c1fa1908dba05137158d0e9f40b991a983a4c9d8
[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         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT3_I(inode)->i_disksize = inode->i_size;
63                 ext3_journal_get_write_access(handle,bh);
64         }
65         return bh;
66 }
67
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u16 offs;
144         u16 size;
145 };
146
147 #ifdef CONFIG_EXT3_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
159                                  struct inode *dir,
160                                  struct dx_hash_info *hinfo,
161                                  struct dx_frame *frame,
162                                  int *err);
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168                 struct dx_map_entry *offsets, int count);
169 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172                                  struct dx_frame *frame,
173                                  struct dx_frame *frames,
174                                  __u32 *start_hash);
175 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176                        struct ext3_dir_entry_2 **res_dir, int *err);
177 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
178                              struct inode *inode);
179
180 /*
181  * Future: use high four bits of block for coalesce-on-delete flags
182  * Mask them off for now.
183  */
184
185 static inline unsigned dx_get_block (struct dx_entry *entry)
186 {
187         return le32_to_cpu(entry->block) & 0x00ffffff;
188 }
189
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191 {
192         entry->block = cpu_to_le32(value);
193 }
194
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
196 {
197         return le32_to_cpu(entry->hash);
198 }
199
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201 {
202         entry->hash = cpu_to_le32(value);
203 }
204
205 static inline unsigned dx_get_count (struct dx_entry *entries)
206 {
207         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208 }
209
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
211 {
212         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213 }
214
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216 {
217         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218 }
219
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221 {
222         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223 }
224
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226 {
227         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228                 EXT3_DIR_REC_LEN(2) - infosize;
229         return 0? 20: entry_space / sizeof(struct dx_entry);
230 }
231
232 static inline unsigned dx_node_limit (struct inode *dir)
233 {
234         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235         return 0? 22: entry_space / sizeof(struct dx_entry);
236 }
237
238 /*
239  * Debug
240  */
241 #ifdef DX_DEBUG
242 static void dx_show_index (char * label, struct dx_entry *entries)
243 {
244         int i, n = dx_get_count (entries);
245         printk("%s index ", label);
246         for (i = 0; i < n; i++)
247         {
248                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
249         }
250         printk("\n");
251 }
252
253 struct stats
254 {
255         unsigned names;
256         unsigned space;
257         unsigned bcount;
258 };
259
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261                                  int size, int show_names)
262 {
263         unsigned names = 0, space = 0;
264         char *base = (char *) de;
265         struct dx_hash_info h = *hinfo;
266
267         printk("names: ");
268         while ((char *) de < base + size)
269         {
270                 if (de->inode)
271                 {
272                         if (show_names)
273                         {
274                                 int len = de->name_len;
275                                 char *name = de->name;
276                                 while (len--) printk("%c", *name++);
277                                 ext3fs_dirhash(de->name, de->name_len, &h);
278                                 printk(":%x.%u ", h.hash,
279                                        ((char *) de - base));
280                         }
281                         space += EXT3_DIR_REC_LEN(de->name_len);
282                         names++;
283                 }
284                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
285         }
286         printk("(%i)\n", names);
287         return (struct stats) { names, space, 1 };
288 }
289
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291                              struct dx_entry *entries, int levels)
292 {
293         unsigned blocksize = dir->i_sb->s_blocksize;
294         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295         unsigned bcount = 0;
296         struct buffer_head *bh;
297         int err;
298         printk("%i indexed blocks...\n", count);
299         for (i = 0; i < count; i++, entries++)
300         {
301                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303                 struct stats stats;
304                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
305                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
306                 stats = levels?
307                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309                 names += stats.names;
310                 space += stats.space;
311                 bcount += stats.bcount;
312                 brelse (bh);
313         }
314         if (bcount)
315                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
316                         names, space/bcount,(space/bcount)*100/blocksize);
317         return (struct stats) { names, space, bcount};
318 }
319 #endif /* DX_DEBUG */
320
321 /*
322  * Probe for a directory leaf block to search.
323  *
324  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325  * error in the directory index, and the caller should fall back to
326  * searching the directory normally.  The callers of dx_probe **MUST**
327  * check for this error code, and make sure it never gets reflected
328  * back to userspace.
329  */
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333 {
334         unsigned count, indirect;
335         struct dx_entry *at, *entries, *p, *q, *m;
336         struct dx_root *root;
337         struct buffer_head *bh;
338         struct dx_frame *frame = frame_in;
339         u32 hash;
340
341         frame->bh = NULL;
342         if (dentry)
343                 dir = dentry->d_parent->d_inode;
344         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
345                 goto fail;
346         root = (struct dx_root *) bh->b_data;
347         if (root->info.hash_version != DX_HASH_TEA &&
348             root->info.hash_version != DX_HASH_HALF_MD4 &&
349             root->info.hash_version != DX_HASH_LEGACY) {
350                 ext3_warning(dir->i_sb, __FUNCTION__,
351                              "Unrecognised inode hash code %d",
352                              root->info.hash_version);
353                 brelse(bh);
354                 *err = ERR_BAD_DX_DIR;
355                 goto fail;
356         }
357         hinfo->hash_version = root->info.hash_version;
358         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
359         if (dentry)
360                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
361         hash = hinfo->hash;
362
363         if (root->info.unused_flags & 1) {
364                 ext3_warning(dir->i_sb, __FUNCTION__,
365                              "Unimplemented inode hash flags: %#06x",
366                              root->info.unused_flags);
367                 brelse(bh);
368                 *err = ERR_BAD_DX_DIR;
369                 goto fail;
370         }
371
372         if ((indirect = root->info.indirect_levels) > 1) {
373                 ext3_warning(dir->i_sb, __FUNCTION__,
374                              "Unimplemented inode hash depth: %#06x",
375                              root->info.indirect_levels);
376                 brelse(bh);
377                 *err = ERR_BAD_DX_DIR;
378                 goto fail;
379         }
380
381         entries = (struct dx_entry *) (((char *)&root->info) +
382                                        root->info.info_length);
383
384         if (dx_get_limit(entries) != dx_root_limit(dir,
385                                                    root->info.info_length)) {
386                 ext3_warning(dir->i_sb, __FUNCTION__,
387                              "dx entry: limit != root limit");
388                 brelse(bh);
389                 *err = ERR_BAD_DX_DIR;
390                 goto fail;
391         }
392
393         dxtrace (printk("Look up %x", hash));
394         while (1)
395         {
396                 count = dx_get_count(entries);
397                 if (!count || count > dx_get_limit(entries)) {
398                         ext3_warning(dir->i_sb, __FUNCTION__,
399                                      "dx entry: no count or count > limit");
400                         brelse(bh);
401                         *err = ERR_BAD_DX_DIR;
402                         goto fail2;
403                 }
404
405                 p = entries + 1;
406                 q = entries + count - 1;
407                 while (p <= q)
408                 {
409                         m = p + (q - p)/2;
410                         dxtrace(printk("."));
411                         if (dx_get_hash(m) > hash)
412                                 q = m - 1;
413                         else
414                                 p = m + 1;
415                 }
416
417                 if (0) // linear search cross check
418                 {
419                         unsigned n = count - 1;
420                         at = entries;
421                         while (n--)
422                         {
423                                 dxtrace(printk(","));
424                                 if (dx_get_hash(++at) > hash)
425                                 {
426                                         at--;
427                                         break;
428                                 }
429                         }
430                         assert (at == p - 1);
431                 }
432
433                 at = p - 1;
434                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
435                 frame->bh = bh;
436                 frame->entries = entries;
437                 frame->at = at;
438                 if (!indirect--) return frame;
439                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
440                         goto fail2;
441                 at = entries = ((struct dx_node *) bh->b_data)->entries;
442                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
443                         ext3_warning(dir->i_sb, __FUNCTION__,
444                                      "dx entry: limit != node limit");
445                         brelse(bh);
446                         *err = ERR_BAD_DX_DIR;
447                         goto fail2;
448                 }
449                 frame++;
450                 frame->bh = NULL;
451         }
452 fail2:
453         while (frame >= frame_in) {
454                 brelse(frame->bh);
455                 frame--;
456         }
457 fail:
458         if (*err == ERR_BAD_DX_DIR)
459                 ext3_warning(dir->i_sb, __FUNCTION__,
460                              "Corrupt dir inode %ld, running e2fsck is "
461                              "recommended.", dir->i_ino);
462         return NULL;
463 }
464
465 static void dx_release (struct dx_frame *frames)
466 {
467         if (frames[0].bh == NULL)
468                 return;
469
470         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
471                 brelse(frames[1].bh);
472         brelse(frames[0].bh);
473 }
474
475 /*
476  * This function increments the frame pointer to search the next leaf
477  * block, and reads in the necessary intervening nodes if the search
478  * should be necessary.  Whether or not the search is necessary is
479  * controlled by the hash parameter.  If the hash value is even, then
480  * the search is only continued if the next block starts with that
481  * hash value.  This is used if we are searching for a specific file.
482  *
483  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
484  *
485  * This function returns 1 if the caller should continue to search,
486  * or 0 if it should not.  If there is an error reading one of the
487  * index blocks, it will a negative error code.
