[PATCH] handle ext3 directory corruption better
[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/bio.h>
39 #include <linux/smp_lock.h>
40
41 #include "namei.h"
42 #include "xattr.h"
43 #include "acl.h"
44
45 /*
46  * define how far ahead to read directories while searching them.
47  */
48 #define NAMEI_RA_CHUNKS  2
49 #define NAMEI_RA_BLOCKS  4
50 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
52
53 static struct buffer_head *ext3_append(handle_t *handle,
54                                         struct inode *inode,
55                                         u32 *block, int *err)
56 {
57         struct buffer_head *bh;
58
59         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60
61         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT3_I(inode)->i_disksize = inode->i_size;
64                 ext3_journal_get_write_access(handle,bh);
65         }
66         return bh;
67 }
68
69 #ifndef assert
70 #define assert(test) J_ASSERT(test)
71 #endif
72
73 #ifndef swap
74 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
75 #endif
76
77 #ifdef DX_DEBUG
78 #define dxtrace(command) command
79 #else
80 #define dxtrace(command)
81 #endif
82
83 struct fake_dirent
84 {
85         __le32 inode;
86         __le16 rec_len;
87         u8 name_len;
88         u8 file_type;
89 };
90
91 struct dx_countlimit
92 {
93         __le16 limit;
94         __le16 count;
95 };
96
97 struct dx_entry
98 {
99         __le32 hash;
100         __le32 block;
101 };
102
103 /*
104  * dx_root_info is laid out so that if it should somehow get overlaid by a
105  * dirent the two low bits of the hash version will be zero.  Therefore, the
106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
107  */
108
109 struct dx_root
110 {
111         struct fake_dirent dot;
112         char dot_name[4];
113         struct fake_dirent dotdot;
114         char dotdot_name[4];
115         struct dx_root_info
116         {
117                 __le32 reserved_zero;
118                 u8 hash_version;
119                 u8 info_length; /* 8 */
120                 u8 indirect_levels;
121                 u8 unused_flags;
122         }
123         info;
124         struct dx_entry entries[0];
125 };
126
127 struct dx_node
128 {
129         struct fake_dirent fake;
130         struct dx_entry entries[0];
131 };
132
133
134 struct dx_frame
135 {
136         struct buffer_head *bh;
137         struct dx_entry *entries;
138         struct dx_entry *at;
139 };
140
141 struct dx_map_entry
142 {
143         u32 hash;
144         u32 offs;
145 };
146
147 #ifdef CONFIG_EXT3_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
159                                  struct inode *dir,
160                                  struct dx_hash_info *hinfo,
161                                  struct dx_frame *frame,
162                                  int *err);
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168                 struct dx_map_entry *offsets, int count);
169 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172                                  struct dx_frame *frame,
173                                  struct dx_frame *frames,
174                                  __u32 *start_hash);
175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176                        struct ext3_dir_entry_2 **res_dir, int *err);
177 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
178                              struct inode *inode);
179
180 /*
181  * Future: use high four bits of block for coalesce-on-delete flags
182  * Mask them off for now.
183  */
184
185 static inline unsigned dx_get_block (struct dx_entry *entry)
186 {
187         return le32_to_cpu(entry->block) & 0x00ffffff;
188 }
189
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191 {
192         entry->block = cpu_to_le32(value);
193 }
194
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
196 {
197         return le32_to_cpu(entry->hash);
198 }
199
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201 {
202         entry->hash = cpu_to_le32(value);
203 }
204
205 static inline unsigned dx_get_count (struct dx_entry *entries)
206 {
207         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208 }
209
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
211 {
212         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213 }
214
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216 {
217         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218 }
219
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221 {
222         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223 }
224
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226 {
227         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228                 EXT3_DIR_REC_LEN(2) - infosize;
229         return 0? 20: entry_space / sizeof(struct dx_entry);
230 }
231
232 static inline unsigned dx_node_limit (struct inode *dir)
233 {
234         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235         return 0? 22: entry_space / sizeof(struct dx_entry);
236 }
237
238 /*
239  * Debug
240  */
241 #ifdef DX_DEBUG
242 static void dx_show_index (char * label, struct dx_entry *entries)
243 {
244         int i, n = dx_get_count (entries);
245         printk("%s index ", label);
246         for (i = 0; i < n; i++)
247         {
248                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
249         }
250         printk("\n");
251 }
252
253 struct stats
254 {
255         unsigned names;
256         unsigned space;
257         unsigned bcount;
258 };
259
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261                                  int size, int show_names)
262 {
263         unsigned names = 0, space = 0;
264         char *base = (char *) de;
265         struct dx_hash_info h = *hinfo;
266
267         printk("names: ");
268         while ((char *) de < base + size)
269         {
270                 if (de->inode)
271                 {
272                         if (show_names)
273                         {
274                                 int len = de->name_len;
275                                 char *name = de->name;
276                                 while (len--) printk("%c", *name++);
277                                 ext3fs_dirhash(de->name, de->name_len, &h);
278                                 printk(":%x.%u ", h.hash,
279                                        ((char *) de - base));
280                         }
281                         space += EXT3_DIR_REC_LEN(de->name_len);
282                         names++;
283                 }
284                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
285         }
286         printk("(%i)\n", names);
287         return (struct stats) { names, space, 1 };
288 }
289
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291                              struct dx_entry *entries, int levels)
292 {
293         unsigned blocksize = dir->i_sb->s_blocksize;
294         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295         unsigned bcount = 0;
296         struct buffer_head *bh;
297         int err;
298         printk("%i indexed blocks...\n", count);
299         for (i = 0; i < count; i++, entries++)
300         {
301                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303                 struct stats stats;
304                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
305                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
306                 stats = levels?
307                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309                 names += stats.names;
310                 space += stats.space;
311                 bcount += stats.bcount;
312                 brelse (bh);
313         }
314         if (bcount)
315                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
316                         names, space/bcount,(space/bcount)*100/blocksize);
317         return (struct stats) { names, space, bcount};
318 }
319 #endif /* DX_DEBUG */
320
321 /*
322  * Probe for a directory leaf block to search.
323  *
324  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325  * error in the directory index, and the caller should fall back to
326  * searching the directory normally.  The callers of dx_probe **MUST**
327  * check for this error code, and make sure it never gets reflected
328  * back to userspace.
329  */
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333 {
334         unsigned count, indirect;
335         struct dx_entry *at, *entries, *p, *q, *m;
336         struct dx_root *root;
337         struct buffer_head *bh;
338         struct dx_frame *frame = frame_in;
339         u32 hash;
340
341         frame->bh = NULL;
342         if (dentry)
343                 dir = dentry->d_parent->d_inode;
344         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
345                 goto fail;
346         root = (struct dx_root *) bh->b_data;
347         if (root->info.hash_version != DX_HASH_TEA &&
348             root->info.hash_version != DX_HASH_HALF_MD4 &&
349             root->info.hash_version != DX_HASH_LEGACY) {
350                 ext3_warning(dir->i_sb, __FUNCTION__,
351                              "Unrecognised inode hash code %d",
352                              root->info.hash_version);
353                 brelse(bh);
354                 *err = ERR_BAD_DX_DIR;
355                 goto fail;
356         }
357         hinfo->hash_version = root->info.hash_version;
358         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
359         if (dentry)
360                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
361         hash = hinfo->hash;
362
363         if (root->info.unused_flags & 1) {
364                 ext3_warning(dir->i_sb, __FUNCTION__,
365                              "Unimplemented inode hash flags: %#06x",
366                              root->info.unused_flags);
367                 brelse(bh);
368                 *err = ERR_BAD_DX_DIR;
369                 goto fail;
370         }
371
372         if ((indirect = root->info.indirect_levels) > 1) {
373                 ext3_warning(dir->i_sb, __FUNCTION__,
374                              "Unimplemented inode hash depth: %#06x",
375                              root->info.indirect_levels);
376                 brelse(bh);
377                 *err = ERR_BAD_DX_DIR;
378                 goto fail;
379         }
380
381         entries = (struct dx_entry *) (((char *)&root->info) +
382                                        root->info.info_length);
383         assert(dx_get_limit(entries) == dx_root_limit(dir,
384                                                       root->info.info_length));
385         dxtrace (printk("Look up %x", hash));
386         while (1)
387         {
388                 count = dx_get_count(entries);
389                 assert (count && count <= dx_get_limit(entries));
390                 p = entries + 1;
391                 q = entries + count - 1;
392                 while (p <= q)
393                 {
394                         m = p + (q - p)/2;
395                         dxtrace(printk("."));
396                         if (dx_get_hash(m) > hash)
397                                 q = m - 1;
398                         else
399                                 p = m + 1;
400                 }
401
402                 if (0) // linear search cross check
403                 {
404                         unsigned n = count - 1;
405                         at = entries;
406                         while (n--)
407                         {
408                                 dxtrace(printk(","));
409                                 if (dx_get_hash(++at) > hash)
410                                 {
411                                         at--;
412                                         break;
413                                 }
414                         }
415                         assert (at == p - 1);
416                 }
417
418                 at = p - 1;
419                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
420                 frame->bh = bh;
421                 frame->entries = entries;
422                 frame->at = at;
423                 if (!indirect--) return frame;
424                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
425                         goto fail2;
426                 at = entries = ((struct dx_node *) bh->b_data)->entries;
427                 assert (dx_get_limit(entries) == dx_node_limit (dir));
428                 frame++;
429         }
430 fail2:
431         while (frame >= frame_in) {
432                 brelse(frame->bh);
433                 frame--;
434         }
435 fail:
436         return NULL;
437 }
438
439 static void dx_release (struct dx_frame *frames)
440 {
441         if (frames[0].bh == NULL)
442                 return;
443
444         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445                 brelse(frames[1].bh);
446         brelse(frames[0].bh);
447 }
448
449 /*
450  * This function increments the frame pointer to search the next leaf
451  * block, and reads in the necessary intervening nodes if the search
452  * should be necessary.  Whether or not the search is necessary is
453  * controlled by the hash parameter.  If the hash value is even, then
454  * the search is only continued if the next block starts with that
455  * hash value.  This is used if we are searching for a specific file.
