1dd2abe6313e3b656f42af6e2ebd63e2ec02a98c
[linux-2.6.git] / 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         bh = ext3_bread(handle, inode, *block, 1, err);
61         if (bh) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT3_I(inode)->i_disksize = inode->i_size;
64                 *err = ext3_journal_get_write_access(handle, bh);
65                 if (*err) {
66                         brelse(bh);
67                         bh = NULL;
68                 }
69         }
70         return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #ifndef swap
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79 #endif
80
81 #ifdef DX_DEBUG
82 #define dxtrace(command) command
83 #else
84 #define dxtrace(command)
85 #endif
86
87 struct fake_dirent
88 {
89         __le32 inode;
90         __le16 rec_len;
91         u8 name_len;
92         u8 file_type;
93 };
94
95 struct dx_countlimit
96 {
97         __le16 limit;
98         __le16 count;
99 };
100
101 struct dx_entry
102 {
103         __le32 hash;
104         __le32 block;
105 };
106
107 /*
108  * dx_root_info is laid out so that if it should somehow get overlaid by a
109  * dirent the two low bits of the hash version will be zero.  Therefore, the
110  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
111  */
112
113 struct dx_root
114 {
115         struct fake_dirent dot;
116         char dot_name[4];
117         struct fake_dirent dotdot;
118         char dotdot_name[4];
119         struct dx_root_info
120         {
121                 __le32 reserved_zero;
122                 u8 hash_version;
123                 u8 info_length; /* 8 */
124                 u8 indirect_levels;
125                 u8 unused_flags;
126         }
127         info;
128         struct dx_entry entries[0];
129 };
130
131 struct dx_node
132 {
133         struct fake_dirent fake;
134         struct dx_entry entries[0];
135 };
136
137
138 struct dx_frame
139 {
140         struct buffer_head *bh;
141         struct dx_entry *entries;
142         struct dx_entry *at;
143 };
144
145 struct dx_map_entry
146 {
147         u32 hash;
148         u16 offs;
149         u16 size;
150 };
151
152 static inline unsigned dx_get_block (struct dx_entry *entry);
153 static void dx_set_block (struct dx_entry *entry, unsigned value);
154 static inline unsigned dx_get_hash (struct dx_entry *entry);
155 static void dx_set_hash (struct dx_entry *entry, unsigned value);
156 static unsigned dx_get_count (struct dx_entry *entries);
157 static unsigned dx_get_limit (struct dx_entry *entries);
158 static void dx_set_count (struct dx_entry *entries, unsigned value);
159 static void dx_set_limit (struct dx_entry *entries, unsigned value);
160 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
161 static unsigned dx_node_limit (struct inode *dir);
162 static struct dx_frame *dx_probe(struct qstr *entry,
163                                  struct inode *dir,
164                                  struct dx_hash_info *hinfo,
165                                  struct dx_frame *frame,
166                                  int *err);
167 static void dx_release (struct dx_frame *frames);
168 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
169                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
170 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
171 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
172                 struct dx_map_entry *offsets, int count);
173 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
174 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
175 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
176                                  struct dx_frame *frame,
177                                  struct dx_frame *frames,
178                                  __u32 *start_hash);
179 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
180                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
181                         int *err);
182 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
183                              struct inode *inode);
184
185 /*
186  * p is at least 6 bytes before the end of page
187  */
188 static inline struct ext3_dir_entry_2 *
189 ext3_next_entry(struct ext3_dir_entry_2 *p)
190 {
191         return (struct ext3_dir_entry_2 *)((char *)p +
192                 ext3_rec_len_from_disk(p->rec_len));
193 }
194
195 /*
196  * Future: use high four bits of block for coalesce-on-delete flags
197  * Mask them off for now.
198  */
199
200 static inline unsigned dx_get_block (struct dx_entry *entry)
201 {
202         return le32_to_cpu(entry->block) & 0x00ffffff;
203 }
204
205 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
206 {
207         entry->block = cpu_to_le32(value);
208 }
209
210 static inline unsigned dx_get_hash (struct dx_entry *entry)
211 {
212         return le32_to_cpu(entry->hash);
213 }
214
215 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
216 {
217         entry->hash = cpu_to_le32(value);
218 }
219
220 static inline unsigned dx_get_count (struct dx_entry *entries)
221 {
222         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
223 }
224
225 static inline unsigned dx_get_limit (struct dx_entry *entries)
226 {
227         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
228 }
229
230 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
231 {
232         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
233 }
234
235 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
236 {
237         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
238 }
239
240 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
241 {
242         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
243                 EXT3_DIR_REC_LEN(2) - infosize;
244         return entry_space / sizeof(struct dx_entry);
245 }
246
247 static inline unsigned dx_node_limit (struct inode *dir)
248 {
249         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
250         return entry_space / sizeof(struct dx_entry);
251 }
252
253 /*
254  * Debug
255  */
256 #ifdef DX_DEBUG
257 static void dx_show_index (char * label, struct dx_entry *entries)
258 {
259         int i, n = dx_get_count (entries);
260         printk("%s index ", label);
261         for (i = 0; i < n; i++)
262         {
263                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
264         }
265         printk("\n");
266 }
267
268 struct stats
269 {
270         unsigned names;
271         unsigned space;
272         unsigned bcount;
273 };
274
275 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
276                                  int size, int show_names)
277 {
278         unsigned names = 0, space = 0;
279         char *base = (char *) de;
280         struct dx_hash_info h = *hinfo;
281
282         printk("names: ");
283         while ((char *) de < base + size)
284         {
285                 if (de->inode)
286                 {
287                         if (show_names)
288                         {
289                                 int len = de->name_len;
290                                 char *name = de->name;
291                                 while (len--) printk("%c", *name++);
292                                 ext3fs_dirhash(de->name, de->name_len, &h);
293                                 printk(":%x.%u ", h.hash,
294                                        ((char *) de - base));
295                         }
296                         space += EXT3_DIR_REC_LEN(de->name_len);
297                         names++;
298                 }
299                 de = ext3_next_entry(de);
300         }
301         printk("(%i)\n", names);
302         return (struct stats) { names, space, 1 };
303 }
304
305 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
306                              struct dx_entry *entries, int levels)
307 {
308         unsigned blocksize = dir->i_sb->s_blocksize;
309         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
310         unsigned bcount = 0;
311         struct buffer_head *bh;
312         int err;
313         printk("%i indexed blocks...\n", count);
314         for (i = 0; i < count; i++, entries++)
315         {
316                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
317                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
318                 struct stats stats;
319                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
320                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
321                 stats = levels?
322                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
323                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
324                 names += stats.names;
325                 space += stats.space;
326                 bcount += stats.bcount;
327                 brelse (bh);
328         }
329         if (bcount)
330                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
331                         names, space/bcount,(space/bcount)*100/blocksize);
332         return (struct stats) { names, space, bcount};
333 }
334 #endif /* DX_DEBUG */
335
336 /*
337  * Probe for a directory leaf block to search.
338  *
339  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340  * error in the directory index, and the caller should fall back to
341  * searching the directory normally.  The callers of dx_probe **MUST**
342  * check for this error code, and make sure it never gets reflected
343  * back to userspace.
