Merge branch 'master' of /usr/src/ntfs-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         if (inode)
1009                 return d_splice_alias(inode, dentry);
1010         d_add(dentry, inode);
1011         return NULL;
1012 }
1013
1014
1015 struct dentry *ext3_get_parent(struct dentry *child)
1016 {
1017         unsigned long ino;
1018         struct dentry *parent;
1019         struct inode *inode;
1020         struct dentry dotdot;
1021         struct ext3_dir_entry_2 * de;
1022         struct buffer_head *bh;
1023
1024         dotdot.d_name.name = "..";
1025         dotdot.d_name.len = 2;
1026         dotdot.d_parent = child; /* confusing, isn't it! */
1027
1028         bh = ext3_find_entry(&dotdot, &de);
1029         inode = NULL;
1030         if (!bh)
1031                 return ERR_PTR(-ENOENT);
1032         ino = le32_to_cpu(de->inode);
1033         brelse(bh);
1034         inode = iget(child->d_inode->i_sb, ino);
1035
1036         if (!inode)
1037                 return ERR_PTR(-EACCES);
1038
1039         parent = d_alloc_anon(inode);
1040         if (!parent) {
1041                 iput(inode);
1042                 parent = ERR_PTR(-ENOMEM);
1043         }
1044         return parent;
1045
1046
1047 #define S_SHIFT 12
1048 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1049         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1050         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1051         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1052         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1053         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1054         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1055         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1056 };
1057
1058 static inline void ext3_set_de_type(struct super_block *sb,
1059                                 struct ext3_dir_entry_2 *de,
1060                                 umode_t mode) {
1061         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1062                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1063 }
1064
1065 #ifdef CONFIG_EXT3_INDEX
1066 static struct ext3_dir_entry_2 *
1067 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1068 {
1069         unsigned rec_len = 0;
1070
1071         while (count--) {
1072                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1073                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1074                 memcpy (to, de, rec_len);
1075                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1076                                 cpu_to_le16(rec_len);
1077                 de->inode = 0;
1078                 map++;
1079                 to += rec_len;
1080         }
1081         return (struct ext3_dir_entry_2 *) (to - rec_len);
1082 }
1083
1084 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1085 {
1086         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1087         unsigned rec_len = 0;
1088
1089         prev = to = de;
1090         while ((char*)de < base + size) {
1091                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1092                                                     le16_to_cpu(de->rec_len));
1093                 if (de->inode && de->name_len) {
1094                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1095                         if (de > to)
1096                                 memmove(to, de, rec_len);
1097                         to->rec_len = cpu_to_le16(rec_len);
1098                         prev = to;
1099                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1100                 }
1101                 de = next;
1102         }
1103         return prev;
1104 }
1105
1106 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1107                         struct buffer_head **bh,struct dx_frame *frame,
1108                         struct dx_hash_info *hinfo, int *error)
1109 {
1110         unsigned blocksize = dir->i_sb->s_blocksize;
1111         unsigned count, continued;
1112         struct buffer_head *bh2;
1113         u32 newblock;
1114         u32 hash2;
1115         struct dx_map_entry *map;
1116         char *data1 = (*bh)->b_data, *data2;
1117         unsigned split;
1118         struct ext3_dir_entry_2 *de = NULL, *de2;
1119         int     err;
1120
1121         bh2 = ext3_append (handle, dir, &newblock, error);
1122         if (!(bh2)) {
1123                 brelse(*bh);
1124                 *bh = NULL;
1125                 goto errout;
1126         }
1127
1128         BUFFER_TRACE(*bh, "get_write_access");
1129         err = ext3_journal_get_write_access(handle, *bh);
1130         if (err) {
1131         journal_error:
1132                 brelse(*bh);
1133                 brelse(bh2);
1134                 *bh = NULL;
1135                 ext3_std_error(dir->i_sb, err);
1136                 goto errout;
1137         }
1138         BUFFER_TRACE(frame->bh, "get_write_access");
1139         err = ext3_journal_get_write_access(handle, frame->bh);
1140         if (err)
1141                 goto journal_error;
1142
1143         data2 = bh2->b_data;
1144
1145         /* create map in the end of data2 block */
1146         map = (struct dx_map_entry *) (data2 + blocksize);
1147         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1148                              blocksize, hinfo, map);
1149         map -= count;
1150         split = count/2; // need to adjust to actual middle
1151         dx_sort_map (map, count);
1152         hash2 = map[split].hash;
1153         continued = hash2 == map[split - 1].hash;
1154         dxtrace(printk("Split block %i at %x, %i/%i\n",
1155                 dx_get_block(frame->at), hash2, split, count-split));
1156
1157         /* Fancy dance to stay within two buffers */
1158         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1159         de = dx_pack_dirents(data1,blocksize);
1160         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1161         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1162         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1163         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1164
1165         /* Which block gets the new entry? */
1166         if (hinfo->hash >= hash2)
1167         {
1168                 swap(*bh, bh2);
1169                 de = de2;
1170         }
1171         dx_insert_block (frame, hash2 + continued, newblock);
1172         err = ext3_journal_dirty_metadata (handle, bh2);
1173         if (err)
1174                 goto journal_error;
1175         err = ext3_journal_dirty_metadata (handle, frame->bh);
1176         if (err)
1177                 goto journal_error;
1178         brelse (bh2);
1179         dxtrace(dx_show_index ("frame", frame->entries));
1180 errout:
1181         return de;
1182 }
1183 #endif
1184
1185
1186 /*
1187  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1188  * it points to a directory entry which is guaranteed to be large
1189  * enough for new directory entry.  If de is NULL, then
1190  * add_dirent_to_buf will attempt search the directory block for
1191  * space.  It will return -ENOSPC if no space is available, and -EIO
1192  * and -EEXIST if directory entry already exists.