456  *
457  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
458  *
459  * This function returns 1 if the caller should continue to search,
460  * or 0 if it should not.  If there is an error reading one of the
461  * index blocks, it will a negative error code.
462  *
463  * If start_hash is non-null, it will be filled in with the starting
464  * hash of the next page.
465  */
466 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
467                                  struct dx_frame *frame,
468                                  struct dx_frame *frames,
469                                  __u32 *start_hash)
470 {
471         struct dx_frame *p;
472         struct buffer_head *bh;
473         int err, num_frames = 0;
474         __u32 bhash;
475
476         p = frame;
477         /*
478          * Find the next leaf page by incrementing the frame pointer.
479          * If we run out of entries in the interior node, loop around and
480          * increment pointer in the parent node.  When we break out of
481          * this loop, num_frames indicates the number of interior
482          * nodes need to be read.
483          */
484         while (1) {
485                 if (++(p->at) < p->entries + dx_get_count(p->entries))
486                         break;
487                 if (p == frames)
488                         return 0;
489                 num_frames++;
490                 p--;
491         }
492
493         /*
494          * If the hash is 1, then continue only if the next page has a
495          * continuation hash of any value.  This is used for readdir
496          * handling.  Otherwise, check to see if the hash matches the
497          * desired contiuation hash.  If it doesn't, return since
498          * there's no point to read in the successive index pages.
499          */
500         bhash = dx_get_hash(p->at);
501         if (start_hash)
502                 *start_hash = bhash;
503         if ((hash & 1) == 0) {
504                 if ((bhash & ~1) != hash)
505                         return 0;
506         }
507         /*
508          * If the hash is HASH_NB_ALWAYS, we always go to the next
509          * block so no check is necessary
510          */
511         while (num_frames--) {
512                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
513                                       0, &err)))
514                         return err; /* Failure */
515                 p++;
516                 brelse (p->bh);
517                 p->bh = bh;
518                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
519         }
520         return 1;
521 }
522
523
524 /*
525  * p is at least 6 bytes before the end of page
526  */
527 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
528 {
529         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
530 }
531
532 /*
533  * This function fills a red-black tree with information from a
534  * directory block.  It returns the number directory entries loaded
535  * into the tree.  If there is an error it is returned in err.
536  */
537 static int htree_dirblock_to_tree(struct file *dir_file,
538                                   struct inode *dir, int block,
539                                   struct dx_hash_info *hinfo,
540                                   __u32 start_hash, __u32 start_minor_hash)
541 {
542         struct buffer_head *bh;
543         struct ext3_dir_entry_2 *de, *top;
544         int err, count = 0;
545
546         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
548                 return err;
549
550         de = (struct ext3_dir_entry_2 *) bh->b_data;
551         top = (struct ext3_dir_entry_2 *) ((char *) de +
552                                            dir->i_sb->s_blocksize -
553                                            EXT3_DIR_REC_LEN(0));
554         for (; de < top; de = ext3_next_entry(de)) {
555                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
556                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
557                                                 +((char *)de - bh->b_data))) {
558                         /* On error, skip the f_pos to the next block. */
559                         dir_file->f_pos = (dir_file->f_pos |
560                                         (dir->i_sb->s_blocksize - 1)) + 1;
561                         brelse (bh);
562                         return count;
563                 }
564                 ext3fs_dirhash(de->name, de->name_len, hinfo);
565                 if ((hinfo->hash < start_hash) ||
566                     ((hinfo->hash == start_hash) &&
567                      (hinfo->minor_hash < start_minor_hash)))
568                         continue;
569                 if (de->inode == 0)
570                         continue;
571                 if ((err = ext3_htree_store_dirent(dir_file,
572                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
573                         brelse(bh);
574                         return err;
575                 }
576                 count++;
577         }
578         brelse(bh);
579         return count;
580 }
581
582
583 /*
584  * This function fills a red-black tree with information from a
585  * directory.  We start scanning the directory in hash order, starting
586  * at start_hash and start_minor_hash.
587  *
588  * This function returns the number of entries inserted into the tree,
589  * or a negative error code.
590  */
591 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
592                          __u32 start_minor_hash, __u32 *next_hash)
593 {
594         struct dx_hash_info hinfo;
595         struct ext3_dir_entry_2 *de;
596         struct dx_frame frames[2], *frame;
597         struct inode *dir;
598         int block, err;
599         int count = 0;
600         int ret;
601         __u32 hashval;
602
603         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
604                        start_minor_hash));
605         dir = dir_file->f_dentry->d_inode;
606         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
607                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
608                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
609                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
610                                                start_hash, start_minor_hash);
611                 *next_hash = ~0;
612                 return count;
613         }
614         hinfo.hash = start_hash;
615         hinfo.minor_hash = 0;
616         frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
617         if (!frame)
618                 return err;
619
620         /* Add '.' and '..' from the htree header */
621         if (!start_hash && !start_minor_hash) {
622                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
623                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
624                         goto errout;
625                 count++;
626         }
627         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
628                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
629                 de = ext3_next_entry(de);
630                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
631                         goto errout;
632                 count++;
633         }
634
635         while (1) {
636                 block = dx_get_block(frame->at);
637                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
638                                              start_hash, start_minor_hash);
639                 if (ret < 0) {
640                         err = ret;
641                         goto errout;
642                 }
643                 count += ret;
644                 hashval = ~0;
645                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
646                                             frame, frames, &hashval);
647                 *next_hash = hashval;
648                 if (ret < 0) {
649                         err = ret;
650                         goto errout;
651                 }
652                 /*
653                  * Stop if:  (a) there are no more entries, or
654                  * (b) we have inserted at least one entry and the
655                  * next hash value is not a continuation
656                  */
657                 if ((ret == 0) ||
658                     (count && ((hashval & 1) == 0)))
659                         break;
660         }
661         dx_release(frames);
662         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
663                        count, *next_hash));
664         return count;
665 errout:
666         dx_release(frames);
667         return (err);
668 }
669
670
671 /*
672  * Directory block splitting, compacting
673  */
674
675 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
676                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
677 {
678         int count = 0;
679         char *base = (char *) de;
680         struct dx_hash_info h = *hinfo;
681
682         while ((char *) de < base + size)
683         {
684                 if (de->name_len && de->inode) {
685                         ext3fs_dirhash(de->name, de->name_len, &h);
686                         map_tail--;
687                         map_tail->hash = h.hash;
688                         map_tail->offs = (u32) ((char *) de - base);
689                         count++;
690                         cond_resched();
691                 }
692                 /* XXX: do we need to check rec_len == 0 case? -Chris */
693                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
694         }
695         return count;
696 }
697
698 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
699 {
700         struct dx_map_entry *p, *q, *top = map + count - 1;
701         int more;
702         /* Combsort until bubble sort doesn't suck */
703         while (count > 2)
704         {
705                 count = count*10/13;
706                 if (count - 9 < 2) /* 9, 10 -> 11 */
707                         count = 11;
708                 for (p = top, q = p - count; q >= map; p--, q--)
709                         if (p->hash < q->hash)
710                                 swap(*p, *q);
711         }
712         /* Garden variety bubble sort */
713         do {
714                 more = 0;
715                 q = top;
716                 while (q-- > map)
717                 {
718                         if (q[1].hash >= q[0].hash)
719                                 continue;
720                         swap(*(q+1), *q);
721                         more = 1;
722                 }
723         } while(more);
724 }
725
726 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
727 {
728         struct dx_entry *entries = frame->entries;
729         struct dx_entry *old = frame->at, *new = old + 1;
730         int count = dx_get_count(entries);
731
732         assert(count < dx_get_limit(entries));
733         assert(old < entries + count);
734         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
735         dx_set_hash(new, hash);
736         dx_set_block(new, block);
737         dx_set_count(entries, count + 1);
738 }
739 #endif
740
741
742 static void ext3_update_dx_flag(struct inode *inode)
743 {
744         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
745                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
746                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
747 }
748
749 /*
750  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
751  *
752  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
753  * `de != NULL' is guaranteed by caller.