344  */
345 static struct dx_frame *
346 dx_probe(struct qstr *entry, struct inode *dir,
347          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
348 {
349         unsigned count, indirect;
350         struct dx_entry *at, *entries, *p, *q, *m;
351         struct dx_root *root;
352         struct buffer_head *bh;
353         struct dx_frame *frame = frame_in;
354         u32 hash;
355
356         frame->bh = NULL;
357         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
358                 goto fail;
359         root = (struct dx_root *) bh->b_data;
360         if (root->info.hash_version != DX_HASH_TEA &&
361             root->info.hash_version != DX_HASH_HALF_MD4 &&
362             root->info.hash_version != DX_HASH_LEGACY) {
363                 ext3_warning(dir->i_sb, __func__,
364                              "Unrecognised inode hash code %d",
365                              root->info.hash_version);
366                 brelse(bh);
367                 *err = ERR_BAD_DX_DIR;
368                 goto fail;
369         }
370         hinfo->hash_version = root->info.hash_version;
371         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
372         if (entry)
373                 ext3fs_dirhash(entry->name, entry->len, hinfo);
374         hash = hinfo->hash;
375
376         if (root->info.unused_flags & 1) {
377                 ext3_warning(dir->i_sb, __func__,
378                              "Unimplemented inode hash flags: %#06x",
379                              root->info.unused_flags);
380                 brelse(bh);
381                 *err = ERR_BAD_DX_DIR;
382                 goto fail;
383         }
384
385         if ((indirect = root->info.indirect_levels) > 1) {
386                 ext3_warning(dir->i_sb, __func__,
387                              "Unimplemented inode hash depth: %#06x",
388                              root->info.indirect_levels);
389                 brelse(bh);
390                 *err = ERR_BAD_DX_DIR;
391                 goto fail;
392         }
393
394         entries = (struct dx_entry *) (((char *)&root->info) +
395                                        root->info.info_length);
396
397         if (dx_get_limit(entries) != dx_root_limit(dir,
398                                                    root->info.info_length)) {
399                 ext3_warning(dir->i_sb, __func__,
400                              "dx entry: limit != root limit");
401                 brelse(bh);
402                 *err = ERR_BAD_DX_DIR;
403                 goto fail;
404         }
405
406         dxtrace (printk("Look up %x", hash));
407         while (1)
408         {
409                 count = dx_get_count(entries);
410                 if (!count || count > dx_get_limit(entries)) {
411                         ext3_warning(dir->i_sb, __func__,
412                                      "dx entry: no count or count > limit");
413                         brelse(bh);
414                         *err = ERR_BAD_DX_DIR;
415                         goto fail2;
416                 }
417
418                 p = entries + 1;
419                 q = entries + count - 1;
420                 while (p <= q)
421                 {
422                         m = p + (q - p)/2;
423                         dxtrace(printk("."));
424                         if (dx_get_hash(m) > hash)
425                                 q = m - 1;
426                         else
427                                 p = m + 1;
428                 }
429
430                 if (0) // linear search cross check
431                 {
432                         unsigned n = count - 1;
433                         at = entries;
434                         while (n--)
435                         {
436                                 dxtrace(printk(","));
437                                 if (dx_get_hash(++at) > hash)
438                                 {
439                                         at--;
440                                         break;
441                                 }
442                         }
443                         assert (at == p - 1);
444                 }
445
446                 at = p - 1;
447                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
448                 frame->bh = bh;
449                 frame->entries = entries;
450                 frame->at = at;
451                 if (!indirect--) return frame;
452                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
453                         goto fail2;
454                 at = entries = ((struct dx_node *) bh->b_data)->entries;
455                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
456                         ext3_warning(dir->i_sb, __func__,
457                                      "dx entry: limit != node limit");
458                         brelse(bh);
459                         *err = ERR_BAD_DX_DIR;
460                         goto fail2;
461                 }
462                 frame++;
463                 frame->bh = NULL;
464         }
465 fail2:
466         while (frame >= frame_in) {
467                 brelse(frame->bh);
468                 frame--;
469         }
470 fail:
471         if (*err == ERR_BAD_DX_DIR)
472                 ext3_warning(dir->i_sb, __func__,
473                              "Corrupt dir inode %ld, running e2fsck is "
474                              "recommended.", dir->i_ino);
475         return NULL;
476 }
477
478 static void dx_release (struct dx_frame *frames)
479 {
480         if (frames[0].bh == NULL)
481                 return;
482
483         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
484                 brelse(frames[1].bh);
485         brelse(frames[0].bh);
486 }
487
488 /*
489  * This function increments the frame pointer to search the next leaf
490  * block, and reads in the necessary intervening nodes if the search
491  * should be necessary.  Whether or not the search is necessary is
492  * controlled by the hash parameter.  If the hash value is even, then
493  * the search is only continued if the next block starts with that
494  * hash value.  This is used if we are searching for a specific file.
495  *
496  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
497  *
498  * This function returns 1 if the caller should continue to search,
499  * or 0 if it should not.  If there is an error reading one of the
500  * index blocks, it will a negative error code.
501  *
502  * If start_hash is non-null, it will be filled in with the starting
503  * hash of the next page.
504  */
505 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
506                                  struct dx_frame *frame,
507                                  struct dx_frame *frames,
508                                  __u32 *start_hash)
509 {
510         struct dx_frame *p;
511         struct buffer_head *bh;
512         int err, num_frames = 0;
513         __u32 bhash;
514
515         p = frame;
516         /*
517          * Find the next leaf page by incrementing the frame pointer.
518          * If we run out of entries in the interior node, loop around and
519          * increment pointer in the parent node.  When we break out of
520          * this loop, num_frames indicates the number of interior
521          * nodes need to be read.
522          */
523         while (1) {
524                 if (++(p->at) < p->entries + dx_get_count(p->entries))
525                         break;
526                 if (p == frames)
527                         return 0;
528                 num_frames++;
529                 p--;
530         }
531
532         /*
533          * If the hash is 1, then continue only if the next page has a
534          * continuation hash of any value.  This is used for readdir
535          * handling.  Otherwise, check to see if the hash matches the
536          * desired contiuation hash.  If it doesn't, return since
537          * there's no point to read in the successive index pages.
538          */
539         bhash = dx_get_hash(p->at);
540         if (start_hash)
541                 *start_hash = bhash;
542         if ((hash & 1) == 0) {
543                 if ((bhash & ~1) != hash)
544                         return 0;
545         }
546         /*
547          * If the hash is HASH_NB_ALWAYS, we always go to the next
548          * block so no check is necessary
549          */
550         while (num_frames--) {
551                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
552                                       0, &err)))
553                         return err; /* Failure */
554                 p++;
555                 brelse (p->bh);
556                 p->bh = bh;
557                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
558         }
559         return 1;
560 }
561
562
563 /*
564  * This function fills a red-black tree with information from a
565  * directory block.  It returns the number directory entries loaded
566  * into the tree.  If there is an error it is returned in err.
567  */
568 static int htree_dirblock_to_tree(struct file *dir_file,
569                                   struct inode *dir, int block,
570                                   struct dx_hash_info *hinfo,
571                                   __u32 start_hash, __u32 start_minor_hash)
572 {
573         struct buffer_head *bh;
574         struct ext3_dir_entry_2 *de, *top;
575         int err, count = 0;
576
577         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
578         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
579                 return err;
580
581         de = (struct ext3_dir_entry_2 *) bh->b_data;
582         top = (struct ext3_dir_entry_2 *) ((char *) de +
583                                            dir->i_sb->s_blocksize -
584                                            EXT3_DIR_REC_LEN(0));
585         for (; de < top; de = ext3_next_entry(de)) {
586                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
587                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
588                                                 +((char *)de - bh->b_data))) {
589                         /* On error, skip the f_pos to the next block. */
590                         dir_file->f_pos = (dir_file->f_pos |
591                                         (dir->i_sb->s_blocksize - 1)) + 1;
592                         brelse (bh);
593                         return count;
594                 }
595                 ext3fs_dirhash(de->name, de->name_len, hinfo);
596                 if ((hinfo->hash < start_hash) ||
597                     ((hinfo->hash == start_hash) &&
598                      (hinfo->minor_hash < start_minor_hash)))
599                         continue;
600                 if (de->inode == 0)
601                         continue;
602                 if ((err = ext3_htree_store_dirent(dir_file,
603                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
604                         brelse(bh);
605                         return err;
606                 }
607                 count++;
608         }
609         brelse(bh);
610         return count;
611 }
612
613
614 /*
615  * This function fills a red-black tree with information from a
616  * directory.  We start scanning the directory in hash order, starting
617  * at start_hash and start_minor_hash.
618  *
619  * This function returns the number of entries inserted into the tree,
620  * or a negative error code.
621  */
622 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
623                          __u32 start_minor_hash, __u32 *next_hash)
624 {
625         struct dx_hash_info hinfo;
626         struct ext3_dir_entry_2 *de;
627         struct dx_frame frames[2], *frame;
628         struct inode *dir;
629         int block, err;
630         int count = 0;
631         int ret;
632         __u32 hashval;
633
634         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
635                        start_minor_hash));
636         dir = dir_file->f_path.dentry->d_inode;
637         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
638                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
639                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
640                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
641                                                start_hash, start_minor_hash);
642                 *next_hash = ~0;
643                 return count;
644         }
645         hinfo.hash = start_hash;
646         hinfo.minor_hash = 0;
647         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
648         if (!frame)
649                 return err;
650
651         /* Add '.' and '..' from the htree header */
652         if (!start_hash && !start_minor_hash) {
653                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
654                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
655                         goto errout;
656                 count++;
657         }
658         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
659                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
660                 de = ext3_next_entry(de);
661                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
662                         goto errout;
663                 count++;
664         }
665
666         while (1) {
667                 block = dx_get_block(frame->at);
668                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
669                                              start_hash, start_minor_hash);
670                 if (ret < 0) {
671                         err = ret;
672                         goto errout;
673                 }
674                 count += ret;
675                 hashval = ~0;
676                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
677                                             frame, frames, &hashval);
678                 *next_hash = hashval;
679                 if (ret < 0) {
680                         err = ret;
681                         goto errout;
682                 }
683                 /*
684                  * Stop if:  (a) there are no more entries, or
685                  * (b) we have inserted at least one entry and the
686                  * next hash value is not a continuation
687                  */
688                 if ((ret == 0) ||
689                     (count && ((hashval & 1) == 0)))
690                         break;
691         }
692         dx_release(frames);
693         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
694                        count, *next_hash));
695         return count;
696 errout:
697         dx_release(frames);
698         return (err);
699 }
700
701
702 /*
703  * Directory block splitting, compacting
704  */
705
706 /*
707  * Create map of hash values, offsets, and sizes, stored at end of block.
708  * Returns number of entries mapped.