488  *
489  * If start_hash is non-null, it will be filled in with the starting
490  * hash of the next page.
491  */
492 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
493                                  struct dx_frame *frame,
494                                  struct dx_frame *frames,
495                                  __u32 *start_hash)
496 {
497         struct dx_frame *p;
498         struct buffer_head *bh;
499         int err, num_frames = 0;
500         __u32 bhash;
501
502         p = frame;
503         /*
504          * Find the next leaf page by incrementing the frame pointer.
505          * If we run out of entries in the interior node, loop around and
506          * increment pointer in the parent node.  When we break out of
507          * this loop, num_frames indicates the number of interior
508          * nodes need to be read.
509          */
510         while (1) {
511                 if (++(p->at) < p->entries + dx_get_count(p->entries))
512                         break;
513                 if (p == frames)
514                         return 0;
515                 num_frames++;
516                 p--;
517         }
518
519         /*
520          * If the hash is 1, then continue only if the next page has a
521          * continuation hash of any value.  This is used for readdir
522          * handling.  Otherwise, check to see if the hash matches the
523          * desired contiuation hash.  If it doesn't, return since
524          * there's no point to read in the successive index pages.
525          */
526         bhash = dx_get_hash(p->at);
527         if (start_hash)
528                 *start_hash = bhash;
529         if ((hash & 1) == 0) {
530                 if ((bhash & ~1) != hash)
531                         return 0;
532         }
533         /*
534          * If the hash is HASH_NB_ALWAYS, we always go to the next
535          * block so no check is necessary
536          */
537         while (num_frames--) {
538                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
539                                       0, &err)))
540                         return err; /* Failure */
541                 p++;
542                 brelse (p->bh);
543                 p->bh = bh;
544                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
545         }
546         return 1;
547 }
548
549
550 /*
551  * p is at least 6 bytes before the end of page
552  */
553 static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
554 {
555         return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
556 }
557
558 /*
559  * This function fills a red-black tree with information from a
560  * directory block.  It returns the number directory entries loaded
561  * into the tree.  If there is an error it is returned in err.
562  */
563 static int htree_dirblock_to_tree(struct file *dir_file,
564                                   struct inode *dir, int block,
565                                   struct dx_hash_info *hinfo,
566                                   __u32 start_hash, __u32 start_minor_hash)
567 {
568         struct buffer_head *bh;
569         struct ext3_dir_entry_2 *de, *top;
570         int err, count = 0;
571
572         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
573         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
574                 return err;
575
576         de = (struct ext3_dir_entry_2 *) bh->b_data;
577         top = (struct ext3_dir_entry_2 *) ((char *) de +
578                                            dir->i_sb->s_blocksize -
579                                            EXT3_DIR_REC_LEN(0));
580         for (; de < top; de = ext3_next_entry(de)) {
581                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
582                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
583                                                 +((char *)de - bh->b_data))) {
584                         /* On error, skip the f_pos to the next block. */
585                         dir_file->f_pos = (dir_file->f_pos |
586                                         (dir->i_sb->s_blocksize - 1)) + 1;
587                         brelse (bh);
588                         return count;
589                 }
590                 ext3fs_dirhash(de->name, de->name_len, hinfo);
591                 if ((hinfo->hash < start_hash) ||
592                     ((hinfo->hash == start_hash) &&
593                      (hinfo->minor_hash < start_minor_hash)))
594                         continue;
595                 if (de->inode == 0)
596                         continue;
597                 if ((err = ext3_htree_store_dirent(dir_file,
598                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
599                         brelse(bh);
600                         return err;
601                 }
602                 count++;
603         }
604         brelse(bh);
605         return count;
606 }
607
608
609 /*
610  * This function fills a red-black tree with information from a
611  * directory.  We start scanning the directory in hash order, starting
612  * at start_hash and start_minor_hash.
613  *
614  * This function returns the number of entries inserted into the tree,
615  * or a negative error code.
616  */
617 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
618                          __u32 start_minor_hash, __u32 *next_hash)
619 {
620         struct dx_hash_info hinfo;
621         struct ext3_dir_entry_2 *de;
622         struct dx_frame frames[2], *frame;
623         struct inode *dir;
624         int block, err;
625         int count = 0;
626         int ret;
627         __u32 hashval;
628
629         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
630                        start_minor_hash));
631         dir = dir_file->f_path.dentry->d_inode;
632         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
633                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
634                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
635                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
636                                                start_hash, start_minor_hash);
637                 *next_hash = ~0;
638                 return count;
639         }
640         hinfo.hash = start_hash;
641         hinfo.minor_hash = 0;
642         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
643         if (!frame)
644                 return err;
645
646         /* Add '.' and '..' from the htree header */
647         if (!start_hash && !start_minor_hash) {
648                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
649                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
650                         goto errout;
651                 count++;
652         }
653         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
654                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655                 de = ext3_next_entry(de);
656                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
657                         goto errout;
658                 count++;
659         }
660
661         while (1) {
662                 block = dx_get_block(frame->at);
663                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
664                                              start_hash, start_minor_hash);
665                 if (ret < 0) {
666                         err = ret;
667                         goto errout;
668                 }
669                 count += ret;
670                 hashval = ~0;
671                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
672                                             frame, frames, &hashval);
673                 *next_hash = hashval;
674                 if (ret < 0) {
675                         err = ret;
676                         goto errout;
677                 }
678                 /*
679                  * Stop if:  (a) there are no more entries, or
680                  * (b) we have inserted at least one entry and the
681                  * next hash value is not a continuation
682                  */
683                 if ((ret == 0) ||
684                     (count && ((hashval & 1) == 0)))
685                         break;
686         }
687         dx_release(frames);
688         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
689                        count, *next_hash));
690         return count;
691 errout:
692         dx_release(frames);
693         return (err);
694 }
695
696
697 /*
698  * Directory block splitting, compacting
699  */
700
701 /*
702  * Create map of hash values, offsets, and sizes, stored at end of block.
703  * Returns number of entries mapped.
704  */
705 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
706                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
707 {
708         int count = 0;
709         char *base = (char *) de;
710         struct dx_hash_info h = *hinfo;
711
712         while ((char *) de < base + size)
713         {
714                 if (de->name_len && de->inode) {
715                         ext3fs_dirhash(de->name, de->name_len, &h);
716                         map_tail--;
717                         map_tail->hash = h.hash;
718                         map_tail->offs = (u16) ((char *) de - base);
719                         map_tail->size = le16_to_cpu(de->rec_len);
720                         count++;
721                         cond_resched();
722                 }
723                 /* XXX: do we need to check rec_len == 0 case? -Chris */
724                 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
725         }
726         return count;
727 }
728
729 /* Sort map by hash value */
730 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
731 {
732         struct dx_map_entry *p, *q, *top = map + count - 1;
733         int more;
734         /* Combsort until bubble sort doesn't suck */
735         while (count > 2)
736         {
737                 count = count*10/13;
738                 if (count - 9 < 2) /* 9, 10 -> 11 */
739                         count = 11;
740                 for (p = top, q = p - count; q >= map; p--, q--)
741                         if (p->hash < q->hash)
742                                 swap(*p, *q);
743         }
744         /* Garden variety bubble sort */
745         do {
746                 more = 0;
747                 q = top;
748                 while (q-- > map)
749                 {
750                         if (q[1].hash >= q[0].hash)
751                                 continue;
752                         swap(*(q+1), *q);
753                         more = 1;
754                 }
755         } while(more);
756 }
757
758 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
759 {
760         struct dx_entry *entries = frame->entries;
761         struct dx_entry *old = frame->at, *new = old + 1;
762         int count = dx_get_count(entries);
763
764         assert(count < dx_get_limit(entries));
765         assert(old < entries + count);
766         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
767         dx_set_hash(new, hash);
768         dx_set_block(new, block);
769         dx_set_count(entries, count + 1);
770 }
771 #endif
772
773
774 static void ext3_update_dx_flag(struct inode *inode)
775 {
776         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
777                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
778                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
779 }
780
781 /*
782  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
783  *
784  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
785  * `de != NULL' is guaranteed by caller.
