Merge ssh://master.kernel.org/pub/scm/linux/kernel/git/tglx/linux-2.6-x86
[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
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext3_append(handle_t *handle,
53                                         struct inode *inode,
54                                         u32 *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT3_I(inode)->i_disksize = inode->i_size;
63                 ext3_journal_get_write_access(handle,bh);
64         }
65         return bh;
66 }
67
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u16 offs;
144         u16 size;
145 };
146
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned value);
149 static inline unsigned dx_get_hash (struct dx_entry *entry);
150 static void dx_set_hash (struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count (struct dx_entry *entries);
152 static unsigned dx_get_limit (struct dx_entry *entries);
153 static void dx_set_count (struct dx_entry *entries, unsigned value);
154 static void dx_set_limit (struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit (struct inode *dir);
157 static struct dx_frame *dx_probe(struct dentry *dentry,
158                                  struct inode *dir,
159                                  struct dx_hash_info *hinfo,
160                                  struct dx_frame *frame,
161                                  int *err);
162 static void dx_release (struct dx_frame *frames);
163 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167                 struct dx_map_entry *offsets, int count);
168 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171                                  struct dx_frame *frame,
172                                  struct dx_frame *frames,
173                                  __u32 *start_hash);
174 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175                        struct ext3_dir_entry_2 **res_dir, int *err);
176 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177                              struct inode *inode);
178
179 /*
180  * Future: use high four bits of block for coalesce-on-delete flags
181  * Mask them off for now.
182  */
183
184 static inline unsigned dx_get_block (struct dx_entry *entry)
185 {
186         return le32_to_cpu(entry->block) & 0x00ffffff;
187 }
188
189 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190 {
191         entry->block = cpu_to_le32(value);
192 }
193
194 static inline unsigned dx_get_hash (struct dx_entry *entry)
195 {
196         return le32_to_cpu(entry->hash);
197 }
198
199 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200 {
201         entry->hash = cpu_to_le32(value);
202 }
203
204 static inline unsigned dx_get_count (struct dx_entry *entries)
205 {
206         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207 }
208
209 static inline unsigned dx_get_limit (struct dx_entry *entries)
210 {
211         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212 }
213
214 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215 {
216         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217 }
218
219 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220 {
221         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222 }
223
224 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225 {
226         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
227                 EXT3_DIR_REC_LEN(2) - infosize;
228         return 0? 20: entry_space / sizeof(struct dx_entry);
229 }
230
231 static inline unsigned dx_node_limit (struct inode *dir)
232 {
233         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
234         return 0? 22: entry_space / sizeof(struct dx_entry);
235 }
236
237 /*
238  * Debug
239  */
240 #ifdef DX_DEBUG
241 static void dx_show_index (char * label, struct dx_entry *entries)
242 {
243         int i, n = dx_get_count (entries);
244         printk("%s index ", label);
245         for (i = 0; i < n; i++)
246         {
247                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
248         }
249         printk("\n");
250 }
251
252 struct stats
253 {
254         unsigned names;
255         unsigned space;
256         unsigned bcount;
257 };
258
259 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
260                                  int size, int show_names)
261 {
262         unsigned names = 0, space = 0;
263         char *base = (char *) de;
264         struct dx_hash_info h = *hinfo;
265
266         printk("names: ");
267         while ((char *) de < base + size)
268         {
269                 if (de->inode)
270                 {
271                         if (show_names)
272                         {
273                                 int len = de->name_len;
274                                 char *name = de->name;
275                                 while (len--) printk("%c", *name++);
276                                 ext3fs_dirhash(de->name, de->name_len, &h);
277                                 printk(":%x.%u ", h.hash,
278                                        ((char *) de - base));
279                         }
280                         space += EXT3_DIR_REC_LEN(de->name_len);
281                         names++;
282                 }
283                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284         }
285         printk("(%i)\n", names);
286         return (struct stats) { names, space, 1 };
287 }
288
289 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290                              struct dx_entry *entries, int levels)
291 {
292         unsigned blocksize = dir->i_sb->s_blocksize;
293         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294         unsigned bcount = 0;
295         struct buffer_head *bh;
296         int err;
297         printk("%i indexed blocks...\n", count);
298         for (i = 0; i < count; i++, entries++)
299         {
300                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302                 struct stats stats;
303                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
304                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
305                 stats = levels?
306                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
308                 names += stats.names;
309                 space += stats.space;
310                 bcount += stats.bcount;
311                 brelse (bh);
312         }
313         if (bcount)
314                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
315                         names, space/bcount,(space/bcount)*100/blocksize);
316         return (struct stats) { names, space, bcount};
317 }
318 #endif /* DX_DEBUG */
319
320 /*
321  * Probe for a directory leaf block to search.
322  *
323  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324  * error in the directory index, and the caller should fall back to
325  * searching the directory normally.  The callers of dx_probe **MUST**
326  * check for this error code, and make sure it never gets reflected
327  * back to userspace.
328  */
329 static struct dx_frame *
330 dx_probe(struct dentry *dentry, struct inode *dir,
331          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332 {
333         unsigned count, indirect;
334         struct dx_entry *at, *entries, *p, *q, *m;
335         struct dx_root *root;
336         struct buffer_head *bh;
337         struct dx_frame *frame = frame_in;
338         u32 hash;
339
340         frame->bh = NULL;
341         if (dentry)
342                 dir = dentry->d_parent->d_inode;
343         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
344                 goto fail;
345         root = (struct dx_root *) bh->b_data;
346         if (root->info.hash_version != DX_HASH_TEA &&
347             root->info.hash_version != DX_HASH_HALF_MD4 &&
348             root->info.hash_version != DX_HASH_LEGACY) {
349                 ext3_warning(dir->i_sb, __FUNCTION__,
350                              "Unrecognised inode hash code %d",
351                              root->info.hash_version);
352                 brelse(bh);
353                 *err = ERR_BAD_DX_DIR;
354                 goto fail;
355         }
356         hinfo->hash_version = root->info.hash_version;
357         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
358         if (dentry)
359                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
360         hash = hinfo->hash;
361
362         if (root->info.unused_flags & 1) {
363                 ext3_warning(dir->i_sb, __FUNCTION__,
364                              "Unimplemented inode hash flags: %#06x",
365                              root->info.unused_flags);
366                 brelse(bh);
367                 *err = ERR_BAD_DX_DIR;
368                 goto fail;
369         }
370
371         if ((indirect = root->info.indirect_levels) > 1) {
372                 ext3_warning(dir->i_sb, __FUNCTION__,
373                              "Unimplemented inode hash depth: %#06x",
374                              root->info.indirect_levels);
375                 brelse(bh);
376                 *err = ERR_BAD_DX_DIR;
377                 goto fail;
378         }
379
380         entries = (struct dx_entry *) (((char *)&root->info) +
381                                        root->info.info_length);
382
383         if (dx_get_limit(entries) != dx_root_limit(dir,
384                                                    root->info.info_length)) {
385                 ext3_warning(dir->i_sb, __FUNCTION__,
386                              "dx entry: limit != root limit");
387                 brelse(bh);
388                 *err = ERR_BAD_DX_DIR;
389                 goto fail;
390         }
391
392         dxtrace (printk("Look up %x", hash));
393         while (1)
394         {
395                 count = dx_get_count(entries);
396                 if (!count || count > dx_get_limit(entries)) {
397                         ext3_warning(dir->i_sb, __FUNCTION__,
398                                      "dx entry: no count or count > limit");
399                         brelse(bh);
400                         *err = ERR_BAD_DX_DIR;
401                         goto fail2;
402                 }
403
404                 p = entries + 1;
405                 q = entries + count - 1;
406                 while (p <= q)
407                 {
408                         m = p + (q - p)/2;
409                         dxtrace(printk("."));
410                         if (dx_get_hash(m) > hash)
411                                 q = m - 1;
412                         else
413                                 p = m + 1;
414                 }
415
416                 if (0) // linear search cross check
417                 {
418                         unsigned n = count - 1;
419                         at = entries;
420                         while (n--)
421                         {
422                                 dxtrace(printk(","));
423                                 if (dx_get_hash(++at) > hash)
424                                 {
425                                         at--;
426                                         break;
427                                 }
428                         }
429                         assert (at == p - 1);
430                 }
431
432                 at = p - 1;
433                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
434                 frame->bh = bh;
435                 frame->entries = entries;
436                 frame->at = at;
437                 if (!indirect--) return frame;
438                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
439                         goto fail2;
440                 at = entries = ((struct dx_node *) bh->b_data)->entries;
441                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
442                         ext3_warning(dir->i_sb, __FUNCTION__,
443                                      "dx entry: limit != node limit");
444                         brelse(bh);
445                         *err = ERR_BAD_DX_DIR;
446                         goto fail2;
447                 }
448                 frame++;
449                 frame->bh = NULL;
450         }
451 fail2:
452         while (frame >= frame_in) {
453                 brelse(frame->bh);
454                 frame--;
455         }
456 fail:
457         if (*err == ERR_BAD_DX_DIR)
458                 ext3_warning(dir->i_sb, __FUNCTION__,
459                              "Corrupt dir inode %ld, running e2fsck is "
460                              "recommended.", dir->i_ino);
461         return NULL;
462 }
463
464 static void dx_release (struct dx_frame *frames)
465 {
466         if (frames[0].bh == NULL)
467                 return;
468
469         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
470                 brelse(frames[1].bh);
471         brelse(frames[0].bh);
472 }
473
474 /*
475  * This function increments the frame pointer to search the next leaf
476  * block, and reads in the necessary intervening nodes if the search
477  * should be necessary.  Whether or not the search is necessary is
478  * controlled by the hash parameter.  If the hash value is even, then
479  * the search is only continued if the next block starts with that
480  * hash value.  This is used if we are searching for a specific file.
