Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[linux-2.6] / fs / ext3 / namei.c
1 /*
2  *  linux/fs/ext3/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/smp_lock.h>
39
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext3_append(handle_t *handle,
53                                         struct inode *inode,
54                                         u32 *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT3_I(inode)->i_disksize = inode->i_size;
63                 ext3_journal_get_write_access(handle,bh);
64         }
65         return bh;
66 }
67
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u32 offs;
144 };
145
146 #ifdef CONFIG_EXT3_INDEX
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned value);
149 static inline unsigned dx_get_hash (struct dx_entry *entry);
150 static void dx_set_hash (struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count (struct dx_entry *entries);
152 static unsigned dx_get_limit (struct dx_entry *entries);
153 static void dx_set_count (struct dx_entry *entries, unsigned value);
154 static void dx_set_limit (struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit (struct inode *dir);
157 static struct dx_frame *dx_probe(struct dentry *dentry,
158                                  struct inode *dir,
159                                  struct dx_hash_info *hinfo,
160                                  struct dx_frame *frame,
161                                  int *err);
162 static void dx_release (struct dx_frame *frames);
163 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167                 struct dx_map_entry *offsets, int count);
168 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171                                  struct dx_frame *frame,
172                                  struct dx_frame *frames,
173                                  __u32 *start_hash);
174 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175                        struct ext3_dir_entry_2 **res_dir, int *err);
176 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177                              struct inode *inode);
178
179 /*
180  * Future: use high four bits of block for coalesce-on-delete flags
181  * Mask them off for now.
182  */
183
184 static inline unsigned dx_get_block (struct dx_entry *entry)
185 {
186         return le32_to_cpu(entry->block) & 0x00ffffff;
187 }
188
189 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190 {
191         entry->block = cpu_to_le32(value);
192 }
193
194 static inline unsigned dx_get_hash (struct dx_entry *entry)
195 {
196         return le32_to_cpu(entry->hash);
197 }
198
199 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200 {
201         entry->hash = cpu_to_le32(value);
202 }
203
204 static inline unsigned dx_get_count (struct dx_entry *entries)
205 {
206         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207 }
208
209 static inline unsigned dx_get_limit (struct dx_entry *entries)
210 {
211         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212 }
213
214 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215 {
216         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217 }
218
219 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220 {
221         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222 }
223
224 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225 {
226         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
227                 EXT3_DIR_REC_LEN(2) - infosize;
228         return 0? 20: entry_space / sizeof(struct dx_entry);
229 }
230
231 static inline unsigned dx_node_limit (struct inode *dir)
232 {
233         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
234         return 0? 22: entry_space / sizeof(struct dx_entry);
235 }
236
237 /*
238  * Debug
239  */
240 #ifdef DX_DEBUG
241 static void dx_show_index (char * label, struct dx_entry *entries)
242 {
243         int i, n = dx_get_count (entries);
244         printk("%s index ", label);
245         for (i = 0; i < n; i++)
246         {
247                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
248         }
249         printk("\n");
250 }
251
252 struct stats
253 {
254         unsigned names;
255         unsigned space;
256         unsigned bcount;
257 };
258
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
260                                  int size, int show_names)
261 {
262         unsigned names = 0, space = 0;
263         char *base = (char *) de;
264         struct dx_hash_info h = *hinfo;
265
266         printk("names: ");
267         while ((char *) de < base + size)
268         {
269                 if (de->inode)
270                 {
271                         if (show_names)
272                         {
273                                 int len = de->name_len;
274                                 char *name = de->name;
275                                 while (len--) printk("%c", *name++);
276                                 ext3fs_dirhash(de->name, de->name_len, &h);
277                                 printk(":%x.%u ", h.hash,
278                                        ((char *) de - base));
279                         }
280                         space += EXT3_DIR_REC_LEN(de->name_len);
281                         names++;
282                 }
283                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284         }
285         printk("(%i)\n", names);
286         return (struct stats) { names, space, 1 };
287 }
288
289 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290                              struct dx_entry *entries, int levels)
291 {
292         unsigned blocksize = dir->i_sb->s_blocksize;
293         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294         unsigned bcount = 0;
295         struct buffer_head *bh;
296         int err;
297         printk("%i indexed blocks...\n", count);
298         for (i = 0; i < count; i++, entries++)
299         {
300                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302                 struct stats stats;
303                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
304                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
305                 stats = levels?
306                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
308                 names += stats.names;
309                 space += stats.space;
310                 bcount += stats.bcount;
311                 brelse (bh);
312         }
313         if (bcount)
314                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
315                         names, space/bcount,(space/bcount)*100/blocksize);
316         return (struct stats) { names, space, bcount};
317 }
318 #endif /* DX_DEBUG */
319
320 /*
321  * Probe for a directory leaf block to search.
322  *
323  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324  * error in the directory index, and the caller should fall back to
325  * searching the directory normally.  The callers of dx_probe **MUST**
326  * check for this error code, and make sure it never gets reflected
327  * back to userspace.
328  */
329 static struct dx_frame *
330 dx_probe(struct dentry *dentry, struct inode *dir,
331          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332 {
333         unsigned count, indirect;
334         struct dx_entry *at, *entries, *p, *q, *m;
335         struct dx_root *root;
336         struct buffer_head *bh;
337         struct dx_frame *frame = frame_in;
338         u32 hash;
339
340         frame->bh = NULL;
341         if (dentry)
342                 dir = dentry->d_parent->d_inode;
343         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
344                 goto fail;
345         root = (struct dx_root *) bh->b_data;
346         if (root->info.hash_version != DX_HASH_TEA &&
347             root->info.hash_version != DX_HASH_HALF_MD4 &&
348             root->info.hash_version != DX_HASH_LEGACY) {
349                 ext3_warning(dir->i_sb, __FUNCTION__,
350                              "Unrecognised inode hash code %d",
351                              root->info.hash_version);
352                 brelse(bh);
353                 *err = ERR_BAD_DX_DIR;
354                 goto fail;
355         }
356         hinfo->hash_version = root->info.hash_version;
357         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
358         if (dentry)
359                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
360         hash = hinfo->hash;
361
362         if (root->info.unused_flags & 1) {
363                 ext3_warning(dir->i_sb, __FUNCTION__,
364                              "Unimplemented inode hash flags: %#06x",
365                              root->info.unused_flags);
366                 brelse(bh);
367                 *err = ERR_BAD_DX_DIR;
368                 goto fail;
369         }
370
371         if ((indirect = root->info.indirect_levels) > 1) {
372                 ext3_warning(dir->i_sb, __FUNCTION__,
373                              "Unimplemented inode hash depth: %#06x",
374                              root->info.indirect_levels);
375                 brelse(bh);
376                 *err = ERR_BAD_DX_DIR;
377                 goto fail;
378         }
379
380         entries = (struct dx_entry *) (((char *)&root->info) +
381                                        root->info.info_length);
382         assert(dx_get_limit(entries) == dx_root_limit(dir,
383                                                       root->info.info_length));
384         dxtrace (printk("Look up %x", hash));
385         while (1)
386         {
387                 count = dx_get_count(entries);
388                 assert (count && count <= dx_get_limit(entries));
389                 p = entries + 1;
390                 q = entries + count - 1;
391                 while (p <= q)
392                 {
393                         m = p + (q - p)/2;
394                         dxtrace(printk("."));
395                         if (dx_get_hash(m) > hash)
396                                 q = m - 1;
397                         else
398                                 p = m + 1;
399                 }
400
401                 if (0) // linear search cross check
402                 {
403                         unsigned n = count - 1;
404                         at = entries;
405                         while (n--)
406                         {
407                                 dxtrace(printk(","));
408                                 if (dx_get_hash(++at) > hash)
409                                 {
410                                         at--;
411                                         break;
412                                 }
413                         }
414                         assert (at == p - 1);
415                 }
416
417                 at = p - 1;
418                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
419                 frame->bh = bh;
420                 frame->entries = entries;
421                 frame->at = at;
422                 if (!indirect--) return frame;
423                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
424                         goto fail2;
425                 at = entries = ((struct dx_node *) bh->b_data)->entries;
426                 assert (dx_get_limit(entries) == dx_node_limit (dir));
427                 frame++;
428         }
429 fail2:
430         while (frame >= frame_in) {
431                 brelse(frame->bh);
432                 frame--;
433         }
434 fail:
435         return NULL;
436 }
437
438 static void dx_release (struct dx_frame *frames)
439 {
440         if (frames[0].bh == NULL)
441                 return;
442
443         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
444                 brelse(frames[1].bh);
445         brelse(frames[0].bh);
446 }
447
448 /*
449  * This function increments the frame pointer to search the next leaf
450  * block, and reads in the necessary intervening nodes if the search
451  * should be necessary.  Whether or not the search is necessary is
452  * controlled by the hash parameter.  If the hash value is even, then
453  * the search is only continued if the next block starts with that
454  * hash value.  This is used if we are searching for a specific file.