709  */
710 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
711                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
712 {
713         int count = 0;
714         char *base = (char *) de;
715         struct dx_hash_info h = *hinfo;
716
717         while ((char *) de < base + size)
718         {
719                 if (de->name_len && de->inode) {
720                         ext3fs_dirhash(de->name, de->name_len, &h);
721                         map_tail--;
722                         map_tail->hash = h.hash;
723                         map_tail->offs = (u16) ((char *) de - base);
724                         map_tail->size = le16_to_cpu(de->rec_len);
725                         count++;
726                         cond_resched();
727                 }
728                 /* XXX: do we need to check rec_len == 0 case? -Chris */
729                 de = ext3_next_entry(de);
730         }
731         return count;
732 }
733
734 /* Sort map by hash value */
735 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
736 {
737         struct dx_map_entry *p, *q, *top = map + count - 1;
738         int more;
739         /* Combsort until bubble sort doesn't suck */
740         while (count > 2)
741         {
742                 count = count*10/13;
743                 if (count - 9 < 2) /* 9, 10 -> 11 */
744                         count = 11;
745                 for (p = top, q = p - count; q >= map; p--, q--)
746                         if (p->hash < q->hash)
747                                 swap(*p, *q);
748         }
749         /* Garden variety bubble sort */
750         do {
751                 more = 0;
752                 q = top;
753                 while (q-- > map)
754                 {
755                         if (q[1].hash >= q[0].hash)
756                                 continue;
757                         swap(*(q+1), *q);
758                         more = 1;
759                 }
760         } while(more);
761 }
762
763 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
764 {
765         struct dx_entry *entries = frame->entries;
766         struct dx_entry *old = frame->at, *new = old + 1;
767         int count = dx_get_count(entries);
768
769         assert(count < dx_get_limit(entries));
770         assert(old < entries + count);
771         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
772         dx_set_hash(new, hash);
773         dx_set_block(new, block);
774         dx_set_count(entries, count + 1);
775 }
776
777 static void ext3_update_dx_flag(struct inode *inode)
778 {
779         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
780                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
781                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
782 }
783
784 /*
785  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
786  *
787  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
788  * `de != NULL' is guaranteed by caller.
789  */
790 static inline int ext3_match (int len, const char * const name,
791                               struct ext3_dir_entry_2 * de)
792 {
793         if (len != de->name_len)
794                 return 0;
795         if (!de->inode)
796                 return 0;
797         return !memcmp(name, de->name, len);
798 }
799
800 /*
801  * Returns 0 if not found, -1 on failure, and 1 on success
802  */
803 static inline int search_dirblock(struct buffer_head * bh,
804                                   struct inode *dir,
805                                   struct qstr *child,
806                                   unsigned long offset,
807                                   struct ext3_dir_entry_2 ** res_dir)
808 {
809         struct ext3_dir_entry_2 * de;
810         char * dlimit;
811         int de_len;
812         const char *name = child->name;
813         int namelen = child->len;
814
815         de = (struct ext3_dir_entry_2 *) bh->b_data;
816         dlimit = bh->b_data + dir->i_sb->s_blocksize;
817         while ((char *) de < dlimit) {
818                 /* this code is executed quadratically often */
819                 /* do minimal checking `by hand' */
820
821                 if ((char *) de + namelen <= dlimit &&
822                     ext3_match (namelen, name, de)) {
823                         /* found a match - just to be sure, do a full check */
824                         if (!ext3_check_dir_entry("ext3_find_entry",
825                                                   dir, de, bh, offset))
826                                 return -1;
827                         *res_dir = de;
828                         return 1;
829                 }
830                 /* prevent looping on a bad block */
831                 de_len = ext3_rec_len_from_disk(de->rec_len);
832                 if (de_len <= 0)
833                         return -1;
834                 offset += de_len;
835                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
836         }
837         return 0;
838 }
839
840
841 /*
842  *      ext3_find_entry()
843  *
844  * finds an entry in the specified directory with the wanted name. It
845  * returns the cache buffer in which the entry was found, and the entry
846  * itself (as a parameter - res_dir). It does NOT read the inode of the
847  * entry - you'll have to do that yourself if you want to.
848  *
849  * The returned buffer_head has ->b_count elevated.  The caller is expected
850  * to brelse() it when appropriate.
851  */
852 static struct buffer_head *ext3_find_entry(struct inode *dir,
853                                         struct qstr *entry,
854                                         struct ext3_dir_entry_2 **res_dir)
855 {
856         struct super_block * sb;
857         struct buffer_head * bh_use[NAMEI_RA_SIZE];
858         struct buffer_head * bh, *ret = NULL;
859         unsigned long start, block, b;
860         int ra_max = 0;         /* Number of bh's in the readahead
861                                    buffer, bh_use[] */
862         int ra_ptr = 0;         /* Current index into readahead
863                                    buffer */
864         int num = 0;
865         int nblocks, i, err;
866         int namelen;
867
868         *res_dir = NULL;
869         sb = dir->i_sb;
870         namelen = entry->len;
871         if (namelen > EXT3_NAME_LEN)
872                 return NULL;
873         if (is_dx(dir)) {
874                 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
875                 /*
876                  * On success, or if the error was file not found,
877                  * return.  Otherwise, fall back to doing a search the
878                  * old fashioned way.
879                  */
880                 if (bh || (err != ERR_BAD_DX_DIR))
881                         return bh;
882                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
883         }
884         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
885         start = EXT3_I(dir)->i_dir_start_lookup;
886         if (start >= nblocks)
887                 start = 0;
888         block = start;
889 restart:
890         do {
891                 /*
892                  * We deal with the read-ahead logic here.
893                  */
894                 if (ra_ptr >= ra_max) {
895                         /* Refill the readahead buffer */
896                         ra_ptr = 0;
897                         b = block;
898                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
899                                 /*
900                                  * Terminate if we reach the end of the
901                                  * directory and must wrap, or if our
902                                  * search has finished at this block.
903                                  */
904                                 if (b >= nblocks || (num && block == start)) {
905                                         bh_use[ra_max] = NULL;
906                                         break;
907                                 }
908                                 num++;
909                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
910                                 bh_use[ra_max] = bh;
911                                 if (bh)
912                                         ll_rw_block(READ_META, 1, &bh);
913                         }
914                 }
915                 if ((bh = bh_use[ra_ptr++]) == NULL)
916                         goto next;
917                 wait_on_buffer(bh);
918                 if (!buffer_uptodate(bh)) {
919                         /* read error, skip block & hope for the best */
920                         ext3_error(sb, __func__, "reading directory #%lu "
921                                    "offset %lu", dir->i_ino, block);
922                         brelse(bh);
923                         goto next;
924                 }
925                 i = search_dirblock(bh, dir, entry,
926                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
927                 if (i == 1) {
928                         EXT3_I(dir)->i_dir_start_lookup = block;
929                         ret = bh;
930                         goto cleanup_and_exit;
931                 } else {
932                         brelse(bh);
933                         if (i < 0)
934                                 goto cleanup_and_exit;
935                 }
936         next:
937                 if (++block >= nblocks)
938                         block = 0;
939         } while (block != start);
940
941         /*
942          * If the directory has grown while we were searching, then
943          * search the last part of the directory before giving up.
944          */
945         block = nblocks;
946         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
947         if (block < nblocks) {
948                 start = 0;
949                 goto restart;
950         }
951
952 cleanup_and_exit:
953         /* Clean up the read-ahead blocks */
954         for (; ra_ptr < ra_max; ra_ptr++)
955                 brelse (bh_use[ra_ptr]);
956         return ret;
957 }
958
959 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
960                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
961                         int *err)
962 {
963         struct super_block * sb;
964         struct dx_hash_info     hinfo;
965         u32 hash;
966         struct dx_frame frames[2], *frame;
967         struct ext3_dir_entry_2 *de, *top;
968         struct buffer_head *bh;
969         unsigned long block;
970         int retval;
971         int namelen = entry->len;
972         const u8 *name = entry->name;
973
974         sb = dir->i_sb;
975         /* NFS may look up ".." - look at dx_root directory block */
976         if (namelen > 2 || name[0] != '.'|| (namelen == 2 && name[1] != '.')) {
977                 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
978                         return NULL;
979         } else {
980                 frame = frames;
981                 frame->bh = NULL;                       /* for dx_release() */
982                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
983                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
984         }
985         hash = hinfo.hash;
986         do {
987                 block = dx_get_block(frame->at);
988                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
989                         goto errout;
990                 de = (struct ext3_dir_entry_2 *) bh->b_data;
991                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
992                                        EXT3_DIR_REC_LEN(0));
993                 for (; de < top; de = ext3_next_entry(de)) {
994                         int off = (block << EXT3_BLOCK_SIZE_BITS(sb))
995                                   + ((char *) de - bh->b_data);
996
997                         if (!ext3_check_dir_entry(__func__, dir, de, bh, off)) {
998                                 brelse(bh);
999                                 *err = ERR_BAD_DX_DIR;
1000                                 goto errout;
1001                         }
1002
1003                         if (ext3_match(namelen, name, de)) {
1004                                 *res_dir = de;
1005                                 dx_release(frames);
1006                                 return bh;
1007                         }
1008                 }
1009                 brelse (bh);
1010                 /* Check to see if we should continue to search */
1011                 retval = ext3_htree_next_block(dir, hash, frame,
1012                                                frames, NULL);
1013                 if (retval < 0) {
1014                         ext3_warning(sb, __func__,
1015                              "error reading index page in directory #%lu",
1016                              dir->i_ino);
1017                         *err = retval;
1018                         goto errout;
1019                 }
1020         } while (retval == 1);
1021
1022         *err = -ENOENT;
1023 errout:
1024         dxtrace(printk("%s not found\n", name));
1025         dx_release (frames);
1026         return NULL;
1027 }
1028
1029 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1030 {
1031         struct inode * inode;
1032         struct ext3_dir_entry_2 * de;
1033         struct buffer_head * bh;
1034
1035         if (dentry->d_name.len > EXT3_NAME_LEN)
1036                 return ERR_PTR(-ENAMETOOLONG);
1037
1038         bh = ext3_find_entry(dir, &dentry->d_name, &de);
1039         inode = NULL;
1040         if (bh) {
1041                 unsigned long ino = le32_to_cpu(de->inode);
1042                 brelse (bh);
1043                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1044                         ext3_error(dir->i_sb, "ext3_lookup",
1045                                    "bad inode number: %lu", ino);
1046                         return ERR_PTR(-EIO);
1047                 }
1048                 inode = ext3_iget(dir->i_sb, ino);
1049                 if (IS_ERR(inode))
1050                         return ERR_CAST(inode);
1051         }
1052         return d_splice_alias(inode, dentry);
1053 }
1054
1055
1056 struct dentry *ext3_get_parent(struct dentry *child)
1057 {
1058         unsigned long ino;
1059         struct qstr dotdot = {.name = "..", .len = 2};
1060         struct ext3_dir_entry_2 * de;
1061         struct buffer_head *bh;
1062
1063         bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1064         if (!bh)
1065                 return ERR_PTR(-ENOENT);
1066         ino = le32_to_cpu(de->inode);
1067         brelse(bh);
1068
1069         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1070                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1071                            "bad inode number: %lu", ino);
1072                 return ERR_PTR(-EIO);
1073         }
1074
1075         return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1076 }
1077
1078 #define S_SHIFT 12
1079 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1080         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1081         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1082         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1083         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1084         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1085         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1086         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1087 };
1088
1089 static inline void ext3_set_de_type(struct super_block *sb,
1090                                 struct ext3_dir_entry_2 *de,
1091                                 umode_t mode) {
1092         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1093                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1094 }
1095
1096 /*
1097  * Move count entries from end of map between two memory locations.