481  *
482  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
483  *
484  * This function returns 1 if the caller should continue to search,
485  * or 0 if it should not.  If there is an error reading one of the
486  * index blocks, it will a negative error code.
487  *
488  * If start_hash is non-null, it will be filled in with the starting
489  * hash of the next page.
490  */
491 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
492                                  struct dx_frame *frame,
493                                  struct dx_frame *frames,
494                                  __u32 *start_hash)
495 {
496         struct dx_frame *p;
497         struct buffer_head *bh;
498         int err, num_frames = 0;
499         __u32 bhash;
500
501         p = frame;
502         /*
503          * Find the next leaf page by incrementing the frame pointer.
504          * If we run out of entries in the interior node, loop around and
505          * increment pointer in the parent node.  When we break out of
506          * this loop, num_frames indicates the number of interior
507          * nodes need to be read.
508          */
509         while (1) {
510                 if (++(p->at) < p->entries + dx_get_count(p->entries))
511                         break;
512                 if (p == frames)
513                         return 0;
514                 num_frames++;
515                 p--;
516         }
517
518         /*
519          * If the hash is 1, then continue only if the next page has a
520          * continuation hash of any value.  This is used for readdir
521          * handling.  Otherwise, check to see if the hash matches the
522          * desired contiuation hash.  If it doesn't, return since
523          * there's no point to read in the successive index pages.
524          */
525         bhash = dx_get_hash(p->at);
526         if (start_hash)
527                 *start_hash = bhash;
528         if ((hash & 1) == 0) {
529                 if ((bhash & ~1) != hash)
530                         return 0;
531         }
532         /*
533          * If the hash is HASH_NB_ALWAYS, we always go to the next
534          * block so no check is necessary
535          */
536         while (num_frames--) {
537                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
538                                       0, &err)))
539                         return err; /* Failure */
540                 p++;
541                 brelse (p->bh);
542                 p->bh = bh;
543                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
544         }
545         return 1;
546 }
547
548
549 /*
550  * p is at least 6 bytes before the end of page
551  */
552 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
553 {
554         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
555 }
556
557 /*
558  * This function fills a red-black tree with information from a
559  * directory block.  It returns the number directory entries loaded
560  * into the tree.  If there is an error it is returned in err.
561  */
562 static int htree_dirblock_to_tree(struct file *dir_file,
563                                   struct inode *dir, int block,
564                                   struct dx_hash_info *hinfo,
565                                   __u32 start_hash, __u32 start_minor_hash)
566 {
567         struct buffer_head *bh;
568         struct ext3_dir_entry_2 *de, *top;
569         int err, count = 0;
570
571         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
572         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
573                 return err;
574
575         de = (struct ext3_dir_entry_2 *) bh->b_data;
576         top = (struct ext3_dir_entry_2 *) ((char *) de +
577                                            dir->i_sb->s_blocksize -
578                                            EXT3_DIR_REC_LEN(0));
579         for (; de < top; de = ext3_next_entry(de)) {
580                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
581                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
582                                                 +((char *)de - bh->b_data))) {
583                         /* On error, skip the f_pos to the next block. */
584                         dir_file->f_pos = (dir_file->f_pos |
585                                         (dir->i_sb->s_blocksize - 1)) + 1;
586                         brelse (bh);
587                         return count;
588                 }
589                 ext3fs_dirhash(de->name, de->name_len, hinfo);
590                 if ((hinfo->hash < start_hash) ||
591                     ((hinfo->hash == start_hash) &&
592                      (hinfo->minor_hash < start_minor_hash)))
593                         continue;
594                 if (de->inode == 0)
595                         continue;
596                 if ((err = ext3_htree_store_dirent(dir_file,
597                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
598                         brelse(bh);
599                         return err;
600                 }
601                 count++;
602         }
603         brelse(bh);
604         return count;
605 }
606
607
608 /*
609  * This function fills a red-black tree with information from a
610  * directory.  We start scanning the directory in hash order, starting
611  * at start_hash and start_minor_hash.
612  *
613  * This function returns the number of entries inserted into the tree,
614  * or a negative error code.
615  */
616 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
617                          __u32 start_minor_hash, __u32 *next_hash)
618 {
619         struct dx_hash_info hinfo;
620         struct ext3_dir_entry_2 *de;
621         struct dx_frame frames[2], *frame;
622         struct inode *dir;
623         int block, err;
624         int count = 0;
625         int ret;
626         __u32 hashval;
627
628         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
629                        start_minor_hash));
630         dir = dir_file->f_path.dentry->d_inode;
631         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
632                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
633                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
634                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
635                                                start_hash, start_minor_hash);
636                 *next_hash = ~0;
637                 return count;
638         }
639         hinfo.hash = start_hash;
640         hinfo.minor_hash = 0;
641         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
642         if (!frame)
643                 return err;
644
645         /* Add '.' and '..' from the htree header */
646         if (!start_hash && !start_minor_hash) {
647                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
648                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
649                         goto errout;
650                 count++;
651         }
652         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
653                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
654                 de = ext3_next_entry(de);
655                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
656                         goto errout;
657                 count++;
658         }
659
660         while (1) {
661                 block = dx_get_block(frame->at);
662                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
663                                              start_hash, start_minor_hash);
664                 if (ret < 0) {
665                         err = ret;
666                         goto errout;
667                 }
668                 count += ret;
669                 hashval = ~0;
670                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
671                                             frame, frames, &hashval);
672                 *next_hash = hashval;
673                 if (ret < 0) {
674                         err = ret;
675                         goto errout;
676                 }
677                 /*
678                  * Stop if:  (a) there are no more entries, or
679                  * (b) we have inserted at least one entry and the
680                  * next hash value is not a continuation
681                  */
682                 if ((ret == 0) ||
683                     (count && ((hashval & 1) == 0)))
684                         break;
685         }
686         dx_release(frames);
687         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
688                        count, *next_hash));
689         return count;
690 errout:
691         dx_release(frames);
692         return (err);
693 }
694
695
696 /*
697  * Directory block splitting, compacting
698  */
699
700 /*
701  * Create map of hash values, offsets, and sizes, stored at end of block.
702  * Returns number of entries mapped.
703  */
704 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
705                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
706 {
707         int count = 0;
708         char *base = (char *) de;
709         struct dx_hash_info h = *hinfo;
710
711         while ((char *) de < base + size)
712         {
713                 if (de->name_len && de->inode) {
714                         ext3fs_dirhash(de->name, de->name_len, &h);
715                         map_tail--;
716                         map_tail->hash = h.hash;
717                         map_tail->offs = (u16) ((char *) de - base);
718                         map_tail->size = le16_to_cpu(de->rec_len);
719                         count++;
720                         cond_resched();
721                 }
722                 /* XXX: do we need to check rec_len == 0 case? -Chris */
723                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
724         }
725         return count;
726 }
727
728 /* Sort map by hash value */
729 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
730 {
731         struct dx_map_entry *p, *q, *top = map + count - 1;
732         int more;
733         /* Combsort until bubble sort doesn't suck */
734         while (count > 2)
735         {
736                 count = count*10/13;
737                 if (count - 9 < 2) /* 9, 10 -> 11 */
738                         count = 11;
739                 for (p = top, q = p - count; q >= map; p--, q--)
740                         if (p->hash < q->hash)
741                                 swap(*p, *q);
742         }
743         /* Garden variety bubble sort */
744         do {
745                 more = 0;
746                 q = top;
747                 while (q-- > map)
748                 {
749                         if (q[1].hash >= q[0].hash)
750                                 continue;
751                         swap(*(q+1), *q);
752                         more = 1;
753                 }
754         } while(more);
755 }
756
757 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
758 {
759         struct dx_entry *entries = frame->entries;
760         struct dx_entry *old = frame->at, *new = old + 1;
761         int count = dx_get_count(entries);
762
763         assert(count < dx_get_limit(entries));
764         assert(old < entries + count);
765         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
766         dx_set_hash(new, hash);
767         dx_set_block(new, block);
768         dx_set_count(entries, count + 1);
769 }
770
771 static void ext3_update_dx_flag(struct inode *inode)
772 {
773         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
774                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
775                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
776 }
777
778 /*
779  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
780  *
781  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
782  * `de != NULL' is guaranteed by caller.