455  *
456  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
457  *
458  * This function returns 1 if the caller should continue to search,
459  * or 0 if it should not.  If there is an error reading one of the
460  * index blocks, it will a negative error code.
461  *
462  * If start_hash is non-null, it will be filled in with the starting
463  * hash of the next page.
464  */
465 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
466                                  struct dx_frame *frame,
467                                  struct dx_frame *frames,
468                                  __u32 *start_hash)
469 {
470         struct dx_frame *p;
471         struct buffer_head *bh;
472         int err, num_frames = 0;
473         __u32 bhash;
474
475         p = frame;
476         /*
477          * Find the next leaf page by incrementing the frame pointer.
478          * If we run out of entries in the interior node, loop around and
479          * increment pointer in the parent node.  When we break out of
480          * this loop, num_frames indicates the number of interior
481          * nodes need to be read.
482          */
483         while (1) {
484                 if (++(p->at) < p->entries + dx_get_count(p->entries))
485                         break;
486                 if (p == frames)
487                         return 0;
488                 num_frames++;
489                 p--;
490         }
491
492         /*
493          * If the hash is 1, then continue only if the next page has a
494          * continuation hash of any value.  This is used for readdir
495          * handling.  Otherwise, check to see if the hash matches the
496          * desired contiuation hash.  If it doesn't, return since
497          * there's no point to read in the successive index pages.
498          */
499         bhash = dx_get_hash(p->at);
500         if (start_hash)
501                 *start_hash = bhash;
502         if ((hash & 1) == 0) {
503                 if ((bhash & ~1) != hash)
504                         return 0;
505         }
506         /*
507          * If the hash is HASH_NB_ALWAYS, we always go to the next
508          * block so no check is necessary
509          */
510         while (num_frames--) {
511                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
512                                       0, &err)))
513                         return err; /* Failure */
514                 p++;
515                 brelse (p->bh);
516                 p->bh = bh;
517                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
518         }
519         return 1;
520 }
521
522
523 /*
524  * p is at least 6 bytes before the end of page
525  */
526 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
527 {
528         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
529 }
530
531 /*
532  * This function fills a red-black tree with information from a
533  * directory block.  It returns the number directory entries loaded
534  * into the tree.  If there is an error it is returned in err.
535  */
536 static int htree_dirblock_to_tree(struct file *dir_file,
537                                   struct inode *dir, int block,
538                                   struct dx_hash_info *hinfo,
539                                   __u32 start_hash, __u32 start_minor_hash)
540 {
541         struct buffer_head *bh;
542         struct ext3_dir_entry_2 *de, *top;
543         int err, count = 0;
544
545         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
546         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
547                 return err;
548
549         de = (struct ext3_dir_entry_2 *) bh->b_data;
550         top = (struct ext3_dir_entry_2 *) ((char *) de +
551                                            dir->i_sb->s_blocksize -
552                                            EXT3_DIR_REC_LEN(0));
553         for (; de < top; de = ext3_next_entry(de)) {
554                 ext3fs_dirhash(de->name, de->name_len, hinfo);
555                 if ((hinfo->hash < start_hash) ||
556                     ((hinfo->hash == start_hash) &&
557                      (hinfo->minor_hash < start_minor_hash)))
558                         continue;
559                 if (de->inode == 0)
560                         continue;
561                 if ((err = ext3_htree_store_dirent(dir_file,
562                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
563                         brelse(bh);
564                         return err;
565                 }
566                 count++;
567         }
568         brelse(bh);
569         return count;
570 }
571
572
573 /*
574  * This function fills a red-black tree with information from a
575  * directory.  We start scanning the directory in hash order, starting
576  * at start_hash and start_minor_hash.
577  *
578  * This function returns the number of entries inserted into the tree,
579  * or a negative error code.
580  */
581 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
582                          __u32 start_minor_hash, __u32 *next_hash)
583 {
584         struct dx_hash_info hinfo;
585         struct ext3_dir_entry_2 *de;
586         struct dx_frame frames[2], *frame;
587         struct inode *dir;
588         int block, err;
589         int count = 0;
590         int ret;
591         __u32 hashval;
592
593         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
594                        start_minor_hash));
595         dir = dir_file->f_dentry->d_inode;
596         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
597                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
598                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
599                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
600                                                start_hash, start_minor_hash);
601                 *next_hash = ~0;
602                 return count;
603         }
604         hinfo.hash = start_hash;
605         hinfo.minor_hash = 0;
606         frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
607         if (!frame)
608                 return err;
609
610         /* Add '.' and '..' from the htree header */
611         if (!start_hash && !start_minor_hash) {
612                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
613                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
614                         goto errout;
615                 count++;
616         }
617         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
618                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
619                 de = ext3_next_entry(de);
620                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
621                         goto errout;
622                 count++;
623         }
624
625         while (1) {
626                 block = dx_get_block(frame->at);
627                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
628                                              start_hash, start_minor_hash);
629                 if (ret < 0) {
630                         err = ret;
631                         goto errout;
632                 }
633                 count += ret;
634                 hashval = ~0;
635                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
636                                             frame, frames, &hashval);
637                 *next_hash = hashval;
638                 if (ret < 0) {
639                         err = ret;
640                         goto errout;
641                 }
642                 /*
643                  * Stop if:  (a) there are no more entries, or
644                  * (b) we have inserted at least one entry and the
645                  * next hash value is not a continuation
646                  */
647                 if ((ret == 0) ||
648                     (count && ((hashval & 1) == 0)))
649                         break;
650         }
651         dx_release(frames);
652         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
653                        count, *next_hash));
654         return count;
655 errout:
656         dx_release(frames);
657         return (err);
658 }
659
660
661 /*
662  * Directory block splitting, compacting
663  */
664
665 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
666                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
667 {
668         int count = 0;
669         char *base = (char *) de;
670         struct dx_hash_info h = *hinfo;
671
672         while ((char *) de < base + size)
673         {
674                 if (de->name_len && de->inode) {
675                         ext3fs_dirhash(de->name, de->name_len, &h);
676                         map_tail--;
677                         map_tail->hash = h.hash;
678                         map_tail->offs = (u32) ((char *) de - base);
679                         count++;
680                         cond_resched();
681                 }
682                 /* XXX: do we need to check rec_len == 0 case? -Chris */
683                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
684         }
685         return count;
686 }
687
688 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
689 {
690         struct dx_map_entry *p, *q, *top = map + count - 1;
691         int more;
692         /* Combsort until bubble sort doesn't suck */
693         while (count > 2)
694         {
695                 count = count*10/13;
696                 if (count - 9 < 2) /* 9, 10 -> 11 */
697                         count = 11;
698                 for (p = top, q = p - count; q >= map; p--, q--)
699                         if (p->hash < q->hash)
700                                 swap(*p, *q);
701         }
702         /* Garden variety bubble sort */
703         do {
704                 more = 0;
705                 q = top;
706                 while (q-- > map)
707                 {
708                         if (q[1].hash >= q[0].hash)
709                                 continue;
710                         swap(*(q+1), *q);
711                         more = 1;
712                 }
713         } while(more);
714 }
715
716 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
717 {
718         struct dx_entry *entries = frame->entries;
719         struct dx_entry *old = frame->at, *new = old + 1;
720         int count = dx_get_count(entries);
721
722         assert(count < dx_get_limit(entries));
723         assert(old < entries + count);
724         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
725         dx_set_hash(new, hash);
726         dx_set_block(new, block);
727         dx_set_count(entries, count + 1);
728 }
729 #endif
730
731
732 static void ext3_update_dx_flag(struct inode *inode)
733 {
734         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
735                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
736                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
737 }
738
739 /*
740  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
741  *
742  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
743  * `de != NULL' is guaranteed by caller.
744  */
745 static inline int ext3_match (int len, const char * const name,
746                               struct ext3_dir_entry_2 * de)
747 {
748         if (len != de->name_len)
749                 return 0;
750         if (!de->inode)
751                 return 0;
752         return !memcmp(name, de->name, len);
753 }
754
755 /*
756  * Returns 0 if not found, -1 on failure, and 1 on success
757  */
758 static inline int search_dirblock(struct buffer_head * bh,
759                                   struct inode *dir,
760                                   struct dentry *dentry,
761                                   unsigned long offset,
762                                   struct ext3_dir_entry_2 ** res_dir)
763 {
764         struct ext3_dir_entry_2 * de;
765         char * dlimit;
766         int de_len;
767         const char *name = dentry->d_name.name;
768         int namelen = dentry->d_name.len;
769
770         de = (struct ext3_dir_entry_2 *) bh->b_data;
771         dlimit = bh->b_data + dir->i_sb->s_blocksize;
772         while ((char *) de < dlimit) {
773                 /* this code is executed quadratically often */
774                 /* do minimal checking `by hand' */
775
776                 if ((char *) de + namelen <= dlimit &&
777                     ext3_match (namelen, name, de)) {
778                         /* found a match - just to be sure, do a full check */
779                         if (!ext3_check_dir_entry("ext3_find_entry",
780                                                   dir, de, bh, offset))
781                                 return -1;
782                         *res_dir = de;
783                         return 1;
784                 }
785                 /* prevent looping on a bad block */
786                 de_len = le16_to_cpu(de->rec_len);
787                 if (de_len <= 0)
788                         return -1;
789                 offset += de_len;
790                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
791         }
792         return 0;
793 }
794
795
796 /*
797  *      ext3_find_entry()
798  *
799  * finds an entry in the specified directory with the wanted name. It
800  * returns the cache buffer in which the entry was found, and the entry
801  * itself (as a parameter - res_dir). It does NOT read the inode of the
802  * entry - you'll have to do that yourself if you want to.