786  */
787 static inline int ext3_match (int len, const char * const name,
788                               struct ext3_dir_entry_2 * de)
789 {
790         if (len != de->name_len)
791                 return 0;
792         if (!de->inode)
793                 return 0;
794         return !memcmp(name, de->name, len);
795 }
796
797 /*
798  * Returns 0 if not found, -1 on failure, and 1 on success
799  */
800 static inline int search_dirblock(struct buffer_head * bh,
801                                   struct inode *dir,
802                                   struct dentry *dentry,
803                                   unsigned long offset,
804                                   struct ext3_dir_entry_2 ** res_dir)
805 {
806         struct ext3_dir_entry_2 * de;
807         char * dlimit;
808         int de_len;
809         const char *name = dentry->d_name.name;
810         int namelen = dentry->d_name.len;
811
812         de = (struct ext3_dir_entry_2 *) bh->b_data;
813         dlimit = bh->b_data + dir->i_sb->s_blocksize;
814         while ((char *) de < dlimit) {
815                 /* this code is executed quadratically often */
816                 /* do minimal checking `by hand' */
817
818                 if ((char *) de + namelen <= dlimit &&
819                     ext3_match (namelen, name, de)) {
820                         /* found a match - just to be sure, do a full check */
821                         if (!ext3_check_dir_entry("ext3_find_entry",
822                                                   dir, de, bh, offset))
823                                 return -1;
824                         *res_dir = de;
825                         return 1;
826                 }
827                 /* prevent looping on a bad block */
828                 de_len = le16_to_cpu(de->rec_len);
829                 if (de_len <= 0)
830                         return -1;
831                 offset += de_len;
832                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
833         }
834         return 0;
835 }
836
837
838 /*
839  *      ext3_find_entry()
840  *
841  * finds an entry in the specified directory with the wanted name. It
842  * returns the cache buffer in which the entry was found, and the entry
843  * itself (as a parameter - res_dir). It does NOT read the inode of the
844  * entry - you'll have to do that yourself if you want to.
845  *
846  * The returned buffer_head has ->b_count elevated.  The caller is expected
847  * to brelse() it when appropriate.
848  */
849 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
850                                         struct ext3_dir_entry_2 ** res_dir)
851 {
852         struct super_block * sb;
853         struct buffer_head * bh_use[NAMEI_RA_SIZE];
854         struct buffer_head * bh, *ret = NULL;
855         unsigned long start, block, b;
856         int ra_max = 0;         /* Number of bh's in the readahead
857                                    buffer, bh_use[] */
858         int ra_ptr = 0;         /* Current index into readahead
859                                    buffer */
860         int num = 0;
861         int nblocks, i, err;
862         struct inode *dir = dentry->d_parent->d_inode;
863         int namelen;
864         const u8 *name;
865         unsigned blocksize;
866
867         *res_dir = NULL;
868         sb = dir->i_sb;
869         blocksize = sb->s_blocksize;
870         namelen = dentry->d_name.len;
871         name = dentry->d_name.name;
872         if (namelen > EXT3_NAME_LEN)
873                 return NULL;
874 #ifdef CONFIG_EXT3_INDEX
875         if (is_dx(dir)) {
876                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
877                 /*
878                  * On success, or if the error was file not found,
879                  * return.  Otherwise, fall back to doing a search the
880                  * old fashioned way.
881                  */
882                 if (bh || (err != ERR_BAD_DX_DIR))
883                         return bh;
884                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
885         }
886 #endif
887         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
888         start = EXT3_I(dir)->i_dir_start_lookup;
889         if (start >= nblocks)
890                 start = 0;
891         block = start;
892 restart:
893         do {
894                 /*
895                  * We deal with the read-ahead logic here.
896                  */
897                 if (ra_ptr >= ra_max) {
898                         /* Refill the readahead buffer */
899                         ra_ptr = 0;
900                         b = block;
901                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
902                                 /*
903                                  * Terminate if we reach the end of the
904                                  * directory and must wrap, or if our
905                                  * search has finished at this block.
906                                  */
907                                 if (b >= nblocks || (num && block == start)) {
908                                         bh_use[ra_max] = NULL;
909                                         break;
910                                 }
911                                 num++;
912                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
913                                 bh_use[ra_max] = bh;
914                                 if (bh)
915                                         ll_rw_block(READ_META, 1, &bh);
916                         }
917                 }
918                 if ((bh = bh_use[ra_ptr++]) == NULL)
919                         goto next;
920                 wait_on_buffer(bh);
921                 if (!buffer_uptodate(bh)) {
922                         /* read error, skip block & hope for the best */
923                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
924                                    "offset %lu", dir->i_ino, block);
925                         brelse(bh);
926                         goto next;
927                 }
928                 i = search_dirblock(bh, dir, dentry,
929                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
930                 if (i == 1) {
931                         EXT3_I(dir)->i_dir_start_lookup = block;
932                         ret = bh;
933                         goto cleanup_and_exit;
934                 } else {
935                         brelse(bh);
936                         if (i < 0)
937                                 goto cleanup_and_exit;
938                 }
939         next:
940                 if (++block >= nblocks)
941                         block = 0;
942         } while (block != start);
943
944         /*
945          * If the directory has grown while we were searching, then
946          * search the last part of the directory before giving up.
947          */
948         block = nblocks;
949         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
950         if (block < nblocks) {
951                 start = 0;
952                 goto restart;
953         }
954
955 cleanup_and_exit:
956         /* Clean up the read-ahead blocks */
957         for (; ra_ptr < ra_max; ra_ptr++)
958                 brelse (bh_use[ra_ptr]);
959         return ret;
960 }
961
962 #ifdef CONFIG_EXT3_INDEX
963 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
964                        struct ext3_dir_entry_2 **res_dir, int *err)
965 {
966         struct super_block * sb;
967         struct dx_hash_info     hinfo;
968         u32 hash;
969         struct dx_frame frames[2], *frame;
970         struct ext3_dir_entry_2 *de, *top;
971         struct buffer_head *bh;
972         unsigned long block;
973         int retval;
974         int namelen = dentry->d_name.len;
975         const u8 *name = dentry->d_name.name;
976         struct inode *dir = dentry->d_parent->d_inode;
977
978         sb = dir->i_sb;
979         /* NFS may look up ".." - look at dx_root directory block */
980         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
981                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
982                         return NULL;
983         } else {
984                 frame = frames;
985                 frame->bh = NULL;                       /* for dx_release() */
986                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
987                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
988         }
989         hash = hinfo.hash;
990         do {
991                 block = dx_get_block(frame->at);
992                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
993                         goto errout;
994                 de = (struct ext3_dir_entry_2 *) bh->b_data;
995                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
996                                        EXT3_DIR_REC_LEN(0));
997                 for (; de < top; de = ext3_next_entry(de))
998                 if (ext3_match (namelen, name, de)) {
999                         if (!ext3_check_dir_entry("ext3_find_entry",
1000                                                   dir, de, bh,
1001                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
1002                                           +((char *)de - bh->b_data))) {
1003                                 brelse (bh);
1004                                 *err = ERR_BAD_DX_DIR;
1005                                 goto errout;
1006                         }
1007                         *res_dir = de;
1008                         dx_release (frames);
1009                         return bh;
1010                 }
1011                 brelse (bh);
1012                 /* Check to see if we should continue to search */
1013                 retval = ext3_htree_next_block(dir, hash, frame,
1014                                                frames, NULL);
1015                 if (retval < 0) {
1016                         ext3_warning(sb, __FUNCTION__,
1017                              "error reading index page in directory #%lu",
1018                              dir->i_ino);
1019                         *err = retval;
1020                         goto errout;
1021                 }
1022         } while (retval == 1);
1023
1024         *err = -ENOENT;
1025 errout:
1026         dxtrace(printk("%s not found\n", name));
1027         dx_release (frames);
1028         return NULL;
1029 }
1030 #endif
1031
1032 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1033 {
1034         struct inode * inode;
1035         struct ext3_dir_entry_2 * de;
1036         struct buffer_head * bh;
1037
1038         if (dentry->d_name.len > EXT3_NAME_LEN)
1039                 return ERR_PTR(-ENAMETOOLONG);
1040
1041         bh = ext3_find_entry(dentry, &de);
1042         inode = NULL;
1043         if (bh) {
1044                 unsigned long ino = le32_to_cpu(de->inode);
1045                 brelse (bh);
1046                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1047                         ext3_error(dir->i_sb, "ext3_lookup",
1048                                    "bad inode number: %lu", ino);
1049                         inode = NULL;
1050                 } else
1051                         inode = iget(dir->i_sb, ino);
1052
1053                 if (!inode)
1054                         return ERR_PTR(-EACCES);
1055
1056                 if (is_bad_inode(inode)) {
1057                         iput(inode);
1058                         return ERR_PTR(-ENOENT);
1059                 }
1060         }
1061         return d_splice_alias(inode, dentry);
1062 }
1063
1064
1065 struct dentry *ext3_get_parent(struct dentry *child)
1066 {
1067         unsigned long ino;
1068         struct dentry *parent;
1069         struct inode *inode;
1070         struct dentry dotdot;
1071         struct ext3_dir_entry_2 * de;
1072         struct buffer_head *bh;
1073
1074         dotdot.d_name.name = "..";
1075         dotdot.d_name.len = 2;
1076         dotdot.d_parent = child; /* confusing, isn't it! */
1077
1078         bh = ext3_find_entry(&dotdot, &de);
1079         inode = NULL;
1080         if (!bh)
1081                 return ERR_PTR(-ENOENT);
1082         ino = le32_to_cpu(de->inode);
1083         brelse(bh);
1084
1085         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1086                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1087                            "bad inode number: %lu", ino);
1088                 inode = NULL;
1089         } else
1090                 inode = iget(child->d_inode->i_sb, ino);
1091
1092         if (!inode)
1093                 return ERR_PTR(-EACCES);
1094
1095         if (is_bad_inode(inode)) {
1096                 iput(inode);
1097                 return ERR_PTR(-ENOENT);
1098         }
1099
1100         parent = d_alloc_anon(inode);
1101         if (!parent) {
1102                 iput(inode);
1103                 parent = ERR_PTR(-ENOMEM);
1104         }
1105         return parent;
1106 }
1107
1108 #define S_SHIFT 12
1109 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1110         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1111         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1112         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1113         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1114         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1115         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1116         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1117 };
1118
1119 static inline void ext3_set_de_type(struct super_block *sb,
1120                                 struct ext3_dir_entry_2 *de,
1121                                 umode_t mode) {
1122         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1123                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1124 }
1125
1126 #ifdef CONFIG_EXT3_INDEX
1127 /*
1128  * Move count entries from end of map between two memory locations.
