Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / fs / ext3 / namei.c
1 /*
2  *  linux/fs/ext3/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.h>
39
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext3_append(handle_t *handle,
53                                         struct inode *inode,
54                                         u32 *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         bh = ext3_bread(handle, inode, *block, 1, err);
61         if (bh) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT3_I(inode)->i_disksize = inode->i_size;
64                 *err = ext3_journal_get_write_access(handle, bh);
65                 if (*err) {
66                         brelse(bh);
67                         bh = NULL;
68                 }
69         }
70         return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #ifndef swap
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79 #endif
80
81 #ifdef DX_DEBUG
82 #define dxtrace(command) command
83 #else
84 #define dxtrace(command)
85 #endif
86
87 struct fake_dirent
88 {
89         __le32 inode;
90         __le16 rec_len;
91         u8 name_len;
92         u8 file_type;
93 };
94
95 struct dx_countlimit
96 {
97         __le16 limit;
98         __le16 count;
99 };
100
101 struct dx_entry
102 {
103         __le32 hash;
104         __le32 block;
105 };
106
107 /*
108  * dx_root_info is laid out so that if it should somehow get overlaid by a
109  * dirent the two low bits of the hash version will be zero.  Therefore, the
110  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
111  */
112
113 struct dx_root
114 {
115         struct fake_dirent dot;
116         char dot_name[4];
117         struct fake_dirent dotdot;
118         char dotdot_name[4];
119         struct dx_root_info
120         {
121                 __le32 reserved_zero;
122                 u8 hash_version;
123                 u8 info_length; /* 8 */
124                 u8 indirect_levels;
125                 u8 unused_flags;
126         }
127         info;
128         struct dx_entry entries[0];
129 };
130
131 struct dx_node
132 {
133         struct fake_dirent fake;
134         struct dx_entry entries[0];
135 };
136
137
138 struct dx_frame
139 {
140         struct buffer_head *bh;
141         struct dx_entry *entries;
142         struct dx_entry *at;
143 };
144
145 struct dx_map_entry
146 {
147         u32 hash;
148         u16 offs;
149         u16 size;
150 };
151
152 static inline unsigned dx_get_block (struct dx_entry *entry);
153 static void dx_set_block (struct dx_entry *entry, unsigned value);
154 static inline unsigned dx_get_hash (struct dx_entry *entry);
155 static void dx_set_hash (struct dx_entry *entry, unsigned value);
156 static unsigned dx_get_count (struct dx_entry *entries);
157 static unsigned dx_get_limit (struct dx_entry *entries);
158 static void dx_set_count (struct dx_entry *entries, unsigned value);
159 static void dx_set_limit (struct dx_entry *entries, unsigned value);
160 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
161 static unsigned dx_node_limit (struct inode *dir);
162 static struct dx_frame *dx_probe(struct dentry *dentry,
163                                  struct inode *dir,
164                                  struct dx_hash_info *hinfo,
165                                  struct dx_frame *frame,
166                                  int *err);
167 static void dx_release (struct dx_frame *frames);
168 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
169                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
170 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
171 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
172                 struct dx_map_entry *offsets, int count);
173 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
174 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
175 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
176                                  struct dx_frame *frame,
177                                  struct dx_frame *frames,
178                                  __u32 *start_hash);
179 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
180                        struct ext3_dir_entry_2 **res_dir, int *err);
181 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
182                              struct inode *inode);
183
184 /*
185  * p is at least 6 bytes before the end of page
186  */
187 static inline struct ext3_dir_entry_2 *
188 ext3_next_entry(struct ext3_dir_entry_2 *p)
189 {
190         return (struct ext3_dir_entry_2 *)((char *)p +
191                 ext3_rec_len_from_disk(p->rec_len));
192 }
193
194 /*
195  * Future: use high four bits of block for coalesce-on-delete flags
196  * Mask them off for now.