803  *
804  * The returned buffer_head has ->b_count elevated.  The caller is expected
805  * to brelse() it when appropriate.
806  */
807 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
808                                         struct ext3_dir_entry_2 ** res_dir)
809 {
810         struct super_block * sb;
811         struct buffer_head * bh_use[NAMEI_RA_SIZE];
812         struct buffer_head * bh, *ret = NULL;
813         unsigned long start, block, b;
814         int ra_max = 0;         /* Number of bh's in the readahead
815                                    buffer, bh_use[] */
816         int ra_ptr = 0;         /* Current index into readahead
817                                    buffer */
818         int num = 0;
819         int nblocks, i, err;
820         struct inode *dir = dentry->d_parent->d_inode;
821         int namelen;
822         const u8 *name;
823         unsigned blocksize;
824
825         *res_dir = NULL;
826         sb = dir->i_sb;
827         blocksize = sb->s_blocksize;
828         namelen = dentry->d_name.len;
829         name = dentry->d_name.name;
830         if (namelen > EXT3_NAME_LEN)
831                 return NULL;
832 #ifdef CONFIG_EXT3_INDEX
833         if (is_dx(dir)) {
834                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
835                 /*
836                  * On success, or if the error was file not found,
837                  * return.  Otherwise, fall back to doing a search the
838                  * old fashioned way.
839                  */
840                 if (bh || (err != ERR_BAD_DX_DIR))
841                         return bh;
842                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
843         }
844 #endif
845         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
846         start = EXT3_I(dir)->i_dir_start_lookup;
847         if (start >= nblocks)
848                 start = 0;
849         block = start;
850 restart:
851         do {
852                 /*
853                  * We deal with the read-ahead logic here.
854                  */
855                 if (ra_ptr >= ra_max) {
856                         /* Refill the readahead buffer */
857                         ra_ptr = 0;
858                         b = block;
859                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
860                                 /*
861                                  * Terminate if we reach the end of the
862                                  * directory and must wrap, or if our
863                                  * search has finished at this block.
864                                  */
865                                 if (b >= nblocks || (num && block == start)) {
866                                         bh_use[ra_max] = NULL;
867                                         break;
868                                 }
869                                 num++;
870                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
871                                 bh_use[ra_max] = bh;
872                                 if (bh)
873                                         ll_rw_block(READ, 1, &bh);
874                         }
875                 }
876                 if ((bh = bh_use[ra_ptr++]) == NULL)
877                         goto next;
878                 wait_on_buffer(bh);
879                 if (!buffer_uptodate(bh)) {
880                         /* read error, skip block & hope for the best */
881                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
882                                    "offset %lu", dir->i_ino, block);
883                         brelse(bh);
884                         goto next;
885                 }
886                 i = search_dirblock(bh, dir, dentry,
887                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
888                 if (i == 1) {
889                         EXT3_I(dir)->i_dir_start_lookup = block;
890                         ret = bh;
891                         goto cleanup_and_exit;
892                 } else {
893                         brelse(bh);
894                         if (i < 0)
895                                 goto cleanup_and_exit;
896                 }
897         next:
898                 if (++block >= nblocks)
899                         block = 0;
900         } while (block != start);
901
902         /*
903          * If the directory has grown while we were searching, then
904          * search the last part of the directory before giving up.
905          */
906         block = nblocks;
907         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
908         if (block < nblocks) {
909                 start = 0;
910                 goto restart;
911         }
912
913 cleanup_and_exit:
914         /* Clean up the read-ahead blocks */
915         for (; ra_ptr < ra_max; ra_ptr++)
916                 brelse (bh_use[ra_ptr]);
917         return ret;
918 }
919
920 #ifdef CONFIG_EXT3_INDEX
921 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
922                        struct ext3_dir_entry_2 **res_dir, int *err)
923 {
924         struct super_block * sb;
925         struct dx_hash_info     hinfo;
926         u32 hash;
927         struct dx_frame frames[2], *frame;
928         struct ext3_dir_entry_2 *de, *top;
929         struct buffer_head *bh;
930         unsigned long block;
931         int retval;
932         int namelen = dentry->d_name.len;
933         const u8 *name = dentry->d_name.name;
934         struct inode *dir = dentry->d_parent->d_inode;
935
936         sb = dir->i_sb;
937         /* NFS may look up ".." - look at dx_root directory block */
938         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
939                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
940                         return NULL;
941         } else {
942                 frame = frames;
943                 frame->bh = NULL;                       /* for dx_release() */
944                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
945                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
946         }
947         hash = hinfo.hash;
948         do {
949                 block = dx_get_block(frame->at);
950                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
951                         goto errout;
952                 de = (struct ext3_dir_entry_2 *) bh->b_data;
953                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
954                                        EXT3_DIR_REC_LEN(0));
955                 for (; de < top; de = ext3_next_entry(de))
956                 if (ext3_match (namelen, name, de)) {
957                         if (!ext3_check_dir_entry("ext3_find_entry",
958                                                   dir, de, bh,
959                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
960                                           +((char *)de - bh->b_data))) {
961                                 brelse (bh);
962                                 goto errout;
963                         }
964                         *res_dir = de;
965                         dx_release (frames);
966                         return bh;
967                 }
968                 brelse (bh);
969                 /* Check to see if we should continue to search */
970                 retval = ext3_htree_next_block(dir, hash, frame,
971                                                frames, NULL);
972                 if (retval < 0) {
973                         ext3_warning(sb, __FUNCTION__,
974                              "error reading index page in directory #%lu",
975                              dir->i_ino);
976                         *err = retval;
977                         goto errout;
978                 }
979         } while (retval == 1);
980
981         *err = -ENOENT;
982 errout:
983         dxtrace(printk("%s not found\n", name));
984         dx_release (frames);
985         return NULL;
986 }
987 #endif
988
989 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
990 {
991         struct inode * inode;
992         struct ext3_dir_entry_2 * de;
993         struct buffer_head * bh;
994
995         if (dentry->d_name.len > EXT3_NAME_LEN)
996                 return ERR_PTR(-ENAMETOOLONG);
997
998         bh = ext3_find_entry(dentry, &de);
999         inode = NULL;
1000         if (bh) {
1001                 unsigned long ino = le32_to_cpu(de->inode);
1002                 brelse (bh);
1003                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1004                         ext3_error(dir->i_sb, "ext3_lookup",
1005                                    "bad inode number: %lu", ino);
1006                         inode = NULL;
1007                 } else
1008                         inode = iget(dir->i_sb, ino);
1009
1010                 if (!inode)
1011                         return ERR_PTR(-EACCES);
1012         }
1013         return d_splice_alias(inode, dentry);
1014 }
1015
1016
1017 struct dentry *ext3_get_parent(struct dentry *child)
1018 {
1019         unsigned long ino;
1020         struct dentry *parent;
1021         struct inode *inode;
1022         struct dentry dotdot;
1023         struct ext3_dir_entry_2 * de;
1024         struct buffer_head *bh;
1025
1026         dotdot.d_name.name = "..";
1027         dotdot.d_name.len = 2;
1028         dotdot.d_parent = child; /* confusing, isn't it! */
1029
1030         bh = ext3_find_entry(&dotdot, &de);
1031         inode = NULL;
1032         if (!bh)
1033                 return ERR_PTR(-ENOENT);
1034         ino = le32_to_cpu(de->inode);
1035         brelse(bh);
1036
1037         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1038                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1039                            "bad inode number: %lu", ino);
1040                 inode = NULL;
1041         } else
1042                 inode = iget(child->d_inode->i_sb, ino);
1043
1044         if (!inode)
1045                 return ERR_PTR(-EACCES);
1046
1047         parent = d_alloc_anon(inode);
1048         if (!parent) {
1049                 iput(inode);
1050                 parent = ERR_PTR(-ENOMEM);