1098  * Returns pointer to last entry moved.
1099  */
1100 static struct ext3_dir_entry_2 *
1101 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1102 {
1103         unsigned rec_len = 0;
1104
1105         while (count--) {
1106                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1107                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1108                 memcpy (to, de, rec_len);
1109                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1110                                 ext3_rec_len_to_disk(rec_len);
1111                 de->inode = 0;
1112                 map++;
1113                 to += rec_len;
1114         }
1115         return (struct ext3_dir_entry_2 *) (to - rec_len);
1116 }
1117
1118 /*
1119  * Compact each dir entry in the range to the minimal rec_len.
1120  * Returns pointer to last entry in range.
1121  */
1122 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1123 {
1124         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1125         unsigned rec_len = 0;
1126
1127         prev = to = de;
1128         while ((char*)de < base + size) {
1129                 next = ext3_next_entry(de);
1130                 if (de->inode && de->name_len) {
1131                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1132                         if (de > to)
1133                                 memmove(to, de, rec_len);
1134                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1135                         prev = to;
1136                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1137                 }
1138                 de = next;
1139         }
1140         return prev;
1141 }
1142
1143 /*
1144  * Split a full leaf block to make room for a new dir entry.
1145  * Allocate a new block, and move entries so that they are approx. equally full.
1146  * Returns pointer to de in block into which the new entry will be inserted.
1147  */
1148 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1149                         struct buffer_head **bh,struct dx_frame *frame,
1150                         struct dx_hash_info *hinfo, int *error)
1151 {
1152         unsigned blocksize = dir->i_sb->s_blocksize;
1153         unsigned count, continued;
1154         struct buffer_head *bh2;
1155         u32 newblock;
1156         u32 hash2;
1157         struct dx_map_entry *map;
1158         char *data1 = (*bh)->b_data, *data2;
1159         unsigned split, move, size, i;
1160         struct ext3_dir_entry_2 *de = NULL, *de2;
1161         int     err = 0;
1162
1163         bh2 = ext3_append (handle, dir, &newblock, &err);
1164         if (!(bh2)) {
1165                 brelse(*bh);
1166                 *bh = NULL;
1167                 goto errout;
1168         }
1169
1170         BUFFER_TRACE(*bh, "get_write_access");
1171         err = ext3_journal_get_write_access(handle, *bh);
1172         if (err)
1173                 goto journal_error;
1174
1175         BUFFER_TRACE(frame->bh, "get_write_access");
1176         err = ext3_journal_get_write_access(handle, frame->bh);
1177         if (err)
1178                 goto journal_error;
1179
1180         data2 = bh2->b_data;
1181
1182         /* create map in the end of data2 block */
1183         map = (struct dx_map_entry *) (data2 + blocksize);
1184         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1185                              blocksize, hinfo, map);
1186         map -= count;
1187         dx_sort_map (map, count);
1188         /* Split the existing block in the middle, size-wise */
1189         size = 0;
1190         move = 0;
1191         for (i = count-1; i >= 0; i--) {
1192                 /* is more than half of this entry in 2nd half of the block? */
1193                 if (size + map[i].size/2 > blocksize/2)
1194                         break;
1195                 size += map[i].size;
1196                 move++;
1197         }
1198         /* map index at which we will split */
1199         split = count - move;
1200         hash2 = map[split].hash;
1201         continued = hash2 == map[split - 1].hash;
1202         dxtrace(printk("Split block %i at %x, %i/%i\n",
1203                 dx_get_block(frame->at), hash2, split, count-split));
1204
1205         /* Fancy dance to stay within two buffers */
1206         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1207         de = dx_pack_dirents(data1,blocksize);
1208         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1209         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1210         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1211         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1212
1213         /* Which block gets the new entry? */
1214         if (hinfo->hash >= hash2)
1215         {
1216                 swap(*bh, bh2);
1217                 de = de2;
1218         }
1219         dx_insert_block (frame, hash2 + continued, newblock);
1220         err = ext3_journal_dirty_metadata (handle, bh2);
1221         if (err)
1222                 goto journal_error;
1223         err = ext3_journal_dirty_metadata (handle, frame->bh);
1224         if (err)
1225                 goto journal_error;
1226         brelse (bh2);
1227         dxtrace(dx_show_index ("frame", frame->entries));
1228         return de;
1229
1230 journal_error:
1231         brelse(*bh);
1232         brelse(bh2);
1233         *bh = NULL;
1234         ext3_std_error(dir->i_sb, err);
1235 errout:
1236         *error = err;
1237         return NULL;
1238 }
1239
1240
1241 /*
1242  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1243  * it points to a directory entry which is guaranteed to be large
1244  * enough for new directory entry.  If de is NULL, then
1245  * add_dirent_to_buf will attempt search the directory block for
1246  * space.  It will return -ENOSPC if no space is available, and -EIO
1247  * and -EEXIST if directory entry already exists.
1248  *
1249  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1250  * all other cases bh is released.
1251  */
1252 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1253                              struct inode *inode, struct ext3_dir_entry_2 *de,
1254                              struct buffer_head * bh)
1255 {
1256         struct inode    *dir = dentry->d_parent->d_inode;
1257         const char      *name = dentry->d_name.name;
1258         int             namelen = dentry->d_name.len;
1259         unsigned long   offset = 0;
1260         unsigned short  reclen;
1261         int             nlen, rlen, err;
1262         char            *top;
1263
1264         reclen = EXT3_DIR_REC_LEN(namelen);
1265         if (!de) {
1266                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1267                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1268                 while ((char *) de <= top) {
1269                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1270                                                   bh, offset)) {
1271                                 brelse (bh);
1272                                 return -EIO;
1273                         }
1274                         if (ext3_match (namelen, name, de)) {
1275                                 brelse (bh);
1276                                 return -EEXIST;
1277                         }
1278                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1279                         rlen = ext3_rec_len_from_disk(de->rec_len);
1280                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1281                                 break;
1282                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1283                         offset += rlen;
1284                 }
1285                 if ((char *) de > top)
1286                         return -ENOSPC;
1287         }
1288         BUFFER_TRACE(bh, "get_write_access");
1289         err = ext3_journal_get_write_access(handle, bh);
1290         if (err) {
1291                 ext3_std_error(dir->i_sb, err);
1292                 brelse(bh);
1293                 return err;
1294         }
1295
1296         /* By now the buffer is marked for journaling */
1297         nlen = EXT3_DIR_REC_LEN(de->name_len);
1298         rlen = ext3_rec_len_from_disk(de->rec_len);
1299         if (de->inode) {
1300                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1301                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1302                 de->rec_len = ext3_rec_len_to_disk(nlen);
1303                 de = de1;
1304         }
1305         de->file_type = EXT3_FT_UNKNOWN;
1306         if (inode) {
1307                 de->inode = cpu_to_le32(inode->i_ino);
1308                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1309         } else
1310                 de->inode = 0;
1311         de->name_len = namelen;
1312         memcpy (de->name, name, namelen);
1313         /*
1314          * XXX shouldn't update any times until successful
1315          * completion of syscall, but too many callers depend
1316          * on this.
1317          *
1318          * XXX similarly, too many callers depend on
1319          * ext3_new_inode() setting the times, but error
1320          * recovery deletes the inode, so the worst that can
1321          * happen is that the times are slightly out of date
1322          * and/or different from the directory change time.