197  */
198
199 static inline unsigned dx_get_block (struct dx_entry *entry)
200 {
201         return le32_to_cpu(entry->block) & 0x00ffffff;
202 }
203
204 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
205 {
206         entry->block = cpu_to_le32(value);
207 }
208
209 static inline unsigned dx_get_hash (struct dx_entry *entry)
210 {
211         return le32_to_cpu(entry->hash);
212 }
213
214 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
215 {
216         entry->hash = cpu_to_le32(value);
217 }
218
219 static inline unsigned dx_get_count (struct dx_entry *entries)
220 {
221         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
222 }
223
224 static inline unsigned dx_get_limit (struct dx_entry *entries)
225 {
226         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
227 }
228
229 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
230 {
231         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
232 }
233
234 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
235 {
236         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
237 }
238
239 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
240 {
241         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
242                 EXT3_DIR_REC_LEN(2) - infosize;
243         return entry_space / sizeof(struct dx_entry);
244 }
245
246 static inline unsigned dx_node_limit (struct inode *dir)
247 {
248         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
249         return entry_space / sizeof(struct dx_entry);
250 }
251
252 /*
253  * Debug
254  */
255 #ifdef DX_DEBUG
256 static void dx_show_index (char * label, struct dx_entry *entries)
257 {
258         int i, n = dx_get_count (entries);
259         printk("%s index ", label);
260         for (i = 0; i < n; i++)
261         {
262                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
263         }
264         printk("\n");
265 }
266
267 struct stats
268 {
269         unsigned names;
270         unsigned space;
271         unsigned bcount;
272 };
273
274 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
275                                  int size, int show_names)
276 {
277         unsigned names = 0, space = 0;
278         char *base = (char *) de;
279         struct dx_hash_info h = *hinfo;
280
281         printk("names: ");
282         while ((char *) de < base + size)
283         {
284                 if (de->inode)
285                 {
286                         if (show_names)
287                         {
288                                 int len = de->name_len;
289                                 char *name = de->name;
290                                 while (len--) printk("%c", *name++);
291                                 ext3fs_dirhash(de->name, de->name_len, &h);
292                                 printk(":%x.%u ", h.hash,
293                                        ((char *) de - base));
294                         }
295                         space += EXT3_DIR_REC_LEN(de->name_len);
296                         names++;
297                 }
298                 de = ext3_next_entry(de);
299         }
300         printk("(%i)\n", names);
301         return (struct stats) { names, space, 1 };
302 }
303
304 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
305                              struct dx_entry *entries, int levels)
306 {
307         unsigned blocksize = dir->i_sb->s_blocksize;
308         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
309         unsigned bcount = 0;
310         struct buffer_head *bh;
311         int err;
312         printk("%i indexed blocks...\n", count);
313         for (i = 0; i < count; i++, entries++)
314         {
315                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
316                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317                 struct stats stats;
318                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
319                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
320                 stats = levels?
321                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
323                 names += stats.names;
324                 space += stats.space;
325                 bcount += stats.bcount;
326                 brelse (bh);
327         }
328         if (bcount)
329                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
330                         names, space/bcount,(space/bcount)*100/blocksize);
331         return (struct stats) { names, space, bcount};
332 }
333 #endif /* DX_DEBUG */
334
335 /*
336  * Probe for a directory leaf block to search.
337  *
338  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
339  * error in the directory index, and the caller should fall back to
340  * searching the directory normally.  The callers of dx_probe **MUST**
341  * check for this error code, and make sure it never gets reflected
342  * back to userspace.
343  */
344 static struct dx_frame *
345 dx_probe(struct dentry *dentry, struct inode *dir,
346          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
347 {
348         unsigned count, indirect;
349         struct dx_entry *at, *entries, *p, *q, *m;
350         struct dx_root *root;
351         struct buffer_head *bh;
352         struct dx_frame *frame = frame_in;
353         u32 hash;
354
355         frame->bh = NULL;
356         if (dentry)
357                 dir = dentry->d_parent->d_inode;
358         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
359                 goto fail;
360         root = (struct dx_root *) bh->b_data;
361         if (root->info.hash_version != DX_HASH_TEA &&
362             root->info.hash_version != DX_HASH_HALF_MD4 &&
363             root->info.hash_version != DX_HASH_LEGACY) {
364                 ext3_warning(dir->i_sb, __func__,
365                              "Unrecognised inode hash code %d",
366                              root->info.hash_version);
367                 brelse(bh);
368                 *err = ERR_BAD_DX_DIR;
369                 goto fail;
370         }
371         hinfo->hash_version = root->info.hash_version;
372         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
373         if (dentry)
374                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
375         hash = hinfo->hash;
376
377         if (root->info.unused_flags & 1) {
378                 ext3_warning(dir->i_sb, __func__,
379                              "Unimplemented inode hash flags: %#06x",
380                              root->info.unused_flags);
381                 brelse(bh);
382                 *err = ERR_BAD_DX_DIR;
383                 goto fail;
384         }
385
386         if ((indirect = root->info.indirect_levels) > 1) {
387                 ext3_warning(dir->i_sb, __func__,
388                              "Unimplemented inode hash depth: %#06x",
389                              root->info.indirect_levels);
390                 brelse(bh);
391                 *err = ERR_BAD_DX_DIR;