754  */
755 static inline int ext3_match (int len, const char * const name,
756                               struct ext3_dir_entry_2 * de)
757 {
758         if (len != de->name_len)
759                 return 0;
760         if (!de->inode)
761                 return 0;
762         return !memcmp(name, de->name, len);
763 }
764
765 /*
766  * Returns 0 if not found, -1 on failure, and 1 on success
767  */
768 static inline int search_dirblock(struct buffer_head * bh,
769                                   struct inode *dir,
770                                   struct dentry *dentry,
771                                   unsigned long offset,
772                                   struct ext3_dir_entry_2 ** res_dir)
773 {
774         struct ext3_dir_entry_2 * de;
775         char * dlimit;
776         int de_len;
777         const char *name = dentry->d_name.name;
778         int namelen = dentry->d_name.len;
779
780         de = (struct ext3_dir_entry_2 *) bh->b_data;
781         dlimit = bh->b_data + dir->i_sb->s_blocksize;
782         while ((char *) de < dlimit) {
783                 /* this code is executed quadratically often */
784                 /* do minimal checking `by hand' */
785
786                 if ((char *) de + namelen <= dlimit &&
787                     ext3_match (namelen, name, de)) {
788                         /* found a match - just to be sure, do a full check */
789                         if (!ext3_check_dir_entry("ext3_find_entry",
790                                                   dir, de, bh, offset))
791                                 return -1;
792                         *res_dir = de;
793                         return 1;
794                 }
795                 /* prevent looping on a bad block */
796                 de_len = le16_to_cpu(de->rec_len);
797                 if (de_len <= 0)
798                         return -1;
799                 offset += de_len;
800                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
801         }
802         return 0;
803 }
804
805
806 /*
807  *      ext3_find_entry()
808  *
809  * finds an entry in the specified directory with the wanted name. It
810  * returns the cache buffer in which the entry was found, and the entry
811  * itself (as a parameter - res_dir). It does NOT read the inode of the
812  * entry - you'll have to do that yourself if you want to.
813  *
814  * The returned buffer_head has ->b_count elevated.  The caller is expected
815  * to brelse() it when appropriate.
816  */
817 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
818                                         struct ext3_dir_entry_2 ** res_dir)
819 {
820         struct super_block * sb;
821         struct buffer_head * bh_use[NAMEI_RA_SIZE];
822         struct buffer_head * bh, *ret = NULL;
823         unsigned long start, block, b;
824         int ra_max = 0;         /* Number of bh's in the readahead
825                                    buffer, bh_use[] */
826         int ra_ptr = 0;         /* Current index into readahead
827                                    buffer */
828         int num = 0;
829         int nblocks, i, err;
830         struct inode *dir = dentry->d_parent->d_inode;
831         int namelen;
832         const u8 *name;
833         unsigned blocksize;
834
835         *res_dir = NULL;
836         sb = dir->i_sb;
837         blocksize = sb->s_blocksize;
838         namelen = dentry->d_name.len;
839         name = dentry->d_name.name;
840         if (namelen > EXT3_NAME_LEN)
841                 return NULL;
842 #ifdef CONFIG_EXT3_INDEX
843         if (is_dx(dir)) {
844                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
845                 /*
846                  * On success, or if the error was file not found,
847                  * return.  Otherwise, fall back to doing a search the
848                  * old fashioned way.
849                  */
850                 if (bh || (err != ERR_BAD_DX_DIR))
851                         return bh;
852                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
853         }
854 #endif
855         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
856         start = EXT3_I(dir)->i_dir_start_lookup;
857         if (start >= nblocks)
858                 start = 0;
859         block = start;
860 restart:
861         do {
862                 /*
863                  * We deal with the read-ahead logic here.
864                  */
865                 if (ra_ptr >= ra_max) {
866                         /* Refill the readahead buffer */
867                         ra_ptr = 0;
868                         b = block;
869                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
870                                 /*
871                                  * Terminate if we reach the end of the
872                                  * directory and must wrap, or if our
873                                  * search has finished at this block.
874                                  */
875                                 if (b >= nblocks || (num && block == start)) {
876                                         bh_use[ra_max] = NULL;
877                                         break;
878                                 }
879                                 num++;
880                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
881                                 bh_use[ra_max] = bh;
882                                 if (bh)
883                                         ll_rw_block(READ_META, 1, &bh);
884                         }
885                 }
886                 if ((bh = bh_use[ra_ptr++]) == NULL)
887                         goto next;
888                 wait_on_buffer(bh);
889                 if (!buffer_uptodate(bh)) {
890                         /* read error, skip block & hope for the best */
891                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
892                                    "offset %lu", dir->i_ino, block);
893                         brelse(bh);
894                         goto next;
895                 }
896                 i = search_dirblock(bh, dir, dentry,
897                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
898                 if (i == 1) {
899                         EXT3_I(dir)->i_dir_start_lookup = block;
900                         ret = bh;
901                         goto cleanup_and_exit;
902                 } else {
903                         brelse(bh);
904                         if (i < 0)
905                                 goto cleanup_and_exit;
906                 }
907         next:
908                 if (++block >= nblocks)
909                         block = 0;
910         } while (block != start);
911
912         /*
913          * If the directory has grown while we were searching, then
914          * search the last part of the directory before giving up.
915          */
916         block = nblocks;
917         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
918         if (block < nblocks) {
919                 start = 0;
920                 goto restart;
921         }
922
923 cleanup_and_exit:
924         /* Clean up the read-ahead blocks */
925         for (; ra_ptr < ra_max; ra_ptr++)
926                 brelse (bh_use[ra_ptr]);
927         return ret;
928 }
929
930 #ifdef CONFIG_EXT3_INDEX
931 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
932                        struct ext3_dir_entry_2 **res_dir, int *err)
933 {
934         struct super_block * sb;
935         struct dx_hash_info     hinfo;
936         u32 hash;
937         struct dx_frame frames[2], *frame;
938         struct ext3_dir_entry_2 *de, *top;
939         struct buffer_head *bh;
940         unsigned long block;
941         int retval;
942         int namelen = dentry->d_name.len;
943         const u8 *name = dentry->d_name.name;
944         struct inode *dir = dentry->d_parent->d_inode;
945
946         sb = dir->i_sb;
947         /* NFS may look up ".." - look at dx_root directory block */
948         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
949                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
950                         return NULL;
951         } else {
952                 frame = frames;
953                 frame->bh = NULL;                       /* for dx_release() */
954                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
955                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
956         }
957         hash = hinfo.hash;
958         do {
959                 block = dx_get_block(frame->at);
960                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
961                         goto errout;
962                 de = (struct ext3_dir_entry_2 *) bh->b_data;
963                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
964                                        EXT3_DIR_REC_LEN(0));
965                 for (; de < top; de = ext3_next_entry(de))
966                 if (ext3_match (namelen, name, de)) {
967                         if (!ext3_check_dir_entry("ext3_find_entry",
968                                                   dir, de, bh,
969                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
970                                           +((char *)de - bh->b_data))) {
971                                 brelse (bh);
972                                 goto errout;
973                         }
974                         *res_dir = de;
975                         dx_release (frames);
976                         return bh;
977                 }
978                 brelse (bh);
979                 /* Check to see if we should continue to search */
980                 retval = ext3_htree_next_block(dir, hash, frame,
981                                                frames, NULL);
982                 if (retval < 0) {
983                         ext3_warning(sb, __FUNCTION__,
984                              "error reading index page in directory #%lu",
985                              dir->i_ino);
986                         *err = retval;
987                         goto errout;
988                 }
989         } while (retval == 1);
990
991         *err = -ENOENT;
992 errout:
993         dxtrace(printk("%s not found\n", name));
994         dx_release (frames);
995         return NULL;
996 }
997 #endif
998
999 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1000 {
1001         struct inode * inode;
1002         struct ext3_dir_entry_2 * de;
1003         struct buffer_head * bh;
1004
1005         if (dentry->d_name.len > EXT3_NAME_LEN)
1006                 return ERR_PTR(-ENAMETOOLONG);
1007
1008         bh = ext3_find_entry(dentry, &de);
1009         inode = NULL;
1010         if (bh) {
1011                 unsigned long ino = le32_to_cpu(de->inode);
1012                 brelse (bh);
1013                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1014                         ext3_error(dir->i_sb, "ext3_lookup",
1015                                    "bad inode number: %lu", ino);
1016                         inode = NULL;
1017                 } else
1018                         inode = iget(dir->i_sb, ino);
1019
1020                 if (!inode)
1021                         return ERR_PTR(-EACCES);
1022         }
1023         return d_splice_alias(inode, dentry);
1024 }