1193  * 
1194  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1195  * all other cases bh is released.
1196  */
1197 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1198                              struct inode *inode, struct ext3_dir_entry_2 *de,
1199                              struct buffer_head * bh)
1200 {
1201         struct inode    *dir = dentry->d_parent->d_inode;
1202         const char      *name = dentry->d_name.name;
1203         int             namelen = dentry->d_name.len;
1204         unsigned long   offset = 0;
1205         unsigned short  reclen;
1206         int             nlen, rlen, err;
1207         char            *top;
1208
1209         reclen = EXT3_DIR_REC_LEN(namelen);
1210         if (!de) {
1211                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1212                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1213                 while ((char *) de <= top) {
1214                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1215                                                   bh, offset)) {
1216                                 brelse (bh);
1217                                 return -EIO;
1218                         }
1219                         if (ext3_match (namelen, name, de)) {
1220                                 brelse (bh);
1221                                 return -EEXIST;
1222                         }
1223                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1224                         rlen = le16_to_cpu(de->rec_len);
1225                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1226                                 break;
1227                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1228                         offset += rlen;
1229                 }
1230                 if ((char *) de > top)
1231                         return -ENOSPC;
1232         }
1233         BUFFER_TRACE(bh, "get_write_access");
1234         err = ext3_journal_get_write_access(handle, bh);
1235         if (err) {
1236                 ext3_std_error(dir->i_sb, err);
1237                 brelse(bh);
1238                 return err;
1239         }
1240
1241         /* By now the buffer is marked for journaling */
1242         nlen = EXT3_DIR_REC_LEN(de->name_len);
1243         rlen = le16_to_cpu(de->rec_len);
1244         if (de->inode) {
1245                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1246                 de1->rec_len = cpu_to_le16(rlen - nlen);
1247                 de->rec_len = cpu_to_le16(nlen);
1248                 de = de1;
1249         }
1250         de->file_type = EXT3_FT_UNKNOWN;
1251         if (inode) {
1252                 de->inode = cpu_to_le32(inode->i_ino);
1253                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1254         } else
1255                 de->inode = 0;
1256         de->name_len = namelen;
1257         memcpy (de->name, name, namelen);
1258         /*
1259          * XXX shouldn't update any times until successful
1260          * completion of syscall, but too many callers depend
1261          * on this.
1262          *
1263          * XXX similarly, too many callers depend on
1264          * ext3_new_inode() setting the times, but error
1265          * recovery deletes the inode, so the worst that can
1266          * happen is that the times are slightly out of date
1267          * and/or different from the directory change time.
1268          */
1269         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1270         ext3_update_dx_flag(dir);
1271         dir->i_version++;
1272         ext3_mark_inode_dirty(handle, dir);
1273         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1274         err = ext3_journal_dirty_metadata(handle, bh);
1275         if (err)
1276                 ext3_std_error(dir->i_sb, err);
1277         brelse(bh);
1278         return 0;
1279 }
1280
1281 #ifdef CONFIG_EXT3_INDEX
1282 /*
1283  * This converts a one block unindexed directory to a 3 block indexed
1284  * directory, and adds the dentry to the indexed directory.
1285  */
1286 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1287                             struct inode *inode, struct buffer_head *bh)
1288 {
1289         struct inode    *dir = dentry->d_parent->d_inode;
1290         const char      *name = dentry->d_name.name;
1291         int             namelen = dentry->d_name.len;
1292         struct buffer_head *bh2;
1293         struct dx_root  *root;
1294         struct dx_frame frames[2], *frame;
1295         struct dx_entry *entries;
1296         struct ext3_dir_entry_2 *de, *de2;
1297         char            *data1, *top;
1298         unsigned        len;
1299         int             retval;
1300         unsigned        blocksize;
1301         struct dx_hash_info hinfo;
1302         u32             block;
1303         struct fake_dirent *fde;
1304
1305         blocksize =  dir->i_sb->s_blocksize;
1306         dxtrace(printk("Creating index\n"));
1307         retval = ext3_journal_get_write_access(handle, bh);
1308         if (retval) {
1309                 ext3_std_error(dir->i_sb, retval);
1310                 brelse(bh);
1311                 return retval;
1312         }
1313         root = (struct dx_root *) bh->b_data;
1314
1315         bh2 = ext3_append (handle, dir, &block, &retval);
1316         if (!(bh2)) {
1317                 brelse(bh);
1318                 return retval;
1319         }
1320         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1321         data1 = bh2->b_data;
1322
1323         /* The 0th block becomes the root, move the dirents out */
1324         fde = &root->dotdot;
1325         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1326         len = ((char *) root) + blocksize - (char *) de;
1327         memcpy (data1, de, len);
1328         de = (struct ext3_dir_entry_2 *) data1;
1329         top = data1 + len;
1330         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1331                 de = de2;
1332         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1333         /* Initialize the root; the dot dirents already exist */
1334         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1335         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1336         memset (&root->info, 0, sizeof(root->info));
1337         root->info.info_length = sizeof(root->info);
1338         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1339         entries = root->entries;
1340         dx_set_block (entries, 1);
1341         dx_set_count (entries, 1);
1342         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1343
1344         /* Initialize as for dx_probe */
1345         hinfo.hash_version = root->info.hash_version;
1346         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1347         ext3fs_dirhash(name, namelen, &hinfo);
1348         frame = frames;
1349         frame->entries = entries;
1350         frame->at = entries;
1351         frame->bh = bh;
1352         bh = bh2;
1353         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1354         dx_release (frames);
1355         if (!(de))
1356                 return retval;
1357
1358         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1359 }
1360 #endif
1361
1362 /*
1363  *      ext3_add_entry()
1364  *
1365  * adds a file entry to the specified directory, using the same
1366  * semantics as ext3_find_entry(). It returns NULL if it failed.