783  */
784 static inline int ext3_match (int len, const char * const name,
785                               struct ext3_dir_entry_2 * de)
786 {
787         if (len != de->name_len)
788                 return 0;
789         if (!de->inode)
790                 return 0;
791         return !memcmp(name, de->name, len);
792 }
793
794 /*
795  * Returns 0 if not found, -1 on failure, and 1 on success
796  */
797 static inline int search_dirblock(struct buffer_head * bh,
798                                   struct inode *dir,
799                                   struct dentry *dentry,
800                                   unsigned long offset,
801                                   struct ext3_dir_entry_2 ** res_dir)
802 {
803         struct ext3_dir_entry_2 * de;
804         char * dlimit;
805         int de_len;
806         const char *name = dentry->d_name.name;
807         int namelen = dentry->d_name.len;
808
809         de = (struct ext3_dir_entry_2 *) bh->b_data;
810         dlimit = bh->b_data + dir->i_sb->s_blocksize;
811         while ((char *) de < dlimit) {
812                 /* this code is executed quadratically often */
813                 /* do minimal checking `by hand' */
814
815                 if ((char *) de + namelen <= dlimit &&
816                     ext3_match (namelen, name, de)) {
817                         /* found a match - just to be sure, do a full check */
818                         if (!ext3_check_dir_entry("ext3_find_entry",
819                                                   dir, de, bh, offset))
820                                 return -1;
821                         *res_dir = de;
822                         return 1;
823                 }
824                 /* prevent looping on a bad block */
825                 de_len = le16_to_cpu(de->rec_len);
826                 if (de_len <= 0)
827                         return -1;
828                 offset += de_len;
829                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
830         }
831         return 0;
832 }
833
834
835 /*
836  *      ext3_find_entry()
837  *
838  * finds an entry in the specified directory with the wanted name. It
839  * returns the cache buffer in which the entry was found, and the entry
840  * itself (as a parameter - res_dir). It does NOT read the inode of the
841  * entry - you'll have to do that yourself if you want to.
842  *
843  * The returned buffer_head has ->b_count elevated.  The caller is expected
844  * to brelse() it when appropriate.
845  */
846 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
847                                         struct ext3_dir_entry_2 ** res_dir)
848 {
849         struct super_block * sb;
850         struct buffer_head * bh_use[NAMEI_RA_SIZE];
851         struct buffer_head * bh, *ret = NULL;
852         unsigned long start, block, b;
853         int ra_max = 0;         /* Number of bh's in the readahead
854                                    buffer, bh_use[] */
855         int ra_ptr = 0;         /* Current index into readahead
856                                    buffer */
857         int num = 0;
858         int nblocks, i, err;
859         struct inode *dir = dentry->d_parent->d_inode;
860         int namelen;
861         const u8 *name;
862         unsigned blocksize;
863
864         *res_dir = NULL;
865         sb = dir->i_sb;
866         blocksize = sb->s_blocksize;
867         namelen = dentry->d_name.len;
868         name = dentry->d_name.name;
869         if (namelen > EXT3_NAME_LEN)
870                 return NULL;
871         if (is_dx(dir)) {
872                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
873                 /*
874                  * On success, or if the error was file not found,
875                  * return.  Otherwise, fall back to doing a search the
876                  * old fashioned way.
877                  */
878                 if (bh || (err != ERR_BAD_DX_DIR))
879                         return bh;
880                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
881         }
882         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
883         start = EXT3_I(dir)->i_dir_start_lookup;
884         if (start >= nblocks)
885                 start = 0;
886         block = start;
887 restart:
888         do {
889                 /*
890                  * We deal with the read-ahead logic here.
891                  */
892                 if (ra_ptr >= ra_max) {
893                         /* Refill the readahead buffer */
894                         ra_ptr = 0;
895                         b = block;
896                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
897                                 /*
898                                  * Terminate if we reach the end of the
899                                  * directory and must wrap, or if our
900                                  * search has finished at this block.
901                                  */
902                                 if (b >= nblocks || (num && block == start)) {
903                                         bh_use[ra_max] = NULL;
904                                         break;
905                                 }
906                                 num++;
907                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
908                                 bh_use[ra_max] = bh;
909                                 if (bh)
910                                         ll_rw_block(READ_META, 1, &bh);
911                         }
912                 }
913                 if ((bh = bh_use[ra_ptr++]) == NULL)
914                         goto next;
915                 wait_on_buffer(bh);
916                 if (!buffer_uptodate(bh)) {
917                         /* read error, skip block & hope for the best */
918                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
919                                    "offset %lu", dir->i_ino, block);
920                         brelse(bh);
921                         goto next;
922                 }
923                 i = search_dirblock(bh, dir, dentry,
924                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
925                 if (i == 1) {
926                         EXT3_I(dir)->i_dir_start_lookup = block;
927                         ret = bh;
928                         goto cleanup_and_exit;
929                 } else {
930                         brelse(bh);
931                         if (i < 0)
932                                 goto cleanup_and_exit;
933                 }
934         next:
935                 if (++block >= nblocks)
936                         block = 0;
937         } while (block != start);
938
939         /*
940          * If the directory has grown while we were searching, then
941          * search the last part of the directory before giving up.
942          */
943         block = nblocks;
944         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
945         if (block < nblocks) {
946                 start = 0;
947                 goto restart;
948         }
949
950 cleanup_and_exit:
951         /* Clean up the read-ahead blocks */
952         for (; ra_ptr < ra_max; ra_ptr++)
953                 brelse (bh_use[ra_ptr]);
954         return ret;
955 }
956
957 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
958                        struct ext3_dir_entry_2 **res_dir, int *err)
959 {
960         struct super_block * sb;
961         struct dx_hash_info     hinfo;
962         u32 hash;
963         struct dx_frame frames[2], *frame;
964         struct ext3_dir_entry_2 *de, *top;
965         struct buffer_head *bh;
966         unsigned long block;
967         int retval;
968         int namelen = dentry->d_name.len;
969         const u8 *name = dentry->d_name.name;
970         struct inode *dir = dentry->d_parent->d_inode;
971
972         sb = dir->i_sb;
973         /* NFS may look up ".." - look at dx_root directory block */
974         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
975                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
976                         return NULL;
977         } else {
978                 frame = frames;
979                 frame->bh = NULL;                       /* for dx_release() */
980                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
981                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
982         }
983         hash = hinfo.hash;
984         do {
985                 block = dx_get_block(frame->at);
986                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
987                         goto errout;
988                 de = (struct ext3_dir_entry_2 *) bh->b_data;
989                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
990                                        EXT3_DIR_REC_LEN(0));
991                 for (; de < top; de = ext3_next_entry(de))
992                 if (ext3_match (namelen, name, de)) {
993                         if (!ext3_check_dir_entry("ext3_find_entry",
994                                                   dir, de, bh,
995                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
996                                           +((char *)de - bh->b_data))) {
997                                 brelse (bh);
998                                 *err = ERR_BAD_DX_DIR;
999                                 goto errout;
1000                         }
1001                         *res_dir = de;
1002                         dx_release (frames);
1003                         return bh;
1004                 }
1005                 brelse (bh);
1006                 /* Check to see if we should continue to search */
1007                 retval = ext3_htree_next_block(dir, hash, frame,
1008                                                frames, NULL);
1009                 if (retval < 0) {
1010                         ext3_warning(sb, __FUNCTION__,
1011                              "error reading index page in directory #%lu",
1012                              dir->i_ino);
1013                         *err = retval;
1014                         goto errout;
1015                 }
1016         } while (retval == 1);
1017
1018         *err = -ENOENT;
1019 errout:
1020         dxtrace(printk("%s not found\n", name));
1021         dx_release (frames);
1022         return NULL;
1023 }
1024
1025 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1026 {
1027         struct inode * inode;
1028         struct ext3_dir_entry_2 * de;
1029         struct buffer_head * bh;
1030
1031         if (dentry->d_name.len > EXT3_NAME_LEN)
1032                 return ERR_PTR(-ENAMETOOLONG);
1033
1034         bh = ext3_find_entry(dentry, &de);
1035         inode = NULL;
1036         if (bh) {
1037                 unsigned long ino = le32_to_cpu(de->inode);
1038                 brelse (bh);
1039                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1040                         ext3_error(dir->i_sb, "ext3_lookup",
1041                                    "bad inode number: %lu", ino);
1042                         inode = NULL;
1043                 } else
1044                         inode = iget(dir->i_sb, ino);
1045
1046                 if (!inode)
1047                         return ERR_PTR(-EACCES);
1048
1049                 if (is_bad_inode(inode)) {
1050                         iput(inode);
1051                         return ERR_PTR(-ENOENT);
1052                 }
1053         }
1054         return d_splice_alias(inode, dentry);
1055 }
1056
1057
1058 struct dentry *ext3_get_parent(struct dentry *child)
1059 {
1060         unsigned long ino;
1061         struct dentry *parent;
1062         struct inode *inode;
1063         struct dentry dotdot;
1064         struct ext3_dir_entry_2 * de;
1065         struct buffer_head *bh;
1066
1067         dotdot.d_name.name = "..";
1068         dotdot.d_name.len = 2;
1069         dotdot.d_parent = child; /* confusing, isn't it! */
1070
1071         bh = ext3_find_entry(&dotdot, &de);
1072         inode = NULL;
1073         if (!bh)
1074                 return ERR_PTR(-ENOENT);
1075         ino = le32_to_cpu(de->inode);
1076         brelse(bh);
1077
1078         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1079                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1080                            "bad inode number: %lu", ino);
1081                 inode = NULL;
1082         } else
1083                 inode = iget(child->d_inode->i_sb, ino);
1084
1085         if (!inode)
1086                 return ERR_PTR(-EACCES);
1087
1088         if (is_bad_inode(inode)) {
1089                 iput(inode);
1090                 return ERR_PTR(-ENOENT);
1091         }
1092
1093         parent = d_alloc_anon(inode);
1094         if (!parent) {
1095                 iput(inode);
1096                 parent = ERR_PTR(-ENOMEM);
1097         }
1098         return parent;
1099 }
1100
1101 #define S_SHIFT 12
1102 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1103         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1104         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1105         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1106         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1107         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1108         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1109         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1110 };
1111
1112 static inline void ext3_set_de_type(struct super_block *sb,
1113                                 struct ext3_dir_entry_2 *de,
1114                                 umode_t mode) {
1115         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1116                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1117 }
1118
1119 /*
1120  * Move count entries from end of map between two memory locations.