1025
1026
1027 struct dentry *ext3_get_parent(struct dentry *child)
1028 {
1029         unsigned long ino;
1030         struct dentry *parent;
1031         struct inode *inode;
1032         struct dentry dotdot;
1033         struct ext3_dir_entry_2 * de;
1034         struct buffer_head *bh;
1035
1036         dotdot.d_name.name = "..";
1037         dotdot.d_name.len = 2;
1038         dotdot.d_parent = child; /* confusing, isn't it! */
1039
1040         bh = ext3_find_entry(&dotdot, &de);
1041         inode = NULL;
1042         if (!bh)
1043                 return ERR_PTR(-ENOENT);
1044         ino = le32_to_cpu(de->inode);
1045         brelse(bh);
1046
1047         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1048                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1049                            "bad inode number: %lu", ino);
1050                 inode = NULL;
1051         } else
1052                 inode = iget(child->d_inode->i_sb, ino);
1053
1054         if (!inode)
1055                 return ERR_PTR(-EACCES);
1056
1057         parent = d_alloc_anon(inode);
1058         if (!parent) {
1059                 iput(inode);
1060                 parent = ERR_PTR(-ENOMEM);
1061         }
1062         return parent;
1063 }
1064
1065 #define S_SHIFT 12
1066 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1067         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1068         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1069         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1070         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1071         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1072         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1073         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1074 };
1075
1076 static inline void ext3_set_de_type(struct super_block *sb,
1077                                 struct ext3_dir_entry_2 *de,
1078                                 umode_t mode) {
1079         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1080                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1081 }
1082
1083 #ifdef CONFIG_EXT3_INDEX
1084 static struct ext3_dir_entry_2 *
1085 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1086 {
1087         unsigned rec_len = 0;
1088
1089         while (count--) {
1090                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1091                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1092                 memcpy (to, de, rec_len);
1093                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1094                                 cpu_to_le16(rec_len);
1095                 de->inode = 0;
1096                 map++;
1097                 to += rec_len;
1098         }
1099         return (struct ext3_dir_entry_2 *) (to - rec_len);
1100 }
1101
1102 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1103 {
1104         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1105         unsigned rec_len = 0;
1106
1107         prev = to = de;
1108         while ((char*)de < base + size) {
1109                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1110                                                     le16_to_cpu(de->rec_len));
1111                 if (de->inode && de->name_len) {
1112                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1113                         if (de > to)
1114                                 memmove(to, de, rec_len);
1115                         to->rec_len = cpu_to_le16(rec_len);
1116                         prev = to;
1117                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1118                 }
1119                 de = next;
1120         }
1121         return prev;
1122 }
1123
1124 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1125                         struct buffer_head **bh,struct dx_frame *frame,
1126                         struct dx_hash_info *hinfo, int *error)
1127 {
1128         unsigned blocksize = dir->i_sb->s_blocksize;
1129         unsigned count, continued;
1130         struct buffer_head *bh2;
1131         u32 newblock;
1132         u32 hash2;
1133         struct dx_map_entry *map;
1134         char *data1 = (*bh)->b_data, *data2;
1135         unsigned split;
1136         struct ext3_dir_entry_2 *de = NULL, *de2;
1137         int     err;
1138
1139         bh2 = ext3_append (handle, dir, &newblock, error);
1140         if (!(bh2)) {
1141                 brelse(*bh);
1142                 *bh = NULL;
1143                 goto errout;
1144         }
1145
1146         BUFFER_TRACE(*bh, "get_write_access");
1147         err = ext3_journal_get_write_access(handle, *bh);
1148         if (err) {
1149         journal_error:
1150                 brelse(*bh);
1151                 brelse(bh2);
1152                 *bh = NULL;
1153                 ext3_std_error(dir->i_sb, err);
1154                 goto errout;
1155         }
1156         BUFFER_TRACE(frame->bh, "get_write_access");
1157         err = ext3_journal_get_write_access(handle, frame->bh);
1158         if (err)
1159                 goto journal_error;
1160
1161         data2 = bh2->b_data;
1162
1163         /* create map in the end of data2 block */
1164         map = (struct dx_map_entry *) (data2 + blocksize);
1165         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1166                              blocksize, hinfo, map);
1167         map -= count;
1168         split = count/2; // need to adjust to actual middle
1169         dx_sort_map (map, count);
1170         hash2 = map[split].hash;
1171         continued = hash2 == map[split - 1].hash;
1172         dxtrace(printk("Split block %i at %x, %i/%i\n",
1173                 dx_get_block(frame->at), hash2, split, count-split));
1174
1175         /* Fancy dance to stay within two buffers */
1176         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1177         de = dx_pack_dirents(data1,blocksize);
1178         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1179         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1180         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1181         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1182
1183         /* Which block gets the new entry? */
1184         if (hinfo->hash >= hash2)
1185         {
1186                 swap(*bh, bh2);
1187                 de = de2;
1188         }
1189         dx_insert_block (frame, hash2 + continued, newblock);
1190         err = ext3_journal_dirty_metadata (handle, bh2);
1191         if (err)
1192                 goto journal_error;
1193         err = ext3_journal_dirty_metadata (handle, frame->bh);
1194         if (err)
1195                 goto journal_error;
1196         brelse (bh2);
1197         dxtrace(dx_show_index ("frame", frame->entries));
1198 errout:
1199         return de;
1200 }
1201 #endif
1202
1203
1204 /*
1205  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1206  * it points to a directory entry which is guaranteed to be large
1207  * enough for new directory entry.  If de is NULL, then
1208  * add_dirent_to_buf will attempt search the directory block for
1209  * space.  It will return -ENOSPC if no space is available, and -EIO
1210  * and -EEXIST if directory entry already exists.
1211  *
1212  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1213  * all other cases bh is released.
1214  */
1215 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1216                              struct inode *inode, struct ext3_dir_entry_2 *de,
1217                              struct buffer_head * bh)
1218 {
1219         struct inode    *dir = dentry->d_parent->d_inode;
1220         const char      *name = dentry->d_name.name;
1221         int             namelen = dentry->d_name.len;
1222         unsigned long   offset = 0;
1223         unsigned short  reclen;
1224         int             nlen, rlen, err;
1225         char            *top;
1226
1227         reclen = EXT3_DIR_REC_LEN(namelen);
1228         if (!de) {
1229                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1230                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1231                 while ((char *) de <= top) {
1232                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1233                                                   bh, offset)) {
1234                                 brelse (bh);
1235                                 return -EIO;
1236                         }
1237                         if (ext3_match (namelen, name, de)) {
1238                                 brelse (bh);
1239                                 return -EEXIST;
1240                         }
1241                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1242                         rlen = le16_to_cpu(de->rec_len);
1243                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1244                                 break;
1245                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1246                         offset += rlen;
1247                 }
1248                 if ((char *) de > top)
1249                         return -ENOSPC;
1250         }
1251         BUFFER_TRACE(bh, "get_write_access");
1252         err = ext3_journal_get_write_access(handle, bh);
1253         if (err) {
1254                 ext3_std_error(dir->i_sb, err);
1255                 brelse(bh);
1256                 return err;
1257         }
1258
1259         /* By now the buffer is marked for journaling */
1260         nlen = EXT3_DIR_REC_LEN(de->name_len);
1261         rlen = le16_to_cpu(de->rec_len);
1262         if (de->inode) {
1263                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1264                 de1->rec_len = cpu_to_le16(rlen - nlen);
1265                 de->rec_len = cpu_to_le16(nlen);
1266                 de = de1;
1267         }
1268         de->file_type = EXT3_FT_UNKNOWN;
1269         if (inode) {
1270                 de->inode = cpu_to_le32(inode->i_ino);
1271                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1272         } else
1273                 de->inode = 0;
1274         de->name_len = namelen;
1275         memcpy (de->name, name, namelen);
1276         /*
1277          * XXX shouldn't update any times until successful
1278          * completion of syscall, but too many callers depend
1279          * on this.
1280          *
1281          * XXX similarly, too many callers depend on
1282          * ext3_new_inode() setting the times, but error
1283          * recovery deletes the inode, so the worst that can
1284          * happen is that the times are slightly out of date
1285          * and/or different from the directory change time.
1286          */
1287         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1288         ext3_update_dx_flag(dir);
1289         dir->i_version++;
1290         ext3_mark_inode_dirty(handle, dir);
1291         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1292         err = ext3_journal_dirty_metadata(handle, bh);
1293         if (err)
1294                 ext3_std_error(dir->i_sb, err);
1295         brelse(bh);
1296         return 0;
1297 }
1298
1299 #ifdef CONFIG_EXT3_INDEX
1300 /*
1301  * This converts a one block unindexed directory to a 3 block indexed
1302  * directory, and adds the dentry to the indexed directory.
