD_DX_DIR))
841                         return bh;
842                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
843         }
844 #endif
845         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
846         start = EXT3_I(dir)->i_dir_start_lookup;
847         if (start >= nblocks)
848                 start = 0;
849         block = start;
850 restart:
851         do {
852                 /*
853                  * We deal with the read-ahead logic here.
854                  */
855                 if (ra_ptr >= ra_max) {
856                         /* Refill the readahead buffer */
857                         ra_ptr = 0;
858                         b = block;
859                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
860                                 /*
861                                  * Terminate if we reach the end of the
862                                  * directory and must wrap, or if our
863                                  * search has finished at this block.
864                                  */
865                                 if (b >= nblocks || (num && block == start)) {
866                                         bh_use[ra_max] = NULL;
867                                         break;
868                                 }
869                                 num++;
870                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
871                                 bh_use[ra_max] = bh;
872                                 if (bh)
873                                         ll_rw_block(READ, 1, &bh);
874                         }
875                 }
876                 if ((bh = bh_use[ra_ptr++]) == NULL)
877                         goto next;
878                 wait_on_buffer(bh);
879                 if (!buffer_uptodate(bh)) {
880                         /* read error, skip block & hope for the best */
881                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
882                                    "offset %lu", dir->i_ino, block);
883                         brelse(bh);
884                         goto next;
885                 }
886                 i = search_dirblock(bh, dir, dentry,
887                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
888                 if (i == 1) {
889                         EXT3_I(dir)->i_dir_start_lookup = block;
890                         ret = bh;
891                         goto cleanup_and_exit;
892                 } else {
893                         brelse(bh);
894                         if (i < 0)
895                                 goto cleanup_and_exit;
896                 }
897         next:
898                 if (++block >= nblocks)
899                         block = 0;
900         } while (block != start);
901
902         /*
903          * If the directory has grown while we were searching, then
904          * search the last part of the directory before giving up.
905          */
906         block = nblocks;
907         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
908         if (block < nblocks) {
909                 start = 0;
910                 goto restart;
911         }
912
913 cleanup_and_exit:
914         /* Clean up the read-ahead blocks */
915         for (; ra_ptr < ra_max; ra_ptr++)
916                 brelse (bh_use[ra_ptr]);
917         return ret;
918 }
919
920 #ifdef CONFIG_EXT3_INDEX
921 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
922                        struct ext3_dir_entry_2 **res_dir, int *err)
923 {
924         struct super_block * sb;
925         struct dx_hash_info     hinfo;
926         u32 hash;
927         struct dx_frame frames[2], *frame;
928         struct ext3_dir_entry_2 *de, *top;
929         struct buffer_head *bh;
930         unsigned long block;
931         int retval;
932         int namelen = dentry->d_name.len;
933         const u8 *name = dentry->d_name.name;
934         struct inode *dir = dentry->d_parent->d_inode;
935
936         sb = dir->i_sb;
937         /* NFS may look up ".." - look at dx_root directory block */
938         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
939                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
940                         return NULL;
941         } else {
942                 frame = frames;
943                 frame->bh = NULL;                       /* for dx_release() */
944                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
945                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
946         }
947         hash = hinfo.hash;
948         do {
949                 block = dx_get_block(frame->at);
950                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
951                         goto errout;
952                 de = (struct ext3_dir_entry_2 *) bh->b_data;
953                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
954                                        EXT3_DIR_REC_LEN(0));
955                 for (; de < top; de = ext3_next_entry(de))
956                 if (ext3_match (namelen, name, de)) {
957                         if (!ext3_check_dir_entry("ext3_find_entry",
958                                                   dir, de, bh,
959                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
960                                           +((char *)de - bh->b_data))) {
961                                 brelse (bh);
962                                 goto errout;
963                         }
964                         *res_dir = de;
965                         dx_release (frames);
966                         return bh;
967                 }
968                 brelse (bh);
969                 /* Check to see if we should continue to search */
970                 retval = ext3_htree_next_block(dir, hash, frame,
971                                                frames, NULL);
972                 if (retval < 0) {
973                         ext3_warning(sb, __FUNCTION__,
974                              "error reading index page in directory #%lu",
975                              dir->i_ino);
976                         *err = retval;
977                         goto errout;
978                 }
979         } while (retval == 1);
980
981         *err = -ENOENT;
982 errout:
983         dxtrace(printk("%s not found\n", name));
984         dx_release (frames);
985         return NULL;
986 }
987 #endif
988
989 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
990 {
991         struct inode * inode;
992         struct ext3_dir_entry_2 * de;
993         struct buffer_head * bh;
994
995         if (dentry->d_name.len > EXT3_NAME_LEN)
996                 return ERR_PTR(-ENAMETOOLONG);
997
998         bh = ext3_find_entry(dentry, &de);
999         inode = NULL;
1000         if (bh) {
1001                 unsigned long ino = le32_to_cpu(de->inode);
1002                 brelse (bh);
1003                 inode = iget(dir->i_sb, ino);
1004
1005                 if (!inode)
1006                         return ERR_PTR(-EACCES);
1007         }
1008         return d_splice_alias(inode, dentry);
1009 }
1010
1011
1012 struct dentry *ext3_get_parent(struct dentry *child)
1013 {
1014         unsigned long ino;
1015         struct dentry *parent;
1016         struct inode *inode;
1017         struct dentry dotdot;
1018         struct ext3_dir_entry_2 * de;
1019         struct buffer_head *bh;
1020
1021         dotdot.d_name.name = "..";
1022         dotdot.d_name.len = 2;
1023         dotdot.d_parent = child; /* confusing, isn't it! */
1024
1025         bh = ext3_find_entry(&dotdot, &de);
1026         inode = NULL;
1027         if (!bh)
1028                 return ERR_PTR(-ENOENT);
1029         ino = le32_to_cpu(de->inode);
1030         brelse(bh);
1031         inode = iget(child->d_inode->i_sb, ino);
1032
1033         if (!inode)
1034                 return ERR_PTR(-EACCES);
1035
1036         parent = d_alloc_anon(inode);
1037         if (!parent) {
1038                 iput(inode);
1039                 parent = ERR_PTR(-ENOMEM);
1040         }
1041         return parent;
1042
1043
1044 #define S_SHIFT 12
1045 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1046         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1047         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1048         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1049         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1050         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1051         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1052         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1053 };