1121  * Returns pointer to last entry moved.
1122  */
1123 static struct ext3_dir_entry_2 *
1124 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1125 {
1126         unsigned rec_len = 0;
1127
1128         while (count--) {
1129                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1130                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1131                 memcpy (to, de, rec_len);
1132                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1133                                 cpu_to_le16(rec_len);
1134                 de->inode = 0;
1135                 map++;
1136                 to += rec_len;
1137         }
1138         return (struct ext3_dir_entry_2 *) (to - rec_len);
1139 }
1140
1141 /*
1142  * Compact each dir entry in the range to the minimal rec_len.
1143  * Returns pointer to last entry in range.
1144  */
1145 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1146 {
1147         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1148         unsigned rec_len = 0;
1149
1150         prev = to = de;
1151         while ((char*)de < base + size) {
1152                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1153                                                     le16_to_cpu(de->rec_len));
1154                 if (de->inode && de->name_len) {
1155                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1156                         if (de > to)
1157                                 memmove(to, de, rec_len);
1158                         to->rec_len = cpu_to_le16(rec_len);
1159                         prev = to;
1160                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1161                 }
1162                 de = next;
1163         }
1164         return prev;
1165 }
1166
1167 /*
1168  * Split a full leaf block to make room for a new dir entry.
1169  * Allocate a new block, and move entries so that they are approx. equally full.
1170  * Returns pointer to de in block into which the new entry will be inserted.
1171  */
1172 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1173                         struct buffer_head **bh,struct dx_frame *frame,
1174                         struct dx_hash_info *hinfo, int *error)
1175 {
1176         unsigned blocksize = dir->i_sb->s_blocksize;
1177         unsigned count, continued;
1178         struct buffer_head *bh2;
1179         u32 newblock;
1180         u32 hash2;
1181         struct dx_map_entry *map;
1182         char *data1 = (*bh)->b_data, *data2;
1183         unsigned split, move, size, i;
1184         struct ext3_dir_entry_2 *de = NULL, *de2;
1185         int     err = 0;
1186
1187         bh2 = ext3_append (handle, dir, &newblock, &err);
1188         if (!(bh2)) {
1189                 brelse(*bh);
1190                 *bh = NULL;
1191                 goto errout;
1192         }
1193
1194         BUFFER_TRACE(*bh, "get_write_access");
1195         err = ext3_journal_get_write_access(handle, *bh);
1196         if (err)
1197                 goto journal_error;
1198
1199         BUFFER_TRACE(frame->bh, "get_write_access");
1200         err = ext3_journal_get_write_access(handle, frame->bh);
1201         if (err)
1202                 goto journal_error;
1203
1204         data2 = bh2->b_data;
1205
1206         /* create map in the end of data2 block */
1207         map = (struct dx_map_entry *) (data2 + blocksize);
1208         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1209                              blocksize, hinfo, map);
1210         map -= count;
1211         dx_sort_map (map, count);
1212         /* Split the existing block in the middle, size-wise */
1213         size = 0;
1214         move = 0;
1215         for (i = count-1; i >= 0; i--) {
1216                 /* is more than half of this entry in 2nd half of the block? */
1217                 if (size + map[i].size/2 > blocksize/2)
1218                         break;
1219                 size += map[i].size;
1220                 move++;
1221         }
1222         /* map index at which we will split */
1223         split = count - move;
1224         hash2 = map[split].hash;
1225         continued = hash2 == map[split - 1].hash;
1226         dxtrace(printk("Split block %i at %x, %i/%i\n",
1227                 dx_get_block(frame->at), hash2, split, count-split));
1228
1229         /* Fancy dance to stay within two buffers */
1230         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1231         de = dx_pack_dirents(data1,blocksize);
1232         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1233         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1234         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1235         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1236
1237         /* Which block gets the new entry? */
1238         if (hinfo->hash >= hash2)
1239         {
1240                 swap(*bh, bh2);
1241                 de = de2;
1242         }
1243         dx_insert_block (frame, hash2 + continued, newblock);
1244         err = ext3_journal_dirty_metadata (handle, bh2);
1245         if (err)
1246                 goto journal_error;
1247         err = ext3_journal_dirty_metadata (handle, frame->bh);
1248         if (err)
1249                 goto journal_error;
1250         brelse (bh2);
1251         dxtrace(dx_show_index ("frame", frame->entries));
1252         return de;
1253
1254 journal_error:
1255         brelse(*bh);
1256         brelse(bh2);
1257         *bh = NULL;
1258         ext3_std_error(dir->i_sb, err);
1259 errout:
1260         *error = err;
1261         return NULL;
1262 }
1263
1264
1265 /*
1266  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1267  * it points to a directory entry which is guaranteed to be large
1268  * enough for new directory entry.  If de is NULL, then
1269  * add_dirent_to_buf will attempt search the directory block for
1270  * space.  It will return -ENOSPC if no space is available, and -EIO
1271  * and -EEXIST if directory entry already exists.
1272  *
1273  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1274  * all other cases bh is released.
1275  */
1276 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1277                              struct inode *inode, struct ext3_dir_entry_2 *de,
1278                              struct buffer_head * bh)
1279 {
1280         struct inode    *dir = dentry->d_parent->d_inode;
1281         const char      *name = dentry->d_name.name;
1282         int             namelen = dentry->d_name.len;
1283         unsigned long   offset = 0;
1284         unsigned short  reclen;
1285         int             nlen, rlen, err;
1286         char            *top;
1287
1288         reclen = EXT3_DIR_REC_LEN(namelen);
1289         if (!de) {
1290                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1291                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1292                 while ((char *) de <= top) {
1293                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1294                                                   bh, offset)) {
1295                                 brelse (bh);
1296                                 return -EIO;
1297                         }
1298                         if (ext3_match (namelen, name, de)) {
1299                                 brelse (bh);
1300                                 return -EEXIST;
1301                         }
1302                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1303                         rlen = le16_to_cpu(de->rec_len);
1304                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1305                                 break;
1306                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1307                         offset += rlen;
1308                 }
1309                 if ((char *) de > top)
1310                         return -ENOSPC;
1311         }
1312         BUFFER_TRACE(bh, "get_write_access");
1313         err = ext3_journal_get_write_access(handle, bh);
1314         if (err) {
1315                 ext3_std_error(dir->i_sb, err);
1316                 brelse(bh);
1317                 return err;
1318         }
1319
1320         /* By now the buffer is marked for journaling */
1321         nlen = EXT3_DIR_REC_LEN(de->name_len);
1322         rlen = le16_to_cpu(de->rec_len);
1323         if (de->inode) {
1324                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1325                 de1->rec_len = cpu_to_le16(rlen - nlen);
1326                 de->rec_len = cpu_to_le16(nlen);
1327                 de = de1;
1328         }
1329         de->file_type = EXT3_FT_UNKNOWN;
1330         if (inode) {
1331                 de->inode = cpu_to_le32(inode->i_ino);
1332                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1333         } else
1334                 de->inode = 0;
1335         de->name_len = namelen;
1336         memcpy (de->name, name, namelen);
1337         /*
1338          * XXX shouldn't update any times until successful
1339          * completion of syscall, but too many callers depend
1340          * on this.
1341          *
1342          * XXX similarly, too many callers depend on
1343          * ext3_new_inode() setting the times, but error
1344          * recovery deletes the inode, so the worst that can
1345          * happen is that the times are slightly out of date
1346          * and/or different from the directory change time.