1129  * Returns pointer to last entry moved.
1130  */
1131 static struct ext3_dir_entry_2 *
1132 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1133 {
1134         unsigned rec_len = 0;
1135
1136         while (count--) {
1137                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1138                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1139                 memcpy (to, de, rec_len);
1140                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1141                                 cpu_to_le16(rec_len);
1142                 de->inode = 0;
1143                 map++;
1144                 to += rec_len;
1145         }
1146         return (struct ext3_dir_entry_2 *) (to - rec_len);
1147 }
1148
1149 /*
1150  * Compact each dir entry in the range to the minimal rec_len.
1151  * Returns pointer to last entry in range.
1152  */
1153 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1154 {
1155         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1156         unsigned rec_len = 0;
1157
1158         prev = to = de;
1159         while ((char*)de < base + size) {
1160                 next = (struct ext3_dir_entry_2 *) ((char *) de +
1161                                                     le16_to_cpu(de->rec_len));
1162                 if (de->inode && de->name_len) {
1163                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1164                         if (de > to)
1165                                 memmove(to, de, rec_len);
1166                         to->rec_len = cpu_to_le16(rec_len);
1167                         prev = to;
1168                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1169                 }
1170                 de = next;
1171         }
1172         return prev;
1173 }
1174
1175 /*
1176  * Split a full leaf block to make room for a new dir entry.
1177  * Allocate a new block, and move entries so that they are approx. equally full.
1178  * Returns pointer to de in block into which the new entry will be inserted.
1179  */
1180 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1181                         struct buffer_head **bh,struct dx_frame *frame,
1182                         struct dx_hash_info *hinfo, int *error)
1183 {
1184         unsigned blocksize = dir->i_sb->s_blocksize;
1185         unsigned count, continued;
1186         struct buffer_head *bh2;
1187         u32 newblock;
1188         u32 hash2;
1189         struct dx_map_entry *map;
1190         char *data1 = (*bh)->b_data, *data2;
1191         unsigned split, move, size, i;
1192         struct ext3_dir_entry_2 *de = NULL, *de2;
1193         int     err = 0;
1194
1195         bh2 = ext3_append (handle, dir, &newblock, &err);
1196         if (!(bh2)) {
1197                 brelse(*bh);
1198                 *bh = NULL;
1199                 goto errout;
1200         }
1201
1202         BUFFER_TRACE(*bh, "get_write_access");
1203         err = ext3_journal_get_write_access(handle, *bh);
1204         if (err)
1205                 goto journal_error;
1206
1207         BUFFER_TRACE(frame->bh, "get_write_access");
1208         err = ext3_journal_get_write_access(handle, frame->bh);
1209         if (err)
1210                 goto journal_error;
1211
1212         data2 = bh2->b_data;
1213
1214         /* create map in the end of data2 block */
1215         map = (struct dx_map_entry *) (data2 + blocksize);
1216         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1217                              blocksize, hinfo, map);
1218         map -= count;
1219         dx_sort_map (map, count);
1220         /* Split the existing block in the middle, size-wise */
1221         size = 0;
1222         move = 0;
1223         for (i = count-1; i >= 0; i--) {
1224                 /* is more than half of this entry in 2nd half of the block? */
1225                 if (size + map[i].size/2 > blocksize/2)
1226                         break;
1227                 size += map[i].size;
1228                 move++;
1229         }
1230         /* map index at which we will split */
1231         split = count - move;
1232         hash2 = map[split].hash;
1233         continued = hash2 == map[split - 1].hash;
1234         dxtrace(printk("Split block %i at %x, %i/%i\n",
1235                 dx_get_block(frame->at), hash2, split, count-split));
1236
1237         /* Fancy dance to stay within two buffers */
1238         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1239         de = dx_pack_dirents(data1,blocksize);
1240         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1241         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1242         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1243         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1244
1245         /* Which block gets the new entry? */
1246         if (hinfo->hash >= hash2)
1247         {
1248                 swap(*bh, bh2);
1249                 de = de2;
1250         }
1251         dx_insert_block (frame, hash2 + continued, newblock);
1252         err = ext3_journal_dirty_metadata (handle, bh2);
1253         if (err)
1254                 goto journal_error;
1255         err = ext3_journal_dirty_metadata (handle, frame->bh);
1256         if (err)
1257                 goto journal_error;
1258         brelse (bh2);
1259         dxtrace(dx_show_index ("frame", frame->entries));
1260         return de;
1261
1262 journal_error:
1263         brelse(*bh);
1264         brelse(bh2);
1265         *bh = NULL;
1266         ext3_std_error(dir->i_sb, err);
1267 errout:
1268         *error = err;
1269         return NULL;
1270 }
1271 #endif
1272
1273
1274 /*
1275  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1276  * it points to a directory entry which is guaranteed to be large
1277  * enough for new directory entry.  If de is NULL, then
1278  * add_dirent_to_buf will attempt search the directory block for
1279  * space.  It will return -ENOSPC if no space is available, and -EIO
1280  * and -EEXIST if directory entry already exists.
1281  *
1282  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1283  * all other cases bh is released.
1284  */
1285 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1286                              struct inode *inode, struct ext3_dir_entry_2 *de,
1287                              struct buffer_head * bh)
1288 {
1289         struct inode    *dir = dentry->d_parent->d_inode;
1290         const char      *name = dentry->d_name.name;
1291         int             namelen = dentry->d_name.len;
1292         unsigned long   offset = 0;
1293         unsigned short  reclen;
1294         int             nlen, rlen, err;
1295         char            *top;
1296
1297         reclen = EXT3_DIR_REC_LEN(namelen);
1298         if (!de) {
1299                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1300                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1301                 while ((char *) de <= top) {
1302                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1303                                                   bh, offset)) {
1304                                 brelse (bh);
1305                                 return -EIO;
1306                         }
1307                         if (ext3_match (namelen, name, de)) {
1308                                 brelse (bh);
1309                                 return -EEXIST;
1310                         }
1311                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1312                         rlen = le16_to_cpu(de->rec_len);
1313                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1314                                 break;
1315                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1316                         offset += rlen;
1317                 }
1318                 if ((char *) de > top)
1319                         return -ENOSPC;
1320         }
1321         BUFFER_TRACE(bh, "get_write_access");
1322         err = ext3_journal_get_write_access(handle, bh);
1323         if (err) {
1324                 ext3_std_error(dir->i_sb, err);
1325                 brelse(bh);
1326                 return err;
1327         }
1328
1329         /* By now the buffer is marked for journaling */
1330         nlen = EXT3_DIR_REC_LEN(de->name_len);
1331         rlen = le16_to_cpu(de->rec_len);
1332         if (de->inode) {
1333                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1334                 de1->rec_len = cpu_to_le16(rlen - nlen);
1335                 de->rec_len = cpu_to_le16(nlen);
1336                 de = de1;
1337         }
1338         de->file_type = EXT3_FT_UNKNOWN;
1339         if (inode) {
1340                 de->inode = cpu_to_le32(inode->i_ino);
1341                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1342         } else
1343                 de->inode = 0;
1344         de->name_len = namelen;
1345         memcpy (de->name, name, namelen);
1346         /*
1347          * XXX shouldn't update any times until successful
1348          * completion of syscall, but too many callers depend
1349          * on this.
1350          *
1351          * XXX similarly, too many callers depend on
1352          * ext3_new_inode() setting the times, but error
1353          * recovery deletes the inode, so the worst that can
1354          * happen is that the times are slightly out of date
1355          * and/or different from the directory change time.
1356          */
1357         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1358         ext3_update_dx_flag(dir);
1359         dir->i_version++;
1360         ext3_mark_inode_dirty(handle, dir);
1361         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1362         err = ext3_journal_dirty_metadata(handle, bh);
1363         if (err)
1364                 ext3_std_error(dir->i_sb, err);
1365         brelse(bh);
1366         return 0;
1367 }
1368
1369 #ifdef CONFIG_EXT3_INDEX
1370 /*
1371  * This converts a one block unindexed directory to a 3 block indexed
1372  * directory, and adds the dentry to the indexed directory.