1323          */
1324         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1325         ext3_update_dx_flag(dir);
1326         dir->i_version++;
1327         ext3_mark_inode_dirty(handle, dir);
1328         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1329         err = ext3_journal_dirty_metadata(handle, bh);
1330         if (err)
1331                 ext3_std_error(dir->i_sb, err);
1332         brelse(bh);
1333         return 0;
1334 }
1335
1336 /*
1337  * This converts a one block unindexed directory to a 3 block indexed
1338  * directory, and adds the dentry to the indexed directory.
1339  */
1340 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1341                             struct inode *inode, struct buffer_head *bh)
1342 {
1343         struct inode    *dir = dentry->d_parent->d_inode;
1344         const char      *name = dentry->d_name.name;
1345         int             namelen = dentry->d_name.len;
1346         struct buffer_head *bh2;
1347         struct dx_root  *root;
1348         struct dx_frame frames[2], *frame;
1349         struct dx_entry *entries;
1350         struct ext3_dir_entry_2 *de, *de2;
1351         char            *data1, *top;
1352         unsigned        len;
1353         int             retval;
1354         unsigned        blocksize;
1355         struct dx_hash_info hinfo;
1356         u32             block;
1357         struct fake_dirent *fde;
1358
1359         blocksize =  dir->i_sb->s_blocksize;
1360         dxtrace(printk("Creating index\n"));
1361         retval = ext3_journal_get_write_access(handle, bh);
1362         if (retval) {
1363                 ext3_std_error(dir->i_sb, retval);
1364                 brelse(bh);
1365                 return retval;
1366         }
1367         root = (struct dx_root *) bh->b_data;
1368
1369         bh2 = ext3_append (handle, dir, &block, &retval);
1370         if (!(bh2)) {
1371                 brelse(bh);
1372                 return retval;
1373         }
1374         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1375         data1 = bh2->b_data;
1376
1377         /* The 0th block becomes the root, move the dirents out */
1378         fde = &root->dotdot;
1379         de = (struct ext3_dir_entry_2 *)((char *)fde +
1380                         ext3_rec_len_from_disk(fde->rec_len));
1381         len = ((char *) root) + blocksize - (char *) de;
1382         memcpy (data1, de, len);
1383         de = (struct ext3_dir_entry_2 *) data1;
1384         top = data1 + len;
1385         while ((char *)(de2 = ext3_next_entry(de)) < top)
1386                 de = de2;
1387         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1388         /* Initialize the root; the dot dirents already exist */
1389         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1390         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1391         memset (&root->info, 0, sizeof(root->info));
1392         root->info.info_length = sizeof(root->info);
1393         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1394         entries = root->entries;
1395         dx_set_block (entries, 1);
1396         dx_set_count (entries, 1);
1397         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1398
1399         /* Initialize as for dx_probe */
1400         hinfo.hash_version = root->info.hash_version;
1401         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1402         ext3fs_dirhash(name, namelen, &hinfo);
1403         frame = frames;
1404         frame->entries = entries;
1405         frame->at = entries;
1406         frame->bh = bh;
1407         bh = bh2;
1408         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1409         dx_release (frames);
1410         if (!(de))
1411                 return retval;
1412
1413         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1414 }
1415
1416 /*
1417  *      ext3_add_entry()
1418  *
1419  * adds a file entry to the specified directory, using the same
1420  * semantics as ext3_find_entry(). It returns NULL if it failed.
1421  *
1422  * NOTE!! The inode part of 'de' is left at 0 - which means you
1423  * may not sleep between calling this and putting something into
1424  * the entry, as someone else might have used it while you slept.
1425  */
1426 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1427         struct inode *inode)
1428 {
1429         struct inode *dir = dentry->d_parent->d_inode;
1430         unsigned long offset;
1431         struct buffer_head * bh;
1432         struct ext3_dir_entry_2 *de;
1433         struct super_block * sb;
1434         int     retval;
1435         int     dx_fallback=0;
1436         unsigned blocksize;
1437         u32 block, blocks;
1438
1439         sb = dir->i_sb;
1440         blocksize = sb->s_blocksize;
1441         if (!dentry->d_name.len)
1442                 return -EINVAL;
1443         if (is_dx(dir)) {
1444                 retval = ext3_dx_add_entry(handle, dentry, inode);
1445                 if (!retval || (retval != ERR_BAD_DX_DIR))
1446                         return retval;
1447                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1448                 dx_fallback++;
1449                 ext3_mark_inode_dirty(handle, dir);
1450         }
1451         blocks = dir->i_size >> sb->s_blocksize_bits;
1452         for (block = 0, offset = 0; block < blocks; block++) {
1453                 bh = ext3_bread(handle, dir, block, 0, &retval);
1454                 if(!bh)
1455                         return retval;
1456                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1457                 if (retval != -ENOSPC)
1458                         return retval;
1459
1460                 if (blocks == 1 && !dx_fallback &&
1461                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1462                         return make_indexed_dir(handle, dentry, inode, bh);
1463                 brelse(bh);
1464         }
1465         bh = ext3_append(handle, dir, &block, &retval);
1466         if (!bh)
1467                 return retval;
1468         de = (struct ext3_dir_entry_2 *) bh->b_data;
1469         de->inode = 0;
1470         de->rec_len = ext3_rec_len_to_disk(blocksize);
1471         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1472 }
1473
1474 /*
1475  * Returns 0 for success, or a negative error value
1476  */
1477 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1478                              struct inode *inode)
1479 {
1480         struct dx_frame frames[2], *frame;
1481         struct dx_entry *entries, *at;
1482         struct dx_hash_info hinfo;
1483         struct buffer_head * bh;
1484         struct inode *dir = dentry->d_parent->d_inode;
1485         struct super_block * sb = dir->i_sb;
1486         struct ext3_dir_entry_2 *de;
1487         int err;
1488
1489         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1490         if (!frame)
1491                 return err;
1492         entries = frame->entries;
1493         at = frame->at;
1494
1495         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1496                 goto cleanup;
1497
1498         BUFFER_TRACE(bh, "get_write_access");
1499         err = ext3_journal_get_write_access(handle, bh);
1500         if (err)
1501                 goto journal_error;
1502
1503         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1504         if (err != -ENOSPC) {
1505                 bh = NULL;
1506                 goto cleanup;
1507         }
1508
1509         /* Block full, should compress but for now just split */
1510         dxtrace(printk("using %u of %u node entries\n",
1511                        dx_get_count(entries), dx_get_limit(entries)));
1512         /* Need to split index? */
1513         if (dx_get_count(entries) == dx_get_limit(entries)) {
1514                 u32 newblock;
1515                 unsigned icount = dx_get_count(entries);
1516                 int levels = frame - frames;
1517                 struct dx_entry *entries2;
1518                 struct dx_node *node2;
1519                 struct buffer_head *bh2;
1520
1521                 if (levels && (dx_get_count(frames->entries) ==
1522                                dx_get_limit(frames->entries))) {
1523                         ext3_warning(sb, __func__,
1524                                      "Directory index full!");
1525                         err = -ENOSPC;
1526                         goto cleanup;
1527                 }
1528                 bh2 = ext3_append (handle, dir, &newblock, &err);
1529                 if (!(bh2))
1530                         goto cleanup;
1531                 node2 = (struct dx_node *)(bh2->b_data);
1532                 entries2 = node2->entries;
1533                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1534                 node2->fake.inode = 0;
1535                 BUFFER_TRACE(frame->bh, "get_write_access");
1536                 err = ext3_journal_get_write_access(handle, frame->bh);
1537                 if (err)
1538                         goto journal_error;
1539                 if (levels) {
1540                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1541                         unsigned hash2 = dx_get_hash(entries + icount1);
1542                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1543
1544                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1545                         err = ext3_journal_get_write_access(handle,
1546                                                              frames[0].bh);
1547                         if (err)
1548                                 goto journal_error;
1549
1550                         memcpy ((char *) entries2, (char *) (entries + icount1),
1551                                 icount2 * sizeof(struct dx_entry));
1552                         dx_set_count (entries, icount1);
1553                         dx_set_count (entries2, icount2);
1554                         dx_set_limit (entries2, dx_node_limit(dir));
1555
1556                         /* Which index block gets the new entry? */
1557                         if (at - entries >= icount1) {
1558                                 frame->at = at = at - entries - icount1 + entries2;
1559                                 frame->entries = entries = entries2;
1560                                 swap(frame->bh, bh2);
1561                         }
1562                         dx_insert_block (frames + 0, hash2, newblock);
1563                         dxtrace(dx_show_index ("node", frames[1].entries));
1564                         dxtrace(dx_show_index ("node",
1565                                ((struct dx_node *) bh2->b_data)->entries));
1566                         err = ext3_journal_dirty_metadata(handle, bh2);
1567                         if (err)
1568                                 goto journal_error;
1569                         brelse (bh2);
1570                 } else {
1571                         dxtrace(printk("Creating second level index...\n"));
1572                         memcpy((char *) entries2, (char *) entries,
1573                                icount * sizeof(struct dx_entry));
1574                         dx_set_limit(entries2, dx_node_limit(dir));
1575
1576                         /* Set up root */
1577                         dx_set_count(entries, 1);
1578                         dx_set_block(entries + 0, newblock);
1579                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1580
1581                         /* Add new access path frame */
1582                         frame = frames + 1;
1583                         frame->at = at = at - entries + entries2;
1584                         frame->entries = entries = entries2;
1585                         frame->bh = bh2;
1586                         err = ext3_journal_get_write_access(handle,
1587                                                              frame->bh);
1588                         if (err)
1589                                 goto journal_error;
1590                 }
1591                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1592         }
1593         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1594         if (!de)