1367  *
1368  * NOTE!! The inode part of 'de' is left at 0 - which means you
1369  * may not sleep between calling this and putting something into
1370  * the entry, as someone else might have used it while you slept.
1371  */
1372 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1373         struct inode *inode)
1374 {
1375         struct inode *dir = dentry->d_parent->d_inode;
1376         unsigned long offset;
1377         struct buffer_head * bh;
1378         struct ext3_dir_entry_2 *de;
1379         struct super_block * sb;
1380         int     retval;
1381 #ifdef CONFIG_EXT3_INDEX
1382         int     dx_fallback=0;
1383 #endif
1384         unsigned blocksize;
1385         unsigned nlen, rlen;
1386         u32 block, blocks;
1387
1388         sb = dir->i_sb;
1389         blocksize = sb->s_blocksize;
1390         if (!dentry->d_name.len)
1391                 return -EINVAL;
1392 #ifdef CONFIG_EXT3_INDEX
1393         if (is_dx(dir)) {
1394                 retval = ext3_dx_add_entry(handle, dentry, inode);
1395                 if (!retval || (retval != ERR_BAD_DX_DIR))
1396                         return retval;
1397                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1398                 dx_fallback++;
1399                 ext3_mark_inode_dirty(handle, dir);
1400         }
1401 #endif
1402         blocks = dir->i_size >> sb->s_blocksize_bits;
1403         for (block = 0, offset = 0; block < blocks; block++) {
1404                 bh = ext3_bread(handle, dir, block, 0, &retval);
1405                 if(!bh)
1406                         return retval;
1407                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1408                 if (retval != -ENOSPC)
1409                         return retval;
1410
1411 #ifdef CONFIG_EXT3_INDEX
1412                 if (blocks == 1 && !dx_fallback &&
1413                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1414                         return make_indexed_dir(handle, dentry, inode, bh);
1415 #endif
1416                 brelse(bh);
1417         }
1418         bh = ext3_append(handle, dir, &block, &retval);
1419         if (!bh)
1420                 return retval;
1421         de = (struct ext3_dir_entry_2 *) bh->b_data;
1422         de->inode = 0;
1423         de->rec_len = cpu_to_le16(rlen = blocksize);
1424         nlen = 0;
1425         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1426 }
1427
1428 #ifdef CONFIG_EXT3_INDEX
1429 /*
1430  * Returns 0 for success, or a negative error value
1431  */
1432 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1433                              struct inode *inode)
1434 {
1435         struct dx_frame frames[2], *frame;
1436         struct dx_entry *entries, *at;
1437         struct dx_hash_info hinfo;
1438         struct buffer_head * bh;
1439         struct inode *dir = dentry->d_parent->d_inode;
1440         struct super_block * sb = dir->i_sb;
1441         struct ext3_dir_entry_2 *de;
1442         int err;
1443
1444         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1445         if (!frame)
1446                 return err;
1447         entries = frame->entries;
1448         at = frame->at;
1449
1450         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1451                 goto cleanup;
1452
1453         BUFFER_TRACE(bh, "get_write_access");
1454         err = ext3_journal_get_write_access(handle, bh);
1455         if (err)
1456                 goto journal_error;
1457
1458         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1459         if (err != -ENOSPC) {
1460                 bh = NULL;
1461                 goto cleanup;
1462         }
1463
1464         /* Block full, should compress but for now just split */
1465         dxtrace(printk("using %u of %u node entries\n",
1466                        dx_get_count(entries), dx_get_limit(entries)));
1467         /* Need to split index? */
1468         if (dx_get_count(entries) == dx_get_limit(entries)) {
1469                 u32 newblock;
1470                 unsigned icount = dx_get_count(entries);
1471                 int levels = frame - frames;
1472                 struct dx_entry *entries2;
1473                 struct dx_node *node2;
1474                 struct buffer_head *bh2;
1475
1476                 if (levels && (dx_get_count(frames->entries) ==
1477                                dx_get_limit(frames->entries))) {
1478                         ext3_warning(sb, __FUNCTION__,
1479                                      "Directory index full!\n");
1480                         err = -ENOSPC;
1481                         goto cleanup;
1482                 }
1483                 bh2 = ext3_append (handle, dir, &newblock, &err);
1484                 if (!(bh2))
1485                         goto cleanup;
1486                 node2 = (struct dx_node *)(bh2->b_data);
1487                 entries2 = node2->entries;
1488                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1489                 node2->fake.inode = 0;
1490                 BUFFER_TRACE(frame->bh, "get_write_access");
1491                 err = ext3_journal_get_write_access(handle, frame->bh);
1492                 if (err)
1493                         goto journal_error;
1494                 if (levels) {
1495                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1496                         unsigned hash2 = dx_get_hash(entries + icount1);
1497                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1498
1499                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1500                         err = ext3_journal_get_write_access(handle,
1501                                                              frames[0].bh);
1502                         if (err)
1503                                 goto journal_error;
1504
1505                         memcpy ((char *) entries2, (char *) (entries + icount1),
1506                                 icount2 * sizeof(struct dx_entry));
1507                         dx_set_count (entries, icount1);
1508                         dx_set_count (entries2, icount2);
1509                         dx_set_limit (entries2, dx_node_limit(dir));
1510
1511                         /* Which index block gets the new entry? */
1512                         if (at - entries >= icount1) {
1513                                 frame->at = at = at - entries - icount1 + entries2;
1514                                 frame->entries = entries = entries2;
1515                                 swap(frame->bh, bh2);
1516                         }
1517                         dx_insert_block (frames + 0, hash2, newblock);
1518                         dxtrace(dx_show_index ("node", frames[1].entries));
1519                         dxtrace(dx_show_index ("node",
1520                                ((struct dx_node *) bh2->b_data)->entries));
1521                         err = ext3_journal_dirty_metadata(handle, bh2);
1522                         if (err)
1523                                 goto journal_error;