1347          */
1348         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1349         ext3_update_dx_flag(dir);
1350         dir->i_version++;
1351         ext3_mark_inode_dirty(handle, dir);
1352         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1353         err = ext3_journal_dirty_metadata(handle, bh);
1354         if (err)
1355                 ext3_std_error(dir->i_sb, err);
1356         brelse(bh);
1357         return 0;
1358 }
1359
1360 /*
1361  * This converts a one block unindexed directory to a 3 block indexed
1362  * directory, and adds the dentry to the indexed directory.
1363  */
1364 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1365                             struct inode *inode, struct buffer_head *bh)
1366 {
1367         struct inode    *dir = dentry->d_parent->d_inode;
1368         const char      *name = dentry->d_name.name;
1369         int             namelen = dentry->d_name.len;
1370         struct buffer_head *bh2;
1371         struct dx_root  *root;
1372         struct dx_frame frames[2], *frame;
1373         struct dx_entry *entries;
1374         struct ext3_dir_entry_2 *de, *de2;
1375         char            *data1, *top;
1376         unsigned        len;
1377         int             retval;
1378         unsigned        blocksize;
1379         struct dx_hash_info hinfo;
1380         u32             block;
1381         struct fake_dirent *fde;
1382
1383         blocksize =  dir->i_sb->s_blocksize;
1384         dxtrace(printk("Creating index\n"));
1385         retval = ext3_journal_get_write_access(handle, bh);
1386         if (retval) {
1387                 ext3_std_error(dir->i_sb, retval);
1388                 brelse(bh);
1389                 return retval;
1390         }
1391         root = (struct dx_root *) bh->b_data;
1392
1393         bh2 = ext3_append (handle, dir, &block, &retval);
1394         if (!(bh2)) {
1395                 brelse(bh);
1396                 return retval;
1397         }
1398         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1399         data1 = bh2->b_data;
1400
1401         /* The 0th block becomes the root, move the dirents out */
1402         fde = &root->dotdot;
1403         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1404         len = ((char *) root) + blocksize - (char *) de;
1405         memcpy (data1, de, len);
1406         de = (struct ext3_dir_entry_2 *) data1;
1407         top = data1 + len;
1408         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1409                 de = de2;
1410         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1411         /* Initialize the root; the dot dirents already exist */
1412         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1413         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1414         memset (&root->info, 0, sizeof(root->info));
1415         root->info.info_length = sizeof(root->info);
1416         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1417         entries = root->entries;
1418         dx_set_block (entries, 1);
1419         dx_set_count (entries, 1);
1420         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1421
1422         /* Initialize as for dx_probe */
1423         hinfo.hash_version = root->info.hash_version;
1424         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1425         ext3fs_dirhash(name, namelen, &hinfo);
1426         frame = frames;
1427         frame->entries = entries;
1428         frame->at = entries;
1429         frame->bh = bh;
1430         bh = bh2;
1431         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1432         dx_release (frames);
1433         if (!(de))
1434                 return retval;
1435
1436         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1437 }
1438
1439 /*
1440  *      ext3_add_entry()
1441  *
1442  * adds a file entry to the specified directory, using the same
1443  * semantics as ext3_find_entry(). It returns NULL if it failed.
1444  *
1445  * NOTE!! The inode part of 'de' is left at 0 - which means you
1446  * may not sleep between calling this and putting something into
1447  * the entry, as someone else might have used it while you slept.
1448  */
1449 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1450         struct inode *inode)
1451 {
1452         struct inode *dir = dentry->d_parent->d_inode;
1453         unsigned long offset;
1454         struct buffer_head * bh;
1455         struct ext3_dir_entry_2 *de;
1456         struct super_block * sb;
1457         int     retval;
1458         int     dx_fallback=0;
1459         unsigned blocksize;
1460         u32 block, blocks;
1461
1462         sb = dir->i_sb;
1463         blocksize = sb->s_blocksize;
1464         if (!dentry->d_name.len)
1465                 return -EINVAL;
1466         if (is_dx(dir)) {
1467                 retval = ext3_dx_add_entry(handle, dentry, inode);
1468                 if (!retval || (retval != ERR_BAD_DX_DIR))
1469                         return retval;
1470                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1471                 dx_fallback++;
1472                 ext3_mark_inode_dirty(handle, dir);
1473         }
1474         blocks = dir->i_size >> sb->s_blocksize_bits;
1475         for (block = 0, offset = 0; block < blocks; block++) {
1476                 bh = ext3_bread(handle, dir, block, 0, &retval);
1477                 if(!bh)
1478                         return retval;
1479                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1480                 if (retval != -ENOSPC)
1481                         return retval;
1482
1483                 if (blocks == 1 && !dx_fallback &&
1484                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1485                         return make_indexed_dir(handle, dentry, inode, bh);
1486                 brelse(bh);
1487         }
1488         bh = ext3_append(handle, dir, &block, &retval);
1489         if (!bh)
1490                 return retval;
1491         de = (struct ext3_dir_entry_2 *) bh->b_data;
1492         de->inode = 0;
1493         de->rec_len = cpu_to_le16(blocksize);
1494         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1495 }
1496
1497 /*
1498  * Returns 0 for success, or a negative error value
1499  */
1500 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1501                              struct inode *inode)
1502 {
1503         struct dx_frame frames[2], *frame;
1504         struct dx_entry *entries, *at;
1505         struct dx_hash_info hinfo;
1506         struct buffer_head * bh;
1507         struct inode *dir = dentry->d_parent->d_inode;
1508         struct super_block * sb = dir->i_sb;
1509         struct ext3_dir_entry_2 *de;
1510         int err;
1511
1512         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1513         if (!frame)
1514                 return err;
1515         entries = frame->entries;
1516         at = frame->at;
1517
1518         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1519                 goto cleanup;
1520
1521         BUFFER_TRACE(bh, "get_write_access");
1522         err = ext3_journal_get_write_access(handle, bh);
1523         if (err)
1524                 goto journal_error;
1525
1526         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1527         if (err != -ENOSPC) {
1528                 bh = NULL;
1529                 goto cleanup;
1530         }
1531
1532         /* Block full, should compress but for now just split */
1533         dxtrace(printk("using %u of %u node entries\n",
1534                        dx_get_count(entries), dx_get_limit(entries)));
1535         /* Need to split index? */
1536         if (dx_get_count(entries) == dx_get_limit(entries)) {
1537                 u32 newblock;
1538                 unsigned icount = dx_get_count(entries);
1539                 int levels = frame - frames;
1540                 struct dx_entry *entries2;
1541                 struct dx_node *node2;
1542                 struct buffer_head *bh2;
1543
1544                 if (levels && (dx_get_count(frames->entries) ==
1545                                dx_get_limit(frames->entries))) {
1546                         ext3_warning(sb, __FUNCTION__,
1547                                      "Directory index full!");
1548                         err = -ENOSPC;
1549                         goto cleanup;
1550                 }
1551                 bh2 = ext3_append (handle, dir, &newblock, &err);
1552                 if (!(bh2))
1553                         goto cleanup;
1554                 node2 = (struct dx_node *)(bh2->b_data);
1555                 entries2 = node2->entries;
1556                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1557                 node2->fake.inode = 0;
1558                 BUFFER_TRACE(frame->bh, "get_write_access");
1559                 err = ext3_journal_get_write_access(handle, frame->bh);
1560                 if (err)
1561                         goto journal_error;
1562                 if (levels) {
1563                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1564                         unsigned hash2 = dx_get_hash(entries + icount1);
1565                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1566
1567                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1568                         err = ext3_journal_get_write_access(handle,
1569                                                              frames[0].bh);
1570                         if (err)
1571                                 goto journal_error;
1572
1573                         memcpy ((char *) entries2, (char *) (entries + icount1),
1574                                 icount2 * sizeof(struct dx_entry));
1575                         dx_set_count (entries, icount1);
1576                         dx_set_count (entries2, icount2);
1577                         dx_set_limit (entries2, dx_node_limit(dir));
1578
1579                         /* Which index block gets the new entry? */
1580                         if (at - entries >= icount1) {
1581                                 frame->at = at = at - entries - icount1 + entries2;
1582                                 frame->entries = entries = entries2;
1583                                 swap(frame->bh, bh2);
1584                         }
1585                         dx_insert_block (frames + 0, hash2, newblock);
1586                         dxtrace(dx_show_index ("node", frames[1].entries));
1587                         dxtrace(dx_show_index ("node",
1588                                ((struct dx_node *) bh2->b_data)->entries));
1589                         err = ext3_journal_dirty_metadata(handle, bh2);
1590                         if (err)
1591                                 goto journal_error;
1592                         brelse (bh2);
1593                 } else {
1594                         dxtrace(printk("Creating second level index...\n"));
1595                         memcpy((char *) entries2, (char *) entries,
1596                                icount * sizeof(struct dx_entry));
1597                         dx_set_limit(entries2, dx_node_limit(dir));
1598
1599                         /* Set up root */
1600                         dx_set_count(entries, 1);
1601                         dx_set_block(entries + 0, newblock);
1602                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1603
1604                         /* Add new access path frame */
1605                         frame = frames + 1;
1606                         frame->at = at = at - entries + entries2;
1607                         frame->entries = entries = entries2;
1608                         frame->bh = bh2;
1609                         err = ext3_journal_get_write_access(handle,
1610                                                              frame->bh);
1611                         if (err)
1612                                 goto journal_error;
1613                 }
1614                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1615         }
1616         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1617         if (!de)
1618                 goto cleanup;
1619         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1620         bh = NULL;
1621         goto cleanup;
1622
1623 journal_error:
1624         ext3_std_error(dir->i_sb, err);
1625 cleanup:
1626         if (bh)
1627                 brelse(bh);
1628         dx_release(frames);
1629         return err;
1630 }
1631
1632 /*
1633  * ext3_delete_entry deletes a directory entry by merging it with the
1634  * previous entry
1635  */
1636 static int ext3_delete_entry (handle_t *handle,
1637                               struct inode * dir,
1638                               struct ext3_dir_entry_2 * de_del,
1639                               struct buffer_head * bh)
1640 {
1641         struct ext3_dir_entry_2 * de, * pde;
1642         int i;
1643
1644         i = 0;
1645         pde = NULL;
1646         de = (struct ext3_dir_entry_2 *) bh->b_data;
1647         while (i < bh->b_size) {
1648                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1649                         return -EIO;
1650                 if (de == de_del)  {
1651                         BUFFER_TRACE(bh, "get_write_access");
1652                         ext3_journal_get_write_access(handle, bh);
1653                         if (pde)
1654                                 pde->rec_len =
1655                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1656                                                     le16_to_cpu(de->rec_len));
1657                         else
1658                                 de->inode = 0;
1659                         dir->i_version++;
1660                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1661                         ext3_journal_dirty_metadata(handle, bh);
1662                         return 0;
1663                 }
1664                 i += le16_to_cpu(de->rec_len);
1665                 pde = de;
1666                 de = (struct ext3_dir_entry_2 *)
1667                         ((char *) de + le16_to_cpu(de->rec_len));
1668         }
1669         return -ENOENT;
1670 }
1671
1672 static int ext3_add_nondir(handle_t *handle,
1673                 struct dentry *dentry, struct inode *inode)
1674 {
1675         int err = ext3_add_entry(handle, dentry, inode);
1676         if (!err) {
1677                 ext3_mark_inode_dirty(handle, inode);
1678                 d_instantiate(dentry, inode);
1679                 return 0;
1680         }
1681         drop_nlink(inode);
1682         iput(inode);
1683         return err;
1684 }
1685
1686 /*
1687  * By the time this is called, we already have created
1688  * the directory cache entry for the new file, but it
1689  * is so far negative - it has no inode.
