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