1524                         brelse (bh2);
1525                 } else {
1526                         dxtrace(printk("Creating second level index...\n"));
1527                         memcpy((char *) entries2, (char *) entries,
1528                                icount * sizeof(struct dx_entry));
1529                         dx_set_limit(entries2, dx_node_limit(dir));
1530
1531                         /* Set up root */
1532                         dx_set_count(entries, 1);
1533                         dx_set_block(entries + 0, newblock);
1534                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1535
1536                         /* Add new access path frame */
1537                         frame = frames + 1;
1538                         frame->at = at = at - entries + entries2;
1539                         frame->entries = entries = entries2;
1540                         frame->bh = bh2;
1541                         err = ext3_journal_get_write_access(handle,
1542                                                              frame->bh);
1543                         if (err)
1544                                 goto journal_error;
1545                 }
1546                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1547         }
1548         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1549         if (!de)
1550                 goto cleanup;
1551         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1552         bh = NULL;
1553         goto cleanup;
1554
1555 journal_error:
1556         ext3_std_error(dir->i_sb, err);
1557 cleanup:
1558         if (bh)
1559                 brelse(bh);
1560         dx_release(frames);
1561         return err;
1562 }
1563 #endif
1564
1565 /*
1566  * ext3_delete_entry deletes a directory entry by merging it with the
1567  * previous entry
1568  */
1569 static int ext3_delete_entry (handle_t *handle, 
1570                               struct inode * dir,
1571                               struct ext3_dir_entry_2 * de_del,
1572                               struct buffer_head * bh)
1573 {
1574         struct ext3_dir_entry_2 * de, * pde;
1575         int i;
1576
1577         i = 0;
1578         pde = NULL;
1579         de = (struct ext3_dir_entry_2 *) bh->b_data;
1580         while (i < bh->b_size) {
1581                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1582                         return -EIO;
1583                 if (de == de_del)  {
1584                         BUFFER_TRACE(bh, "get_write_access");
1585                         ext3_journal_get_write_access(handle, bh);
1586                         if (pde)
1587                                 pde->rec_len =
1588                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1589                                                     le16_to_cpu(de->rec_len));
1590                         else
1591                                 de->inode = 0;
1592                         dir->i_version++;
1593                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1594                         ext3_journal_dirty_metadata(handle, bh);
1595                         return 0;
1596                 }
1597                 i += le16_to_cpu(de->rec_len);
1598                 pde = de;
1599                 de = (struct ext3_dir_entry_2 *)
1600                         ((char *) de + le16_to_cpu(de->rec_len));
1601         }
1602         return -ENOENT;
1603 }
1604
1605 /*
1606  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1607  * do not perform it in these functions.  We perform it at the call site,
1608  * if it is needed.
1609  */
1610 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1611 {
1612         inode->i_nlink++;
1613 }
1614
1615 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1616 {
1617         inode->i_nlink--;
1618 }
1619
1620 static int ext3_add_nondir(handle_t *handle,
1621                 struct dentry *dentry, struct inode *inode)
1622 {
1623         int err = ext3_add_entry(handle, dentry, inode);
1624         if (!err) {
1625                 ext3_mark_inode_dirty(handle, inode);
1626                 d_instantiate(dentry, inode);
1627                 return 0;
1628         }
1629         ext3_dec_count(handle, inode);
1630         iput(inode);
1631         return err;
1632 }
1633
1634 /*
1635  * By the time this is called, we already have created
1636  * the directory cache entry for the new file, but it
1637  * is so far negative - it has no inode.
1638  *
1639  * If the create succeeds, we fill in the inode information
1640  * with d_instantiate(). 
1641  */
1642 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1643                 struct nameidata *nd)
1644 {
1645         handle_t *handle; 
1646         struct inode * inode;
1647         int err, retries = 0;
1648
1649 retry:
1650         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1651                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1652                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1653         if (IS_ERR(handle))
1654                 return PTR_ERR(handle);
1655
1656         if (IS_DIRSYNC(dir))
1657                 handle->h_sync = 1;
1658
1659         inode = ext3_new_inode (handle, dir, mode);
1660         err = PTR_ERR(inode);
1661         if (!IS_ERR(inode)) {
1662                 inode->i_op = &ext3_file_inode_operations;
1663                 inode->i_fop = &ext3_file_operations;
1664                 ext3_set_aops(inode);
1665                 err = ext3_add_nondir(handle, dentry, inode);
1666         }
1667         ext3_journal_stop(handle);
1668         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1669                 goto retry;
1670         return err;
1671 }
1672
1673 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1674                         int mode, dev_t rdev)
1675 {
1676         handle_t *handle;
1677         struct inode *inode;
1678         int err, retries = 0;
1679
1680         if (!new_valid_dev(rdev))
1681                 return -EINVAL;
1682
1683 retry:
1684         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1685                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1686                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1687         if (IS_ERR(handle))
1688                 return PTR_ERR(handle);
1689
1690         if (IS_DIRSYNC(dir))
1691                 handle->h_sync = 1;
1692
1693         inode = ext3_new_inode (handle, dir, mode);
1694         err = PTR_ERR(inode);
1695         if (!IS_ERR(inode)) {
1696                 init_special_inode(inode, inode->i_mode, rdev);
1697 #ifdef CONFIG_EXT3_FS_XATTR
1698                 inode->i_op = &ext3_special_inode_operations;
1699 #endif
1700                 err = ext3_add_nondir(handle, dentry, inode);
1701         }
1702         ext3_journal_stop(handle);
1703         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1704                 goto retry;
1705         return err;
1706 }