1054
1055 static inline void ext3_set_de_type(struct super_block *sb,
1056                                 struct ext3_dir_entry_2 *de,
1057                                 umode_t mode) {
1058         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1059                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1060 }
1061
1062 #ifdef CONFIG_EXT3_INDEX
1063 static struct ext3_dir_entry_2 *
1064 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1065 {
1066         unsigned rec_len = 0;
1067
1068         while (count--) {
1069                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1070                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1071                 memcpy (to, de, rec_len);
1072                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1073                                 cpu_to_le16(rec_len);
1074                 de->inode = 0;
1075                 map++;
1076                 to += rec_len;
1077         }
1078         return (struct ext3_dir_entry_2 *) (to - rec_len);
1079 }
1080
1081 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1082 {
1083         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1084         unsigned rec_len = 0;
1085
1086         prev = to = de;
1087         while ((char*)de < base + size) {
1088                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1089                                                     le16_to_cpu(de->rec_len));
1090                 if (de->inode && de->name_len) {
1091                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1092                         if (de > to)
1093                                 memmove(to, de, rec_len);
1094                         to->rec_len = cpu_to_le16(rec_len);
1095                         prev = to;
1096                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1097                 }
1098                 de = next;
1099         }
1100         return prev;
1101 }
1102
1103 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1104                         struct buffer_head **bh,struct dx_frame *frame,
1105                         struct dx_hash_info *hinfo, int *error)
1106 {
1107         unsigned blocksize = dir->i_sb->s_blocksize;
1108         unsigned count, continued;
1109         struct buffer_head *bh2;
1110         u32 newblock;
1111         u32 hash2;
1112         struct dx_map_entry *map;
1113         char *data1 = (*bh)->b_data, *data2;
1114         unsigned split;
1115         struct ext3_dir_entry_2 *de = NULL, *de2;
1116         int     err;
1117
1118         bh2 = ext3_append (handle, dir, &newblock, error);
1119         if (!(bh2)) {
1120                 brelse(*bh);
1121                 *bh = NULL;
1122                 goto errout;
1123         }
1124
1125         BUFFER_TRACE(*bh, "get_write_access");
1126         err = ext3_journal_get_write_access(handle, *bh);
1127         if (err) {
1128         journal_error:
1129                 brelse(*bh);
1130                 brelse(bh2);
1131                 *bh = NULL;
1132                 ext3_std_error(dir->i_sb, err);
1133                 goto errout;
1134         }
1135         BUFFER_TRACE(frame->bh, "get_write_access");
1136         err = ext3_journal_get_write_access(handle, frame->bh);
1137         if (err)
1138                 goto journal_error;
1139
1140         data2 = bh2->b_data;
1141
1142         /* create map in the end of data2 block */
1143         map = (struct dx_map_entry *) (data2 + blocksize);
1144         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1145                              blocksize, hinfo, map);
1146         map -= count;
1147         split = count/2; // need to adjust to actual middle
1148         dx_sort_map (map, count);
1149         hash2 = map[split].hash;
1150         continued = hash2 == map[split - 1].hash;
1151         dxtrace(printk("Split block %i at %x, %i/%i\n",
1152                 dx_get_block(frame->at), hash2, split, count-split));
1153
1154         /* Fancy dance to stay within two buffers */
1155         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1156         de = dx_pack_dirents(data1,blocksize);
1157         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1158         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1159         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1160         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1161
1162         /* Which block gets the new entry? */
1163         if (hinfo->hash >= hash2)
1164         {
1165                 swap(*bh, bh2);
1166                 de = de2;
1167         }
1168         dx_insert_block (frame, hash2 + continued, newblock);
1169         err = ext3_journal_dirty_metadata (handle, bh2);
1170         if (err)
1171                 goto journal_error;
1172         err = ext3_journal_dirty_metadata (handle, frame->bh);
1173         if (err)
1174                 goto journal_error;
1175         brelse (bh2);
1176         dxtrace(dx_show_index ("frame", frame->entries));
1177 errout:
1178         return de;
1179 }
1180 #endif
1181
1182
1183 /*
1184  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1185  * it points to a directory entry which is guaranteed to be large
1186  * enough for new directory entry.  If de is NULL, then
1187  * add_dirent_to_buf will attempt search the directory block for
1188  * space.  It will return -ENOSPC if no space is available, and -EIO
1189  * and -EEXIST if directory entry already exists.
1190  * 
1191  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1192  * all other cases bh is released.
1193  */
1194 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1195                              struct inode *inode, struct ext3_dir_entry_2 *de,
1196                              struct buffer_head * bh)
1197 {
1198         struct inode    *dir = dentry->d_parent->d_inode;
1199         const char      *name = dentry->d_name.name;
1200         int             namelen = dentry->d_name.len;
1201         unsigned long   offset = 0;
1202         unsigned short  reclen;
1203         int             nlen, rlen, err;
1204         char            *top;
1205
1206         reclen = EXT3_DIR_REC_LEN(namelen);
1207         if (!de) {
1208                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1209                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1210                 while ((char *) de <= top) {
1211                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1212                                                   bh, offset)) {
1213                                 brelse (bh);
1214                                 return -EIO;
1215                         }
1216                         if (ext3_match (namelen, name, de)) {
1217                                 brelse (bh);
1218                                 return -EEXIST;
1219                         }
1220                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1221                         rlen = le16_to_cpu(de->rec_len);
1222                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1223                                 break;
1224                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1225                         offset += rlen;
1226                 }
1227                 if ((char *) de > top)
1228                         return -ENOSPC;
1229         }
1230         BUFFER_TRACE(bh, "get_write_access");
1231         err = ext3_journal_get_write_access(handle, bh);
1232         if (err) {
1233                 ext3_std_error(dir->i_sb, err);
1234                 brelse(bh);
1235                 return err;
1236         }
1237
1238         /* By now the buffer is marked for journaling */
1239         nlen = EXT3_DIR_REC_LEN(de->name_len);
1240         rlen = le16_to_cpu(de->rec_len);
1241         if (de->inode) {
1242                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1243                 de1->rec_len = cpu_to_le16(rlen - nlen);
1244                 de->rec_len = cpu_to_le16(nlen);
1245                 de = de1;
1246         }
1247         de->file_type = EXT3_FT_UNKNOWN;
1248         if (inode) {
1249                 de->inode = cpu_to_le32(inode->i_ino);
1250                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1251         } else
1252                 de->inode = 0;
1253         de->name_len = namelen;
1254         memcpy (de->name, name, namelen);
1255         /*
1256          * XXX shouldn't update any times until successful
1257          * completion of syscall, but too many callers depend
1258          * on this.