1051         }
1052         return parent;
1053 }
1054
1055 #define S_SHIFT 12
1056 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1057         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1058         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1059         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1060         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1061         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1062         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1063         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1064 };
1065
1066 static inline void ext3_set_de_type(struct super_block *sb,
1067                                 struct ext3_dir_entry_2 *de,
1068                                 umode_t mode) {
1069         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1070                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1071 }
1072
1073 #ifdef CONFIG_EXT3_INDEX
1074 static struct ext3_dir_entry_2 *
1075 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1076 {
1077         unsigned rec_len = 0;
1078
1079         while (count--) {
1080                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1081                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1082                 memcpy (to, de, rec_len);
1083                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1084                                 cpu_to_le16(rec_len);
1085                 de->inode = 0;
1086                 map++;
1087                 to += rec_len;
1088         }
1089         return (struct ext3_dir_entry_2 *) (to - rec_len);
1090 }
1091
1092 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1093 {
1094         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1095         unsigned rec_len = 0;
1096
1097         prev = to = de;
1098         while ((char*)de < base + size) {
1099                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1100                                                     le16_to_cpu(de->rec_len));
1101                 if (de->inode && de->name_len) {
1102                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1103                         if (de > to)
1104                                 memmove(to, de, rec_len);
1105                         to->rec_len = cpu_to_le16(rec_len);
1106                         prev = to;
1107                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1108                 }
1109                 de = next;
1110         }
1111         return prev;
1112 }
1113
1114 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1115                         struct buffer_head **bh,struct dx_frame *frame,
1116                         struct dx_hash_info *hinfo, int *error)
1117 {
1118         unsigned blocksize = dir->i_sb->s_blocksize;
1119         unsigned count, continued;
1120         struct buffer_head *bh2;
1121         u32 newblock;
1122         u32 hash2;
1123         struct dx_map_entry *map;
1124         char *data1 = (*bh)->b_data, *data2;
1125         unsigned split;
1126         struct ext3_dir_entry_2 *de = NULL, *de2;
1127         int     err;
1128
1129         bh2 = ext3_append (handle, dir, &newblock, error);
1130         if (!(bh2)) {
1131                 brelse(*bh);
1132                 *bh = NULL;
1133                 goto errout;
1134         }
1135
1136         BUFFER_TRACE(*bh, "get_write_access");
1137         err = ext3_journal_get_write_access(handle, *bh);
1138         if (err) {
1139         journal_error:
1140                 brelse(*bh);
1141                 brelse(bh2);
1142                 *bh = NULL;
1143                 ext3_std_error(dir->i_sb, err);
1144                 goto errout;
1145         }
1146         BUFFER_TRACE(frame->bh, "get_write_access");
1147         err = ext3_journal_get_write_access(handle, frame->bh);
1148         if (err)
1149                 goto journal_error;
1150
1151         data2 = bh2->b_data;
1152
1153         /* create map in the end of data2 block */
1154         map = (struct dx_map_entry *) (data2 + blocksize);
1155         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1156                              blocksize, hinfo, map);
1157         map -= count;
1158         split = count/2; // need to adjust to actual middle
1159         dx_sort_map (map, count);
1160         hash2 = map[split].hash;
1161         continued = hash2 == map[split - 1].hash;
1162         dxtrace(printk("Split block %i at %x, %i/%i\n",
1163                 dx_get_block(frame->at), hash2, split, count-split));
1164
1165         /* Fancy dance to stay within two buffers */
1166         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1167         de = dx_pack_dirents(data1,blocksize);
1168         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1169         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1170         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1171         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1172
1173         /* Which block gets the new entry? */
1174         if (hinfo->hash >= hash2)
1175         {
1176                 swap(*bh, bh2);
1177                 de = de2;
1178         }
1179         dx_insert_block (frame, hash2 + continued, newblock);
1180         err = ext3_journal_dirty_metadata (handle, bh2);
1181         if (err)
1182                 goto journal_error;
1183         err = ext3_journal_dirty_metadata (handle, frame->bh);
1184         if (err)
1185                 goto journal_error;
1186         brelse (bh2);
1187         dxtrace(dx_show_index ("frame", frame->entries));
1188 errout:
1189         return de;
1190 }
1191 #endif
1192
1193
1194 /*
1195  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1196  * it points to a directory entry which is guaranteed to be large
1197  * enough for new directory entry.  If de is NULL, then
1198  * add_dirent_to_buf will attempt search the directory block for
1199  * space.  It will return -ENOSPC if no space is available, and -EIO
1200  * and -EEXIST if directory entry already exists.
1201  *
1202  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1203  * all other cases bh is released.
1204  */
1205 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1206                              struct inode *inode, struct ext3_dir_entry_2 *de,
1207                              struct buffer_head * bh)
1208 {
1209         struct inode    *dir = dentry->d_parent->d_inode;
1210         const char      *name = dentry->d_name.name;
1211         int             namelen = dentry->d_name.len;
1212         unsigned long   offset = 0;
1213         unsigned short  reclen;
1214         int             nlen, rlen, err;
1215         char            *top;
1216
1217         reclen = EXT3_DIR_REC_LEN(namelen);
1218         if (!de) {
1219                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1220                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1221                 while ((char *) de <= top) {
1222                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1223                                                   bh, offset)) {
1224                                 brelse (bh);
1225                                 return -EIO;
1226                         }
1227                         if (ext3_match (namelen, name, de)) {
1228                                 brelse (bh);
1229                                 return -EEXIST;
1230                         }
1231                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1232                         rlen = le16_to_cpu(de->rec_len);
1233                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1234                                 break;
1235                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1236                         offset += rlen;
1237                 }
1238                 if ((char *) de > top)
1239                         return -ENOSPC;
1240         }
1241         BUFFER_TRACE(bh, "get_write_access");
1242         err = ext3_journal_get_write_access(handle, bh);
1243         if (err) {
1244                 ext3_std_error(dir->i_sb, err);
1245                 brelse(bh);
1246                 return err;
1247         }
1248
1249         /* By now the buffer is marked for journaling */
1250         nlen = EXT3_DIR_REC_LEN(de->name_len);
1251         rlen = le16_to_cpu(de->rec_len);
1252         if (de->inode) {
1253                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1254                 de1->rec_len = cpu_to_le16(rlen - nlen);
1255                 de->rec_len = cpu_to_le16(nlen);
1256                 de = de1;
1257         }
1258         de->file_type = EXT3_FT_UNKNOWN;
1259         if (inode) {
1260                 de->inode = cpu_to_le32(inode->i_ino);
1261                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1262         } else
1263                 de->inode = 0;
1264         de->name_len = namelen;
1265         memcpy (de->name, name, namelen);
1266         /*
1267          * XXX shouldn't update any times until successful
1268          * completion of syscall, but too many callers depend
1269          * on this.
1270          *
1271          * XXX similarly, too many callers depend on
1272          * ext3_new_inode() setting the times, but error
1273          * recovery deletes the inode, so the worst that can
1274          * happen is that the times are slightly out of date
1275          * and/or different from the directory change time.
1276          */
1277         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1278         ext3_update_dx_flag(dir);
1279         dir->i_version++;
1280         ext3_mark_inode_dirty(handle, dir);
1281         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1282         err = ext3_journal_dirty_metadata(handle, bh);
1283         if (err)
1284                 ext3_std_error(dir->i_sb, err);
1285         brelse(bh);
1286         return 0;
1287 }
1288
1289 #ifdef CONFIG_EXT3_INDEX
1290 /*
1291  * This converts a one block unindexed directory to a 3 block indexed
1292  * directory, and adds the dentry to the indexed directory.