1690  *
1691  * If the create succeeds, we fill in the inode information
1692  * with d_instantiate().
1693  */
1694 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1695                 struct nameidata *nd)
1696 {
1697         handle_t *handle;
1698         struct inode * inode;
1699         int err, retries = 0;
1700
1701 retry:
1702         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1703                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1704                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1705         if (IS_ERR(handle))
1706                 return PTR_ERR(handle);
1707
1708         if (IS_DIRSYNC(dir))
1709                 handle->h_sync = 1;
1710
1711         inode = ext3_new_inode (handle, dir, mode);
1712         err = PTR_ERR(inode);
1713         if (!IS_ERR(inode)) {
1714                 inode->i_op = &ext3_file_inode_operations;
1715                 inode->i_fop = &ext3_file_operations;
1716                 ext3_set_aops(inode);
1717                 err = ext3_add_nondir(handle, dentry, inode);
1718         }
1719         ext3_journal_stop(handle);
1720         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1721                 goto retry;
1722         return err;
1723 }
1724
1725 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1726                         int mode, dev_t rdev)
1727 {
1728         handle_t *handle;
1729         struct inode *inode;
1730         int err, retries = 0;
1731
1732         if (!new_valid_dev(rdev))
1733                 return -EINVAL;
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, mode);
1746         err = PTR_ERR(inode);
1747         if (!IS_ERR(inode)) {
1748                 init_special_inode(inode, inode->i_mode, rdev);
1749 #ifdef CONFIG_EXT3_FS_XATTR
1750                 inode->i_op = &ext3_special_inode_operations;
1751 #endif
1752                 err = ext3_add_nondir(handle, dentry, inode);
1753         }
1754         ext3_journal_stop(handle);
1755         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1756                 goto retry;
1757         return err;
1758 }
1759
1760 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1761 {
1762         handle_t *handle;
1763         struct inode * inode;
1764         struct buffer_head * dir_block;
1765         struct ext3_dir_entry_2 * de;
1766         int err, retries = 0;
1767
1768         if (dir->i_nlink >= EXT3_LINK_MAX)
1769                 return -EMLINK;
1770
1771 retry:
1772         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1773                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1774                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1775         if (IS_ERR(handle))
1776                 return PTR_ERR(handle);
1777
1778         if (IS_DIRSYNC(dir))
1779                 handle->h_sync = 1;
1780
1781         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1782         err = PTR_ERR(inode);
1783         if (IS_ERR(inode))
1784                 goto out_stop;
1785
1786         inode->i_op = &ext3_dir_inode_operations;
1787         inode->i_fop = &ext3_dir_operations;
1788         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1789         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1790         if (!dir_block) {
1791                 drop_nlink(inode); /* is this nlink == 0? */
1792                 ext3_mark_inode_dirty(handle, inode);
1793                 iput (inode);
1794                 goto out_stop;
1795         }
1796         BUFFER_TRACE(dir_block, "get_write_access");
1797         ext3_journal_get_write_access(handle, dir_block);
1798         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1799         de->inode = cpu_to_le32(inode->i_ino);
1800         de->name_len = 1;
1801         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1802         strcpy (de->name, ".");
1803         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1804         de = (struct ext3_dir_entry_2 *)
1805                         ((char *) de + le16_to_cpu(de->rec_len));
1806         de->inode = cpu_to_le32(dir->i_ino);
1807         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1808         de->name_len = 2;
1809         strcpy (de->name, "..");
1810         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1811         inode->i_nlink = 2;
1812         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1813         ext3_journal_dirty_metadata(handle, dir_block);
1814         brelse (dir_block);
1815         ext3_mark_inode_dirty(handle, inode);
1816         err = ext3_add_entry (handle, dentry, inode);
1817         if (err) {
1818                 inode->i_nlink = 0;
1819                 ext3_mark_inode_dirty(handle, inode);
1820                 iput (inode);
1821                 goto out_stop;
1822         }
1823         inc_nlink(dir);
1824         ext3_update_dx_flag(dir);
1825         ext3_mark_inode_dirty(handle, dir);
1826         d_instantiate(dentry, inode);
1827 out_stop:
1828         ext3_journal_stop(handle);
1829         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1830                 goto retry;
1831         return err;
1832 }
1833
1834 /*
1835  * routine to check that the specified directory is empty (for rmdir)
1836  */
1837 static int empty_dir (struct inode * inode)
1838 {
1839         unsigned long offset;
1840         struct buffer_head * bh;
1841         struct ext3_dir_entry_2 * de, * de1;
1842         struct super_block * sb;
1843         int err = 0;
1844
1845         sb = inode->i_sb;
1846         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1847             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1848                 if (err)
1849                         ext3_error(inode->i_sb, __FUNCTION__,
1850                                    "error %d reading directory #%lu offset 0",
1851                                    err, inode->i_ino);
1852                 else
1853                         ext3_warning(inode->i_sb, __FUNCTION__,
1854                                      "bad directory (dir #%lu) - no data block",
1855                                      inode->i_ino);
1856                 return 1;
1857         }
1858         de = (struct ext3_dir_entry_2 *) bh->b_data;
1859         de1 = (struct ext3_dir_entry_2 *)
1860                         ((char *) de + le16_to_cpu(de->rec_len));
1861         if (le32_to_cpu(de->inode) != inode->i_ino ||
1862                         !le32_to_cpu(de1->inode) ||
1863                         strcmp (".", de->name) ||
1864                         strcmp ("..", de1->name)) {
1865                 ext3_warning (inode->i_sb, "empty_dir",
1866                               "bad directory (dir #%lu) - no `.' or `..'",
1867                               inode->i_ino);
1868                 brelse (bh);
1869                 return 1;
1870         }
1871         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1872         de = (struct ext3_dir_entry_2 *)
1873                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1874         while (offset < inode->i_size ) {
1875                 if (!bh ||
1876                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1877                         err = 0;
1878                         brelse (bh);
1879                         bh = ext3_bread (NULL, inode,
1880                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1881                         if (!bh) {
1882                                 if (err)
1883                                         ext3_error(sb, __FUNCTION__,
1884                                                    "error %d reading directory"
1885                                                    " #%lu offset %lu",
1886                                                    err, inode->i_ino, offset);
1887                                 offset += sb->s_blocksize;
1888                                 continue;
1889                         }
1890                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1891                 }
1892                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1893                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1894                                                          sb->s_blocksize);
1895                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1896                         continue;
1897                 }
1898                 if (le32_to_cpu(de->inode)) {
1899                         brelse (bh);
1900                         return 0;
1901                 }
1902                 offset += le16_to_cpu(de->rec_len);
1903                 de = (struct ext3_dir_entry_2 *)
1904                                 ((char *) de + le16_to_cpu(de->rec_len));
1905         }
1906         brelse (bh);
1907         return 1;
1908 }
1909
1910 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1911  * such inodes, starting at the superblock, in case we crash before the
1912  * file is closed/deleted, or in case the inode truncate spans multiple
1913  * transactions and the last transaction is not recovered after a crash.