1707
1708 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1709 {
1710         handle_t *handle;
1711         struct inode * inode;
1712         struct buffer_head * dir_block;
1713         struct ext3_dir_entry_2 * de;
1714         int err, retries = 0;
1715
1716         if (dir->i_nlink >= EXT3_LINK_MAX)
1717                 return -EMLINK;
1718
1719 retry:
1720         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1721                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1722                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1723         if (IS_ERR(handle))
1724                 return PTR_ERR(handle);
1725
1726         if (IS_DIRSYNC(dir))
1727                 handle->h_sync = 1;
1728
1729         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1730         err = PTR_ERR(inode);
1731         if (IS_ERR(inode))
1732                 goto out_stop;
1733
1734         inode->i_op = &ext3_dir_inode_operations;
1735         inode->i_fop = &ext3_dir_operations;
1736         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1737         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1738         if (!dir_block) {
1739                 inode->i_nlink--; /* is this nlink == 0? */
1740                 ext3_mark_inode_dirty(handle, inode);
1741                 iput (inode);
1742                 goto out_stop;
1743         }
1744         BUFFER_TRACE(dir_block, "get_write_access");
1745         ext3_journal_get_write_access(handle, dir_block);
1746         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1747         de->inode = cpu_to_le32(inode->i_ino);
1748         de->name_len = 1;
1749         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1750         strcpy (de->name, ".");
1751         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1752         de = (struct ext3_dir_entry_2 *)
1753                         ((char *) de + le16_to_cpu(de->rec_len));
1754         de->inode = cpu_to_le32(dir->i_ino);
1755         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1756         de->name_len = 2;
1757         strcpy (de->name, "..");
1758         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1759         inode->i_nlink = 2;
1760         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1761         ext3_journal_dirty_metadata(handle, dir_block);
1762         brelse (dir_block);
1763         ext3_mark_inode_dirty(handle, inode);
1764         err = ext3_add_entry (handle, dentry, inode);
1765         if (err) {
1766                 inode->i_nlink = 0;
1767                 ext3_mark_inode_dirty(handle, inode);
1768                 iput (inode);
1769                 goto out_stop;
1770         }
1771         dir->i_nlink++;
1772         ext3_update_dx_flag(dir);
1773         ext3_mark_inode_dirty(handle, dir);
1774         d_instantiate(dentry, inode);
1775 out_stop:
1776         ext3_journal_stop(handle);
1777         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1778                 goto retry;
1779         return err;
1780 }
1781
1782 /*
1783  * routine to check that the specified directory is empty (for rmdir)
1784  */
1785 static int empty_dir (struct inode * inode)
1786 {
1787         unsigned long offset;
1788         struct buffer_head * bh;
1789         struct ext3_dir_entry_2 * de, * de1;
1790         struct super_block * sb;
1791         int err = 0;
1792
1793         sb = inode->i_sb;
1794         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1795             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1796                 if (err)
1797                         ext3_error(inode->i_sb, __FUNCTION__,
1798                                    "error %d reading directory #%lu offset 0",
1799                                    err, inode->i_ino);
1800                 else
1801                         ext3_warning(inode->i_sb, __FUNCTION__,
1802                                      "bad directory (dir #%lu) - no data block",
1803                                      inode->i_ino);
1804                 return 1;
1805         }
1806         de = (struct ext3_dir_entry_2 *) bh->b_data;
1807         de1 = (struct ext3_dir_entry_2 *)
1808                         ((char *) de + le16_to_cpu(de->rec_len));
1809         if (le32_to_cpu(de->inode) != inode->i_ino ||
1810                         !le32_to_cpu(de1->inode) || 
1811                         strcmp (".", de->name) ||
1812                         strcmp ("..", de1->name)) {
1813                 ext3_warning (inode->i_sb, "empty_dir",
1814                               "bad directory (dir #%lu) - no `.' or `..'",
1815                               inode->i_ino);
1816                 brelse (bh);
1817                 return 1;
1818         }
1819         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1820         de = (struct ext3_dir_entry_2 *)
1821                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1822         while (offset < inode->i_size ) {
1823                 if (!bh ||
1824                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1825                         err = 0;
1826                         brelse (bh);
1827                         bh = ext3_bread (NULL, inode,
1828                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1829                         if (!bh) {
1830                                 if (err)
1831                                         ext3_error(sb, __FUNCTION__,
1832                                                    "error %d reading directory"
1833                                                    " #%lu offset %lu",
1834                                                    err, inode->i_ino, offset);
1835                                 offset += sb->s_blocksize;
1836                                 continue;
1837                         }
1838                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1839                 }
1840                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1841                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1842                                                          sb->s_blocksize);
1843                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1844                         continue;
1845                 }
1846                 if (le32_to_cpu(de->inode)) {
1847                         brelse (bh);
1848                         return 0;
1849                 }
1850                 offset += le16_to_cpu(de->rec_len);
1851                 de = (struct ext3_dir_entry_2 *)
1852                                 ((char *) de + le16_to_cpu(de->rec_len));
1853         }
1854         brelse (bh);
1855         return 1;
1856 }
1857
1858 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1859  * such inodes, starting at the superblock, in case we crash before the
1860  * file is closed/deleted, or in case the inode truncate spans multiple
1861  * transactions and the last transaction is not recovered after a crash.