1373  */
1374 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1375                             struct inode *inode, struct buffer_head *bh)
1376 {
1377         struct inode    *dir = dentry->d_parent->d_inode;
1378         const char      *name = dentry->d_name.name;
1379         int             namelen = dentry->d_name.len;
1380         struct buffer_head *bh2;
1381         struct dx_root  *root;
1382         struct dx_frame frames[2], *frame;
1383         struct dx_entry *entries;
1384         struct ext3_dir_entry_2 *de, *de2;
1385         char            *data1, *top;
1386         unsigned        len;
1387         int             retval;
1388         unsigned        blocksize;
1389         struct dx_hash_info hinfo;
1390         u32             block;
1391         struct fake_dirent *fde;
1392
1393         blocksize =  dir->i_sb->s_blocksize;
1394         dxtrace(printk("Creating index\n"));
1395         retval = ext3_journal_get_write_access(handle, bh);
1396         if (retval) {
1397                 ext3_std_error(dir->i_sb, retval);
1398                 brelse(bh);
1399                 return retval;
1400         }
1401         root = (struct dx_root *) bh->b_data;
1402
1403         bh2 = ext3_append (handle, dir, &block, &retval);
1404         if (!(bh2)) {
1405                 brelse(bh);
1406                 return retval;
1407         }
1408         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1409         data1 = bh2->b_data;
1410
1411         /* The 0th block becomes the root, move the dirents out */
1412         fde = &root->dotdot;
1413         de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1414         len = ((char *) root) + blocksize - (char *) de;
1415         memcpy (data1, de, len);
1416         de = (struct ext3_dir_entry_2 *) data1;
1417         top = data1 + len;
1418         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1419                 de = de2;
1420         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1421         /* Initialize the root; the dot dirents already exist */
1422         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1423         de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1424         memset (&root->info, 0, sizeof(root->info));
1425         root->info.info_length = sizeof(root->info);
1426         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1427         entries = root->entries;
1428         dx_set_block (entries, 1);
1429         dx_set_count (entries, 1);
1430         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1431
1432         /* Initialize as for dx_probe */
1433         hinfo.hash_version = root->info.hash_version;
1434         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1435         ext3fs_dirhash(name, namelen, &hinfo);
1436         frame = frames;
1437         frame->entries = entries;
1438         frame->at = entries;
1439         frame->bh = bh;
1440         bh = bh2;
1441         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1442         dx_release (frames);
1443         if (!(de))
1444                 return retval;
1445
1446         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1447 }
1448 #endif
1449
1450 /*
1451  *      ext3_add_entry()
1452  *
1453  * adds a file entry to the specified directory, using the same
1454  * semantics as ext3_find_entry(). It returns NULL if it failed.
1455  *
1456  * NOTE!! The inode part of 'de' is left at 0 - which means you
1457  * may not sleep between calling this and putting something into
1458  * the entry, as someone else might have used it while you slept.
1459  */
1460 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1461         struct inode *inode)
1462 {
1463         struct inode *dir = dentry->d_parent->d_inode;
1464         unsigned long offset;
1465         struct buffer_head * bh;
1466         struct ext3_dir_entry_2 *de;
1467         struct super_block * sb;
1468         int     retval;
1469 #ifdef CONFIG_EXT3_INDEX
1470         int     dx_fallback=0;
1471 #endif
1472         unsigned blocksize;
1473         u32 block, blocks;
1474
1475         sb = dir->i_sb;
1476         blocksize = sb->s_blocksize;
1477         if (!dentry->d_name.len)
1478                 return -EINVAL;
1479 #ifdef CONFIG_EXT3_INDEX
1480         if (is_dx(dir)) {
1481                 retval = ext3_dx_add_entry(handle, dentry, inode);
1482                 if (!retval || (retval != ERR_BAD_DX_DIR))
1483                         return retval;
1484                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1485                 dx_fallback++;
1486                 ext3_mark_inode_dirty(handle, dir);
1487         }
1488 #endif
1489         blocks = dir->i_size >> sb->s_blocksize_bits;
1490         for (block = 0, offset = 0; block < blocks; block++) {
1491                 bh = ext3_bread(handle, dir, block, 0, &retval);
1492                 if(!bh)
1493                         return retval;
1494                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1495                 if (retval != -ENOSPC)
1496                         return retval;
1497
1498 #ifdef CONFIG_EXT3_INDEX
1499                 if (blocks == 1 && !dx_fallback &&
1500                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1501                         return make_indexed_dir(handle, dentry, inode, bh);
1502 #endif
1503                 brelse(bh);
1504         }
1505         bh = ext3_append(handle, dir, &block, &retval);
1506         if (!bh)
1507                 return retval;
1508         de = (struct ext3_dir_entry_2 *) bh->b_data;
1509         de->inode = 0;
1510         de->rec_len = cpu_to_le16(blocksize);
1511         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1512 }
1513
1514 #ifdef CONFIG_EXT3_INDEX
1515 /*
1516  * Returns 0 for success, or a negative error value
1517  */
1518 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1519                              struct inode *inode)
1520 {
1521         struct dx_frame frames[2], *frame;
1522         struct dx_entry *entries, *at;
1523         struct dx_hash_info hinfo;
1524         struct buffer_head * bh;
1525         struct inode *dir = dentry->d_parent->d_inode;
1526         struct super_block * sb = dir->i_sb;
1527         struct ext3_dir_entry_2 *de;
1528         int err;
1529
1530         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1531         if (!frame)
1532                 return err;
1533         entries = frame->entries;
1534         at = frame->at;
1535
1536         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1537                 goto cleanup;
1538
1539         BUFFER_TRACE(bh, "get_write_access");
1540         err = ext3_journal_get_write_access(handle, bh);
1541         if (err)
1542                 goto journal_error;
1543
1544         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1545         if (err != -ENOSPC) {
1546                 bh = NULL;
1547                 goto cleanup;
1548         }
1549
1550         /* Block full, should compress but for now just split */
1551         dxtrace(printk("using %u of %u node entries\n",
1552                        dx_get_count(entries), dx_get_limit(entries)));
1553         /* Need to split index? */
1554         if (dx_get_count(entries) == dx_get_limit(entries)) {
1555                 u32 newblock;
1556                 unsigned icount = dx_get_count(entries);
1557                 int levels = frame - frames;
1558                 struct dx_entry *entries2;
1559                 struct dx_node *node2;
1560                 struct buffer_head *bh2;
1561
1562                 if (levels && (dx_get_count(frames->entries) ==
1563                                dx_get_limit(frames->entries))) {
1564                         ext3_warning(sb, __FUNCTION__,
1565                                      "Directory index full!");
1566                         err = -ENOSPC;
1567                         goto cleanup;
1568                 }
1569                 bh2 = ext3_append (handle, dir, &newblock, &err);
1570                 if (!(bh2))
1571                         goto cleanup;
1572                 node2 = (struct dx_node *)(bh2->b_data);
1573                 entries2 = node2->entries;
1574                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1575                 node2->fake.inode = 0;
1576                 BUFFER_TRACE(frame->bh, "get_write_access");
1577                 err = ext3_journal_get_write_access(handle, frame->bh);
1578                 if (err)
1579                         goto journal_error;
1580                 if (levels) {
1581                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1582                         unsigned hash2 = dx_get_hash(entries + icount1);
1583                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1584
1585                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1586                         err = ext3_journal_get_write_access(handle,
1587                                                              frames[0].bh);
1588                         if (err)
1589                                 goto journal_error;
1590
1591                         memcpy ((char *) entries2, (char *) (entries + icount1),
1592                                 icount2 * sizeof(struct dx_entry));
1593                         dx_set_count (entries, icount1);
1594                         dx_set_count (entries2, icount2);
1595                         dx_set_limit (entries2, dx_node_limit(dir));
1596
1597                         /* Which index block gets the new entry? */
1598                         if (at - entries >= icount1) {
1599                                 frame->at = at = at - entries - icount1 + entries2;
1600                                 frame->entries = entries = entries2;
1601                                 swap(frame->bh, bh2);
1602                         }
1603                         dx_insert_block (frames + 0, hash2, newblock);
1604                         dxtrace(dx_show_index ("node", frames[1].entries));
1605                         dxtrace(dx_show_index ("node",
1606                                ((struct dx_node *) bh2->b_data)->entries));
1607                         err = ext3_journal_dirty_metadata(handle, bh2);
1608                         if (err)
1609                                 goto journal_error;
1610                         brelse (bh2);
1611                 } else {
1612                         dxtrace(printk("Creating second level index...\n"));
1613                         memcpy((char *) entries2, (char *) entries,
1614                                icount * sizeof(struct dx_entry));
1615                         dx_set_limit(entries2, dx_node_limit(dir));
1616
1617                         /* Set up root */
1618                         dx_set_count(entries, 1);
1619                         dx_set_block(entries + 0, newblock);
1620                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1621
1622                         /* Add new access path frame */
1623                         frame = frames + 1;
1624                         frame->at = at = at - entries + entries2;
1625                         frame->entries = entries = entries2;
1626                         frame->bh = bh2;
1627                         err = ext3_journal_get_write_access(handle,
1628                                                              frame->bh);
1629                         if (err)
1630                                 goto journal_error;
1631                 }
1632                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1633         }
1634         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1635         if (!de)
1636                 goto cleanup;
1637         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1638         bh = NULL;
1639         goto cleanup;
1640
1641 journal_error:
1642         ext3_std_error(dir->i_sb, err);
1643 cleanup:
1644         if (bh)
1645                 brelse(bh);
1646         dx_release(frames);
1647         return err;
1648 }
1649 #endif
1650
1651 /*
1652  * ext3_delete_entry deletes a directory entry by merging it with the
1653  * previous entry
1654  */
1655 static int ext3_delete_entry (handle_t *handle,
1656                               struct inode * dir,
1657                               struct ext3_dir_entry_2 * de_del,
1658                               struct buffer_head * bh)
1659 {
1660         struct ext3_dir_entry_2 * de, * pde;
1661         int i;
1662
1663         i = 0;
1664         pde = NULL;
1665         de = (struct ext3_dir_entry_2 *) bh->b_data;
1666         while (i < bh->b_size) {
1667                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1668                         return -EIO;
1669                 if (de == de_del)  {
1670                         BUFFER_TRACE(bh, "get_write_access");
1671                         ext3_journal_get_write_access(handle, bh);
1672                         if (pde)
1673                                 pde->rec_len =
1674                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1675                                                     le16_to_cpu(de->rec_len));
1676                         else
1677                                 de->inode = 0;
1678                         dir->i_version++;
1679                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1680                         ext3_journal_dirty_metadata(handle, bh);
1681                         return 0;
1682                 }
1683                 i += le16_to_cpu(de->rec_len);
1684                 pde = de;
1685                 de = (struct ext3_dir_entry_2 *)
1686                         ((char *) de + le16_to_cpu(de->rec_len));
1687         }
1688         return -ENOENT;
1689 }
1690
1691 static int ext3_add_nondir(handle_t *handle,
1692                 struct dentry *dentry, struct inode *inode)
1693 {
1694         int err = ext3_add_entry(handle, dentry, inode);
1695         if (!err) {
1696                 ext3_mark_inode_dirty(handle, inode);
1697                 d_instantiate(dentry, inode);
1698                 return 0;
1699         }
1700         drop_nlink(inode);
1701         iput(inode);
1702         return err;
1703 }
1704
1705 /*
1706  * By the time this is called, we already have created
1707  * the directory cache entry for the new file, but it
1708  * is so far negative - it has no inode.