392                 goto fail;
393         }
394
395         entries = (struct dx_entry *) (((char *)&root->info) +
396                                        root->info.info_length);
397
398         if (dx_get_limit(entries) != dx_root_limit(dir,
399                                                    root->info.info_length)) {
400                 ext3_warning(dir->i_sb, __func__,
401                              "dx entry: limit != root limit");
402                 brelse(bh);
403                 *err = ERR_BAD_DX_DIR;
404                 goto fail;
405         }
406
407         dxtrace (printk("Look up %x", hash));
408         while (1)
409         {
410                 count = dx_get_count(entries);
411                 if (!count || count > dx_get_limit(entries)) {
412                         ext3_warning(dir->i_sb, __func__,
413                                      "dx entry: no count or count > limit");
414                         brelse(bh);
415                         *err = ERR_BAD_DX_DIR;
416                         goto fail2;
417                 }
418
419                 p = entries + 1;
420                 q = entries + count - 1;
421                 while (p <= q)
422                 {
423                         m = p + (q - p)/2;
424                         dxtrace(printk("."));
425                         if (dx_get_hash(m) > hash)
426                                 q = m - 1;
427                         else
428                                 p = m + 1;
429                 }
430
431                 if (0) // linear search cross check
432                 {
433                         unsigned n = count - 1;
434                         at = entries;
435                         while (n--)
436                         {
437                                 dxtrace(printk(","));
438                                 if (dx_get_hash(++at) > hash)
439                                 {
440                                         at--;
441                                         break;
442                                 }
443                         }
444                         assert (at == p - 1);
445                 }
446
447                 at = p - 1;
448                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
449                 frame->bh = bh;
450                 frame->entries = entries;
451                 frame->at = at;
452                 if (!indirect--) return frame;
453                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
454                         goto fail2;
455                 at = entries = ((struct dx_node *) bh->b_data)->entries;
456                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
457                         ext3_warning(dir->i_sb, __func__,
458                                      "dx entry: limit != node limit");
459                         brelse(bh);
460                         *err = ERR_BAD_DX_DIR;
461                         goto fail2;
462                 }
463                 frame++;
464                 frame->bh = NULL;
465         }
466 fail2:
467         while (frame >= frame_in) {
468                 brelse(frame->bh);
469                 frame--;
470         }
471 fail:
472         if (*err == ERR_BAD_DX_DIR)
473                 ext3_warning(dir->i_sb, __func__,
474                              "Corrupt dir inode %ld, running e2fsck is "
475                              "recommended.", dir->i_ino);
476         return NULL;
477 }
478
479 static void dx_release (struct dx_frame *frames)
480 {
481         if (frames[0].bh == NULL)
482                 return;
483
484         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
485                 brelse(frames[1].bh);
486         brelse(frames[0].bh);
487 }
488
489 /*
490  * This function increments the frame pointer to search the next leaf
491  * block, and reads in the necessary intervening nodes if the search
492  * should be necessary.  Whether or not the search is necessary is
493  * controlled by the hash parameter.  If the hash value is even, then
494  * the search is only continued if the next block starts with that
495  * hash value.  This is used if we are searching for a specific file.
496  *
497  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
498  *
499  * This function returns 1 if the caller should continue to search,
500  * or 0 if it should not.  If there is an error reading one of the
501  * index blocks, it will a negative error code.
502  *
503  * If start_hash is non-null, it will be filled in with the starting
504  * hash of the next page.
505  */
506 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
507                                  struct dx_frame *frame,
508                                  struct dx_frame *frames,
509                                  __u32 *start_hash)
510 {
511         struct dx_frame *p;
512         struct buffer_head *bh;
513         int err, num_frames = 0;
514         __u32 bhash;
515
516         p = frame;
517         /*
518          * Find the next leaf page by incrementing the frame pointer.
519          * If we run out of entries in the interior node, loop around and
520          * increment pointer in the parent node.  When we break out of
521          * this loop, num_frames indicates the number of interior
522          * nodes need to be read.
523          */
524         while (1) {
525                 if (++(p->at) < p->entries + dx_get_count(p->entries))
526                         break;
527                 if (p == frames)
528                         return 0;
529                 num_frames++;
530                 p--;
531         }
532
533         /*
534          * If the hash is 1, then continue only if the next page has a
535          * continuation hash of any value.  This is used for readdir
536          * handling.  Otherwise, check to see if the hash matches the
537          * desired contiuation hash.  If it doesn't, return since
538          * there's no point to read in the successive index pages.
539          */
540         bhash = dx_get_hash(p->at);
541         if (start_hash)
542                 *start_hash = bhash;
543         if ((hash & 1) == 0) {
544                 if ((bhash & ~1) != hash)
545                         return 0;
546         }
547         /*
548          * If the hash is HASH_NB_ALWAYS, we always go to the next
549          * block so no check is necessary
550          */
551         while (num_frames--) {
552                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
553                                       0, &err)))
554                         return err; /* Failure */
555                 p++;
556                 brelse (p->bh);
557                 p->bh = bh;
558                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
559         }
560         return 1;
561 }
562
563
564 /*
565  * This function fills a red-black tree with information from a
566  * directory block.  It returns the number directory entries loaded
567  * into the tree.  If there is an error it is returned in err.