1595                 goto cleanup;
1596         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1597         bh = NULL;
1598         goto cleanup;
1599
1600 journal_error:
1601         ext3_std_error(dir->i_sb, err);
1602 cleanup:
1603         if (bh)
1604                 brelse(bh);
1605         dx_release(frames);
1606         return err;
1607 }
1608
1609 /*
1610  * ext3_delete_entry deletes a directory entry by merging it with the
1611  * previous entry
1612  */
1613 static int ext3_delete_entry (handle_t *handle,
1614                               struct inode * dir,
1615                               struct ext3_dir_entry_2 * de_del,
1616                               struct buffer_head * bh)
1617 {
1618         struct ext3_dir_entry_2 * de, * pde;
1619         int i;
1620
1621         i = 0;
1622         pde = NULL;
1623         de = (struct ext3_dir_entry_2 *) bh->b_data;
1624         while (i < bh->b_size) {
1625                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1626                         return -EIO;
1627                 if (de == de_del)  {
1628                         BUFFER_TRACE(bh, "get_write_access");
1629                         ext3_journal_get_write_access(handle, bh);
1630                         if (pde)
1631                                 pde->rec_len = ext3_rec_len_to_disk(
1632                                         ext3_rec_len_from_disk(pde->rec_len) +
1633                                         ext3_rec_len_from_disk(de->rec_len));
1634                         else
1635                                 de->inode = 0;
1636                         dir->i_version++;
1637                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1638                         ext3_journal_dirty_metadata(handle, bh);
1639                         return 0;
1640                 }
1641                 i += ext3_rec_len_from_disk(de->rec_len);
1642                 pde = de;
1643                 de = ext3_next_entry(de);
1644         }
1645         return -ENOENT;
1646 }
1647
1648 static int ext3_add_nondir(handle_t *handle,
1649                 struct dentry *dentry, struct inode *inode)
1650 {
1651         int err = ext3_add_entry(handle, dentry, inode);
1652         if (!err) {
1653                 ext3_mark_inode_dirty(handle, inode);
1654                 d_instantiate(dentry, inode);
1655                 unlock_new_inode(inode);
1656                 return 0;
1657         }
1658         drop_nlink(inode);
1659         unlock_new_inode(inode);
1660         iput(inode);
1661         return err;
1662 }
1663
1664 /*
1665  * By the time this is called, we already have created
1666  * the directory cache entry for the new file, but it
1667  * is so far negative - it has no inode.
1668  *
1669  * If the create succeeds, we fill in the inode information
1670  * with d_instantiate().
1671  */
1672 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1673                 struct nameidata *nd)
1674 {
1675         handle_t *handle;
1676         struct inode * inode;
1677         int err, retries = 0;
1678
1679 retry:
1680         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1681                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1682                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1683         if (IS_ERR(handle))
1684                 return PTR_ERR(handle);
1685
1686         if (IS_DIRSYNC(dir))
1687                 handle->h_sync = 1;
1688
1689         inode = ext3_new_inode (handle, dir, mode);
1690         err = PTR_ERR(inode);
1691         if (!IS_ERR(inode)) {
1692                 inode->i_op = &ext3_file_inode_operations;
1693                 inode->i_fop = &ext3_file_operations;
1694                 ext3_set_aops(inode);
1695                 err = ext3_add_nondir(handle, dentry, inode);
1696         }
1697         ext3_journal_stop(handle);
1698         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1699                 goto retry;
1700         return err;
1701 }
1702
1703 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1704                         int mode, dev_t rdev)
1705 {
1706         handle_t *handle;
1707         struct inode *inode;
1708         int err, retries = 0;
1709
1710         if (!new_valid_dev(rdev))
1711                 return -EINVAL;
1712
1713 retry:
1714         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1715                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1716                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1717         if (IS_ERR(handle))
1718                 return PTR_ERR(handle);
1719
1720         if (IS_DIRSYNC(dir))
1721                 handle->h_sync = 1;
1722
1723         inode = ext3_new_inode (handle, dir, mode);
1724         err = PTR_ERR(inode);
1725         if (!IS_ERR(inode)) {
1726                 init_special_inode(inode, inode->i_mode, rdev);
1727 #ifdef CONFIG_EXT3_FS_XATTR
1728                 inode->i_op = &ext3_special_inode_operations;
1729 #endif
1730                 err = ext3_add_nondir(handle, dentry, inode);
1731         }
1732         ext3_journal_stop(handle);
1733         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1734                 goto retry;
1735         return err;
1736 }
1737
1738 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1739 {
1740         handle_t *handle;
1741         struct inode * inode;
1742         struct buffer_head * dir_block;
1743         struct ext3_dir_entry_2 * de;
1744         int err, retries = 0;
1745
1746         if (dir->i_nlink >= EXT3_LINK_MAX)
1747                 return -EMLINK;
1748
1749 retry:
1750         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1751                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1752                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1753         if (IS_ERR(handle))
1754                 return PTR_ERR(handle);
1755
1756         if (IS_DIRSYNC(dir))
1757                 handle->h_sync = 1;
1758
1759         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1760         err = PTR_ERR(inode);
1761         if (IS_ERR(inode))
1762                 goto out_stop;
1763
1764         inode->i_op = &ext3_dir_inode_operations;
1765         inode->i_fop = &ext3_dir_operations;
1766         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1767         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1768         if (!dir_block) {
1769                 drop_nlink(inode); /* is this nlink == 0? */
1770                 unlock_new_inode(inode);
1771                 ext3_mark_inode_dirty(handle, inode);
1772                 iput (inode);
1773                 goto out_stop;
1774         }
1775         BUFFER_TRACE(dir_block, "get_write_access");
1776         ext3_journal_get_write_access(handle, dir_block);
1777         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1778         de->inode = cpu_to_le32(inode->i_ino);
1779         de->name_len = 1;
1780         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1781         strcpy (de->name, ".");
1782         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1783         de = ext3_next_entry(de);
1784         de->inode = cpu_to_le32(dir->i_ino);
1785         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1786                                         EXT3_DIR_REC_LEN(1));
1787         de->name_len = 2;
1788         strcpy (de->name, "..");
1789         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1790         inode->i_nlink = 2;
1791         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1792         ext3_journal_dirty_metadata(handle, dir_block);
1793         brelse (dir_block);
1794         ext3_mark_inode_dirty(handle, inode);
1795         err = ext3_add_entry (handle, dentry, inode);
1796         if (err) {
1797                 inode->i_nlink = 0;
1798                 unlock_new_inode(inode);
1799                 ext3_mark_inode_dirty(handle, inode);
1800                 iput (inode);
1801                 goto out_stop;
1802         }
1803         inc_nlink(dir);
1804         ext3_update_dx_flag(dir);
1805         ext3_mark_inode_dirty(handle, dir);
1806         d_instantiate(dentry, inode);
1807         unlock_new_inode(inode);
1808 out_stop:
1809         ext3_journal_stop(handle);
1810         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1811                 goto retry;
1812         return err;
1813 }
1814
1815 /*
1816  * routine to check that the specified directory is empty (for rmdir)
1817  */
1818 static int empty_dir (struct inode * inode)
1819 {
1820         unsigned long offset;
1821         struct buffer_head * bh;
1822         struct ext3_dir_entry_2 * de, * de1;
1823         struct super_block * sb;
1824         int err = 0;
1825
1826         sb = inode->i_sb;
1827         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1828             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1829                 if (err)
1830                         ext3_error(inode->i_sb, __func__,
1831                                    "error %d reading directory #%lu offset 0",
1832                                    err, inode->i_ino);
1833                 else
1834                         ext3_warning(inode->i_sb, __func__,
1835                                      "bad directory (dir #%lu) - no data block",
1836                                      inode->i_ino);
1837                 return 1;
1838         }
1839         de = (struct ext3_dir_entry_2 *) bh->b_data;
1840         de1 = ext3_next_entry(de);
1841         if (le32_to_cpu(de->inode) != inode->i_ino ||
1842                         !le32_to_cpu(de1->inode) ||
1843                         strcmp (".", de->name) ||
1844                         strcmp ("..", de1->name)) {
1845                 ext3_warning (inode->i_sb, "empty_dir",
1846                               "bad directory (dir #%lu) - no `.' or `..'",
1847                               inode->i_ino);
1848                 brelse (bh);
1849                 return 1;
1850         }
1851         offset = ext3_rec_len_from_disk(de->rec_len) +
1852                         ext3_rec_len_from_disk(de1->rec_len);
1853         de = ext3_next_entry(de1);
1854         while (offset < inode->i_size ) {
1855                 if (!bh ||
1856                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1857                         err = 0;
1858                         brelse (bh);
1859                         bh = ext3_bread (NULL, inode,
1860                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1861                         if (!bh) {
1862                                 if (err)
1863                                         ext3_error(sb, __func__,
1864                                                    "error %d reading directory"
1865                                                    " #%lu offset %lu",
1866                                                    err, inode->i_ino, offset);
1867                                 offset += sb->s_blocksize;
1868                                 continue;
1869                         }
1870                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1871                 }
1872                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1873                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1874                                                          sb->s_blocksize);
1875                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1876                         continue;
1877                 }
1878                 if (le32_to_cpu(de->inode)) {
1879                         brelse (bh);
1880                         return 0;
1881                 }
1882                 offset += ext3_rec_len_from_disk(de->rec_len);
1883                 de = ext3_next_entry(de);
1884         }
1885         brelse (bh);
1886         return 1;
1887 }
1888
1889 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1890  * such inodes, starting at the superblock, in case we crash before the
1891  * file is closed/deleted, or in case the inode truncate spans multiple
1892  * transactions and the last transaction is not recovered after a crash.