1709  *
1710  * If the create succeeds, we fill in the inode information
1711  * with d_instantiate().
1712  */
1713 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1714                 struct nameidata *nd)
1715 {
1716         handle_t *handle;
1717         struct inode * inode;
1718         int err, retries = 0;
1719
1720 retry:
1721         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1722                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1723                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1724         if (IS_ERR(handle))
1725                 return PTR_ERR(handle);
1726
1727         if (IS_DIRSYNC(dir))
1728                 handle->h_sync = 1;
1729
1730         inode = ext3_new_inode (handle, dir, mode);
1731         err = PTR_ERR(inode);
1732         if (!IS_ERR(inode)) {
1733                 inode->i_op = &ext3_file_inode_operations;
1734                 inode->i_fop = &ext3_file_operations;
1735                 ext3_set_aops(inode);
1736                 err = ext3_add_nondir(handle, dentry, inode);
1737         }
1738         ext3_journal_stop(handle);
1739         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1740                 goto retry;
1741         return err;
1742 }
1743
1744 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1745                         int mode, dev_t rdev)
1746 {
1747         handle_t *handle;
1748         struct inode *inode;
1749         int err, retries = 0;
1750
1751         if (!new_valid_dev(rdev))
1752                 return -EINVAL;
1753
1754 retry:
1755         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1756                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1757                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1758         if (IS_ERR(handle))
1759                 return PTR_ERR(handle);
1760
1761         if (IS_DIRSYNC(dir))
1762                 handle->h_sync = 1;
1763
1764         inode = ext3_new_inode (handle, dir, mode);
1765         err = PTR_ERR(inode);
1766         if (!IS_ERR(inode)) {
1767                 init_special_inode(inode, inode->i_mode, rdev);
1768 #ifdef CONFIG_EXT3_FS_XATTR
1769                 inode->i_op = &ext3_special_inode_operations;
1770 #endif
1771                 err = ext3_add_nondir(handle, dentry, inode);
1772         }
1773         ext3_journal_stop(handle);
1774         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1775                 goto retry;
1776         return err;
1777 }
1778
1779 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1780 {
1781         handle_t *handle;
1782         struct inode * inode;
1783         struct buffer_head * dir_block;
1784         struct ext3_dir_entry_2 * de;
1785         int err, retries = 0;
1786
1787         if (dir->i_nlink >= EXT3_LINK_MAX)
1788                 return -EMLINK;
1789
1790 retry:
1791         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1792                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1793                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1794         if (IS_ERR(handle))
1795                 return PTR_ERR(handle);
1796
1797         if (IS_DIRSYNC(dir))
1798                 handle->h_sync = 1;
1799
1800         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1801         err = PTR_ERR(inode);
1802         if (IS_ERR(inode))
1803                 goto out_stop;
1804
1805         inode->i_op = &ext3_dir_inode_operations;
1806         inode->i_fop = &ext3_dir_operations;
1807         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1808         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1809         if (!dir_block) {
1810                 drop_nlink(inode); /* is this nlink == 0? */
1811                 ext3_mark_inode_dirty(handle, inode);
1812                 iput (inode);
1813                 goto out_stop;
1814         }
1815         BUFFER_TRACE(dir_block, "get_write_access");
1816         ext3_journal_get_write_access(handle, dir_block);
1817         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1818         de->inode = cpu_to_le32(inode->i_ino);
1819         de->name_len = 1;
1820         de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1821         strcpy (de->name, ".");
1822         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1823         de = (struct ext3_dir_entry_2 *)
1824                         ((char *) de + le16_to_cpu(de->rec_len));
1825         de->inode = cpu_to_le32(dir->i_ino);
1826         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1827         de->name_len = 2;
1828         strcpy (de->name, "..");
1829         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1830         inode->i_nlink = 2;
1831         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1832         ext3_journal_dirty_metadata(handle, dir_block);
1833         brelse (dir_block);
1834         ext3_mark_inode_dirty(handle, inode);
1835         err = ext3_add_entry (handle, dentry, inode);
1836         if (err) {
1837                 inode->i_nlink = 0;
1838                 ext3_mark_inode_dirty(handle, inode);
1839                 iput (inode);
1840                 goto out_stop;
1841         }
1842         inc_nlink(dir);
1843         ext3_update_dx_flag(dir);
1844         ext3_mark_inode_dirty(handle, dir);
1845         d_instantiate(dentry, inode);
1846 out_stop:
1847         ext3_journal_stop(handle);
1848         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1849                 goto retry;
1850         return err;
1851 }
1852
1853 /*
1854  * routine to check that the specified directory is empty (for rmdir)
1855  */
1856 static int empty_dir (struct inode * inode)
1857 {
1858         unsigned long offset;
1859         struct buffer_head * bh;
1860         struct ext3_dir_entry_2 * de, * de1;
1861         struct super_block * sb;
1862         int err = 0;
1863
1864         sb = inode->i_sb;
1865         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1866             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1867                 if (err)
1868                         ext3_error(inode->i_sb, __FUNCTION__,
1869                                    "error %d reading directory #%lu offset 0",
1870                                    err, inode->i_ino);
1871                 else
1872                         ext3_warning(inode->i_sb, __FUNCTION__,
1873                                      "bad directory (dir #%lu) - no data block",
1874                                      inode->i_ino);
1875                 return 1;
1876         }
1877         de = (struct ext3_dir_entry_2 *) bh->b_data;
1878         de1 = (struct ext3_dir_entry_2 *)
1879                         ((char *) de + le16_to_cpu(de->rec_len));
1880         if (le32_to_cpu(de->inode) != inode->i_ino ||
1881                         !le32_to_cpu(de1->inode) ||
1882                         strcmp (".", de->name) ||
1883                         strcmp ("..", de1->name)) {
1884                 ext3_warning (inode->i_sb, "empty_dir",
1885                               "bad directory (dir #%lu) - no `.' or `..'",
1886                               inode->i_ino);
1887                 brelse (bh);
1888                 return 1;
1889         }
1890         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1891         de = (struct ext3_dir_entry_2 *)
1892                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1893         while (offset < inode->i_size ) {
1894                 if (!bh ||
1895                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1896                         err = 0;
1897                         brelse (bh);
1898                         bh = ext3_bread (NULL, inode,
1899                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1900                         if (!bh) {
1901                                 if (err)
1902                                         ext3_error(sb, __FUNCTION__,
1903                                                    "error %d reading directory"
1904                                                    " #%lu offset %lu",
1905                                                    err, inode->i_ino, offset);
1906                                 offset += sb->s_blocksize;
1907                                 continue;
1908                         }
1909                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1910                 }
1911                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1912                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1913                                                          sb->s_blocksize);
1914                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1915                         continue;
1916                 }
1917                 if (le32_to_cpu(de->inode)) {
1918                         brelse (bh);
1919                         return 0;
1920                 }
1921                 offset += le16_to_cpu(de->rec_len);
1922                 de = (struct ext3_dir_entry_2 *)
1923                                 ((char *) de + le16_to_cpu(de->rec_len));
1924         }
1925         brelse (bh);
1926         return 1;
1927 }
1928
1929 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1930  * such inodes, starting at the superblock, in case we crash before the
1931  * file is closed/deleted, or in case the inode truncate spans multiple
1932  * transactions and the last transaction is not recovered after a crash.