1259          *
1260          * XXX similarly, too many callers depend on
1261          * ext3_new_inode() setting the times, but error
1262          * recovery deletes the inode, so the worst that can
1263          * happen is that the times are slightly out of date
1264          * and/or different from the directory change time.
1265          */
1266         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1267         ext3_update_dx_flag(dir);
1268         dir->i_version++;
1269         ext3_mark_inode_dirty(handle, dir);
1270         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1271         err = ext3_journal_dirty_metadata(handle, bh);
1272         if (err)
1273                 ext3_std_error(dir->i_sb, err);
1274         brelse(bh);
1275         return 0;
1276 }
1277
1278 #ifdef CONFIG_EXT3_INDEX
1279 /*
1280  * This converts a one block unindexed directory to a 3 block indexed
1281  * directory, and adds the dentry to the indexed directory.
1282  */
1283 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1284                             struct inode *inode, struct buffer_head *bh)
1285 {
1286         struct inode    *dir = dentry->d_parent->d_inode;
1287         const char      *name = dentry->d_name.name;
1288         int             namelen = dentry->d_name.len;
1289         struct buffer_head *bh2;
1290         struct dx_root  *root;
1291         struct dx_frame frames[2], *frame;
1292         struct dx_entry *entries;
1293         struct ext3_dir_entry_2 *de, *de2;
1294         char            *data1, *top;
1295         unsigned        len;
1296         int             retval;
1297         unsigned        blocksize;
1298         struct dx_hash_info hinfo;
1299         u32             block;
1300         struct fake_dirent *fde;
1301
1302         blocksize =  dir->i_sb->s_blocksize;
1303         dxtrace(printk("Creating index\n"));
1304         retval = ext3_journal_get_write_access(handle, bh);
1305         if (retval) {
1306                 ext3_std_error(dir->i_sb, retval);
1307                 brelse(bh);
1308                 return retval;
1309         }
1310         root = (struct dx_root *) bh->b_data;
1311
1312         bh2 = ext3_append (handle, dir, &block, &retval);
1313         if (!(bh2)) {
1314                 brelse(bh);
1315                 return retval;
1316         }
1317         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1318         data1 = bh2->b_data;
1319
1320         /* The 0th block becomes the root, move the dirents out */
1321         fde = &root->dotdot;
1322         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1323         len = ((char *) root) + blocksize - (char *) de;
1324         memcpy (data1, de, len);
1325         de = (struct ext3_dir_entry_2 *) data1;
1326         top = data1 + len;
1327         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1328                 de = de2;
1329         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1330         /* Initialize the root; the dot dirents already exist */
1331         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1332         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1333         memset (&root->info, 0, sizeof(root->info));
1334         root->info.info_length = sizeof(root->info);
1335         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1336         entries = root->entries;
1337         dx_set_block (entries, 1);
1338         dx_set_count (entries, 1);
1339         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1340
1341         /* Initialize as for dx_probe */
1342         hinfo.hash_version = root->info.hash_version;
1343         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1344         ext3fs_dirhash(name, namelen, &hinfo);
1345         frame = frames;
1346         frame->entries = entries;
1347         frame->at = entries;
1348         frame->bh = bh;
1349         bh = bh2;
1350         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1351         dx_release (frames);
1352         if (!(de))
1353                 return retval;
1354
1355         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1356 }
1357 #endif
1358
1359 /*
1360  *      ext3_add_entry()
1361  *
1362  * adds a file entry to the specified directory, using the same
1363  * semantics as ext3_find_entry(). It returns NULL if it failed.
1364  *
1365  * NOTE!! The inode part of 'de' is left at 0 - which means you
1366  * may not sleep between calling this and putting something into
1367  * the entry, as someone else might have used it while you slept.
1368  */
1369 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1370         struct inode *inode)
1371 {
1372         struct inode *dir = dentry->d_parent->d_inode;
1373         unsigned long offset;
1374         struct buffer_head * bh;
1375         struct ext3_dir_entry_2 *de;
1376         struct super_block * sb;
1377         int     retval;
1378 #ifdef CONFIG_EXT3_INDEX
1379         int     dx_fallback=0;
1380 #endif
1381         unsigned blocksize;
1382         u32 block, blocks;
1383
1384         sb = dir->i_sb;
1385         blocksize = sb->s_blocksize;
1386         if (!dentry->d_name.len)
1387                 return -EINVAL;
1388 #ifdef CONFIG_EXT3_INDEX
1389         if (is_dx(dir)) {
1390                 retval = ext3_dx_add_entry(handle, dentry, inode);
1391                 if (!retval || (retval != ERR_BAD_DX_DIR))
1392                         return retval;
1393                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1394                 dx_fallback++;
1395                 ext3_mark_inode_dirty(handle, dir);
1396         }
1397 #endif
1398         blocks = dir->i_size >> sb->s_blocksize_bits;
1399         for (block = 0, offset = 0; block < blocks; block++) {
1400                 bh = ext3_bread(handle, dir, block, 0, &retval);
1401                 if(!bh)
1402                         return retval;
1403                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1404                 if (retval != -ENOSPC)
1405                         return retval;
1406
1407 #ifdef CONFIG_EXT3_INDEX
1408                 if (blocks == 1 && !dx_fallback &&
1409                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1410                         return make_indexed_dir(handle, dentry, inode, bh);
1411 #endif
1412                 brelse(bh);
1413         }
1414         bh = ext3_append(handle, dir, &block, &retval);
1415         if (!bh)
1416                 return retval;
1417         de = (struct ext3_dir_entry_2 *) bh->b_data;
1418         de->inode = 0;
1419         de->rec_len = cpu_to_le16(blocksize);
1420         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1421 }
1422
1423 #ifdef CONFIG_EXT3_INDEX
1424 /*
1425  * Returns 0 for success, or a negative error value
1426  */
1427 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1428                              struct inode *inode)
1429 {
1430         struct dx_frame frames[2], *frame;
1431         struct dx_entry *entries, *at;
1432         struct dx_hash_info hinfo;
1433         struct buffer_head * bh;
1434         struct inode *dir = dentry->d_parent->d_inode;
1435         struct super_block * sb = dir->i_sb;
1436         struct ext3_dir_entry_2 *de;
1437         int err;
1438
1439         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1440         if (!frame)
1441                 return err;
1442         entries = frame->entries;
1443         at = frame->at;