1303  */
1304 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1305                             struct inode *inode, struct buffer_head *bh)
1306 {
1307         struct inode    *dir = dentry->d_parent->d_inode;
1308         const char      *name = dentry->d_name.name;
1309         int             namelen = dentry->d_name.len;
1310         struct buffer_head *bh2;
1311         struct dx_root  *root;
1312         struct dx_frame frames[2], *frame;
1313         struct dx_entry *entries;
1314         struct ext3_dir_entry_2 *de, *de2;
1315         char            *data1, *top;
1316         unsigned        len;
1317         int             retval;
1318         unsigned        blocksize;
1319         struct dx_hash_info hinfo;
1320         u32             block;
1321         struct fake_dirent *fde;
1322
1323         blocksize =  dir->i_sb->s_blocksize;
1324         dxtrace(printk("Creating index\n"));
1325         retval = ext3_journal_get_write_access(handle, bh);
1326         if (retval) {
1327                 ext3_std_error(dir->i_sb, retval);
1328                 brelse(bh);
1329                 return retval;
1330         }
1331         root = (struct dx_root *) bh->b_data;
1332
1333         bh2 = ext3_append (handle, dir, &block, &retval);
1334         if (!(bh2)) {
1335                 brelse(bh);
1336                 return retval;
1337         }
1338         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1339         data1 = bh2->b_data;
1340
1341         /* The 0th block becomes the root, move the dirents out */
1342         fde = &root->dotdot;
1343         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1344         len = ((char *) root) + blocksize - (char *) de;
1345         memcpy (data1, de, len);
1346         de = (struct ext3_dir_entry_2 *) data1;
1347         top = data1 + len;
1348         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1349                 de = de2;
1350         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1351         /* Initialize the root; the dot dirents already exist */
1352         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1353         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1354         memset (&root->info, 0, sizeof(root->info));
1355         root->info.info_length = sizeof(root->info);
1356         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1357         entries = root->entries;
1358         dx_set_block (entries, 1);
1359         dx_set_count (entries, 1);
1360         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1361
1362         /* Initialize as for dx_probe */
1363         hinfo.hash_version = root->info.hash_version;
1364         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1365         ext3fs_dirhash(name, namelen, &hinfo);
1366         frame = frames;
1367         frame->entries = entries;
1368         frame->at = entries;
1369         frame->bh = bh;
1370         bh = bh2;
1371         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1372         dx_release (frames);
1373         if (!(de))
1374                 return retval;
1375
1376         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1377 }
1378 #endif
1379
1380 /*
1381  *      ext3_add_entry()
1382  *
1383  * adds a file entry to the specified directory, using the same
1384  * semantics as ext3_find_entry(). It returns NULL if it failed.
1385  *
1386  * NOTE!! The inode part of 'de' is left at 0 - which means you
1387  * may not sleep between calling this and putting something into
1388  * the entry, as someone else might have used it while you slept.
1389  */
1390 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1391         struct inode *inode)
1392 {
1393         struct inode *dir = dentry->d_parent->d_inode;
1394         unsigned long offset;
1395         struct buffer_head * bh;
1396         struct ext3_dir_entry_2 *de;
1397         struct super_block * sb;
1398         int     retval;
1399 #ifdef CONFIG_EXT3_INDEX
1400         int     dx_fallback=0;
1401 #endif
1402         unsigned blocksize;
1403         u32 block, blocks;
1404
1405         sb = dir->i_sb;
1406         blocksize = sb->s_blocksize;
1407         if (!dentry->d_name.len)
1408                 return -EINVAL;
1409 #ifdef CONFIG_EXT3_INDEX
1410         if (is_dx(dir)) {
1411                 retval = ext3_dx_add_entry(handle, dentry, inode);
1412                 if (!retval || (retval != ERR_BAD_DX_DIR))
1413                         return retval;
1414                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1415                 dx_fallback++;
1416                 ext3_mark_inode_dirty(handle, dir);
1417         }
1418 #endif
1419         blocks = dir->i_size >> sb->s_blocksize_bits;
1420         for (block = 0, offset = 0; block < blocks; block++) {
1421                 bh = ext3_bread(handle, dir, block, 0, &retval);
1422                 if(!bh)
1423                         return retval;
1424                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1425                 if (retval != -ENOSPC)
1426                         return retval;
1427
1428 #ifdef CONFIG_EXT3_INDEX
1429                 if (blocks == 1 && !dx_fallback &&
1430                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1431                         return make_indexed_dir(handle, dentry, inode, bh);
1432 #endif
1433                 brelse(bh);
1434         }
1435         bh = ext3_append(handle, dir, &block, &retval);
1436         if (!bh)
1437                 return retval;
1438         de = (struct ext3_dir_entry_2 *) bh->b_data;
1439         de->inode = 0;
1440         de->rec_len = cpu_to_le16(blocksize);
1441         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1442 }
1443
1444 #ifdef CONFIG_EXT3_INDEX
1445 /*
1446  * Returns 0 for success, or a negative error value
1447  */
1448 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1449                              struct inode *inode)
1450 {
1451         struct dx_frame frames[2], *frame;
1452         struct dx_entry *entries, *at;
1453         struct dx_hash_info hinfo;
1454         struct buffer_head * bh;
1455         struct inode *dir = dentry->d_parent->d_inode;
1456         struct super_block * sb = dir->i_sb;
1457         struct ext3_dir_entry_2 *de;
1458         int err;
1459
1460         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1461         if (!frame)
1462                 return err;
1463         entries = frame->entries;
1464         at = frame->at;
1465
1466         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1467                 goto cleanup;
1468
1469         BUFFER_TRACE(bh, "get_write_access");
1470         err = ext3_journal_get_write_access(handle, bh);
1471         if (err)
1472                 goto journal_error;
1473
1474         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1475         if (err != -ENOSPC) {
1476                 bh = NULL;
1477                 goto cleanup;
1478         }
1479
1480         /* Block full, should compress but for now just split */
1481         dxtrace(printk("using %u of %u node entries\n",
1482                        dx_get_count(entries), dx_get_limit(entries)));
1483         /* Need to split index? */
1484         if (dx_get_count(entries) == dx_get_limit(entries)) {
1485                 u32 newblock;
1486                 unsigned icount = dx_get_count(entries);
1487                 int levels = frame - frames;
1488                 struct dx_entry *entries2;
1489                 struct dx_node *node2;
1490                 struct buffer_head *bh2;
1491
1492                 if (levels && (dx_get_count(frames->entries) ==
1493                                dx_get_limit(frames->entries))) {
1494                         ext3_warning(sb, __FUNCTION__,
1495                                      "Directory index full!");
1496                         err = -ENOSPC;
1497                         goto cleanup;
1498                 }
1499                 bh2 = ext3_append (handle, dir, &newblock, &err);
1500                 if (!(bh2))
1501                         goto cleanup;
1502                 node2 = (struct dx_node *)(bh2->b_data);
1503                 entries2 = node2->entries;
1504                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1505                 node2->fake.inode = 0;
1506                 BUFFER_TRACE(frame->bh, "get_write_access");
1507                 err = ext3_journal_get_write_access(handle, frame->bh);
1508                 if (err)
1509                         goto journal_error;
1510                 if (levels) {
1511                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1512                         unsigned hash2 = dx_get_hash(entries + icount1);
1513                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1514
1515                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1516                         err = ext3_journal_get_write_access(handle,
1517                                                              frames[0].bh);
1518                         if (err)
1519                                 goto journal_error;
1520
1521                         memcpy ((char *) entries2, (char *) (entries + icount1),
1522                                 icount2 * sizeof(struct dx_entry));
1523                         dx_set_count (entries, icount1);
1524                         dx_set_count (entries2, icount2);
1525                         dx_set_limit (entries2, dx_node_limit(dir));
1526
1527                         /* Which index block gets the new entry? */
1528                         if (at - entries >= icount1) {
1529                                 frame->at = at = at - entries - icount1 + entries2;
1530                                 frame->entries = entries = entries2;
1531                                 swap(frame->bh, bh2);
1532                         }
1533                         dx_insert_block (frames + 0, hash2, newblock);
1534                         dxtrace(dx_show_index ("node", frames[1].entries));
1535                         dxtrace(dx_show_index ("node",
1536                                ((struct dx_node *) bh2->b_data)->entries));
1537                         err = ext3_journal_dirty_metadata(handle, bh2);
1538                         if (err)
1539                                 goto journal_error;
1540                         brelse (bh2);
1541                 } else {
1542                         dxtrace(printk("Creating second level index...\n"));
1543                         memcpy((char *) entries2, (char *) entries,
1544                                icount * sizeof(struct dx_entry));
1545                         dx_set_limit(entries2, dx_node_limit(dir));
1546
1547                         /* Set up root */
1548                         dx_set_count(entries, 1);
1549                         dx_set_block(entries + 0, newblock);
1550                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1551
1552                         /* Add new access path frame */
1553                         frame = frames + 1;
1554                         frame->at = at = at - entries + entries2;
1555                         frame->entries = entries = entries2;
1556                         frame->bh = bh2;
1557                         err = ext3_journal_get_write_access(handle,
1558                                                              frame->bh);
1559                         if (err)
1560                                 goto journal_error;
1561                 }
1562                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1563         }
1564         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1565         if (!de)
1566                 goto cleanup;
1567         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1568         bh = NULL;
1569         goto cleanup;
1570
1571 journal_error:
1572         ext3_std_error(dir->i_sb, err);
1573 cleanup:
1574         if (bh)
1575                 brelse(bh);
1576         dx_release(frames);
1577         return err;
1578 }
1579 #endif
1580
1581 /*
1582  * ext3_delete_entry deletes a directory entry by merging it with the
1583  * previous entry
1584  */
1585 static int ext3_delete_entry (handle_t *handle,
1586                               struct inode * dir,
1587                               struct ext3_dir_entry_2 * de_del,
1588                               struct buffer_head * bh)
1589 {
1590         struct ext3_dir_entry_2 * de, * pde;
1591         int i;
1592
1593         i = 0;
1594         pde = NULL;
1595         de = (struct ext3_dir_entry_2 *) bh->b_data;
1596         while (i < bh->b_size) {
1597                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1598                         return -EIO;
1599                 if (de == de_del)  {
1600                         BUFFER_TRACE(bh, "get_write_access");
1601                         ext3_journal_get_write_access(handle, bh);
1602                         if (pde)
1603                                 pde->rec_len =
1604                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1605                                                     le16_to_cpu(de->rec_len));
1606                         else
1607                                 de->inode = 0;
1608                         dir->i_version++;
1609                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1610                         ext3_journal_dirty_metadata(handle, bh);
1611                         return 0;
1612                 }
1613                 i += le16_to_cpu(de->rec_len);
1614                 pde = de;
1615                 de = (struct ext3_dir_entry_2 *)
1616                         ((char *) de + le16_to_cpu(de->rec_len));
1617         }
1618         return -ENOENT;
1619 }
1620
1621 /*
1622  * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1623  * do not perform it in these functions.  We perform it at the call site,
1624  * if it is needed.