1862  *
1863  * At filesystem recovery time, we walk this list deleting unlinked
1864  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1865  */
1866 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1867 {
1868         struct super_block *sb = inode->i_sb;
1869         struct ext3_iloc iloc;
1870         int err = 0, rc;
1871
1872         lock_super(sb);
1873         if (!list_empty(&EXT3_I(inode)->i_orphan))
1874                 goto out_unlock;
1875
1876         /* Orphan handling is only valid for files with data blocks
1877          * being truncated, or files being unlinked. */
1878
1879         /* @@@ FIXME: Observation from aviro:
1880          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block 
1881          * here (on lock_super()), so race with ext3_link() which might bump
1882          * ->i_nlink. For, say it, character device. Not a regular file,
1883          * not a directory, not a symlink and ->i_nlink > 0.
1884          */
1885         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1886                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1887
1888         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1889         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1890         if (err)
1891                 goto out_unlock;
1892
1893         err = ext3_reserve_inode_write(handle, inode, &iloc);
1894         if (err)
1895                 goto out_unlock;
1896
1897         /* Insert this inode at the head of the on-disk orphan list... */
1898         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1899         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1900         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1901         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1902         if (!err)
1903                 err = rc;
1904
1905         /* Only add to the head of the in-memory list if all the
1906          * previous operations succeeded.  If the orphan_add is going to
1907          * fail (possibly taking the journal offline), we can't risk
1908          * leaving the inode on the orphan list: stray orphan-list
1909          * entries can cause panics at unmount time.
1910          *
1911          * This is safe: on error we're going to ignore the orphan list
1912          * anyway on the next recovery. */
1913         if (!err)
1914                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1915
1916         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
1917         jbd_debug(4, "orphan inode %ld will point to %d\n",
1918                         inode->i_ino, NEXT_ORPHAN(inode));
1919 out_unlock:
1920         unlock_super(sb);
1921         ext3_std_error(inode->i_sb, err);
1922         return err;
1923 }
1924
1925 /*
1926  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1927  * of such inodes stored on disk, because it is finally being cleaned up.
1928  */
1929 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1930 {
1931         struct list_head *prev;
1932         struct ext3_inode_info *ei = EXT3_I(inode);
1933         struct ext3_sb_info *sbi;
1934         unsigned long ino_next;
1935         struct ext3_iloc iloc;
1936         int err = 0;
1937
1938         lock_super(inode->i_sb);
1939         if (list_empty(&ei->i_orphan)) {
1940                 unlock_super(inode->i_sb);
1941                 return 0;
1942         }
1943
1944         ino_next = NEXT_ORPHAN(inode);
1945         prev = ei->i_orphan.prev;
1946         sbi = EXT3_SB(inode->i_sb);
1947
1948         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1949
1950         list_del_init(&ei->i_orphan);
1951
1952         /* If we're on an error path, we may not have a valid
1953          * transaction handle with which to update the orphan list on
1954          * disk, but we still need to remove the inode from the linked
1955          * list in memory. */
1956         if (!handle)
1957                 goto out;
1958
1959         err = ext3_reserve_inode_write(handle, inode, &iloc);
1960         if (err)
1961                 goto out_err;
1962
1963         if (prev == &sbi->s_orphan) {
1964                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1965                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1966                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1967                 if (err)
1968                         goto out_brelse;
1969                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1970                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1971         } else {
1972                 struct ext3_iloc iloc2;
1973                 struct inode *i_prev =
1974                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1975
1976                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1977                           i_prev->i_ino, ino_next);
1978                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1979                 if (err)
1980                         goto out_brelse;
1981                 NEXT_ORPHAN(i_prev) = ino_next;
1982                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1983         }
1984         if (err)
1985                 goto out_brelse;
1986         NEXT_ORPHAN(inode) = 0;
1987         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1988
1989 out_err:
1990         ext3_std_error(inode->i_sb, err);
1991 out:
1992         unlock_super(inode->i_sb);
1993         return err;
1994
1995 out_brelse:
1996         brelse(iloc.bh);
1997         goto out_err;
1998 }
1999
2000 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2001 {
2002         int retval;
2003         struct inode * inode;
2004         struct buffer_head * bh;
2005         struct ext3_dir_entry_2 * de;
2006         handle_t *handle;
2007
2008         /* Initialize quotas before so that eventual writes go in
2009          * separate transaction */
2010         DQUOT_INIT(dentry->d_inode);
2011         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2012         if (IS_ERR(handle))
2013                 return PTR_ERR(handle);
2014
2015         retval = -ENOENT;
2016         bh = ext3_find_entry (dentry, &de);
2017         if (!bh)
2018                 goto end_rmdir;
2019
2020         if (IS_DIRSYNC(dir))
2021                 handle->h_sync = 1;
2022
2023         inode = dentry->d_inode;
2024
2025         retval = -EIO;
2026         if (le32_to_cpu(de->inode) != inode->i_ino)
2027                 goto end_rmdir;
2028
2029         retval = -ENOTEMPTY;
2030         if (!empty_dir (inode))
2031                 goto end_rmdir;
2032
2033         retval = ext3_delete_entry(handle, dir, de, bh);
2034         if (retval)
2035                 goto end_rmdir;
2036         if (inode->i_nlink != 2)
2037                 ext3_warning (inode->i_sb, "ext3_rmdir",
2038                               "empty directory has nlink!=2 (%d)",
2039                               inode->i_nlink);
2040         inode->i_version++;
2041         inode->i_nlink = 0;
2042         /* There's no need to set i_disksize: the fact that i_nlink is