1293  */
1294 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1295                             struct inode *inode, struct buffer_head *bh)
1296 {
1297         struct inode    *dir = dentry->d_parent->d_inode;
1298         const char      *name = dentry->d_name.name;
1299         int             namelen = dentry->d_name.len;
1300         struct buffer_head *bh2;
1301         struct dx_root  *root;
1302         struct dx_frame frames[2], *frame;
1303         struct dx_entry *entries;
1304         struct ext3_dir_entry_2 *de, *de2;
1305         char            *data1, *top;
1306         unsigned        len;
1307         int             retval;
1308         unsigned        blocksize;
1309         struct dx_hash_info hinfo;
1310         u32             block;
1311         struct fake_dirent *fde;
1312
1313         blocksize =  dir->i_sb->s_blocksize;
1314         dxtrace(printk("Creating index\n"));
1315         retval = ext3_journal_get_write_access(handle, bh);
1316         if (retval) {
1317                 ext3_std_error(dir->i_sb, retval);
1318                 brelse(bh);
1319                 return retval;
1320         }
1321         root = (struct dx_root *) bh->b_data;
1322
1323         bh2 = ext3_append (handle, dir, &block, &retval);
1324         if (!(bh2)) {
1325                 brelse(bh);
1326                 return retval;
1327         }
1328         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1329         data1 = bh2->b_data;
1330
1331         /* The 0th block becomes the root, move the dirents out */
1332         fde = &root->dotdot;
1333         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1334         len = ((char *) root) + blocksize - (char *) de;
1335         memcpy (data1, de, len);
1336         de = (struct ext3_dir_entry_2 *) data1;
1337         top = data1 + len;
1338         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1339                 de = de2;
1340         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1341         /* Initialize the root; the dot dirents already exist */
1342         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1343         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1344         memset (&root->info, 0, sizeof(root->info));
1345         root->info.info_length = sizeof(root->info);
1346         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1347         entries = root->entries;
1348         dx_set_block (entries, 1);
1349         dx_set_count (entries, 1);
1350         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1351
1352         /* Initialize as for dx_probe */
1353         hinfo.hash_version = root->info.hash_version;
1354         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1355         ext3fs_dirhash(name, namelen, &hinfo);
1356         frame = frames;
1357         frame->entries = entries;
1358         frame->at = entries;
1359         frame->bh = bh;
1360         bh = bh2;
1361         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1362         dx_release (frames);
1363         if (!(de))
1364                 return retval;
1365
1366         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1367 }
1368 #endif
1369
1370 /*
1371  *      ext3_add_entry()
1372  *
1373  * adds a file entry to the specified directory, using the same
1374  * semantics as ext3_find_entry(). It returns NULL if it failed.
1375  *
1376  * NOTE!! The inode part of 'de' is left at 0 - which means you
1377  * may not sleep between calling this and putting something into
1378  * the entry, as someone else might have used it while you slept.
1379  */
1380 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1381         struct inode *inode)
1382 {
1383         struct inode *dir = dentry->d_parent->d_inode;
1384         unsigned long offset;
1385         struct buffer_head * bh;
1386         struct ext3_dir_entry_2 *de;
1387         struct super_block * sb;
1388         int     retval;
1389 #ifdef CONFIG_EXT3_INDEX
1390         int     dx_fallback=0;
1391 #endif
1392         unsigned blocksize;
1393         u32 block, blocks;
1394
1395         sb = dir->i_sb;
1396         blocksize = sb->s_blocksize;
1397         if (!dentry->d_name.len)
1398                 return -EINVAL;
1399 #ifdef CONFIG_EXT3_INDEX
1400         if (is_dx(dir)) {
1401                 retval = ext3_dx_add_entry(handle, dentry, inode);
1402                 if (!retval || (retval != ERR_BAD_DX_DIR))
1403                         return retval;
1404                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1405                 dx_fallback++;
1406                 ext3_mark_inode_dirty(handle, dir);
1407         }
1408 #endif
1409         blocks = dir->i_size >> sb->s_blocksize_bits;
1410         for (block = 0, offset = 0; block < blocks; block++) {
1411                 bh = ext3_bread(handle, dir, block, 0, &retval);
1412                 if(!bh)
1413                         return retval;
1414                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1415                 if (retval != -ENOSPC)
1416                         return retval;
1417
1418 #ifdef CONFIG_EXT3_INDEX
1419                 if (blocks == 1 && !dx_fallback &&
1420                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1421                         return make_indexed_dir(handle, dentry, inode, bh);
1422 #endif
1423                 brelse(bh);
1424         }
1425         bh = ext3_append(handle, dir, &block, &retval);
1426         if (!bh)
1427                 return retval;
1428         de = (struct ext3_dir_entry_2 *) bh->b_data;
1429         de->inode = 0;
1430         de->rec_len = cpu_to_le16(blocksize);
1431         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1432 }
1433
1434 #ifdef CONFIG_EXT3_INDEX
1435 /*
1436  * Returns 0 for success, or a negative error value
1437  */
1438 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1439                              struct inode *inode)
1440 {
1441         struct dx_frame frames[2], *frame;
1442         struct dx_entry *entries, *at;
1443         struct dx_hash_info hinfo;
1444         struct buffer_head * bh;
1445         struct inode *dir = dentry->d_parent->d_inode;
1446         struct super_block * sb = dir->i_sb;
1447         struct ext3_dir_entry_2 *de;
1448         int err;
1449
1450         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1451         if (!frame)
1452                 return err;
1453         entries = frame->entries;
1454         at = frame->at;
1455
1456         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1457                 goto cleanup;
1458
1459         BUFFER_TRACE(bh, "get_write_access");
1460         err = ext3_journal_get_write_access(handle, bh);
1461         if (err)
1462                 goto journal_error;
1463
1464         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1465         if (err != -ENOSPC) {
1466                 bh = NULL;
1467                 goto cleanup;
1468         }
1469
1470         /* Block full, should compress but for now just split */
1471         dxtrace(printk("using %u of %u node entries\n",
1472                        dx_get_count(entries), dx_get_limit(entries)));
1473         /* Need to split index? */
1474         if (dx_get_count(entries) == dx_get_limit(entries)) {
1475                 u32 newblock;
1476                 unsigned icount = dx_get_count(entries);
1477                 int levels = frame - frames;
1478                 struct dx_entry *entries2;
1479                 struct dx_node *node2;
1480                 struct buffer_head *bh2;
1481
1482                 if (levels && (dx_get_count(frames->entries) ==
1483                                dx_get_limit(frames->entries))) {
1484                         ext3_warning(sb, __FUNCTION__,
1485                                      "Directory index full!");
1486                         err = -ENOSPC;
1487                         goto cleanup;
1488                 }
1489                 bh2 = ext3_append (handle, dir, &newblock, &err);
1490                 if (!(bh2))
1491                         goto cleanup;
1492                 node2 = (struct dx_node *)(bh2->b_data);
1493                 entries2 = node2->entries;
1494                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1495                 node2->fake.inode = 0;
1496                 BUFFER_TRACE(frame->bh, "get_write_access");
1497                 err = ext3_journal_get_write_access(handle, frame->bh);
1498                 if (err)
1499                         goto journal_error;
1500                 if (levels) {
1501                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1502                         unsigned hash2 = dx_get_hash(entries + icount1);
1503                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1504
1505                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1506                         err = ext3_journal_get_write_access(handle,
1507                                                              frames[0].bh);
1508                         if (err)
1509                                 goto journal_error;
1510
1511                         memcpy ((char *) entries2, (char *) (entries + icount1),
1512                                 icount2 * sizeof(struct dx_entry));
1513                         dx_set_count (entries, icount1);
1514                         dx_set_count (entries2, icount2);
1515                         dx_set_limit (entries2, dx_node_limit(dir));
1516
1517                         /* Which index block gets the new entry? */
1518                         if (at - entries >= icount1) {
1519                                 frame->at = at = at - entries - icount1 + entries2;
1520                                 frame->entries = entries = entries2;
1521                                 swap(frame->bh, bh2);
1522                         }
1523                         dx_insert_block (frames + 0, hash2, newblock);
1524                         dxtrace(dx_show_index ("node", frames[1].entries));
1525                         dxtrace(dx_show_index ("node",
1526                                ((struct dx_node *) bh2->b_data)->entries));
1527                         err = ext3_journal_dirty_metadata(handle, bh2);
1528                         if (err)
1529                                 goto journal_error;
1530                         brelse (bh2);
1531                 } else {
1532                         dxtrace(printk("Creating second level index...\n"));
1533                         memcpy((char *) entries2, (char *) entries,
1534                                icount * sizeof(struct dx_entry));
1535                         dx_set_limit(entries2, dx_node_limit(dir));
1536
1537                         /* Set up root */
1538                         dx_set_count(entries, 1);
1539                         dx_set_block(entries + 0, newblock);
1540                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1541
1542                         /* Add new access path frame */
1543                         frame = frames + 1;
1544                         frame->at = at = at - entries + entries2;
1545                         frame->entries = entries = entries2;
1546                         frame->bh = bh2;
1547                         err = ext3_journal_get_write_access(handle,
1548                                                              frame->bh);
1549                         if (err)
1550                                 goto journal_error;
1551                 }
1552                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1553         }
1554         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1555         if (!de)
1556                 goto cleanup;
1557         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1558         bh = NULL;
1559         goto cleanup;
1560
1561 journal_error:
1562         ext3_std_error(dir->i_sb, err);
1563 cleanup:
1564         if (bh)
1565                 brelse(bh);
1566         dx_release(frames);
1567         return err;
1568 }
1569 #endif
1570
1571 /*
1572  * ext3_delete_entry deletes a directory entry by merging it with the
1573  * previous entry
1574  */
1575 static int ext3_delete_entry (handle_t *handle,
1576                               struct inode * dir,
1577                               struct ext3_dir_entry_2 * de_del,
1578                               struct buffer_head * bh)
1579 {
1580         struct ext3_dir_entry_2 * de, * pde;
1581         int i;
1582
1583         i = 0;
1584         pde = NULL;
1585         de = (struct ext3_dir_entry_2 *) bh->b_data;
1586         while (i < bh->b_size) {
1587                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1588                         return -EIO;
1589                 if (de == de_del)  {
1590                         BUFFER_TRACE(bh, "get_write_access");
1591                         ext3_journal_get_write_access(handle, bh);
1592                         if (pde)
1593                                 pde->rec_len =
1594                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1595                                                     le16_to_cpu(de->rec_len));
1596                         else
1597                                 de->inode = 0;
1598                         dir->i_version++;
1599                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1600                         ext3_journal_dirty_metadata(handle, bh);
1601                         return 0;
1602                 }
1603                 i += le16_to_cpu(de->rec_len);
1604                 pde = de;
1605                 de = (struct ext3_dir_entry_2 *)
1606                         ((char *) de + le16_to_cpu(de->rec_len));
1607         }
1608         return -ENOENT;
1609 }
1610
1611 /*
1612  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1613  * do not perform it in these functions.  We perform it at the call site,
1614  * if it is needed.