1625  */
1626 static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1627 {
1628         inc_nlink(inode);
1629 }
1630
1631 static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1632 {
1633         drop_nlink(inode);
1634 }
1635
1636 static int ext3_add_nondir(handle_t *handle,
1637                 struct dentry *dentry, struct inode *inode)
1638 {
1639         int err = ext3_add_entry(handle, dentry, inode);
1640         if (!err) {
1641                 ext3_mark_inode_dirty(handle, inode);
1642                 d_instantiate(dentry, inode);
1643                 return 0;
1644         }
1645         ext3_dec_count(handle, inode);
1646         iput(inode);
1647         return err;
1648 }
1649
1650 /*
1651  * By the time this is called, we already have created
1652  * the directory cache entry for the new file, but it
1653  * is so far negative - it has no inode.
1654  *
1655  * If the create succeeds, we fill in the inode information
1656  * with d_instantiate().
1657  */
1658 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1659                 struct nameidata *nd)
1660 {
1661         handle_t *handle;
1662         struct inode * inode;
1663         int err, retries = 0;
1664
1665 retry:
1666         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1667                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1668                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1669         if (IS_ERR(handle))
1670                 return PTR_ERR(handle);
1671
1672         if (IS_DIRSYNC(dir))
1673                 handle->h_sync = 1;
1674
1675         inode = ext3_new_inode (handle, dir, mode);
1676         err = PTR_ERR(inode);
1677         if (!IS_ERR(inode)) {
1678                 inode->i_op = &ext3_file_inode_operations;
1679                 inode->i_fop = &ext3_file_operations;
1680                 ext3_set_aops(inode);
1681                 err = ext3_add_nondir(handle, dentry, inode);
1682         }
1683         ext3_journal_stop(handle);
1684         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1685                 goto retry;
1686         return err;
1687 }
1688
1689 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1690                         int mode, dev_t rdev)
1691 {
1692         handle_t *handle;
1693         struct inode *inode;
1694         int err, retries = 0;
1695
1696         if (!new_valid_dev(rdev))
1697                 return -EINVAL;
1698
1699 retry:
1700         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1701                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1702                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1703         if (IS_ERR(handle))
1704                 return PTR_ERR(handle);
1705
1706         if (IS_DIRSYNC(dir))
1707                 handle->h_sync = 1;
1708
1709         inode = ext3_new_inode (handle, dir, mode);
1710         err = PTR_ERR(inode);
1711         if (!IS_ERR(inode)) {
1712                 init_special_inode(inode, inode->i_mode, rdev);
1713 #ifdef CONFIG_EXT3_FS_XATTR
1714                 inode->i_op = &ext3_special_inode_operations;
1715 #endif
1716                 err = ext3_add_nondir(handle, dentry, inode);
1717         }
1718         ext3_journal_stop(handle);
1719         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1720                 goto retry;
1721         return err;
1722 }
1723
1724 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1725 {
1726         handle_t *handle;
1727         struct inode * inode;
1728         struct buffer_head * dir_block;
1729         struct ext3_dir_entry_2 * de;
1730         int err, retries = 0;
1731
1732         if (dir->i_nlink >= EXT3_LINK_MAX)
1733                 return -EMLINK;
1734
1735 retry:
1736         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1737                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1738                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1739         if (IS_ERR(handle))
1740                 return PTR_ERR(handle);
1741
1742         if (IS_DIRSYNC(dir))
1743                 handle->h_sync = 1;
1744
1745         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1746         err = PTR_ERR(inode);
1747         if (IS_ERR(inode))
1748                 goto out_stop;
1749
1750         inode->i_op = &ext3_dir_inode_operations;
1751         inode->i_fop = &ext3_dir_operations;
1752         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1753         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1754         if (!dir_block) {
1755                 drop_nlink(inode); /* is this nlink == 0? */
1756                 ext3_mark_inode_dirty(handle, inode);
1757                 iput (inode);
1758                 goto out_stop;
1759         }
1760         BUFFER_TRACE(dir_block, "get_write_access");
1761         ext3_journal_get_write_access(handle, dir_block);
1762         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1763         de->inode = cpu_to_le32(inode->i_ino);
1764         de->name_len = 1;
1765         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1766         strcpy (de->name, ".");
1767         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1768         de = (struct ext3_dir_entry_2 *)
1769                         ((char *) de + le16_to_cpu(de->rec_len));
1770         de->inode = cpu_to_le32(dir->i_ino);
1771         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1772         de->name_len = 2;
1773         strcpy (de->name, "..");
1774         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1775         inode->i_nlink = 2;
1776         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1777         ext3_journal_dirty_metadata(handle, dir_block);
1778         brelse (dir_block);
1779         ext3_mark_inode_dirty(handle, inode);
1780         err = ext3_add_entry (handle, dentry, inode);
1781         if (err) {
1782                 inode->i_nlink = 0;
1783                 ext3_mark_inode_dirty(handle, inode);
1784                 iput (inode);
1785                 goto out_stop;
1786         }
1787         inc_nlink(dir);
1788         ext3_update_dx_flag(dir);
1789         ext3_mark_inode_dirty(handle, dir);
1790         d_instantiate(dentry, inode);
1791 out_stop:
1792         ext3_journal_stop(handle);
1793         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1794                 goto retry;
1795         return err;
1796 }
1797
1798 /*
1799  * routine to check that the specified directory is empty (for rmdir)
1800  */
1801 static int empty_dir (struct inode * inode)
1802 {
1803         unsigned long offset;
1804         struct buffer_head * bh;
1805         struct ext3_dir_entry_2 * de, * de1;
1806         struct super_block * sb;
1807         int err = 0;
1808
1809         sb = inode->i_sb;
1810         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1811             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1812                 if (err)
1813                         ext3_error(inode->i_sb, __FUNCTION__,
1814                                    "error %d reading directory #%lu offset 0",
1815                                    err, inode->i_ino);
1816                 else
1817                         ext3_warning(inode->i_sb, __FUNCTION__,
1818                                      "bad directory (dir #%lu) - no data block",
1819                                      inode->i_ino);
1820                 return 1;
1821         }
1822         de = (struct ext3_dir_entry_2 *) bh->b_data;
1823         de1 = (struct ext3_dir_entry_2 *)
1824                         ((char *) de + le16_to_cpu(de->rec_len));
1825         if (le32_to_cpu(de->inode) != inode->i_ino ||
1826                         !le32_to_cpu(de1->inode) ||
1827                         strcmp (".", de->name) ||
1828                         strcmp ("..", de1->name)) {
1829                 ext3_warning (inode->i_sb, "empty_dir",
1830                               "bad directory (dir #%lu) - no `.' or `..'",
1831                               inode->i_ino);
1832                 brelse (bh);
1833                 return 1;
1834         }
1835         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1836         de = (struct ext3_dir_entry_2 *)
1837                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1838         while (offset < inode->i_size ) {
1839                 if (!bh ||
1840                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1841                         err = 0;
1842                         brelse (bh);
1843                         bh = ext3_bread (NULL, inode,
1844                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1845                         if (!bh) {
1846                                 if (err)
1847                                         ext3_error(sb, __FUNCTION__,
1848                                                    "error %d reading directory"
1849                                                    " #%lu offset %lu",
1850                                                    err, inode->i_ino, offset);
1851                                 offset += sb->s_blocksize;
1852                                 continue;
1853                         }
1854                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1855                 }
1856                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1857                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1858                                                          sb->s_blocksize);
1859                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1860                         continue;
1861                 }
1862                 if (le32_to_cpu(de->inode)) {
1863                         brelse (bh);
1864                         return 0;
1865                 }
1866                 offset += le16_to_cpu(de->rec_len);
1867                 de = (struct ext3_dir_entry_2 *)
1868                                 ((char *) de + le16_to_cpu(de->rec_len));
1869         }
1870         brelse (bh);
1871         return 1;
1872 }
1873
1874 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1875  * such inodes, starting at the superblock, in case we crash before the
1876  * file is closed/deleted, or in case the inode truncate spans multiple
1877  * transactions and the last transaction is not recovered after a crash.