1914  *
1915  * At filesystem recovery time, we walk this list deleting unlinked
1916  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1917  */
1918 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1919 {
1920         struct super_block *sb = inode->i_sb;
1921         struct ext3_iloc iloc;
1922         int err = 0, rc;
1923
1924         lock_super(sb);
1925         if (!list_empty(&EXT3_I(inode)->i_orphan))
1926                 goto out_unlock;
1927
1928         /* Orphan handling is only valid for files with data blocks
1929          * being truncated, or files being unlinked. */
1930
1931         /* @@@ FIXME: Observation from aviro:
1932          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1933          * here (on lock_super()), so race with ext3_link() which might bump
1934          * ->i_nlink. For, say it, character device. Not a regular file,
1935          * not a directory, not a symlink and ->i_nlink > 0.
1936          */
1937         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1938                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1939
1940         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1941         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1942         if (err)
1943                 goto out_unlock;
1944
1945         err = ext3_reserve_inode_write(handle, inode, &iloc);
1946         if (err)
1947                 goto out_unlock;
1948
1949         /* Insert this inode at the head of the on-disk orphan list... */
1950         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1951         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1952         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1953         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1954         if (!err)
1955                 err = rc;
1956
1957         /* Only add to the head of the in-memory list if all the
1958          * previous operations succeeded.  If the orphan_add is going to
1959          * fail (possibly taking the journal offline), we can't risk
1960          * leaving the inode on the orphan list: stray orphan-list
1961          * entries can cause panics at unmount time.
1962          *
1963          * This is safe: on error we're going to ignore the orphan list
1964          * anyway on the next recovery. */
1965         if (!err)
1966                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1967
1968         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1969         jbd_debug(4, "orphan inode %lu will point to %d\n",
1970                         inode->i_ino, NEXT_ORPHAN(inode));
1971 out_unlock:
1972         unlock_super(sb);
1973         ext3_std_error(inode->i_sb, err);
1974         return err;
1975 }
1976
1977 /*
1978  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1979  * of such inodes stored on disk, because it is finally being cleaned up.
1980  */
1981 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1982 {
1983         struct list_head *prev;
1984         struct ext3_inode_info *ei = EXT3_I(inode);
1985         struct ext3_sb_info *sbi;
1986         unsigned long ino_next;
1987         struct ext3_iloc iloc;
1988         int err = 0;
1989
1990         lock_super(inode->i_sb);
1991         if (list_empty(&ei->i_orphan)) {
1992                 unlock_super(inode->i_sb);
1993                 return 0;
1994         }
1995
1996         ino_next = NEXT_ORPHAN(inode);
1997         prev = ei->i_orphan.prev;
1998         sbi = EXT3_SB(inode->i_sb);
1999
2000         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2001
2002         list_del_init(&ei->i_orphan);
2003
2004         /* If we're on an error path, we may not have a valid
2005          * transaction handle with which to update the orphan list on
2006          * disk, but we still need to remove the inode from the linked
2007          * list in memory. */
2008         if (!handle)
2009                 goto out;
2010
2011         err = ext3_reserve_inode_write(handle, inode, &iloc);
2012         if (err)
2013                 goto out_err;
2014
2015         if (prev == &sbi->s_orphan) {
2016                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2017                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2018                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2019                 if (err)
2020                         goto out_brelse;
2021                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2022                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2023         } else {
2024                 struct ext3_iloc iloc2;
2025                 struct inode *i_prev =
2026                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2027
2028                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2029                           i_prev->i_ino, ino_next);
2030                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2031                 if (err)
2032                         goto out_brelse;
2033                 NEXT_ORPHAN(i_prev) = ino_next;
2034                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2035         }
2036         if (err)
2037                 goto out_brelse;
2038         NEXT_ORPHAN(inode) = 0;
2039         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2040
2041 out_err:
2042         ext3_std_error(inode->i_sb, err);
2043 out:
2044         unlock_super(inode->i_sb);
2045         return err;
2046
2047 out_brelse:
2048         brelse(iloc.bh);
2049         goto out_err;
2050 }
2051
2052 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2053 {
2054         int retval;
2055         struct inode * inode;
2056         struct buffer_head * bh;
2057         struct ext3_dir_entry_2 * de;
2058         handle_t *handle;
2059
2060         /* Initialize quotas before so that eventual writes go in
2061          * separate transaction */
2062         DQUOT_INIT(dentry->d_inode);
2063         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2064         if (IS_ERR(handle))
2065                 return PTR_ERR(handle);
2066
2067         retval = -ENOENT;
2068         bh = ext3_find_entry (dentry, &de);
2069         if (!bh)
2070                 goto end_rmdir;
2071
2072         if (IS_DIRSYNC(dir))
2073                 handle->h_sync = 1;
2074
2075         inode = dentry->d_inode;
2076
2077         retval = -EIO;
2078         if (le32_to_cpu(de->inode) != inode->i_ino)
2079                 goto end_rmdir;
2080
2081         retval = -ENOTEMPTY;
2082         if (!empty_dir (inode))
2083                 goto end_rmdir;
2084
2085         retval = ext3_delete_entry(handle, dir, de, bh);
2086         if (retval)
2087                 goto end_rmdir;
2088         if (inode->i_nlink != 2)
2089                 ext3_warning (inode->i_sb, "ext3_rmdir",
2090                               "empty directory has nlink!=2 (%d)",
2091                               inode->i_nlink);
2092         inode->i_version++;
2093         clear_nlink(inode);
2094         /* There's no need to set i_disksize: the fact that i_nlink is
2095          * zero will ensure that the right thing happens during any
2096          * recovery. */
2097         inode->i_size = 0;
2098         ext3_orphan_add(handle, inode);
2099         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2100         ext3_mark_inode_dirty(handle, inode);
2101         drop_nlink(dir);
2102         ext3_update_dx_flag(dir);
2103         ext3_mark_inode_dirty(handle, dir);
2104
2105 end_rmdir:
2106         ext3_journal_stop(handle);
2107         brelse (bh);
2108         return retval;
2109 }
2110
2111 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2112 {
2113         int retval;
2114         struct inode * inode;
2115         struct buffer_head * bh;
2116         struct ext3_dir_entry_2 * de;
2117         handle_t *handle;
2118
2119         /* Initialize quotas before so that eventual writes go
2120          * in separate transaction */
2121         DQUOT_INIT(dentry->d_inode);
2122         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2123         if (IS_ERR(handle))
2124                 return PTR_ERR(handle);
2125
2126         if (IS_DIRSYNC(dir))
2127                 handle->h_sync = 1;
2128
2129         retval = -ENOENT;
2130         bh = ext3_find_entry (dentry, &de);
2131         if (!bh)
2132                 goto end_unlink;
2133
2134         inode = dentry->d_inode;
2135
2136         retval = -EIO;
2137         if (le32_to_cpu(de->inode) != inode->i_ino)
2138                 goto end_unlink;
2139
2140         if (!inode->i_nlink) {
2141                 ext3_warning (inode->i_sb, "ext3_unlink",
2142                               "Deleting nonexistent file (%lu), %d",
2143                               inode->i_ino, inode->i_nlink);
2144                 inode->i_nlink = 1;
2145         }
2146         retval = ext3_delete_entry(handle, dir, de, bh);
2147         if (retval)
2148                 goto end_unlink;
2149         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2150         ext3_update_dx_flag(dir);
2151         ext3_mark_inode_dirty(handle, dir);
2152         drop_nlink(inode);
2153         if (!inode->i_nlink)
2154                 ext3_orphan_add(handle, inode);
2155         inode->i_ctime = dir->i_ctime;
2156         ext3_mark_inode_dirty(handle, inode);
2157         retval = 0;
2158
2159 end_unlink:
2160         ext3_journal_stop(handle);
2161         brelse (bh);
2162         return retval;
2163 }
2164
2165 static int ext3_symlink (struct inode * dir,
2166                 struct dentry *dentry, const char * symname)
2167 {
2168         handle_t *handle;
2169         struct inode * inode;
2170         int l, err, retries = 0;
2171
2172         l = strlen(symname)+1;
2173         if (l > dir->i_sb->s_blocksize)
2174                 return -ENAMETOOLONG;
2175
2176 retry:
2177         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2178                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2179                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2180         if (IS_ERR(handle))
2181                 return PTR_ERR(handle);
2182
2183         if (IS_DIRSYNC(dir))
2184                 handle->h_sync = 1;
2185
2186         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2187         err = PTR_ERR(inode);
2188         if (IS_ERR(inode))
2189                 goto out_stop;
2190
2191         if (l > sizeof (EXT3_I(inode)->i_data)) {
2192                 inode->i_op = &ext3_symlink_inode_operations;
2193                 ext3_set_aops(inode);
2194                 /*
2195                  * page_symlink() calls into ext3_prepare/commit_write.