1444
1445         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1446                 goto cleanup;
1447
1448         BUFFER_TRACE(bh, "get_write_access");
1449         err = ext3_journal_get_write_access(handle, bh);
1450         if (err)
1451                 goto journal_error;
1452
1453         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1454         if (err != -ENOSPC) {
1455                 bh = NULL;
1456                 goto cleanup;
1457         }
1458
1459         /* Block full, should compress but for now just split */
1460         dxtrace(printk("using %u of %u node entries\n",
1461                        dx_get_count(entries), dx_get_limit(entries)));
1462         /* Need to split index? */
1463         if (dx_get_count(entries) == dx_get_limit(entries)) {
1464                 u32 newblock;
1465                 unsigned icount = dx_get_count(entries);
1466                 int levels = frame - frames;
1467                 struct dx_entry *entries2;
1468                 struct dx_node *node2;
1469                 struct buffer_head *bh2;
1470
1471                 if (levels && (dx_get_count(frames->entries) ==
1472                                dx_get_limit(frames->entries))) {
1473                         ext3_warning(sb, __FUNCTION__,
1474                                      "Directory index full!");
1475                         err = -ENOSPC;
1476                         goto cleanup;
1477                 }
1478                 bh2 = ext3_append (handle, dir, &newblock, &err);
1479                 if (!(bh2))
1480                         goto cleanup;
1481                 node2 = (struct dx_node *)(bh2->b_data);
1482                 entries2 = node2->entries;
1483                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1484                 node2->fake.inode = 0;
1485                 BUFFER_TRACE(frame->bh, "get_write_access");
1486                 err = ext3_journal_get_write_access(handle, frame->bh);
1487                 if (err)
1488                         goto journal_error;
1489                 if (levels) {
1490                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1491                         unsigned hash2 = dx_get_hash(entries + icount1);
1492                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1493
1494                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1495                         err = ext3_journal_get_write_access(handle,
1496                                                              frames[0].bh);
1497                         if (err)
1498                                 goto journal_error;
1499
1500                         memcpy ((char *) entries2, (char *) (entries + icount1),
1501                                 icount2 * sizeof(struct dx_entry));
1502                         dx_set_count (entries, icount1);
1503                         dx_set_count (entries2, icount2);
1504                         dx_set_limit (entries2, dx_node_limit(dir));
1505
1506                         /* Which index block gets the new entry? */
1507                         if (at - entries >= icount1) {
1508                                 frame->at = at = at - entries - icount1 + entries2;
1509                                 frame->entries = entries = entries2;
1510                                 swap(frame->bh, bh2);
1511                         }
1512                         dx_insert_block (frames + 0, hash2, newblock);
1513                         dxtrace(dx_show_index ("node", frames[1].entries));
1514                         dxtrace(dx_show_index ("node",
1515                                ((struct dx_node *) bh2->b_data)->entries));
1516                         err = ext3_journal_dirty_metadata(handle, bh2);
1517                         if (err)
1518                                 goto journal_error;
1519                         brelse (bh2);
1520                 } else {
1521                         dxtrace(printk("Creating second level index...\n"));
1522                         memcpy((char *) entries2, (char *) entries,
1523                                icount * sizeof(struct dx_entry));
1524                         dx_set_limit(entries2, dx_node_limit(dir));
1525
1526                         /* Set up root */
1527                         dx_set_count(entries, 1);
1528                         dx_set_block(entries + 0, newblock);
1529                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1530
1531                         /* Add new access path frame */
1532                         frame = frames + 1;
1533                         frame->at = at = at - entries + entries2;
1534                         frame->entries = entries = entries2;
1535                         frame->bh = bh2;
1536                         err = ext3_journal_get_write_access(handle,
1537                                                              frame->bh);
1538                         if (err)
1539                                 goto journal_error;
1540                 }
1541                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1542         }
1543         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1544         if (!de)
1545                 goto cleanup;
1546         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1547         bh = NULL;
1548         goto cleanup;
1549
1550 journal_error:
1551         ext3_std_error(dir->i_sb, err);
1552 cleanup:
1553         if (bh)
1554                 brelse(bh);
1555         dx_release(frames);
1556         return err;
1557 }
1558 #endif
1559
1560 /*
1561  * ext3_delete_entry deletes a directory entry by merging it with the
1562  * previous entry
1563  */
1564 static int ext3_delete_entry (handle_t *handle, 
1565                               struct inode * dir,
1566                               struct ext3_dir_entry_2 * de_del,
1567                               struct buffer_head * bh)
1568 {
1569         struct ext3_dir_entry_2 * de, * pde;
1570         int i;
1571
1572         i = 0;
1573         pde = NULL;
1574         de = (struct ext3_dir_entry_2 *) bh->b_data;
1575         while (i < bh->b_size) {
1576                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1577                         return -EIO;
1578                 if (de == de_del)  {
1579                         BUFFER_TRACE(bh, "get_write_access");
1580                         ext3_journal_get_write_access(handle, bh);
1581                         if (pde)
1582                                 pde->rec_len =
1583                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1584                                                     le16_to_cpu(de->rec_len));
1585                         else
1586                                 de->inode = 0;
1587                         dir->i_version++;
1588                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1589                         ext3_journal_dirty_metadata(handle, bh);
1590                         return 0;
1591                 }
1592                 i += le16_to_cpu(de->rec_len);
1593                 pde = de;
1594                 de = (struct ext3_dir_entry_2 *)
1595                         ((char *) de + le16_to_cpu(de->rec_len));
1596         }
1597         return -ENOENT;
1598 }
1599
1600 /*
1601  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1602  * do not perform it in these functions.  We perform it at the call site,
1603  * if it is needed.