1933  *
1934  * At filesystem recovery time, we walk this list deleting unlinked
1935  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1936  */
1937 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1938 {
1939         struct super_block *sb = inode->i_sb;
1940         struct ext3_iloc iloc;
1941         int err = 0, rc;
1942
1943         lock_super(sb);
1944         if (!list_empty(&EXT3_I(inode)->i_orphan))
1945                 goto out_unlock;
1946
1947         /* Orphan handling is only valid for files with data blocks
1948          * being truncated, or files being unlinked. */
1949
1950         /* @@@ FIXME: Observation from aviro:
1951          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1952          * here (on lock_super()), so race with ext3_link() which might bump
1953          * ->i_nlink. For, say it, character device. Not a regular file,
1954          * not a directory, not a symlink and ->i_nlink > 0.
1955          */
1956         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1957                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1958
1959         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1960         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1961         if (err)
1962                 goto out_unlock;
1963
1964         err = ext3_reserve_inode_write(handle, inode, &iloc);
1965         if (err)
1966                 goto out_unlock;
1967
1968         /* Insert this inode at the head of the on-disk orphan list... */
1969         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1970         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1971         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1972         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1973         if (!err)
1974                 err = rc;
1975
1976         /* Only add to the head of the in-memory list if all the
1977          * previous operations succeeded.  If the orphan_add is going to
1978          * fail (possibly taking the journal offline), we can't risk
1979          * leaving the inode on the orphan list: stray orphan-list
1980          * entries can cause panics at unmount time.
1981          *
1982          * This is safe: on error we're going to ignore the orphan list
1983          * anyway on the next recovery. */
1984         if (!err)
1985                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1986
1987         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1988         jbd_debug(4, "orphan inode %lu will point to %d\n",
1989                         inode->i_ino, NEXT_ORPHAN(inode));
1990 out_unlock:
1991         unlock_super(sb);
1992         ext3_std_error(inode->i_sb, err);
1993         return err;
1994 }
1995
1996 /*
1997  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1998  * of such inodes stored on disk, because it is finally being cleaned up.
1999  */
2000 int ext3_orphan_del(handle_t *handle, struct inode *inode)
2001 {
2002         struct list_head *prev;
2003         struct ext3_inode_info *ei = EXT3_I(inode);
2004         struct ext3_sb_info *sbi;
2005         unsigned long ino_next;
2006         struct ext3_iloc iloc;
2007         int err = 0;
2008
2009         lock_super(inode->i_sb);
2010         if (list_empty(&ei->i_orphan)) {
2011                 unlock_super(inode->i_sb);
2012                 return 0;
2013         }
2014
2015         ino_next = NEXT_ORPHAN(inode);
2016         prev = ei->i_orphan.prev;
2017         sbi = EXT3_SB(inode->i_sb);
2018
2019         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2020
2021         list_del_init(&ei->i_orphan);
2022
2023         /* If we're on an error path, we may not have a valid
2024          * transaction handle with which to update the orphan list on
2025          * disk, but we still need to remove the inode from the linked
2026          * list in memory. */
2027         if (!handle)
2028                 goto out;
2029
2030         err = ext3_reserve_inode_write(handle, inode, &iloc);
2031         if (err)
2032                 goto out_err;
2033
2034         if (prev == &sbi->s_orphan) {
2035                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2036                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2037                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2038                 if (err)
2039                         goto out_brelse;
2040                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2041                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2042         } else {
2043                 struct ext3_iloc iloc2;
2044                 struct inode *i_prev =
2045                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2046
2047                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2048                           i_prev->i_ino, ino_next);
2049                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2050                 if (err)
2051                         goto out_brelse;
2052                 NEXT_ORPHAN(i_prev) = ino_next;
2053                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2054         }
2055         if (err)
2056                 goto out_brelse;
2057         NEXT_ORPHAN(inode) = 0;
2058         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2059
2060 out_err:
2061         ext3_std_error(inode->i_sb, err);
2062 out:
2063         unlock_super(inode->i_sb);
2064         return err;
2065
2066 out_brelse:
2067         brelse(iloc.bh);
2068         goto out_err;
2069 }
2070
2071 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2072 {
2073         int retval;
2074         struct inode * inode;
2075         struct buffer_head * bh;
2076         struct ext3_dir_entry_2 * de;
2077         handle_t *handle;
2078
2079         /* Initialize quotas before so that eventual writes go in
2080          * separate transaction */
2081         DQUOT_INIT(dentry->d_inode);
2082         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2083         if (IS_ERR(handle))
2084                 return PTR_ERR(handle);
2085
2086         retval = -ENOENT;
2087         bh = ext3_find_entry (dentry, &de);
2088         if (!bh)
2089                 goto end_rmdir;
2090
2091         if (IS_DIRSYNC(dir))
2092                 handle->h_sync = 1;
2093
2094         inode = dentry->d_inode;
2095
2096         retval = -EIO;
2097         if (le32_to_cpu(de->inode) != inode->i_ino)
2098                 goto end_rmdir;
2099
2100         retval = -ENOTEMPTY;
2101         if (!empty_dir (inode))
2102                 goto end_rmdir;
2103
2104         retval = ext3_delete_entry(handle, dir, de, bh);
2105         if (retval)
2106                 goto end_rmdir;
2107         if (inode->i_nlink != 2)
2108                 ext3_warning (inode->i_sb, "ext3_rmdir",
2109                               "empty directory has nlink!=2 (%d)",
2110                               inode->i_nlink);
2111         inode->i_version++;
2112         clear_nlink(inode);
2113         /* There's no need to set i_disksize: the fact that i_nlink is
2114          * zero will ensure that the right thing happens during any
2115          * recovery. */
2116         inode->i_size = 0;
2117         ext3_orphan_add(handle, inode);
2118         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2119         ext3_mark_inode_dirty(handle, inode);
2120         drop_nlink(dir);
2121         ext3_update_dx_flag(dir);
2122         ext3_mark_inode_dirty(handle, dir);
2123
2124 end_rmdir:
2125         ext3_journal_stop(handle);
2126         brelse (bh);
2127         return retval;
2128 }
2129
2130 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2131 {
2132         int retval;
2133         struct inode * inode;
2134         struct buffer_head * bh;
2135         struct ext3_dir_entry_2 * de;
2136         handle_t *handle;
2137
2138         /* Initialize quotas before so that eventual writes go
2139          * in separate transaction */
2140         DQUOT_INIT(dentry->d_inode);
2141         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2142         if (IS_ERR(handle))
2143                 return PTR_ERR(handle);
2144
2145         if (IS_DIRSYNC(dir))
2146                 handle->h_sync = 1;
2147
2148         retval = -ENOENT;
2149         bh = ext3_find_entry (dentry, &de);
2150         if (!bh)
2151                 goto end_unlink;
2152
2153         inode = dentry->d_inode;
2154
2155         retval = -EIO;
2156         if (le32_to_cpu(de->inode) != inode->i_ino)
2157                 goto end_unlink;
2158
2159         if (!inode->i_nlink) {
2160                 ext3_warning (inode->i_sb, "ext3_unlink",
2161                               "Deleting nonexistent file (%lu), %d",
2162                               inode->i_ino, inode->i_nlink);
2163                 inode->i_nlink = 1;
2164         }
2165         retval = ext3_delete_entry(handle, dir, de, bh);
2166         if (retval)
2167                 goto end_unlink;
2168         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2169         ext3_update_dx_flag(dir);
2170         ext3_mark_inode_dirty(handle, dir);
2171         drop_nlink(inode);
2172         if (!inode->i_nlink)
2173                 ext3_orphan_add(handle, inode);
2174         inode->i_ctime = dir->i_ctime;
2175         ext3_mark_inode_dirty(handle, inode);
2176         retval = 0;
2177
2178 end_unlink:
2179         ext3_journal_stop(handle);
2180         brelse (bh);
2181         return retval;
2182 }
2183
2184 static int ext3_symlink (struct inode * dir,
2185                 struct dentry *dentry, const char * symname)
2186 {
2187         handle_t *handle;
2188         struct inode * inode;
2189         int l, err, retries = 0;
2190
2191         l = strlen(symname)+1;
2192         if (l > dir->i_sb->s_blocksize)
2193                 return -ENAMETOOLONG;
2194
2195 retry:
2196         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2197                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2198                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2199         if (IS_ERR(handle))
2200                 return PTR_ERR(handle);
2201
2202         if (IS_DIRSYNC(dir))
2203                 handle->h_sync = 1;
2204
2205         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2206         err = PTR_ERR(inode);
2207         if (IS_ERR(inode))
2208                 goto out_stop;
2209
2210         if (l > sizeof (EXT3_I(inode)->i_data)) {
2211                 inode->i_op = &ext3_symlink_inode_operations;
2212                 ext3_set_aops(inode);
2213                 /*
2214                  * page_symlink() calls into ext3_prepare/commit_write.