1615  */
1616 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1617 {
1618         inode->i_nlink++;
1619 }
1620
1621 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1622 {
1623         inode->i_nlink--;
1624 }
1625
1626 static int ext3_add_nondir(handle_t *handle,
1627                 struct dentry *dentry, struct inode *inode)
1628 {
1629         int err = ext3_add_entry(handle, dentry, inode);
1630         if (!err) {
1631                 ext3_mark_inode_dirty(handle, inode);
1632                 d_instantiate(dentry, inode);
1633                 return 0;
1634         }
1635         ext3_dec_count(handle, inode);
1636         iput(inode);
1637         return err;
1638 }
1639
1640 /*
1641  * By the time this is called, we already have created
1642  * the directory cache entry for the new file, but it
1643  * is so far negative - it has no inode.
1644  *
1645  * If the create succeeds, we fill in the inode information
1646  * with d_instantiate().
1647  */
1648 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1649                 struct nameidata *nd)
1650 {
1651         handle_t *handle;
1652         struct inode * inode;
1653         int err, retries = 0;
1654
1655 retry:
1656         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1657                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1658                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1659         if (IS_ERR(handle))
1660                 return PTR_ERR(handle);
1661
1662         if (IS_DIRSYNC(dir))
1663                 handle->h_sync = 1;
1664
1665         inode = ext3_new_inode (handle, dir, mode);
1666         err = PTR_ERR(inode);
1667         if (!IS_ERR(inode)) {
1668                 inode->i_op = &ext3_file_inode_operations;
1669                 inode->i_fop = &ext3_file_operations;
1670                 ext3_set_aops(inode);
1671                 err = ext3_add_nondir(handle, dentry, inode);
1672         }
1673         ext3_journal_stop(handle);
1674         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1675                 goto retry;
1676         return err;
1677 }
1678
1679 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1680                         int mode, dev_t rdev)
1681 {
1682         handle_t *handle;
1683         struct inode *inode;
1684         int err, retries = 0;
1685
1686         if (!new_valid_dev(rdev))
1687                 return -EINVAL;
1688
1689 retry:
1690         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1691                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1692                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1693         if (IS_ERR(handle))
1694                 return PTR_ERR(handle);
1695
1696         if (IS_DIRSYNC(dir))
1697                 handle->h_sync = 1;
1698
1699         inode = ext3_new_inode (handle, dir, mode);
1700         err = PTR_ERR(inode);
1701         if (!IS_ERR(inode)) {
1702                 init_special_inode(inode, inode->i_mode, rdev);
1703 #ifdef CONFIG_EXT3_FS_XATTR
1704                 inode->i_op = &ext3_special_inode_operations;
1705 #endif
1706                 err = ext3_add_nondir(handle, dentry, inode);
1707         }
1708         ext3_journal_stop(handle);
1709         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1710                 goto retry;
1711         return err;
1712 }
1713
1714 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1715 {
1716         handle_t *handle;
1717         struct inode * inode;
1718         struct buffer_head * dir_block;
1719         struct ext3_dir_entry_2 * de;
1720         int err, retries = 0;
1721
1722         if (dir->i_nlink >= EXT3_LINK_MAX)
1723                 return -EMLINK;
1724
1725 retry:
1726         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1727                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1728                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1729         if (IS_ERR(handle))
1730                 return PTR_ERR(handle);
1731
1732         if (IS_DIRSYNC(dir))
1733                 handle->h_sync = 1;
1734
1735         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1736         err = PTR_ERR(inode);
1737         if (IS_ERR(inode))
1738                 goto out_stop;
1739
1740         inode->i_op = &ext3_dir_inode_operations;
1741         inode->i_fop = &ext3_dir_operations;
1742         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1743         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1744         if (!dir_block) {
1745                 inode->i_nlink--; /* is this nlink == 0? */
1746                 ext3_mark_inode_dirty(handle, inode);
1747                 iput (inode);
1748                 goto out_stop;
1749         }
1750         BUFFER_TRACE(dir_block, "get_write_access");
1751         ext3_journal_get_write_access(handle, dir_block);
1752         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1753         de->inode = cpu_to_le32(inode->i_ino);
1754         de->name_len = 1;
1755         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1756         strcpy (de->name, ".");
1757         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1758         de = (struct ext3_dir_entry_2 *)
1759                         ((char *) de + le16_to_cpu(de->rec_len));
1760         de->inode = cpu_to_le32(dir->i_ino);
1761         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1762         de->name_len = 2;
1763         strcpy (de->name, "..");
1764         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1765         inode->i_nlink = 2;
1766         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1767         ext3_journal_dirty_metadata(handle, dir_block);
1768         brelse (dir_block);
1769         ext3_mark_inode_dirty(handle, inode);
1770         err = ext3_add_entry (handle, dentry, inode);
1771         if (err) {
1772                 inode->i_nlink = 0;
1773                 ext3_mark_inode_dirty(handle, inode);
1774                 iput (inode);
1775                 goto out_stop;
1776         }
1777         dir->i_nlink++;
1778         ext3_update_dx_flag(dir);
1779         ext3_mark_inode_dirty(handle, dir);
1780         d_instantiate(dentry, inode);
1781 out_stop:
1782         ext3_journal_stop(handle);
1783         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1784                 goto retry;
1785         return err;
1786 }
1787
1788 /*
1789  * routine to check that the specified directory is empty (for rmdir)
1790  */
1791 static int empty_dir (struct inode * inode)
1792 {
1793         unsigned long offset;
1794         struct buffer_head * bh;
1795         struct ext3_dir_entry_2 * de, * de1;
1796         struct super_block * sb;
1797         int err = 0;
1798
1799         sb = inode->i_sb;
1800         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1801             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1802                 if (err)
1803                         ext3_error(inode->i_sb, __FUNCTION__,
1804                                    "error %d reading directory #%lu offset 0",
1805                                    err, inode->i_ino);
1806                 else
1807                         ext3_warning(inode->i_sb, __FUNCTION__,
1808                                      "bad directory (dir #%lu) - no data block",
1809                                      inode->i_ino);
1810                 return 1;
1811         }
1812         de = (struct ext3_dir_entry_2 *) bh->b_data;
1813         de1 = (struct ext3_dir_entry_2 *)
1814                         ((char *) de + le16_to_cpu(de->rec_len));
1815         if (le32_to_cpu(de->inode) != inode->i_ino ||
1816                         !le32_to_cpu(de1->inode) ||
1817                         strcmp (".", de->name) ||
1818                         strcmp ("..", de1->name)) {
1819                 ext3_warning (inode->i_sb, "empty_dir",
1820                               "bad directory (dir #%lu) - no `.' or `..'",
1821                               inode->i_ino);
1822                 brelse (bh);
1823                 return 1;
1824         }
1825         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1826         de = (struct ext3_dir_entry_2 *)
1827                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1828         while (offset < inode->i_size ) {
1829                 if (!bh ||
1830                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1831                         err = 0;
1832                         brelse (bh);
1833                         bh = ext3_bread (NULL, inode,
1834                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1835                         if (!bh) {
1836                                 if (err)
1837                                         ext3_error(sb, __FUNCTION__,
1838                                                    "error %d reading directory"
1839                                                    " #%lu offset %lu",
1840                                                    err, inode->i_ino, offset);
1841                                 offset += sb->s_blocksize;
1842                                 continue;
1843                         }
1844                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1845                 }
1846                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1847                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1848                                                          sb->s_blocksize);
1849                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1850                         continue;
1851                 }
1852                 if (le32_to_cpu(de->inode)) {
1853                         brelse (bh);
1854                         return 0;
1855                 }
1856                 offset += le16_to_cpu(de->rec_len);
1857                 de = (struct ext3_dir_entry_2 *)
1858                                 ((char *) de + le16_to_cpu(de->rec_len));
1859         }
1860         brelse (bh);
1861         return 1;
1862 }
1863
1864 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1865  * such inodes, starting at the superblock, in case we crash before the
1866  * file is closed/deleted, or in case the inode truncate spans multiple
1867  * transactions and the last transaction is not recovered after a crash.