2043          * zero will ensure that the right thing happens during any
2044          * recovery. */
2045         inode->i_size = 0;
2046         ext3_orphan_add(handle, inode);
2047         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2048         ext3_mark_inode_dirty(handle, inode);
2049         dir->i_nlink--;
2050         ext3_update_dx_flag(dir);
2051         ext3_mark_inode_dirty(handle, dir);
2052
2053 end_rmdir:
2054         ext3_journal_stop(handle);
2055         brelse (bh);
2056         return retval;
2057 }
2058
2059 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2060 {
2061         int retval;
2062         struct inode * inode;
2063         struct buffer_head * bh;
2064         struct ext3_dir_entry_2 * de;
2065         handle_t *handle;
2066
2067         /* Initialize quotas before so that eventual writes go
2068          * in separate transaction */
2069         DQUOT_INIT(dentry->d_inode);
2070         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2071         if (IS_ERR(handle))
2072                 return PTR_ERR(handle);
2073
2074         if (IS_DIRSYNC(dir))
2075                 handle->h_sync = 1;
2076
2077         retval = -ENOENT;
2078         bh = ext3_find_entry (dentry, &de);
2079         if (!bh)
2080                 goto end_unlink;
2081
2082         inode = dentry->d_inode;
2083
2084         retval = -EIO;
2085         if (le32_to_cpu(de->inode) != inode->i_ino)
2086                 goto end_unlink;
2087
2088         if (!inode->i_nlink) {
2089                 ext3_warning (inode->i_sb, "ext3_unlink",
2090                               "Deleting nonexistent file (%lu), %d",
2091                               inode->i_ino, inode->i_nlink);
2092                 inode->i_nlink = 1;
2093         }
2094         retval = ext3_delete_entry(handle, dir, de, bh);
2095         if (retval)
2096                 goto end_unlink;
2097         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2098         ext3_update_dx_flag(dir);
2099         ext3_mark_inode_dirty(handle, dir);
2100         inode->i_nlink--;
2101         if (!inode->i_nlink)
2102                 ext3_orphan_add(handle, inode);
2103         inode->i_ctime = dir->i_ctime;
2104         ext3_mark_inode_dirty(handle, inode);
2105         retval = 0;
2106
2107 end_unlink:
2108         ext3_journal_stop(handle);
2109         brelse (bh);
2110         return retval;
2111 }
2112
2113 static int ext3_symlink (struct inode * dir,
2114                 struct dentry *dentry, const char * symname)
2115 {
2116         handle_t *handle;
2117         struct inode * inode;
2118         int l, err, retries = 0;
2119
2120         l = strlen(symname)+1;
2121         if (l > dir->i_sb->s_blocksize)
2122                 return -ENAMETOOLONG;
2123
2124 retry:
2125         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2126                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2127                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2128         if (IS_ERR(handle))
2129                 return PTR_ERR(handle);
2130
2131         if (IS_DIRSYNC(dir))
2132                 handle->h_sync = 1;
2133
2134         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2135         err = PTR_ERR(inode);
2136         if (IS_ERR(inode))
2137                 goto out_stop;
2138
2139         if (l > sizeof (EXT3_I(inode)->i_data)) {
2140                 inode->i_op = &ext3_symlink_inode_operations;
2141                 ext3_set_aops(inode);
2142                 /*
2143                  * page_symlink() calls into ext3_prepare/commit_write.
2144                  * We have a transaction open.  All is sweetness.  It also sets
2145                  * i_size in generic_commit_write().
2146                  */
2147                 err = page_symlink(inode, symname, l);
2148                 if (err) {
2149                         ext3_dec_count(handle, inode);
2150                         ext3_mark_inode_dirty(handle, inode);
2151                         iput (inode);
2152                         goto out_stop;
2153                 }
2154         } else {
2155                 inode->i_op = &ext3_fast_symlink_inode_operations;
2156                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2157                 inode->i_size = l-1;
2158         }
2159         EXT3_I(inode)->i_disksize = inode->i_size;
2160         err = ext3_add_nondir(handle, dentry, inode);
2161 out_stop:
2162         ext3_journal_stop(handle);
2163         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2164                 goto retry;
2165         return err;
2166 }
2167
2168 static int ext3_link (struct dentry * old_dentry,
2169                 struct inode * dir, struct dentry *dentry)
2170 {
2171         handle_t *handle;
2172         struct inode *inode = old_dentry->d_inode;
2173         int err, retries = 0;
2174
2175         if (inode->i_nlink >= EXT3_LINK_MAX)
2176                 return -EMLINK;
2177
2178 retry:
2179         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2180                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2181         if (IS_ERR(handle))
2182                 return PTR_ERR(handle);
2183
2184         if (IS_DIRSYNC(dir))
2185                 handle->h_sync = 1;
2186
2187         inode->i_ctime = CURRENT_TIME_SEC;
2188         ext3_inc_count(handle, inode);
2189         atomic_inc(&inode->i_count);
2190
2191         err = ext3_add_nondir(handle, dentry, inode);
2192         ext3_journal_stop(handle);
2193         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2194                 goto retry;
2195         return err;
2196 }
2197
2198 #define PARENT_INO(buffer) \
2199         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2200         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2201
2202 /*
2203  * Anybody can rename anything with this: the permission checks are left to the
2204  * higher-level routines.
2205  */
2206 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2207                            struct inode * new_dir,struct dentry *new_dentry)
2208 {
2209         handle_t *handle;
2210         struct inode * old_inode, * new_inode;
2211         struct buffer_head * old_bh, * new_bh, * dir_bh;
2212         struct ext3_dir_entry_2 * old_de, * new_de;
2213         int retval;
2214
2215         old_bh = new_bh = dir_bh = NULL;
2216
2217         /* Initialize quotas before so that eventual writes go
2218          * in separate transaction */
2219         if (new_dentry->d_inode)
2220                 DQUOT_INIT(new_dentry->d_inode);
2221         handle = ext3_journal_start(old_dir, 2 *
2222                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2223                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2224         if (IS_ERR(handle))
2225                 return PTR_ERR(handle);
2226
2227         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2228                 handle->h_sync = 1;
2229
2230         old_bh = ext3_find_entry (old_dentry, &old_de);
2231         /*
2232          *  Check for inode number is _not_ due to possible IO errors.