2196                  * We have a transaction open.  All is sweetness.  It also sets
2197                  * i_size in generic_commit_write().
2198                  */
2199                 err = __page_symlink(inode, symname, l,
2200                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2201                 if (err) {
2202                         drop_nlink(inode);
2203                         ext3_mark_inode_dirty(handle, inode);
2204                         iput (inode);
2205                         goto out_stop;
2206                 }
2207         } else {
2208                 inode->i_op = &ext3_fast_symlink_inode_operations;
2209                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2210                 inode->i_size = l-1;
2211         }
2212         EXT3_I(inode)->i_disksize = inode->i_size;
2213         err = ext3_add_nondir(handle, dentry, inode);
2214 out_stop:
2215         ext3_journal_stop(handle);
2216         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2217                 goto retry;
2218         return err;
2219 }
2220
2221 static int ext3_link (struct dentry * old_dentry,
2222                 struct inode * dir, struct dentry *dentry)
2223 {
2224         handle_t *handle;
2225         struct inode *inode = old_dentry->d_inode;
2226         int err, retries = 0;
2227
2228         if (inode->i_nlink >= EXT3_LINK_MAX)
2229                 return -EMLINK;
2230         /*
2231          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2232          * otherwise has the potential to corrupt the orphan inode list.
2233          */
2234         if (inode->i_nlink == 0)
2235                 return -ENOENT;
2236
2237 retry:
2238         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2239                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2240         if (IS_ERR(handle))
2241                 return PTR_ERR(handle);
2242
2243         if (IS_DIRSYNC(dir))
2244                 handle->h_sync = 1;
2245
2246         inode->i_ctime = CURRENT_TIME_SEC;
2247         inc_nlink(inode);
2248         atomic_inc(&inode->i_count);
2249
2250         err = ext3_add_nondir(handle, dentry, inode);
2251         ext3_journal_stop(handle);
2252         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2253                 goto retry;
2254         return err;
2255 }
2256
2257 #define PARENT_INO(buffer) \
2258         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2259         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2260
2261 /*
2262  * Anybody can rename anything with this: the permission checks are left to the
2263  * higher-level routines.
2264  */
2265 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2266                            struct inode * new_dir,struct dentry *new_dentry)
2267 {
2268         handle_t *handle;
2269         struct inode * old_inode, * new_inode;
2270         struct buffer_head * old_bh, * new_bh, * dir_bh;
2271         struct ext3_dir_entry_2 * old_de, * new_de;
2272         int retval;
2273
2274         old_bh = new_bh = dir_bh = NULL;
2275
2276         /* Initialize quotas before so that eventual writes go
2277          * in separate transaction */
2278         if (new_dentry->d_inode)
2279                 DQUOT_INIT(new_dentry->d_inode);
2280         handle = ext3_journal_start(old_dir, 2 *
2281                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2282                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2283         if (IS_ERR(handle))
2284                 return PTR_ERR(handle);
2285
2286         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2287                 handle->h_sync = 1;
2288
2289         old_bh = ext3_find_entry (old_dentry, &old_de);
2290         /*
2291          *  Check for inode number is _not_ due to possible IO errors.
2292          *  We might rmdir the source, keep it as pwd of some process
2293          *  and merrily kill the link to whatever was created under the
2294          *  same name. Goodbye sticky bit ;-<
2295          */
2296         old_inode = old_dentry->d_inode;
2297         retval = -ENOENT;
2298         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2299                 goto end_rename;
2300
2301         new_inode = new_dentry->d_inode;
2302         new_bh = ext3_find_entry (new_dentry, &new_de);
2303         if (new_bh) {
2304                 if (!new_inode) {
2305                         brelse (new_bh);
2306                         new_bh = NULL;
2307                 }
2308         }
2309         if (S_ISDIR(old_inode->i_mode)) {
2310                 if (new_inode) {
2311                         retval = -ENOTEMPTY;
2312                         if (!empty_dir (new_inode))
2313                                 goto end_rename;
2314                 }
2315                 retval = -EIO;
2316                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2317                 if (!dir_bh)
2318                         goto end_rename;
2319                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2320                         goto end_rename;
2321                 retval = -EMLINK;
2322                 if (!new_inode && new_dir!=old_dir &&
2323                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2324                         goto end_rename;
2325         }
2326         if (!new_bh) {
2327                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2328                 if (retval)
2329                         goto end_rename;
2330         } else {
2331                 BUFFER_TRACE(new_bh, "get write access");
2332                 ext3_journal_get_write_access(handle, new_bh);
2333                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2334                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2335                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2336                         new_de->file_type = old_de->file_type;
2337                 new_dir->i_version++;
2338                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2339                 ext3_journal_dirty_metadata(handle, new_bh);
2340                 brelse(new_bh);
2341                 new_bh = NULL;
2342         }
2343
2344         /*
2345          * Like most other Unix systems, set the ctime for inodes on a
2346          * rename.
2347          */
2348         old_inode->i_ctime = CURRENT_TIME_SEC;
2349         ext3_mark_inode_dirty(handle, old_inode);
2350
2351         /*
2352          * ok, that's it
2353          */
2354         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2355             old_de->name_len != old_dentry->d_name.len ||
2356             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2357             (retval = ext3_delete_entry(handle, old_dir,
2358                                         old_de, old_bh)) == -ENOENT) {
2359                 /* old_de could have moved from under us during htree split, so
2360                  * make sure that we are deleting the right entry.  We might
2361                  * also be pointing to a stale entry in the unused part of
2362                  * old_bh so just checking inum and the name isn't enough. */
2363                 struct buffer_head *old_bh2;
2364                 struct ext3_dir_entry_2 *old_de2;
2365
2366                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2367                 if (old_bh2) {
2368                         retval = ext3_delete_entry(handle, old_dir,
2369                                                    old_de2, old_bh2);
2370                         brelse(old_bh2);
2371                 }
2372         }
2373         if (retval) {
2374                 ext3_warning(old_dir->i_sb, "ext3_rename",
2375                                 "Deleting old file (%lu), %d, error=%d",
2376                                 old_dir->i_ino, old_dir->i_nlink, retval);
2377         }
2378
2379         if (new_inode) {
2380                 drop_nlink(new_inode);
2381                 new_inode->i_ctime = CURRENT_TIME_SEC;
2382         }
2383         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2384         ext3_update_dx_flag(old_dir);
2385         if (dir_bh) {
2386                 BUFFER_TRACE(dir_bh, "get_write_access");
2387                 ext3_journal_get_write_access(handle, dir_bh);
2388                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2389                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2390                 ext3_journal_dirty_metadata(handle, dir_bh);
2391                 drop_nlink(old_dir);
2392                 if (new_inode) {
2393                         drop_nlink(new_inode);
2394                 } else {
2395                         inc_nlink(new_dir);
2396                         ext3_update_dx_flag(new_dir);
2397                         ext3_mark_inode_dirty(handle, new_dir);
2398                 }
2399         }
2400         ext3_mark_inode_dirty(handle, old_dir);
2401         if (new_inode) {
2402                 ext3_mark_inode_dirty(handle, new_inode);
2403                 if (!new_inode->i_nlink)
2404                         ext3_orphan_add(handle, new_inode);
2405         }
2406         retval = 0;
2407
2408 end_rename:
2409         brelse (dir_bh);
2410         brelse (old_bh);
2411         brelse (new_bh);
2412         ext3_journal_stop(handle);
2413         return retval;
2414 }
2415
2416 /*
2417  * directories can handle most operations...
2418  */
2419 const struct inode_operations ext3_dir_inode_operations = {
2420         .create         = ext3_create,
2421         .lookup         = ext3_lookup,
2422         .link           = ext3_link,
2423         .unlink         = ext3_unlink,
2424         .symlink        = ext3_symlink,
2425         .mkdir          = ext3_mkdir,
2426         .rmdir          = ext3_rmdir,
2427         .mknod          = ext3_mknod,
2428         .rename         = ext3_rename,
2429         .setattr        = ext3_setattr,
2430 #ifdef CONFIG_EXT3_FS_XATTR
2431         .setxattr       = generic_setxattr,
2432         .getxattr       = generic_getxattr,
2433         .listxattr      = ext3_listxattr,
2434         .removexattr    = generic_removexattr,
2435 #endif
2436         .permission     = ext3_permission,
2437 };
2438
2439 const struct inode_operations ext3_special_inode_operations = {
2440         .setattr        = ext3_setattr,
2441 #ifdef CONFIG_EXT3_FS_XATTR
2442         .setxattr       = generic_setxattr,
2443         .getxattr       = generic_getxattr,
2444         .listxattr      = ext3_listxattr,
2445         .removexattr    = generic_removexattr,
2446 #endif
2447         .permission     = ext3_permission,
2448 };