1604  */
1605 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1606 {
1607         inode->i_nlink++;
1608 }
1609
1610 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1611 {
1612         inode->i_nlink--;
1613 }
1614
1615 static int ext3_add_nondir(handle_t *handle,
1616                 struct dentry *dentry, struct inode *inode)
1617 {
1618         int err = ext3_add_entry(handle, dentry, inode);
1619         if (!err) {
1620                 ext3_mark_inode_dirty(handle, inode);
1621                 d_instantiate(dentry, inode);
1622                 return 0;
1623         }
1624         ext3_dec_count(handle, inode);
1625         iput(inode);
1626         return err;
1627 }
1628
1629 /*
1630  * By the time this is called, we already have created
1631  * the directory cache entry for the new file, but it
1632  * is so far negative - it has no inode.
1633  *
1634  * If the create succeeds, we fill in the inode information
1635  * with d_instantiate(). 
1636  */
1637 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1638                 struct nameidata *nd)
1639 {
1640         handle_t *handle; 
1641         struct inode * inode;
1642         int err, retries = 0;
1643
1644 retry:
1645         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1646                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1647                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1648         if (IS_ERR(handle))
1649                 return PTR_ERR(handle);
1650
1651         if (IS_DIRSYNC(dir))
1652                 handle->h_sync = 1;
1653
1654         inode = ext3_new_inode (handle, dir, mode);
1655         err = PTR_ERR(inode);
1656         if (!IS_ERR(inode)) {
1657                 inode->i_op = &ext3_file_inode_operations;
1658                 inode->i_fop = &ext3_file_operations;
1659                 ext3_set_aops(inode);
1660                 err = ext3_add_nondir(handle, dentry, inode);
1661         }
1662         ext3_journal_stop(handle);
1663         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1664                 goto retry;
1665         return err;
1666 }
1667
1668 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1669                         int mode, dev_t rdev)
1670 {
1671         handle_t *handle;
1672         struct inode *inode;
1673         int err, retries = 0;
1674
1675         if (!new_valid_dev(rdev))
1676                 return -EINVAL;
1677
1678 retry:
1679         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1680                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1681                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1682         if (IS_ERR(handle))
1683                 return PTR_ERR(handle);
1684
1685         if (IS_DIRSYNC(dir))
1686                 handle->h_sync = 1;
1687
1688         inode = ext3_new_inode (handle, dir, mode);
1689         err = PTR_ERR(inode);
1690         if (!IS_ERR(inode)) {
1691                 init_special_inode(inode, inode->i_mode, rdev);
1692 #ifdef CONFIG_EXT3_FS_XATTR
1693                 inode->i_op = &ext3_special_inode_operations;
1694 #endif
1695                 err = ext3_add_nondir(handle, dentry, inode);
1696         }
1697         ext3_journal_stop(handle);
1698         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1699                 goto retry;
1700         return err;
1701 }
1702
1703 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1704 {
1705         handle_t *handle;
1706         struct inode * inode;
1707         struct buffer_head * dir_block;
1708         struct ext3_dir_entry_2 * de;
1709         int err, retries = 0;
1710
1711         if (dir->i_nlink >= EXT3_LINK_MAX)
1712                 return -EMLINK;
1713
1714 retry:
1715         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1716                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1717                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1718         if (IS_ERR(handle))
1719                 return PTR_ERR(handle);
1720
1721         if (IS_DIRSYNC(dir))
1722                 handle->h_sync = 1;
1723
1724         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1725         err = PTR_ERR(inode);
1726         if (IS_ERR(inode))
1727                 goto out_stop;
1728
1729         inode->i_op = &ext3_dir_inode_operations;
1730         inode->i_fop = &ext3_dir_operations;
1731         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1732         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1733         if (!dir_block) {
1734                 inode->i_nlink--; /* is this nlink == 0? */
1735                 ext3_mark_inode_dirty(handle, inode);
1736                 iput (inode);
1737                 goto out_stop;
1738         }
1739         BUFFER_TRACE(dir_block, "get_write_access");
1740         ext3_journal_get_write_access(handle, dir_block);
1741         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1742         de->inode = cpu_to_le32(inode->i_ino);
1743         de->name_len = 1;
1744         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1745         strcpy (de->name, ".");
1746         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1747         de = (struct ext3_dir_entry_2 *)
1748                         ((char *) de + le16_to_cpu(de->rec_len));
1749         de->inode = cpu_to_le32(dir->i_ino);
1750         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1751         de->name_len = 2;
1752         strcpy (de->name, "..");
1753         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1754         inode->i_nlink = 2;
1755         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1756         ext3_journal_dirty_metadata(handle, dir_block);
1757         brelse (dir_block);
1758         ext3_mark_inode_dirty(handle, inode);
1759         err = ext3_add_entry (handle, dentry, inode);
1760         if (err) {
1761                 inode->i_nlink = 0;
1762                 ext3_mark_inode_dirty(handle, inode);
1763                 iput (inode);
1764                 goto out_stop;
1765         }
1766         dir->i_nlink++;
1767         ext3_update_dx_flag(dir);
1768         ext3_mark_inode_dirty(handle, dir);
1769         d_instantiate(dentry, inode);
1770 out_stop:
1771         ext3_journal_stop(handle);
1772         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1773                 goto retry;
1774         return err;
1775 }
1776
1777 /*
1778  * routine to check that the specified directory is empty (for rmdir)
1779  */
1780 static int empty_dir (struct inode * inode)
1781 {
1782         unsigned long offset;
1783         struct buffer_head * bh;
1784         struct ext3_dir_entry_2 * de, * de1;
1785         struct super_block * sb;
1786         int err = 0;
1787
1788         sb = inode->i_sb;
1789         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1790             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1791                 if (err)
1792                         ext3_error(inode->i_sb, __FUNCTION__,
1793                                    "error %d reading directory #%lu offset 0",
1794                                    err, inode->i_ino);
1795                 else
1796                         ext3_warning(inode->i_sb, __FUNCTION__,
1797                                      "bad directory (dir #%lu) - no data block",
1798                                      inode->i_ino);
1799                 return 1;
1800         }
1801         de = (struct ext3_dir_entry_2 *) bh->b_data;
1802         de1 = (struct ext3_dir_entry_2 *)
1803                         ((char *) de + le16_to_cpu(de->rec_len));
1804         if (le32_to_cpu(de->inode) != inode->i_ino ||
1805                         !le32_to_cpu(de1->inode) || 
1806                         strcmp (".", de->name) ||
1807                         strcmp ("..", de1->name)) {
1808                 ext3_warning (inode->i_sb, "empty_dir",
1809                               "bad directory (dir #%lu) - no `.' or `..'",
1810                               inode->i_ino);
1811                 brelse (bh);
1812                 return 1;
1813         }
1814         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1815         de = (struct ext3_dir_entry_2 *)
1816                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1817         while (offset < inode->i_size ) {
1818                 if (!bh ||
1819                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1820                         err = 0;
1821                         brelse (bh);
1822                         bh = ext3_bread (NULL, inode,
1823                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1824                         if (!bh) {
1825                                 if (err)
1826                                         ext3_error(sb, __FUNCTION__,
1827                                                    "error %d reading directory"
1828                                                    " #%lu offset %lu",
1829                                                    err, inode->i_ino, offset);
1830                                 offset += sb->s_blocksize;
1831                                 continue;
1832                         }
1833                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1834                 }
1835                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1836                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1837                                                          sb->s_blocksize);
1838                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1839                         continue;
1840                 }
1841                 if (le32_to_cpu(de->inode)) {
1842                         brelse (bh);
1843                         return 0;
1844                 }
1845                 offset += le16_to_cpu(de->rec_len);
1846                 de = (struct ext3_dir_entry_2 *)
1847                                 ((char *) de + le16_to_cpu(de->rec_len));
1848         }
1849         brelse (bh);
1850         return 1;
1851 }
1852
1853 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1854  * such inodes, starting at the superblock, in case we crash before the
1855  * file is closed/deleted, or in case the inode truncate spans multiple
1856  * transactions and the last transaction is not recovered after a crash.