1878  *
1879  * At filesystem recovery time, we walk this list deleting unlinked
1880  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1881  */
1882 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1883 {
1884         struct super_block *sb = inode->i_sb;
1885         struct ext3_iloc iloc;
1886         int err = 0, rc;
1887
1888         lock_super(sb);
1889         if (!list_empty(&EXT3_I(inode)->i_orphan))
1890                 goto out_unlock;
1891
1892         /* Orphan handling is only valid for files with data blocks
1893          * being truncated, or files being unlinked. */
1894
1895         /* @@@ FIXME: Observation from aviro:
1896          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1897          * here (on lock_super()), so race with ext3_link() which might bump
1898          * ->i_nlink. For, say it, character device. Not a regular file,
1899          * not a directory, not a symlink and ->i_nlink > 0.
1900          */
1901         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1902                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1903
1904         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1905         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1906         if (err)
1907                 goto out_unlock;
1908
1909         err = ext3_reserve_inode_write(handle, inode, &iloc);
1910         if (err)
1911                 goto out_unlock;
1912
1913         /* Insert this inode at the head of the on-disk orphan list... */
1914         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1915         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1916         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1917         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1918         if (!err)
1919                 err = rc;
1920
1921         /* Only add to the head of the in-memory list if all the
1922          * previous operations succeeded.  If the orphan_add is going to
1923          * fail (possibly taking the journal offline), we can't risk
1924          * leaving the inode on the orphan list: stray orphan-list
1925          * entries can cause panics at unmount time.
1926          *
1927          * This is safe: on error we're going to ignore the orphan list
1928          * anyway on the next recovery. */
1929         if (!err)
1930                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1931
1932         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1933         jbd_debug(4, "orphan inode %lu will point to %d\n",
1934                         inode->i_ino, NEXT_ORPHAN(inode));
1935 out_unlock:
1936         unlock_super(sb);
1937         ext3_std_error(inode->i_sb, err);
1938         return err;
1939 }
1940
1941 /*
1942  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1943  * of such inodes stored on disk, because it is finally being cleaned up.
1944  */
1945 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1946 {
1947         struct list_head *prev;
1948         struct ext3_inode_info *ei = EXT3_I(inode);
1949         struct ext3_sb_info *sbi;
1950         unsigned long ino_next;
1951         struct ext3_iloc iloc;
1952         int err = 0;
1953
1954         lock_super(inode->i_sb);
1955         if (list_empty(&ei->i_orphan)) {
1956                 unlock_super(inode->i_sb);
1957                 return 0;
1958         }
1959
1960         ino_next = NEXT_ORPHAN(inode);
1961         prev = ei->i_orphan.prev;
1962         sbi = EXT3_SB(inode->i_sb);
1963
1964         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1965
1966         list_del_init(&ei->i_orphan);
1967
1968         /* If we're on an error path, we may not have a valid
1969          * transaction handle with which to update the orphan list on
1970          * disk, but we still need to remove the inode from the linked
1971          * list in memory. */
1972         if (!handle)
1973                 goto out;
1974
1975         err = ext3_reserve_inode_write(handle, inode, &iloc);
1976         if (err)
1977                 goto out_err;
1978
1979         if (prev == &sbi->s_orphan) {
1980                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1981                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1982                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1983                 if (err)
1984                         goto out_brelse;
1985                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1986                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1987         } else {
1988                 struct ext3_iloc iloc2;
1989                 struct inode *i_prev =
1990                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1991
1992                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1993                           i_prev->i_ino, ino_next);
1994                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1995                 if (err)
1996                         goto out_brelse;
1997                 NEXT_ORPHAN(i_prev) = ino_next;
1998                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1999         }
2000         if (err)
2001                 goto out_brelse;
2002         NEXT_ORPHAN(inode) = 0;
2003         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2004
2005 out_err:
2006         ext3_std_error(inode->i_sb, err);
2007 out:
2008         unlock_super(inode->i_sb);
2009         return err;
2010
2011 out_brelse:
2012         brelse(iloc.bh);
2013         goto out_err;
2014 }
2015
2016 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2017 {
2018         int retval;
2019         struct inode * inode;
2020         struct buffer_head * bh;
2021         struct ext3_dir_entry_2 * de;
2022         handle_t *handle;
2023
2024         /* Initialize quotas before so that eventual writes go in
2025          * separate transaction */
2026         DQUOT_INIT(dentry->d_inode);
2027         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2028         if (IS_ERR(handle))
2029                 return PTR_ERR(handle);
2030
2031         retval = -ENOENT;
2032         bh = ext3_find_entry (dentry, &de);
2033         if (!bh)
2034                 goto end_rmdir;
2035
2036         if (IS_DIRSYNC(dir))
2037                 handle->h_sync = 1;
2038
2039         inode = dentry->d_inode;
2040
2041         retval = -EIO;
2042         if (le32_to_cpu(de->inode) != inode->i_ino)
2043                 goto end_rmdir;
2044
2045         retval = -ENOTEMPTY;
2046         if (!empty_dir (inode))
2047                 goto end_rmdir;
2048
2049         retval = ext3_delete_entry(handle, dir, de, bh);
2050         if (retval)
2051                 goto end_rmdir;
2052         if (inode->i_nlink != 2)
2053                 ext3_warning (inode->i_sb, "ext3_rmdir",
2054                               "empty directory has nlink!=2 (%d)",
2055                               inode->i_nlink);
2056         inode->i_version++;
2057         clear_nlink(inode);
2058         /* There's no need to set i_disksize: the fact that i_nlink is
2059          * zero will ensure that the right thing happens during any
2060          * recovery. */
2061         inode->i_size = 0;
2062         ext3_orphan_add(handle, inode);
2063         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2064         ext3_mark_inode_dirty(handle, inode);
2065         drop_nlink(dir);
2066         ext3_update_dx_flag(dir);
2067         ext3_mark_inode_dirty(handle, dir);
2068
2069 end_rmdir:
2070         ext3_journal_stop(handle);
2071         brelse (bh);
2072         return retval;
2073 }
2074
2075 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2076 {
2077         int retval;
2078         struct inode * inode;
2079         struct buffer_head * bh;
2080         struct ext3_dir_entry_2 * de;
2081         handle_t *handle;
2082
2083         /* Initialize quotas before so that eventual writes go
2084          * in separate transaction */
2085         DQUOT_INIT(dentry->d_inode);
2086         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2087         if (IS_ERR(handle))
2088                 return PTR_ERR(handle);
2089
2090         if (IS_DIRSYNC(dir))
2091                 handle->h_sync = 1;
2092
2093         retval = -ENOENT;
2094         bh = ext3_find_entry (dentry, &de);
2095         if (!bh)
2096                 goto end_unlink;
2097
2098         inode = dentry->d_inode;
2099
2100         retval = -EIO;
2101         if (le32_to_cpu(de->inode) != inode->i_ino)
2102                 goto end_unlink;
2103
2104         if (!inode->i_nlink) {
2105                 ext3_warning (inode->i_sb, "ext3_unlink",
2106                               "Deleting nonexistent file (%lu), %d",
2107                               inode->i_ino, inode->i_nlink);
2108                 inode->i_nlink = 1;
2109         }
2110         retval = ext3_delete_entry(handle, dir, de, bh);
2111         if (retval)
2112                 goto end_unlink;
2113         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2114         ext3_update_dx_flag(dir);
2115         ext3_mark_inode_dirty(handle, dir);
2116         drop_nlink(inode);
2117         if (!inode->i_nlink)
2118                 ext3_orphan_add(handle, inode);
2119         inode->i_ctime = dir->i_ctime;
2120         ext3_mark_inode_dirty(handle, inode);
2121         retval = 0;
2122
2123 end_unlink:
2124         ext3_journal_stop(handle);
2125         brelse (bh);
2126         return retval;
2127 }
2128
2129 static int ext3_symlink (struct inode * dir,
2130                 struct dentry *dentry, const char * symname)
2131 {
2132         handle_t *handle;
2133         struct inode * inode;
2134         int l, err, retries = 0;
2135
2136         l = strlen(symname)+1;
2137         if (l > dir->i_sb->s_blocksize)
2138                 return -ENAMETOOLONG;
2139
2140 retry:
2141         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2142                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2143                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2144         if (IS_ERR(handle))
2145                 return PTR_ERR(handle);
2146
2147         if (IS_DIRSYNC(dir))
2148                 handle->h_sync = 1;
2149
2150         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2151         err = PTR_ERR(inode);
2152         if (IS_ERR(inode))
2153                 goto out_stop;
2154
2155         if (l > sizeof (EXT3_I(inode)->i_data)) {
2156                 inode->i_op = &ext3_symlink_inode_operations;
2157                 ext3_set_aops(inode);
2158                 /*
2159                  * page_symlink() calls into ext3_prepare/commit_write.