1868  *
1869  * At filesystem recovery time, we walk this list deleting unlinked
1870  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1871  */
1872 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1873 {
1874         struct super_block *sb = inode->i_sb;
1875         struct ext3_iloc iloc;
1876         int err = 0, rc;
1877
1878         lock_super(sb);
1879         if (!list_empty(&EXT3_I(inode)->i_orphan))
1880                 goto out_unlock;
1881
1882         /* Orphan handling is only valid for files with data blocks
1883          * being truncated, or files being unlinked. */
1884
1885         /* @@@ FIXME: Observation from aviro:
1886          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1887          * here (on lock_super()), so race with ext3_link() which might bump
1888          * ->i_nlink. For, say it, character device. Not a regular file,
1889          * not a directory, not a symlink and ->i_nlink > 0.
1890          */
1891         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1892                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1893
1894         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1895         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1896         if (err)
1897                 goto out_unlock;
1898
1899         err = ext3_reserve_inode_write(handle, inode, &iloc);
1900         if (err)
1901                 goto out_unlock;
1902
1903         /* Insert this inode at the head of the on-disk orphan list... */
1904         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1905         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1906         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1907         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1908         if (!err)
1909                 err = rc;
1910
1911         /* Only add to the head of the in-memory list if all the
1912          * previous operations succeeded.  If the orphan_add is going to
1913          * fail (possibly taking the journal offline), we can't risk
1914          * leaving the inode on the orphan list: stray orphan-list
1915          * entries can cause panics at unmount time.
1916          *
1917          * This is safe: on error we're going to ignore the orphan list
1918          * anyway on the next recovery. */
1919         if (!err)
1920                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1921
1922         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1923         jbd_debug(4, "orphan inode %lu will point to %d\n",
1924                         inode->i_ino, NEXT_ORPHAN(inode));
1925 out_unlock:
1926         unlock_super(sb);
1927         ext3_std_error(inode->i_sb, err);
1928         return err;
1929 }
1930
1931 /*
1932  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1933  * of such inodes stored on disk, because it is finally being cleaned up.
1934  */
1935 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1936 {
1937         struct list_head *prev;
1938         struct ext3_inode_info *ei = EXT3_I(inode);
1939         struct ext3_sb_info *sbi;
1940         unsigned long ino_next;
1941         struct ext3_iloc iloc;
1942         int err = 0;
1943
1944         lock_super(inode->i_sb);
1945         if (list_empty(&ei->i_orphan)) {
1946                 unlock_super(inode->i_sb);
1947                 return 0;
1948         }
1949
1950         ino_next = NEXT_ORPHAN(inode);
1951         prev = ei->i_orphan.prev;
1952         sbi = EXT3_SB(inode->i_sb);
1953
1954         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1955
1956         list_del_init(&ei->i_orphan);
1957
1958         /* If we're on an error path, we may not have a valid
1959          * transaction handle with which to update the orphan list on
1960          * disk, but we still need to remove the inode from the linked
1961          * list in memory. */
1962         if (!handle)
1963                 goto out;
1964
1965         err = ext3_reserve_inode_write(handle, inode, &iloc);
1966         if (err)
1967                 goto out_err;
1968
1969         if (prev == &sbi->s_orphan) {
1970                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1971                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1972                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1973                 if (err)
1974                         goto out_brelse;
1975                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1976                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1977         } else {
1978                 struct ext3_iloc iloc2;
1979                 struct inode *i_prev =
1980                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1981
1982                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1983                           i_prev->i_ino, ino_next);
1984                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1985                 if (err)
1986                         goto out_brelse;
1987                 NEXT_ORPHAN(i_prev) = ino_next;
1988                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1989         }
1990         if (err)
1991                 goto out_brelse;
1992         NEXT_ORPHAN(inode) = 0;
1993         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1994
1995 out_err:
1996         ext3_std_error(inode->i_sb, err);
1997 out:
1998         unlock_super(inode->i_sb);
1999         return err;
2000
2001 out_brelse:
2002         brelse(iloc.bh);
2003         goto out_err;
2004 }
2005
2006 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2007 {
2008         int retval;
2009         struct inode * inode;
2010         struct buffer_head * bh;
2011         struct ext3_dir_entry_2 * de;
2012         handle_t *handle;
2013
2014         /* Initialize quotas before so that eventual writes go in
2015          * separate transaction */
2016         DQUOT_INIT(dentry->d_inode);
2017         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2018         if (IS_ERR(handle))
2019                 return PTR_ERR(handle);
2020
2021         retval = -ENOENT;
2022         bh = ext3_find_entry (dentry, &de);
2023         if (!bh)
2024                 goto end_rmdir;
2025
2026         if (IS_DIRSYNC(dir))
2027                 handle->h_sync = 1;
2028
2029         inode = dentry->d_inode;
2030
2031         retval = -EIO;
2032         if (le32_to_cpu(de->inode) != inode->i_ino)
2033                 goto end_rmdir;
2034
2035         retval = -ENOTEMPTY;
2036         if (!empty_dir (inode))
2037                 goto end_rmdir;
2038
2039         retval = ext3_delete_entry(handle, dir, de, bh);
2040         if (retval)
2041                 goto end_rmdir;
2042         if (inode->i_nlink != 2)
2043                 ext3_warning (inode->i_sb, "ext3_rmdir",
2044                               "empty directory has nlink!=2 (%d)",
2045                               inode->i_nlink);
2046         inode->i_version++;
2047         inode->i_nlink = 0;
2048         /* There's no need to set i_disksize: the fact that i_nlink is
2049          * zero will ensure that the right thing happens during any
2050          * recovery. */
2051         inode->i_size = 0;
2052         ext3_orphan_add(handle, inode);
2053         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2054         ext3_mark_inode_dirty(handle, inode);
2055         dir->i_nlink--;
2056         ext3_update_dx_flag(dir);
2057         ext3_mark_inode_dirty(handle, dir);
2058
2059 end_rmdir:
2060         ext3_journal_stop(handle);
2061         brelse (bh);
2062         return retval;
2063 }
2064
2065 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2066 {
2067         int retval;
2068         struct inode * inode;
2069         struct buffer_head * bh;
2070         struct ext3_dir_entry_2 * de;
2071         handle_t *handle;
2072
2073         /* Initialize quotas before so that eventual writes go
2074          * in separate transaction */
2075         DQUOT_INIT(dentry->d_inode);
2076         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2077         if (IS_ERR(handle))
2078                 return PTR_ERR(handle);
2079
2080         if (IS_DIRSYNC(dir))
2081                 handle->h_sync = 1;
2082
2083         retval = -ENOENT;
2084         bh = ext3_find_entry (dentry, &de);
2085         if (!bh)
2086                 goto end_unlink;
2087
2088         inode = dentry->d_inode;
2089
2090         retval = -EIO;
2091         if (le32_to_cpu(de->inode) != inode->i_ino)
2092                 goto end_unlink;
2093
2094         if (!inode->i_nlink) {
2095                 ext3_warning (inode->i_sb, "ext3_unlink",
2096                               "Deleting nonexistent file (%lu), %d",
2097                               inode->i_ino, inode->i_nlink);
2098                 inode->i_nlink = 1;
2099         }
2100         retval = ext3_delete_entry(handle, dir, de, bh);
2101         if (retval)
2102                 goto end_unlink;
2103         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2104         ext3_update_dx_flag(dir);
2105         ext3_mark_inode_dirty(handle, dir);
2106         inode->i_nlink--;
2107         if (!inode->i_nlink)
2108                 ext3_orphan_add(handle, inode);
2109         inode->i_ctime = dir->i_ctime;
2110         ext3_mark_inode_dirty(handle, inode);
2111         retval = 0;
2112
2113 end_unlink:
2114         ext3_journal_stop(handle);
2115         brelse (bh);
2116         return retval;
2117 }
2118
2119 static int ext3_symlink (struct inode * dir,
2120                 struct dentry *dentry, const char * symname)
2121 {
2122         handle_t *handle;
2123         struct inode * inode;
2124         int l, err, retries = 0;
2125
2126         l = strlen(symname)+1;
2127         if (l > dir->i_sb->s_blocksize)
2128                 return -ENAMETOOLONG;
2129
2130 retry:
2131         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2132                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2133                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2134         if (IS_ERR(handle))
2135                 return PTR_ERR(handle);
2136
2137         if (IS_DIRSYNC(dir))
2138                 handle->h_sync = 1;
2139
2140         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2141         err = PTR_ERR(inode);
2142         if (IS_ERR(inode))
2143                 goto out_stop;
2144
2145         if (l > sizeof (EXT3_I(inode)->i_data)) {
2146                 inode->i_op = &ext3_symlink_inode_operations;
2147                 ext3_set_aops(inode);
2148                 /*
2149                  * page_symlink() calls into ext3_prepare/commit_write.