1857  *
1858  * At filesystem recovery time, we walk this list deleting unlinked
1859  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1860  */
1861 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1862 {
1863         struct super_block *sb = inode->i_sb;
1864         struct ext3_iloc iloc;
1865         int err = 0, rc;
1866
1867         lock_super(sb);
1868         if (!list_empty(&EXT3_I(inode)->i_orphan))
1869                 goto out_unlock;
1870
1871         /* Orphan handling is only valid for files with data blocks
1872          * being truncated, or files being unlinked. */
1873
1874         /* @@@ FIXME: Observation from aviro:
1875          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block 
1876          * here (on lock_super()), so race with ext3_link() which might bump
1877          * ->i_nlink. For, say it, character device. Not a regular file,
1878          * not a directory, not a symlink and ->i_nlink > 0.
1879          */
1880         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1881                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1882
1883         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1884         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1885         if (err)
1886                 goto out_unlock;
1887
1888         err = ext3_reserve_inode_write(handle, inode, &iloc);
1889         if (err)
1890                 goto out_unlock;
1891
1892         /* Insert this inode at the head of the on-disk orphan list... */
1893         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1894         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1895         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1896         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1897         if (!err)
1898                 err = rc;
1899
1900         /* Only add to the head of the in-memory list if all the
1901          * previous operations succeeded.  If the orphan_add is going to
1902          * fail (possibly taking the journal offline), we can't risk
1903          * leaving the inode on the orphan list: stray orphan-list
1904          * entries can cause panics at unmount time.
1905          *
1906          * This is safe: on error we're going to ignore the orphan list
1907          * anyway on the next recovery. */
1908         if (!err)
1909                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1910
1911         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1912         jbd_debug(4, "orphan inode %ld will point to %d\n",
1913                         inode->i_ino, NEXT_ORPHAN(inode));
1914 out_unlock:
1915         unlock_super(sb);
1916         ext3_std_error(inode->i_sb, err);
1917         return err;
1918 }
1919
1920 /*
1921  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1922  * of such inodes stored on disk, because it is finally being cleaned up.
1923  */
1924 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1925 {
1926         struct list_head *prev;
1927         struct ext3_inode_info *ei = EXT3_I(inode);
1928         struct ext3_sb_info *sbi;
1929         unsigned long ino_next;
1930         struct ext3_iloc iloc;
1931         int err = 0;
1932
1933         lock_super(inode->i_sb);
1934         if (list_empty(&ei->i_orphan)) {
1935                 unlock_super(inode->i_sb);
1936                 return 0;
1937         }
1938
1939         ino_next = NEXT_ORPHAN(inode);
1940         prev = ei->i_orphan.prev;
1941         sbi = EXT3_SB(inode->i_sb);
1942
1943         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1944
1945         list_del_init(&ei->i_orphan);
1946
1947         /* If we're on an error path, we may not have a valid
1948          * transaction handle with which to update the orphan list on
1949          * disk, but we still need to remove the inode from the linked
1950          * list in memory. */
1951         if (!handle)
1952                 goto out;
1953
1954         err = ext3_reserve_inode_write(handle, inode, &iloc);
1955         if (err)
1956                 goto out_err;
1957
1958         if (prev == &sbi->s_orphan) {
1959                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1960                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1961                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1962                 if (err)
1963                         goto out_brelse;
1964                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1965                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1966         } else {
1967                 struct ext3_iloc iloc2;
1968                 struct inode *i_prev =
1969                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1970
1971                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1972                           i_prev->i_ino, ino_next);
1973                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1974                 if (err)
1975                         goto out_brelse;
1976                 NEXT_ORPHAN(i_prev) = ino_next;
1977                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1978         }
1979         if (err)
1980                 goto out_brelse;
1981         NEXT_ORPHAN(inode) = 0;
1982         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1983
1984 out_err:
1985         ext3_std_error(inode->i_sb, err);
1986 out:
1987         unlock_super(inode->i_sb);
1988         return err;
1989
1990 out_brelse:
1991         brelse(iloc.bh);
1992         goto out_err;
1993 }
1994
1995 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
1996 {
1997         int retval;
1998         struct inode * inode;
1999         struct buffer_head * bh;
2000         struct ext3_dir_entry_2 * de;
2001         handle_t *handle;
2002
2003         /* Initialize quotas before so that eventual writes go in
2004          * separate transaction */
2005         DQUOT_INIT(dentry->d_inode);
2006         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2007         if (IS_ERR(handle))
2008                 return PTR_ERR(handle);
2009
2010         retval = -ENOENT;
2011         bh = ext3_find_entry (dentry, &de);
2012         if (!bh)
2013                 goto end_rmdir;
2014
2015         if (IS_DIRSYNC(dir))
2016                 handle->h_sync = 1;
2017
2018         inode = dentry->d_inode;
2019
2020         retval = -EIO;
2021         if (le32_to_cpu(de->inode) != inode->i_ino)
2022                 goto end_rmdir;
2023
2024         retval = -ENOTEMPTY;
2025         if (!empty_dir (inode))