2215                  * We have a transaction open.  All is sweetness.  It also sets
2216                  * i_size in generic_commit_write().
2217                  */
2218                 err = __page_symlink(inode, symname, l,
2219                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2220                 if (err) {
2221                         drop_nlink(inode);
2222                         ext3_mark_inode_dirty(handle, inode);
2223                         iput (inode);
2224                         goto out_stop;
2225                 }
2226         } else {
2227                 inode->i_op = &ext3_fast_symlink_inode_operations;
2228                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2229                 inode->i_size = l-1;
2230         }
2231         EXT3_I(inode)->i_disksize = inode->i_size;
2232         err = ext3_add_nondir(handle, dentry, inode);
2233 out_stop:
2234         ext3_journal_stop(handle);
2235         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2236                 goto retry;
2237         return err;
2238 }
2239
2240 static int ext3_link (struct dentry * old_dentry,
2241                 struct inode * dir, struct dentry *dentry)
2242 {
2243         handle_t *handle;
2244         struct inode *inode = old_dentry->d_inode;
2245         int err, retries = 0;
2246
2247         if (inode->i_nlink >= EXT3_LINK_MAX)
2248                 return -EMLINK;
2249         /*
2250          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2251          * otherwise has the potential to corrupt the orphan inode list.
2252          */
2253         if (inode->i_nlink == 0)
2254                 return -ENOENT;
2255
2256 retry:
2257         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2258                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2259         if (IS_ERR(handle))
2260                 return PTR_ERR(handle);
2261
2262         if (IS_DIRSYNC(dir))
2263                 handle->h_sync = 1;
2264
2265         inode->i_ctime = CURRENT_TIME_SEC;
2266         inc_nlink(inode);
2267         atomic_inc(&inode->i_count);
2268
2269         err = ext3_add_nondir(handle, dentry, inode);
2270         ext3_journal_stop(handle);
2271         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2272                 goto retry;
2273         return err;
2274 }
2275
2276 #define PARENT_INO(buffer) \
2277         ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2278         le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2279
2280 /*
2281  * Anybody can rename anything with this: the permission checks are left to the
2282  * higher-level routines.
2283  */
2284 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2285                            struct inode * new_dir,struct dentry *new_dentry)
2286 {
2287         handle_t *handle;
2288         struct inode * old_inode, * new_inode;
2289         struct buffer_head * old_bh, * new_bh, * dir_bh;
2290         struct ext3_dir_entry_2 * old_de, * new_de;
2291         int retval;
2292
2293         old_bh = new_bh = dir_bh = NULL;
2294
2295         /* Initialize quotas before so that eventual writes go
2296          * in separate transaction */
2297         if (new_dentry->d_inode)
2298                 DQUOT_INIT(new_dentry->d_inode);
2299         handle = ext3_journal_start(old_dir, 2 *
2300                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2301                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2302         if (IS_ERR(handle))
2303                 return PTR_ERR(handle);
2304
2305         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2306                 handle->h_sync = 1;
2307
2308         old_bh = ext3_find_entry (old_dentry, &old_de);
2309         /*
2310          *  Check for inode number is _not_ due to possible IO errors.
2311          *  We might rmdir the source, keep it as pwd of some process
2312          *  and merrily kill the link to whatever was created under the
2313          *  same name. Goodbye sticky bit ;-<
2314          */
2315         old_inode = old_dentry->d_inode;
2316         retval = -ENOENT;
2317         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2318                 goto end_rename;
2319
2320         new_inode = new_dentry->d_inode;
2321         new_bh = ext3_find_entry (new_dentry, &new_de);
2322         if (new_bh) {
2323                 if (!new_inode) {
2324                         brelse (new_bh);
2325                         new_bh = NULL;
2326                 }
2327         }
2328         if (S_ISDIR(old_inode->i_mode)) {
2329                 if (new_inode) {
2330                         retval = -ENOTEMPTY;
2331                         if (!empty_dir (new_inode))
2332                                 goto end_rename;
2333                 }
2334                 retval = -EIO;
2335                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2336                 if (!dir_bh)
2337                         goto end_rename;
2338                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2339                         goto end_rename;
2340                 retval = -EMLINK;
2341                 if (!new_inode && new_dir!=old_dir &&
2342                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2343                         goto end_rename;
2344         }
2345         if (!new_bh) {
2346                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2347                 if (retval)
2348                         goto end_rename;
2349         } else {
2350                 BUFFER_TRACE(new_bh, "get write access");
2351                 ext3_journal_get_write_access(handle, new_bh);
2352                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2353                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2354                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2355                         new_de->file_type = old_de->file_type;
2356                 new_dir->i_version++;
2357                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2358                 ext3_journal_dirty_metadata(handle, new_bh);
2359                 brelse(new_bh);
2360                 new_bh = NULL;
2361         }
2362
2363         /*
2364          * Like most other Unix systems, set the ctime for inodes on a
2365          * rename.
2366          */
2367         old_inode->i_ctime = CURRENT_TIME_SEC;
2368         ext3_mark_inode_dirty(handle, old_inode);
2369
2370         /*
2371          * ok, that's it
2372          */
2373         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2374             old_de->name_len != old_dentry->d_name.len ||
2375             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2376             (retval = ext3_delete_entry(handle, old_dir,
2377                                         old_de, old_bh)) == -ENOENT) {
2378                 /* old_de could have moved from under us during htree split, so
2379                  * make sure that we are deleting the right entry.  We might
2380                  * also be pointing to a stale entry in the unused part of
2381                  * old_bh so just checking inum and the name isn't enough. */
2382                 struct buffer_head *old_bh2;
2383                 struct ext3_dir_entry_2 *old_de2;
2384
2385                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2386                 if (old_bh2) {
2387                         retval = ext3_delete_entry(handle, old_dir,
2388                                                    old_de2, old_bh2);
2389                         brelse(old_bh2);
2390                 }
2391         }
2392         if (retval) {
2393                 ext3_warning(old_dir->i_sb, "ext3_rename",
2394                                 "Deleting old file (%lu), %d, error=%d",
2395                                 old_dir->i_ino, old_dir->i_nlink, retval);
2396         }
2397
2398         if (new_inode) {
2399                 drop_nlink(new_inode);
2400                 new_inode->i_ctime = CURRENT_TIME_SEC;
2401         }
2402         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2403         ext3_update_dx_flag(old_dir);
2404         if (dir_bh) {
2405                 BUFFER_TRACE(dir_bh, "get_write_access");
2406                 ext3_journal_get_write_access(handle, dir_bh);
2407                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2408                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2409                 ext3_journal_dirty_metadata(handle, dir_bh);
2410                 drop_nlink(old_dir);
2411                 if (new_inode) {
2412                         drop_nlink(new_inode);
2413                 } else {
2414                         inc_nlink(new_dir);
2415                         ext3_update_dx_flag(new_dir);
2416                         ext3_mark_inode_dirty(handle, new_dir);
2417                 }
2418         }
2419         ext3_mark_inode_dirty(handle, old_dir);
2420         if (new_inode) {
2421                 ext3_mark_inode_dirty(handle, new_inode);
2422                 if (!new_inode->i_nlink)
2423                         ext3_orphan_add(handle, new_inode);
2424         }
2425         retval = 0;
2426
2427 end_rename:
2428         brelse (dir_bh);
2429         brelse (old_bh);
2430         brelse (new_bh);
2431         ext3_journal_stop(handle);
2432         return retval;
2433 }
2434
2435 /*
2436  * directories can handle most operations...
2437  */
2438 const struct inode_operations ext3_dir_inode_operations = {
2439         .create         = ext3_create,
2440         .lookup         = ext3_lookup,
2441         .link           = ext3_link,
2442         .unlink         = ext3_unlink,
2443         .symlink        = ext3_symlink,
2444         .mkdir          = ext3_mkdir,
2445         .rmdir          = ext3_rmdir,
2446         .mknod          = ext3_mknod,
2447         .rename         = ext3_rename,
2448         .setattr        = ext3_setattr,
2449 #ifdef CONFIG_EXT3_FS_XATTR
2450         .setxattr       = generic_setxattr,
2451         .getxattr       = generic_getxattr,
2452         .listxattr      = ext3_listxattr,
2453         .removexattr    = generic_removexattr,
2454 #endif
2455         .permission     = ext3_permission,
2456 };
2457
2458 const struct inode_operations ext3_special_inode_operations = {
2459         .setattr        = ext3_setattr,
2460 #ifdef CONFIG_EXT3_FS_XATTR
2461         .setxattr       = generic_setxattr,
2462         .getxattr       = generic_getxattr,
2463         .listxattr      = ext3_listxattr,
2464         .removexattr    = generic_removexattr,
2465 #endif
2466         .permission     = ext3_permission,
2467 };