1893  *
1894  * At filesystem recovery time, we walk this list deleting unlinked
1895  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1896  */
1897 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1898 {
1899         struct super_block *sb = inode->i_sb;
1900         struct ext3_iloc iloc;
1901         int err = 0, rc;
1902
1903         lock_super(sb);
1904         if (!list_empty(&EXT3_I(inode)->i_orphan))
1905                 goto out_unlock;
1906
1907         /* Orphan handling is only valid for files with data blocks
1908          * being truncated, or files being unlinked. */
1909
1910         /* @@@ FIXME: Observation from aviro:
1911          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1912          * here (on lock_super()), so race with ext3_link() which might bump
1913          * ->i_nlink. For, say it, character device. Not a regular file,
1914          * not a directory, not a symlink and ->i_nlink > 0.
1915          */
1916         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1917                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1918
1919         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1920         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1921         if (err)
1922                 goto out_unlock;
1923
1924         err = ext3_reserve_inode_write(handle, inode, &iloc);
1925         if (err)
1926                 goto out_unlock;
1927
1928         /* Insert this inode at the head of the on-disk orphan list... */
1929         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1930         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1931         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1932         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1933         if (!err)
1934                 err = rc;
1935
1936         /* Only add to the head of the in-memory list if all the
1937          * previous operations succeeded.  If the orphan_add is going to
1938          * fail (possibly taking the journal offline), we can't risk
1939          * leaving the inode on the orphan list: stray orphan-list
1940          * entries can cause panics at unmount time.
1941          *
1942          * This is safe: on error we're going to ignore the orphan list
1943          * anyway on the next recovery. */
1944         if (!err)
1945                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1946
1947         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1948         jbd_debug(4, "orphan inode %lu will point to %d\n",
1949                         inode->i_ino, NEXT_ORPHAN(inode));
1950 out_unlock:
1951         unlock_super(sb);
1952         ext3_std_error(inode->i_sb, err);
1953         return err;
1954 }
1955
1956 /*
1957  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1958  * of such inodes stored on disk, because it is finally being cleaned up.
1959  */
1960 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1961 {
1962         struct list_head *prev;
1963         struct ext3_inode_info *ei = EXT3_I(inode);
1964         struct ext3_sb_info *sbi;
1965         unsigned long ino_next;
1966         struct ext3_iloc iloc;
1967         int err = 0;
1968
1969         lock_super(inode->i_sb);
1970         if (list_empty(&ei->i_orphan)) {
1971                 unlock_super(inode->i_sb);
1972                 return 0;
1973         }
1974
1975         ino_next = NEXT_ORPHAN(inode);
1976         prev = ei->i_orphan.prev;
1977         sbi = EXT3_SB(inode->i_sb);
1978
1979         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1980
1981         list_del_init(&ei->i_orphan);
1982
1983         /* If we're on an error path, we may not have a valid
1984          * transaction handle with which to update the orphan list on
1985          * disk, but we still need to remove the inode from the linked
1986          * list in memory. */
1987         if (!handle)
1988                 goto out;
1989
1990         err = ext3_reserve_inode_write(handle, inode, &iloc);
1991         if (err)
1992                 goto out_err;
1993
1994         if (prev == &sbi->s_orphan) {
1995                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1996                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1997                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1998                 if (err)
1999                         goto out_brelse;
2000                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2001                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2002         } else {
2003                 struct ext3_iloc iloc2;
2004                 struct inode *i_prev =
2005                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2006
2007                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2008                           i_prev->i_ino, ino_next);
2009                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2010                 if (err)
2011                         goto out_brelse;
2012                 NEXT_ORPHAN(i_prev) = ino_next;
2013                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2014         }
2015         if (err)
2016                 goto out_brelse;
2017         NEXT_ORPHAN(inode) = 0;
2018         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2019
2020 out_err:
2021         ext3_std_error(inode->i_sb, err);
2022 out:
2023         unlock_super(inode->i_sb);
2024         return err;
2025
2026 out_brelse:
2027         brelse(iloc.bh);
2028         goto out_err;
2029 }
2030
2031 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2032 {
2033         int retval;
2034         struct inode * inode;
2035         struct buffer_head * bh;
2036         struct ext3_dir_entry_2 * de;
2037         handle_t *handle;
2038
2039         /* Initialize quotas before so that eventual writes go in
2040          * separate transaction */
2041         DQUOT_INIT(dentry->d_inode);
2042         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2043         if (IS_ERR(handle))
2044                 return PTR_ERR(handle);
2045
2046         retval = -ENOENT;
2047         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2048         if (!bh)
2049                 goto end_rmdir;
2050
2051         if (IS_DIRSYNC(dir))
2052                 handle->h_sync = 1;
2053
2054         inode = dentry->d_inode;
2055
2056         retval = -EIO;
2057         if (le32_to_cpu(de->inode) != inode->i_ino)
2058                 goto end_rmdir;
2059
2060         retval = -ENOTEMPTY;
2061         if (!empty_dir (inode))
2062                 goto end_rmdir;
2063
2064         retval = ext3_delete_entry(handle, dir, de, bh);
2065         if (retval)
2066                 goto end_rmdir;
2067         if (inode->i_nlink != 2)
2068                 ext3_warning (inode->i_sb, "ext3_rmdir",
2069                               "empty directory has nlink!=2 (%d)",
2070                               inode->i_nlink);
2071         inode->i_version++;
2072         clear_nlink(inode);
2073         /* There's no need to set i_disksize: the fact that i_nlink is
2074          * zero will ensure that the right thing happens during any
2075          * recovery. */
2076         inode->i_size = 0;
2077         ext3_orphan_add(handle, inode);
2078         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2079         ext3_mark_inode_dirty(handle, inode);
2080         drop_nlink(dir);
2081         ext3_update_dx_flag(dir);
2082         ext3_mark_inode_dirty(handle, dir);
2083
2084 end_rmdir:
2085         ext3_journal_stop(handle);
2086         brelse (bh);
2087         return retval;
2088 }
2089
2090 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2091 {
2092         int retval;
2093         struct inode * inode;
2094         struct buffer_head * bh;
2095         struct ext3_dir_entry_2 * de;
2096         handle_t *handle;
2097
2098         /* Initialize quotas before so that eventual writes go
2099          * in separate transaction */
2100         DQUOT_INIT(dentry->d_inode);
2101         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2102         if (IS_ERR(handle))
2103                 return PTR_ERR(handle);
2104
2105         if (IS_DIRSYNC(dir))
2106                 handle->h_sync = 1;
2107
2108         retval = -ENOENT;
2109         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2110         if (!bh)
2111                 goto end_unlink;
2112
2113         inode = dentry->d_inode;
2114
2115         retval = -EIO;
2116         if (le32_to_cpu(de->inode) != inode->i_ino)
2117                 goto end_unlink;
2118
2119         if (!inode->i_nlink) {
2120                 ext3_warning (inode->i_sb, "ext3_unlink",
2121                               "Deleting nonexistent file (%lu), %d",
2122                               inode->i_ino, inode->i_nlink);
2123                 inode->i_nlink = 1;
2124         }
2125         retval = ext3_delete_entry(handle, dir, de, bh);
2126         if (retval)
2127                 goto end_unlink;
2128         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2129         ext3_update_dx_flag(dir);
2130         ext3_mark_inode_dirty(handle, dir);
2131         drop_nlink(inode);
2132         if (!inode->i_nlink)
2133                 ext3_orphan_add(handle, inode);
2134         inode->i_ctime = dir->i_ctime;
2135         ext3_mark_inode_dirty(handle, inode);
2136         retval = 0;
2137
2138 end_unlink:
2139         ext3_journal_stop(handle);
2140         brelse (bh);
2141         return retval;
2142 }
2143
2144 static int ext3_symlink (struct inode * dir,
2145                 struct dentry *dentry, const char * symname)
2146 {
2147         handle_t *handle;
2148         struct inode * inode;
2149         int l, err, retries = 0;
2150
2151         l = strlen(symname)+1;
2152         if (l > dir->i_sb->s_blocksize)
2153                 return -ENAMETOOLONG;
2154
2155 retry:
2156         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2157                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2158                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2159         if (IS_ERR(handle))
2160                 return PTR_ERR(handle);
2161
2162         if (IS_DIRSYNC(dir))
2163                 handle->h_sync = 1;
2164
2165         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2166         err = PTR_ERR(inode);
2167         if (IS_ERR(inode))
2168                 goto out_stop;
2169
2170         if (l > sizeof (EXT3_I(inode)->i_data)) {
2171                 inode->i_op = &ext3_symlink_inode_operations;
2172                 ext3_set_aops(inode);
2173                 /*
2174                  * page_symlink() calls into ext3_prepare/commit_write.