2026                 goto end_rmdir;
2027
2028         retval = ext3_delete_entry(handle, dir, de, bh);
2029         if (retval)
2030                 goto end_rmdir;
2031         if (inode->i_nlink != 2)
2032                 ext3_warning (inode->i_sb, "ext3_rmdir",
2033                               "empty directory has nlink!=2 (%d)",
2034                               inode->i_nlink);
2035         inode->i_version++;
2036         inode->i_nlink = 0;
2037         /* There's no need to set i_disksize: the fact that i_nlink is
2038          * zero will ensure that the right thing happens during any
2039          * recovery. */
2040         inode->i_size = 0;
2041         ext3_orphan_add(handle, inode);
2042         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2043         ext3_mark_inode_dirty(handle, inode);
2044         dir->i_nlink--;
2045         ext3_update_dx_flag(dir);
2046         ext3_mark_inode_dirty(handle, dir);
2047
2048 end_rmdir:
2049         ext3_journal_stop(handle);
2050         brelse (bh);
2051         return retval;
2052 }
2053
2054 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2055 {
2056         int retval;
2057         struct inode * inode;
2058         struct buffer_head * bh;
2059         struct ext3_dir_entry_2 * de;
2060         handle_t *handle;
2061
2062         /* Initialize quotas before so that eventual writes go
2063          * in separate transaction */
2064         DQUOT_INIT(dentry->d_inode);
2065         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2066         if (IS_ERR(handle))
2067                 return PTR_ERR(handle);
2068
2069         if (IS_DIRSYNC(dir))
2070                 handle->h_sync = 1;
2071
2072         retval = -ENOENT;
2073         bh = ext3_find_entry (dentry, &de);
2074         if (!bh)
2075                 goto end_unlink;
2076
2077         inode = dentry->d_inode;
2078
2079         retval = -EIO;
2080         if (le32_to_cpu(de->inode) != inode->i_ino)
2081                 goto end_unlink;
2082
2083         if (!inode->i_nlink) {
2084                 ext3_warning (inode->i_sb, "ext3_unlink",
2085                               "Deleting nonexistent file (%lu), %d",
2086                               inode->i_ino, inode->i_nlink);
2087                 inode->i_nlink = 1;
2088         }
2089         retval = ext3_delete_entry(handle, dir, de, bh);
2090         if (retval)
2091                 goto end_unlink;
2092         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2093         ext3_update_dx_flag(dir);
2094         ext3_mark_inode_dirty(handle, dir);
2095         inode->i_nlink--;
2096         if (!inode->i_nlink)
2097                 ext3_orphan_add(handle, inode);
2098         inode->i_ctime = dir->i_ctime;
2099         ext3_mark_inode_dirty(handle, inode);
2100         retval = 0;
2101
2102 end_unlink:
2103         ext3_journal_stop(handle);
2104         brelse (bh);
2105         return retval;
2106 }
2107
2108 static int ext3_symlink (struct inode * dir,
2109                 struct dentry *dentry, const char * symname)
2110 {
2111         handle_t *handle;
2112         struct inode * inode;
2113         int l, err, retries = 0;
2114
2115         l = strlen(symname)+1;
2116         if (l > dir->i_sb->s_blocksize)
2117                 return -ENAMETOOLONG;
2118
2119 retry:
2120         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2121                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2122                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2123         if (IS_ERR(handle))
2124                 return PTR_ERR(handle);
2125
2126         if (IS_DIRSYNC(dir))
2127                 handle->h_sync = 1;
2128
2129         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2130         err = PTR_ERR(inode);
2131         if (IS_ERR(inode))
2132                 goto out_stop;
2133
2134         if (l > sizeof (EXT3_I(inode)->i_data)) {
2135                 inode->i_op = &ext3_symlink_inode_operations;
2136                 ext3_set_aops(inode);
2137                 /*
2138                  * page_symlink() calls into ext3_prepare/commit_write.
2139                  * We have a transaction open.  All is sweetness.  It also sets
2140                  * i_size in generic_commit_write().
2141                  */
2142                 err = __page_symlink(inode, symname, l,
2143                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2144                 if (err) {
2145                         ext3_dec_count(handle, inode);
2146                         ext3_mark_inode_dirty(handle, inode);
2147                         iput (inode);
2148                         goto out_stop;
2149                 }
2150         } else {
2151                 inode->i_op = &ext3_fast_symlink_inode_operations;
2152                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2153                 inode->i_size = l-1;
2154         }
2155         EXT3_I(inode)->i_disksize = inode->i_size;
2156         err = ext3_add_nondir(handle, dentry, inode);
2157 out_stop:
2158         ext3_journal_stop(handle);
2159         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2160                 goto retry;
2161         return err;
2162 }
2163
2164 static int ext3_link (struct dentry * old_dentry,
2165                 struct inode * dir, struct dentry *dentry)
2166 {
2167         handle_t *handle;
2168         struct inode *inode = old_dentry->d_inode;
2169         int err, retries = 0;
2170
2171         if (inode->i_nlink >= EXT3_LINK_MAX)
2172                 return -EMLINK;
2173
2174 retry:
2175         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2176                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2177         if (IS_ERR(handle))
2178                 return PTR_ERR(handle);
2179
2180         if (IS_DIRSYNC(dir))
2181                 handle->h_sync = 1;
2182
2183         inode->i_ctime = CURRENT_TIME_SEC;
2184         ext3_inc_count(handle, inode);
2185         atomic_inc(&inode->i_count);
2186
2187         err = ext3_add_nondir(handle, dentry, inode);
2188         ext3_journal_stop(handle);
2189         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2190                 goto retry;
2191         return err;
2192 }
2193
2194 #define PARENT_INO(buffer) \
2195         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2196         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2197
2198 /*
2199  * Anybody can rename anything with this: the permission checks are left to the
2200  * higher-level routines.