2150                  * We have a transaction open.  All is sweetness.  It also sets
2151                  * i_size in generic_commit_write().
2152                  */
2153                 err = __page_symlink(inode, symname, l,
2154                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2155                 if (err) {
2156                         ext3_dec_count(handle, inode);
2157                         ext3_mark_inode_dirty(handle, inode);
2158                         iput (inode);
2159                         goto out_stop;
2160                 }
2161         } else {
2162                 inode->i_op = &ext3_fast_symlink_inode_operations;
2163                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2164                 inode->i_size = l-1;
2165         }
2166         EXT3_I(inode)->i_disksize = inode->i_size;
2167         err = ext3_add_nondir(handle, dentry, inode);
2168 out_stop:
2169         ext3_journal_stop(handle);
2170         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2171                 goto retry;
2172         return err;
2173 }
2174
2175 static int ext3_link (struct dentry * old_dentry,
2176                 struct inode * dir, struct dentry *dentry)
2177 {
2178         handle_t *handle;
2179         struct inode *inode = old_dentry->d_inode;
2180         int err, retries = 0;
2181
2182         if (inode->i_nlink >= EXT3_LINK_MAX)
2183                 return -EMLINK;
2184
2185 retry:
2186         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2187                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2188         if (IS_ERR(handle))
2189                 return PTR_ERR(handle);
2190
2191         if (IS_DIRSYNC(dir))
2192                 handle->h_sync = 1;
2193
2194         inode->i_ctime = CURRENT_TIME_SEC;
2195         ext3_inc_count(handle, inode);
2196         atomic_inc(&inode->i_count);
2197
2198         err = ext3_add_nondir(handle, dentry, inode);
2199         ext3_journal_stop(handle);
2200         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2201                 goto retry;
2202         return err;
2203 }
2204
2205 #define PARENT_INO(buffer) \
2206         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2207         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2208
2209 /*
2210  * Anybody can rename anything with this: the permission checks are left to the
2211  * higher-level routines.
2212  */
2213 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2214                            struct inode * new_dir,struct dentry *new_dentry)
2215 {
2216         handle_t *handle;
2217         struct inode * old_inode, * new_inode;
2218         struct buffer_head * old_bh, * new_bh, * dir_bh;
2219         struct ext3_dir_entry_2 * old_de, * new_de;
2220         int retval;
2221
2222         old_bh = new_bh = dir_bh = NULL;
2223
2224         /* Initialize quotas before so that eventual writes go
2225          * in separate transaction */
2226         if (new_dentry->d_inode)
2227                 DQUOT_INIT(new_dentry->d_inode);
2228         handle = ext3_journal_start(old_dir, 2 *
2229                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2230                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2231         if (IS_ERR(handle))
2232                 return PTR_ERR(handle);
2233
2234         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2235                 handle->h_sync = 1;
2236
2237         old_bh = ext3_find_entry (old_dentry, &old_de);
2238         /*
2239          *  Check for inode number is _not_ due to possible IO errors.
2240          *  We might rmdir the source, keep it as pwd of some process
2241          *  and merrily kill the link to whatever was created under the
2242          *  same name. Goodbye sticky bit ;-<
2243          */
2244         old_inode = old_dentry->d_inode;
2245         retval = -ENOENT;
2246         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2247                 goto end_rename;
2248
2249         new_inode = new_dentry->d_inode;
2250         new_bh = ext3_find_entry (new_dentry, &new_de);
2251         if (new_bh) {
2252                 if (!new_inode) {
2253                         brelse (new_bh);
2254                         new_bh = NULL;
2255                 }
2256         }
2257         if (S_ISDIR(old_inode->i_mode)) {
2258                 if (new_inode) {
2259                         retval = -ENOTEMPTY;
2260                         if (!empty_dir (new_inode))
2261                                 goto end_rename;
2262                 }
2263                 retval = -EIO;
2264                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2265                 if (!dir_bh)
2266                         goto end_rename;
2267                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2268                         goto end_rename;
2269                 retval = -EMLINK;
2270                 if (!new_inode && new_dir!=old_dir &&
2271                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2272                         goto end_rename;
2273         }
2274         if (!new_bh) {
2275                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2276                 if (retval)
2277                         goto end_rename;
2278         } else {
2279                 BUFFER_TRACE(new_bh, "get write access");
2280                 ext3_journal_get_write_access(handle, new_bh);
2281                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2282                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2283                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2284                         new_de->file_type = old_de->file_type;
2285                 new_dir->i_version++;
2286                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2287                 ext3_journal_dirty_metadata(handle, new_bh);
2288                 brelse(new_bh);
2289                 new_bh = NULL;
2290         }
2291
2292         /*
2293          * Like most other Unix systems, set the ctime for inodes on a
2294          * rename.
2295          */
2296         old_inode->i_ctime = CURRENT_TIME_SEC;
2297         ext3_mark_inode_dirty(handle, old_inode);
2298
2299         /*
2300          * ok, that's it
2301          */
2302         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2303             old_de->name_len != old_dentry->d_name.len ||
2304             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2305             (retval = ext3_delete_entry(handle, old_dir,
2306                                         old_de, old_bh)) == -ENOENT) {
2307                 /* old_de could have moved from under us during htree split, so
2308                  * make sure that we are deleting the right entry.  We might
2309                  * also be pointing to a stale entry in the unused part of
2310                  * old_bh so just checking inum and the name isn't enough. */
2311                 struct buffer_head *old_bh2;
2312                 struct ext3_dir_entry_2 *old_de2;
2313
2314                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2315                 if (old_bh2) {
2316                         retval = ext3_delete_entry(handle, old_dir,
2317                                                    old_de2, old_bh2);
2318                         brelse(old_bh2);
2319                 }
2320         }
2321         if (retval) {
2322                 ext3_warning(old_dir->i_sb, "ext3_rename",
2323                                 "Deleting old file (%lu), %d, error=%d",
2324                                 old_dir->i_ino, old_dir->i_nlink, retval);
2325         }
2326
2327         if (new_inode) {
2328                 new_inode->i_nlink--;
2329                 new_inode->i_ctime = CURRENT_TIME_SEC;
2330         }
2331         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2332         ext3_update_dx_flag(old_dir);
2333         if (dir_bh) {
2334                 BUFFER_TRACE(dir_bh, "get_write_access");
2335                 ext3_journal_get_write_access(handle, dir_bh);
2336                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2337                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2338                 ext3_journal_dirty_metadata(handle, dir_bh);
2339                 old_dir->i_nlink--;
2340                 if (new_inode) {
2341                         new_inode->i_nlink--;
2342                 } else {
2343                         new_dir->i_nlink++;
2344                         ext3_update_dx_flag(new_dir);
2345                         ext3_mark_inode_dirty(handle, new_dir);
2346                 }
2347         }
2348         ext3_mark_inode_dirty(handle, old_dir);
2349         if (new_inode) {
2350                 ext3_mark_inode_dirty(handle, new_inode);
2351                 if (!new_inode->i_nlink)
2352                         ext3_orphan_add(handle, new_inode);
2353         }
2354         retval = 0;
2355
2356 end_rename:
2357         brelse (dir_bh);
2358         brelse (old_bh);
2359         brelse (new_bh);
2360         ext3_journal_stop(handle);
2361         return retval;
2362 }
2363
2364 /*
2365  * directories can handle most operations...
2366  */
2367 struct inode_operations ext3_dir_inode_operations = {
2368         .create         = ext3_create,
2369         .lookup         = ext3_lookup,
2370         .link           = ext3_link,
2371         .unlink         = ext3_unlink,
2372         .symlink        = ext3_symlink,
2373         .mkdir          = ext3_mkdir,
2374         .rmdir          = ext3_rmdir,
2375         .mknod          = ext3_mknod,
2376         .rename         = ext3_rename,
2377         .setattr        = ext3_setattr,
2378 #ifdef CONFIG_EXT3_FS_XATTR
2379         .setxattr       = generic_setxattr,
2380         .getxattr       = generic_getxattr,
2381         .listxattr      = ext3_listxattr,
2382         .removexattr    = generic_removexattr,
2383 #endif
2384         .permission     = ext3_permission,
2385 };
2386
2387 struct inode_operations ext3_special_inode_operations = {
2388         .setattr        = ext3_setattr,
2389 #ifdef CONFIG_EXT3_FS_XATTR
2390         .setxattr       = generic_setxattr,
2391         .getxattr       = generic_getxattr,
2392         .listxattr      = ext3_listxattr,
2393         .removexattr    = generic_removexattr,
2394 #endif
2395         .permission     = ext3_permission,
2396 };