2175                  * We have a transaction open.  All is sweetness.  It also sets
2176                  * i_size in generic_commit_write().
2177                  */
2178                 err = __page_symlink(inode, symname, l, 1);
2179                 if (err) {
2180                         drop_nlink(inode);
2181                         unlock_new_inode(inode);
2182                         ext3_mark_inode_dirty(handle, inode);
2183                         iput (inode);
2184                         goto out_stop;
2185                 }
2186         } else {
2187                 inode->i_op = &ext3_fast_symlink_inode_operations;
2188                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2189                 inode->i_size = l-1;
2190         }
2191         EXT3_I(inode)->i_disksize = inode->i_size;
2192         err = ext3_add_nondir(handle, dentry, inode);
2193 out_stop:
2194         ext3_journal_stop(handle);
2195         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2196                 goto retry;
2197         return err;
2198 }
2199
2200 static int ext3_link (struct dentry * old_dentry,
2201                 struct inode * dir, struct dentry *dentry)
2202 {
2203         handle_t *handle;
2204         struct inode *inode = old_dentry->d_inode;
2205         int err, retries = 0;
2206
2207         if (inode->i_nlink >= EXT3_LINK_MAX)
2208                 return -EMLINK;
2209         /*
2210          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2211          * otherwise has the potential to corrupt the orphan inode list.
2212          */
2213         if (inode->i_nlink == 0)
2214                 return -ENOENT;
2215
2216 retry:
2217         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2218                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2219         if (IS_ERR(handle))
2220                 return PTR_ERR(handle);
2221
2222         if (IS_DIRSYNC(dir))
2223                 handle->h_sync = 1;
2224
2225         inode->i_ctime = CURRENT_TIME_SEC;
2226         inc_nlink(inode);
2227         atomic_inc(&inode->i_count);
2228
2229         err = ext3_add_entry(handle, dentry, inode);
2230         if (!err) {
2231                 ext3_mark_inode_dirty(handle, inode);
2232                 d_instantiate(dentry, inode);
2233         } else {
2234                 drop_nlink(inode);
2235                 iput(inode);
2236         }
2237         ext3_journal_stop(handle);
2238         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2239                 goto retry;
2240         return err;
2241 }
2242
2243 #define PARENT_INO(buffer) \
2244         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2245
2246 /*
2247  * Anybody can rename anything with this: the permission checks are left to the
2248  * higher-level routines.
2249  */
2250 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2251                            struct inode * new_dir,struct dentry *new_dentry)
2252 {
2253         handle_t *handle;
2254         struct inode * old_inode, * new_inode;
2255         struct buffer_head * old_bh, * new_bh, * dir_bh;
2256         struct ext3_dir_entry_2 * old_de, * new_de;
2257         int retval;
2258
2259         old_bh = new_bh = dir_bh = NULL;
2260
2261         /* Initialize quotas before so that eventual writes go
2262          * in separate transaction */
2263         if (new_dentry->d_inode)
2264                 DQUOT_INIT(new_dentry->d_inode);
2265         handle = ext3_journal_start(old_dir, 2 *
2266                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2267                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2268         if (IS_ERR(handle))
2269                 return PTR_ERR(handle);
2270
2271         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2272                 handle->h_sync = 1;
2273
2274         old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2275         /*
2276          *  Check for inode number is _not_ due to possible IO errors.
2277          *  We might rmdir the source, keep it as pwd of some process
2278          *  and merrily kill the link to whatever was created under the
2279          *  same name. Goodbye sticky bit ;-<
2280          */
2281         old_inode = old_dentry->d_inode;
2282         retval = -ENOENT;
2283         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2284                 goto end_rename;
2285
2286         new_inode = new_dentry->d_inode;
2287         new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2288         if (new_bh) {
2289                 if (!new_inode) {
2290                         brelse (new_bh);
2291                         new_bh = NULL;
2292                 }
2293         }
2294         if (S_ISDIR(old_inode->i_mode)) {
2295                 if (new_inode) {
2296                         retval = -ENOTEMPTY;
2297                         if (!empty_dir (new_inode))
2298                                 goto end_rename;
2299                 }
2300                 retval = -EIO;
2301                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2302                 if (!dir_bh)
2303                         goto end_rename;
2304                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2305                         goto end_rename;
2306                 retval = -EMLINK;
2307                 if (!new_inode && new_dir!=old_dir &&
2308                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2309                         goto end_rename;
2310         }
2311         if (!new_bh) {
2312                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2313                 if (retval)
2314                         goto end_rename;
2315         } else {
2316                 BUFFER_TRACE(new_bh, "get write access");
2317                 ext3_journal_get_write_access(handle, new_bh);
2318                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2319                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2320                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2321                         new_de->file_type = old_de->file_type;
2322                 new_dir->i_version++;
2323                 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2324                 ext3_mark_inode_dirty(handle, new_dir);
2325                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2326                 ext3_journal_dirty_metadata(handle, new_bh);
2327                 brelse(new_bh);
2328                 new_bh = NULL;
2329         }
2330
2331         /*
2332          * Like most other Unix systems, set the ctime for inodes on a
2333          * rename.
2334          */
2335         old_inode->i_ctime = CURRENT_TIME_SEC;
2336         ext3_mark_inode_dirty(handle, old_inode);
2337
2338         /*
2339          * ok, that's it
2340          */
2341         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2342             old_de->name_len != old_dentry->d_name.len ||
2343             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2344             (retval = ext3_delete_entry(handle, old_dir,
2345                                         old_de, old_bh)) == -ENOENT) {
2346                 /* old_de could have moved from under us during htree split, so
2347                  * make sure that we are deleting the right entry.  We might
2348                  * also be pointing to a stale entry in the unused part of
2349                  * old_bh so just checking inum and the name isn't enough. */
2350                 struct buffer_head *old_bh2;
2351                 struct ext3_dir_entry_2 *old_de2;
2352
2353                 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2354                                           &old_de2);
2355                 if (old_bh2) {
2356                         retval = ext3_delete_entry(handle, old_dir,
2357                                                    old_de2, old_bh2);
2358                         brelse(old_bh2);
2359                 }
2360         }
2361         if (retval) {
2362                 ext3_warning(old_dir->i_sb, "ext3_rename",
2363                                 "Deleting old file (%lu), %d, error=%d",
2364                                 old_dir->i_ino, old_dir->i_nlink, retval);
2365         }
2366
2367         if (new_inode) {
2368                 drop_nlink(new_inode);
2369                 new_inode->i_ctime = CURRENT_TIME_SEC;
2370         }
2371         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2372         ext3_update_dx_flag(old_dir);
2373         if (dir_bh) {
2374                 BUFFER_TRACE(dir_bh, "get_write_access");
2375                 ext3_journal_get_write_access(handle, dir_bh);
2376                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2377                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2378                 ext3_journal_dirty_metadata(handle, dir_bh);
2379                 drop_nlink(old_dir);
2380                 if (new_inode) {
2381                         drop_nlink(new_inode);
2382                 } else {
2383                         inc_nlink(new_dir);
2384                         ext3_update_dx_flag(new_dir);
2385                         ext3_mark_inode_dirty(handle, new_dir);
2386                 }
2387         }
2388         ext3_mark_inode_dirty(handle, old_dir);
2389         if (new_inode) {
2390                 ext3_mark_inode_dirty(handle, new_inode);
2391                 if (!new_inode->i_nlink)
2392                         ext3_orphan_add(handle, new_inode);
2393         }
2394         retval = 0;
2395
2396 end_rename:
2397         brelse (dir_bh);
2398         brelse (old_bh);
2399         brelse (new_bh);
2400         ext3_journal_stop(handle);
2401         return retval;
2402 }
2403
2404 /*
2405  * directories can handle most operations...
2406  */
2407 const struct inode_operations ext3_dir_inode_operations = {
2408         .create         = ext3_create,
2409         .lookup         = ext3_lookup,
2410         .link           = ext3_link,
2411         .unlink         = ext3_unlink,
2412         .symlink        = ext3_symlink,
2413         .mkdir          = ext3_mkdir,
2414         .rmdir          = ext3_rmdir,
2415         .mknod          = ext3_mknod,
2416         .rename         = ext3_rename,
2417         .setattr        = ext3_setattr,
2418 #ifdef CONFIG_EXT3_FS_XATTR
2419         .setxattr       = generic_setxattr,
2420         .getxattr       = generic_getxattr,
2421         .listxattr      = ext3_listxattr,
2422         .removexattr    = generic_removexattr,
2423 #endif
2424         .permission     = ext3_permission,
2425 };
2426
2427 const struct inode_operations ext3_special_inode_operations = {
2428         .setattr        = ext3_setattr,
2429 #ifdef CONFIG_EXT3_FS_XATTR
2430         .setxattr       = generic_setxattr,
2431         .getxattr       = generic_getxattr,
2432         .listxattr      = ext3_listxattr,
2433         .removexattr    = generic_removexattr,
2434 #endif
2435         .permission     = ext3_permission,
2436 };