2201  */
2202 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2203                            struct inode * new_dir,struct dentry *new_dentry)
2204 {
2205         handle_t *handle;
2206         struct inode * old_inode, * new_inode;
2207         struct buffer_head * old_bh, * new_bh, * dir_bh;
2208         struct ext3_dir_entry_2 * old_de, * new_de;
2209         int retval;
2210
2211         old_bh = new_bh = dir_bh = NULL;
2212
2213         /* Initialize quotas before so that eventual writes go
2214          * in separate transaction */
2215         if (new_dentry->d_inode)
2216                 DQUOT_INIT(new_dentry->d_inode);
2217         handle = ext3_journal_start(old_dir, 2 *
2218                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2219                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2220         if (IS_ERR(handle))
2221                 return PTR_ERR(handle);
2222
2223         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2224                 handle->h_sync = 1;
2225
2226         old_bh = ext3_find_entry (old_dentry, &old_de);
2227         /*
2228          *  Check for inode number is _not_ due to possible IO errors.
2229          *  We might rmdir the source, keep it as pwd of some process
2230          *  and merrily kill the link to whatever was created under the
2231          *  same name. Goodbye sticky bit ;-<
2232          */
2233         old_inode = old_dentry->d_inode;
2234         retval = -ENOENT;
2235         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2236                 goto end_rename;
2237
2238         new_inode = new_dentry->d_inode;
2239         new_bh = ext3_find_entry (new_dentry, &new_de);
2240         if (new_bh) {
2241                 if (!new_inode) {
2242                         brelse (new_bh);
2243                         new_bh = NULL;
2244                 }
2245         }
2246         if (S_ISDIR(old_inode->i_mode)) {
2247                 if (new_inode) {
2248                         retval = -ENOTEMPTY;
2249                         if (!empty_dir (new_inode))
2250                                 goto end_rename;
2251                 }
2252                 retval = -EIO;
2253                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2254                 if (!dir_bh)
2255                         goto end_rename;
2256                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2257                         goto end_rename;
2258                 retval = -EMLINK;
2259                 if (!new_inode && new_dir!=old_dir &&
2260                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2261                         goto end_rename;
2262         }
2263         if (!new_bh) {
2264                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2265                 if (retval)
2266                         goto end_rename;
2267         } else {
2268                 BUFFER_TRACE(new_bh, "get write access");
2269                 ext3_journal_get_write_access(handle, new_bh);
2270                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2271                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2272                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2273                         new_de->file_type = old_de->file_type;
2274                 new_dir->i_version++;
2275                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2276                 ext3_journal_dirty_metadata(handle, new_bh);
2277                 brelse(new_bh);
2278                 new_bh = NULL;
2279         }
2280
2281         /*
2282          * Like most other Unix systems, set the ctime for inodes on a
2283          * rename.
2284          */
2285         old_inode->i_ctime = CURRENT_TIME_SEC;
2286         ext3_mark_inode_dirty(handle, old_inode);
2287
2288         /*
2289          * ok, that's it
2290          */
2291         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2292             old_de->name_len != old_dentry->d_name.len ||
2293             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2294             (retval = ext3_delete_entry(handle, old_dir,
2295                                         old_de, old_bh)) == -ENOENT) {
2296                 /* old_de could have moved from under us during htree split, so
2297                  * make sure that we are deleting the right entry.  We might
2298                  * also be pointing to a stale entry in the unused part of
2299                  * old_bh so just checking inum and the name isn't enough. */
2300                 struct buffer_head *old_bh2;
2301                 struct ext3_dir_entry_2 *old_de2;
2302
2303                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2304                 if (old_bh2) {
2305                         retval = ext3_delete_entry(handle, old_dir,
2306                                                    old_de2, old_bh2);
2307                         brelse(old_bh2);
2308                 }
2309         }
2310         if (retval) {
2311                 ext3_warning(old_dir->i_sb, "ext3_rename",
2312                                 "Deleting old file (%lu), %d, error=%d",
2313                                 old_dir->i_ino, old_dir->i_nlink, retval);
2314         }
2315
2316         if (new_inode) {
2317                 new_inode->i_nlink--;
2318                 new_inode->i_ctime = CURRENT_TIME_SEC;
2319         }
2320         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2321         ext3_update_dx_flag(old_dir);
2322         if (dir_bh) {
2323                 BUFFER_TRACE(dir_bh, "get_write_access");
2324                 ext3_journal_get_write_access(handle, dir_bh);
2325                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2326                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2327                 ext3_journal_dirty_metadata(handle, dir_bh);
2328                 old_dir->i_nlink--;
2329                 if (new_inode) {
2330                         new_inode->i_nlink--;
2331                 } else {
2332                         new_dir->i_nlink++;
2333                         ext3_update_dx_flag(new_dir);
2334                         ext3_mark_inode_dirty(handle, new_dir);
2335                 }
2336         }
2337         ext3_mark_inode_dirty(handle, old_dir);
2338         if (new_inode) {
2339                 ext3_mark_inode_dirty(handle, new_inode);
2340                 if (!new_inode->i_nlink)
2341                         ext3_orphan_add(handle, new_inode);
2342         }
2343         retval = 0;
2344
2345 end_rename:
2346         brelse (dir_bh);
2347         brelse (old_bh);
2348         brelse (new_bh);
2349         ext3_journal_stop(handle);
2350         return retval;
2351 }
2352
2353 /*
2354  * directories can handle most operations...
2355  */
2356 struct inode_operations ext3_dir_inode_operations = {
2357         .create         = ext3_create,
2358         .lookup         = ext3_lookup,
2359         .link           = ext3_link,
2360         .unlink         = ext3_unlink,
2361         .symlink        = ext3_symlink,
2362         .mkdir          = ext3_mkdir,
2363         .rmdir          = ext3_rmdir,
2364         .mknod          = ext3_mknod,
2365         .rename         = ext3_rename,
2366         .setattr        = ext3_setattr,
2367 #ifdef CONFIG_EXT3_FS_XATTR
2368         .setxattr       = generic_setxattr,
2369         .getxattr       = generic_getxattr,
2370         .listxattr      = ext3_listxattr,
2371         .removexattr    = generic_removexattr,
2372 #endif
2373         .permission     = ext3_permission,
2374 };
2375
2376 struct inode_operations ext3_special_inode_operations = {
2377         .setattr        = ext3_setattr,
2378 #ifdef CONFIG_EXT3_FS_XATTR
2379         .setxattr       = generic_setxattr,
2380         .getxattr       = generic_getxattr,
2381         .listxattr      = ext3_listxattr,
2382         .removexattr    = generic_removexattr,
2383 #endif
2384         .permission     = ext3_permission,
2385 };