2160                  * We have a transaction open.  All is sweetness.  It also sets
2161                  * i_size in generic_commit_write().
2162                  */
2163                 err = __page_symlink(inode, symname, l,
2164                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2165                 if (err) {
2166                         ext3_dec_count(handle, inode);
2167                         ext3_mark_inode_dirty(handle, inode);
2168                         iput (inode);
2169                         goto out_stop;
2170                 }
2171         } else {
2172                 inode->i_op = &ext3_fast_symlink_inode_operations;
2173                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2174                 inode->i_size = l-1;
2175         }
2176         EXT3_I(inode)->i_disksize = inode->i_size;
2177         err = ext3_add_nondir(handle, dentry, inode);
2178 out_stop:
2179         ext3_journal_stop(handle);
2180         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2181                 goto retry;
2182         return err;
2183 }
2184
2185 static int ext3_link (struct dentry * old_dentry,
2186                 struct inode * dir, struct dentry *dentry)
2187 {
2188         handle_t *handle;
2189         struct inode *inode = old_dentry->d_inode;
2190         int err, retries = 0;
2191
2192         if (inode->i_nlink >= EXT3_LINK_MAX)
2193                 return -EMLINK;
2194
2195 retry:
2196         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2197                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2198         if (IS_ERR(handle))
2199                 return PTR_ERR(handle);
2200
2201         if (IS_DIRSYNC(dir))
2202                 handle->h_sync = 1;
2203
2204         inode->i_ctime = CURRENT_TIME_SEC;
2205         ext3_inc_count(handle, inode);
2206         atomic_inc(&inode->i_count);
2207
2208         err = ext3_add_nondir(handle, dentry, inode);
2209         ext3_journal_stop(handle);
2210         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2211                 goto retry;
2212         return err;
2213 }
2214
2215 #define PARENT_INO(buffer) \
2216         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2217         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2218
2219 /*
2220  * Anybody can rename anything with this: the permission checks are left to the
2221  * higher-level routines.
2222  */
2223 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2224                            struct inode * new_dir,struct dentry *new_dentry)
2225 {
2226         handle_t *handle;
2227         struct inode * old_inode, * new_inode;
2228         struct buffer_head * old_bh, * new_bh, * dir_bh;
2229         struct ext3_dir_entry_2 * old_de, * new_de;
2230         int retval;
2231
2232         old_bh = new_bh = dir_bh = NULL;
2233
2234         /* Initialize quotas before so that eventual writes go
2235          * in separate transaction */
2236         if (new_dentry->d_inode)
2237                 DQUOT_INIT(new_dentry->d_inode);
2238         handle = ext3_journal_start(old_dir, 2 *
2239                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2240                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2241         if (IS_ERR(handle))
2242                 return PTR_ERR(handle);
2243
2244         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2245                 handle->h_sync = 1;
2246
2247         old_bh = ext3_find_entry (old_dentry, &old_de);
2248         /*
2249          *  Check for inode number is _not_ due to possible IO errors.
2250          *  We might rmdir the source, keep it as pwd of some process
2251          *  and merrily kill the link to whatever was created under the
2252          *  same name. Goodbye sticky bit ;-<
2253          */
2254         old_inode = old_dentry->d_inode;
2255         retval = -ENOENT;
2256         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2257                 goto end_rename;
2258
2259         new_inode = new_dentry->d_inode;
2260         new_bh = ext3_find_entry (new_dentry, &new_de);
2261         if (new_bh) {
2262                 if (!new_inode) {
2263                         brelse (new_bh);
2264                         new_bh = NULL;
2265                 }
2266         }
2267         if (S_ISDIR(old_inode->i_mode)) {
2268                 if (new_inode) {
2269                         retval = -ENOTEMPTY;
2270                         if (!empty_dir (new_inode))
2271                                 goto end_rename;
2272                 }
2273                 retval = -EIO;
2274                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2275                 if (!dir_bh)
2276                         goto end_rename;
2277                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2278                         goto end_rename;
2279                 retval = -EMLINK;
2280                 if (!new_inode && new_dir!=old_dir &&
2281                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2282                         goto end_rename;
2283         }
2284         if (!new_bh) {
2285                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2286                 if (retval)
2287                         goto end_rename;
2288         } else {
2289                 BUFFER_TRACE(new_bh, "get write access");
2290                 ext3_journal_get_write_access(handle, new_bh);
2291                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2292                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2293                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2294                         new_de->file_type = old_de->file_type;
2295                 new_dir->i_version++;
2296                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2297                 ext3_journal_dirty_metadata(handle, new_bh);
2298                 brelse(new_bh);
2299                 new_bh = NULL;
2300         }
2301
2302         /*
2303          * Like most other Unix systems, set the ctime for inodes on a
2304          * rename.
2305          */
2306         old_inode->i_ctime = CURRENT_TIME_SEC;
2307         ext3_mark_inode_dirty(handle, old_inode);
2308
2309         /*
2310          * ok, that's it
2311          */
2312         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2313             old_de->name_len != old_dentry->d_name.len ||
2314             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2315             (retval = ext3_delete_entry(handle, old_dir,
2316                                         old_de, old_bh)) == -ENOENT) {
2317                 /* old_de could have moved from under us during htree split, so
2318                  * make sure that we are deleting the right entry.  We might
2319                  * also be pointing to a stale entry in the unused part of
2320                  * old_bh so just checking inum and the name isn't enough. */
2321                 struct buffer_head *old_bh2;
2322                 struct ext3_dir_entry_2 *old_de2;
2323
2324                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2325                 if (old_bh2) {
2326                         retval = ext3_delete_entry(handle, old_dir,
2327                                                    old_de2, old_bh2);
2328                         brelse(old_bh2);
2329                 }
2330         }
2331         if (retval) {
2332                 ext3_warning(old_dir->i_sb, "ext3_rename",
2333                                 "Deleting old file (%lu), %d, error=%d",
2334                                 old_dir->i_ino, old_dir->i_nlink, retval);
2335         }
2336
2337         if (new_inode) {
2338                 drop_nlink(new_inode);
2339                 new_inode->i_ctime = CURRENT_TIME_SEC;
2340         }
2341         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2342         ext3_update_dx_flag(old_dir);
2343         if (dir_bh) {
2344                 BUFFER_TRACE(dir_bh, "get_write_access");
2345                 ext3_journal_get_write_access(handle, dir_bh);
2346                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2347                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2348                 ext3_journal_dirty_metadata(handle, dir_bh);
2349                 drop_nlink(old_dir);
2350                 if (new_inode) {
2351                         drop_nlink(new_inode);
2352                 } else {
2353                         inc_nlink(new_dir);
2354                         ext3_update_dx_flag(new_dir);
2355                         ext3_mark_inode_dirty(handle, new_dir);
2356                 }
2357         }
2358         ext3_mark_inode_dirty(handle, old_dir);
2359         if (new_inode) {
2360                 ext3_mark_inode_dirty(handle, new_inode);
2361                 if (!new_inode->i_nlink)
2362                         ext3_orphan_add(handle, new_inode);
2363         }
2364         retval = 0;
2365
2366 end_rename:
2367         brelse (dir_bh);
2368         brelse (old_bh);
2369         brelse (new_bh);
2370         ext3_journal_stop(handle);
2371         return retval;
2372 }
2373
2374 /*
2375  * directories can handle most operations...
2376  */
2377 struct inode_operations ext3_dir_inode_operations = {
2378         .create         = ext3_create,
2379         .lookup         = ext3_lookup,
2380         .link           = ext3_link,
2381         .unlink         = ext3_unlink,
2382         .symlink        = ext3_symlink,
2383         .mkdir          = ext3_mkdir,
2384         .rmdir          = ext3_rmdir,
2385         .mknod          = ext3_mknod,
2386         .rename         = ext3_rename,
2387         .setattr        = ext3_setattr,
2388 #ifdef CONFIG_EXT3_FS_XATTR
2389         .setxattr       = generic_setxattr,
2390         .getxattr       = generic_getxattr,
2391         .listxattr      = ext3_listxattr,
2392         .removexattr    = generic_removexattr,
2393 #endif
2394         .permission     = ext3_permission,
2395 };
2396
2397 struct inode_operations ext3_special_inode_operations = {
2398         .setattr        = ext3_setattr,
2399 #ifdef CONFIG_EXT3_FS_XATTR
2400         .setxattr       = generic_setxattr,
2401         .getxattr       = generic_getxattr,
2402         .listxattr      = ext3_listxattr,
2403         .removexattr    = generic_removexattr,
2404 #endif
2405         .permission     = ext3_permission,
2406 };