2233          *  We might rmdir the source, keep it as pwd of some process
2234          *  and merrily kill the link to whatever was created under the
2235          *  same name. Goodbye sticky bit ;-<
2236          */
2237         old_inode = old_dentry->d_inode;
2238         retval = -ENOENT;
2239         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2240                 goto end_rename;
2241
2242         new_inode = new_dentry->d_inode;
2243         new_bh = ext3_find_entry (new_dentry, &new_de);
2244         if (new_bh) {
2245                 if (!new_inode) {
2246                         brelse (new_bh);
2247                         new_bh = NULL;
2248                 }
2249         }
2250         if (S_ISDIR(old_inode->i_mode)) {
2251                 if (new_inode) {
2252                         retval = -ENOTEMPTY;
2253                         if (!empty_dir (new_inode))
2254                                 goto end_rename;
2255                 }
2256                 retval = -EIO;
2257                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2258                 if (!dir_bh)
2259                         goto end_rename;
2260                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2261                         goto end_rename;
2262                 retval = -EMLINK;
2263                 if (!new_inode && new_dir!=old_dir &&
2264                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2265                         goto end_rename;
2266         }
2267         if (!new_bh) {
2268                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2269                 if (retval)
2270                         goto end_rename;
2271         } else {
2272                 BUFFER_TRACE(new_bh, "get write access");
2273                 ext3_journal_get_write_access(handle, new_bh);
2274                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2275                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2276                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2277                         new_de->file_type = old_de->file_type;
2278                 new_dir->i_version++;
2279                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2280                 ext3_journal_dirty_metadata(handle, new_bh);
2281                 brelse(new_bh);
2282                 new_bh = NULL;
2283         }
2284
2285         /*
2286          * Like most other Unix systems, set the ctime for inodes on a
2287          * rename.
2288          */
2289         old_inode->i_ctime = CURRENT_TIME_SEC;
2290         ext3_mark_inode_dirty(handle, old_inode);
2291
2292         /*
2293          * ok, that's it
2294          */
2295         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2296             old_de->name_len != old_dentry->d_name.len ||
2297             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2298             (retval = ext3_delete_entry(handle, old_dir,
2299                                         old_de, old_bh)) == -ENOENT) {
2300                 /* old_de could have moved from under us during htree split, so
2301                  * make sure that we are deleting the right entry.  We might
2302                  * also be pointing to a stale entry in the unused part of
2303                  * old_bh so just checking inum and the name isn't enough. */
2304                 struct buffer_head *old_bh2;
2305                 struct ext3_dir_entry_2 *old_de2;
2306
2307                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2308                 if (old_bh2) {
2309                         retval = ext3_delete_entry(handle, old_dir,
2310                                                    old_de2, old_bh2);
2311                         brelse(old_bh2);
2312                 }
2313         }
2314         if (retval) {
2315                 ext3_warning(old_dir->i_sb, "ext3_rename",
2316                                 "Deleting old file (%lu), %d, error=%d",
2317                                 old_dir->i_ino, old_dir->i_nlink, retval);
2318         }
2319
2320         if (new_inode) {
2321                 new_inode->i_nlink--;
2322                 new_inode->i_ctime = CURRENT_TIME_SEC;
2323         }
2324         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2325         ext3_update_dx_flag(old_dir);
2326         if (dir_bh) {
2327                 BUFFER_TRACE(dir_bh, "get_write_access");
2328                 ext3_journal_get_write_access(handle, dir_bh);
2329                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2330                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2331                 ext3_journal_dirty_metadata(handle, dir_bh);
2332                 old_dir->i_nlink--;
2333                 if (new_inode) {
2334                         new_inode->i_nlink--;
2335                 } else {
2336                         new_dir->i_nlink++;
2337                         ext3_update_dx_flag(new_dir);
2338                         ext3_mark_inode_dirty(handle, new_dir);
2339                 }
2340         }
2341         ext3_mark_inode_dirty(handle, old_dir);
2342         if (new_inode) {
2343                 ext3_mark_inode_dirty(handle, new_inode);
2344                 if (!new_inode->i_nlink)
2345                         ext3_orphan_add(handle, new_inode);
2346         }
2347         retval = 0;
2348
2349 end_rename:
2350         brelse (dir_bh);
2351         brelse (old_bh);
2352         brelse (new_bh);
2353         ext3_journal_stop(handle);
2354         return retval;
2355 }
2356
2357 /*
2358  * directories can handle most operations...
2359  */
2360 struct inode_operations ext3_dir_inode_operations = {
2361         .create         = ext3_create,
2362         .lookup         = ext3_lookup,
2363         .link           = ext3_link,
2364         .unlink         = ext3_unlink,
2365         .symlink        = ext3_symlink,
2366         .mkdir          = ext3_mkdir,
2367         .rmdir          = ext3_rmdir,
2368         .mknod          = ext3_mknod,
2369         .rename         = ext3_rename,
2370         .setattr        = ext3_setattr,
2371 #ifdef CONFIG_EXT3_FS_XATTR
2372         .setxattr       = generic_setxattr,
2373         .getxattr       = generic_getxattr,
2374         .listxattr      = ext3_listxattr,
2375         .removexattr    = generic_removexattr,
2376 #endif
2377         .permission     = ext3_permission,
2378 };
2379
2380 struct inode_operations ext3_special_inode_operations = {
2381         .setattr        = ext3_setattr,
2382 #ifdef CONFIG_EXT3_FS_XATTR
2383         .setxattr       = generic_setxattr,
2384         .getxattr       = generic_getxattr,
2385         .listxattr      = ext3_listxattr,
2386         .removexattr    = generic_removexattr,
2387 #endif
2388         .permission     = ext3_permission,
2389 };