568  */
569 static int htree_dirblock_to_tree(struct file *dir_file,
570                                   struct inode *dir, int block,
571                                   struct dx_hash_info *hinfo,
572                                   __u32 start_hash, __u32 start_minor_hash)
573 {
574         struct buffer_head *bh;
575         struct ext3_dir_entry_2 *de, *top;
576         int err, count = 0;
577
578         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
579         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
580                 return err;
581
582         de = (struct ext3_dir_entry_2 *) bh->b_data;
583         top = (struct ext3_dir_entry_2 *) ((char *) de +
584                                            dir->i_sb->s_blocksize -
585                                            EXT3_DIR_REC_LEN(0));
586         for (; de < top; de = ext3_next_entry(de)) {
587                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
588                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
589                                                 +((char *)de - bh->b_data))) {
590                         /* On error, skip the f_pos to the next block. */
591                         dir_file->f_pos = (dir_file->f_pos |
592                                         (dir->i_sb->s_blocksize - 1)) + 1;
593                         brelse (bh);
594                         return count;
595                 }
596                 ext3fs_dirhash(de->name, de->name_len, hinfo);
597                 if ((hinfo->hash < start_hash) ||
598                     ((hinfo->hash == start_hash) &&
599                      (hinfo->minor_hash < start_minor_hash)))
600                         continue;
601                 if (de->inode == 0)
602                         continue;
603                 if ((err = ext3_htree_store_dirent(dir_file,
604                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
605                         brelse(bh);
606                         return err;
607                 }
608                 count++;
609         }
610         brelse(bh);
611         return count;
612 }
613
614
615 /*
616  * This function fills a red-black tree with information from a
617  * directory.  We start scanning the directory in hash order, starting
618  * at start_hash and start_minor_hash.
619  *
620  * This function returns the number of entries inserted into the tree,
621  * or a negative error code.
622  */
623 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
624                          __u32 start_minor_hash, __u32 *next_hash)
625 {
626         struct dx_hash_info hinfo;
627         struct ext3_dir_entry_2 *de;
628         struct dx_frame frames[2], *frame;
629         struct inode *dir;
630         int block, err;
631         int count = 0;
632         int ret;
633         __u32 hashval;
634
635         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
636                        start_minor_hash));
637         dir = dir_file->f_path.dentry->d_inode;
638         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
639                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
640                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
641                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642                                                start_hash, start_minor_hash);
643                 *next_hash = ~0;
644                 return count;
645         }
646         hinfo.hash = start_hash;
647         hinfo.minor_hash = 0;
648         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
649         if (!frame)
650                 return err;
651
652         /* Add '.' and '..' from the htree header */
653         if (!start_hash && !start_minor_hash) {
654                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
656                         goto errout;
657                 count++;
658         }
659         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
661                 de = ext3_next_entry(de);
662                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
663                         goto errout;
664                 count++;
665         }
666
667         while (1) {
668                 block = dx_get_block(frame->at);
669                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670                                              start_hash, start_minor_hash);
671                 if (ret < 0) {
672                         err = ret;
673                         goto errout;
674                 }
675                 count += ret;
676                 hashval = ~0;
677                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
678                                             frame, frames, &hashval);
679                 *next_hash = hashval;
680                 if (ret < 0) {
681                         err = ret;
682                         goto errout;
683                 }
684                 /*
685                  * Stop if:  (a) there are no more entries, or
686                  * (b) we have inserted at least one entry and the
687                  * next hash value is not a continuation
688                  */
689                 if ((ret == 0) ||
690                     (count && ((hashval & 1) == 0)))
691                         break;
692         }
693         dx_release(frames);
694         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
695                        count, *next_hash));
696         return count;
697 errout:
698         dx_release(frames);
699         return (err);
700 }
701
702
703 /*
704  * Directory block splitting, compacting
705  */
706
707 /*
708  * Create map of hash values, offsets, and sizes, stored at end of block.
709  * Returns number of entries mapped.
710  */
711 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
712                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
713 {
714         int count = 0;
715         char *base = (char *) de;
716         struct dx_hash_info h = *hinfo;
717
718         while ((char *) de < base + size)
719         {
720                 if (de->name_len && de->inode) {
721                         ext3fs_dirhash(de->name, de->name_len, &h);
722                         map_tail--;
723                         map_tail->hash = h.hash;
724                         map_tail->offs = (u16) ((char *) de - base);
725                         map_tail->size = le16_to_cpu(de->rec_len);
726                         count++;
727                         cond_resched();
728                 }
729                 /* XXX: do we need to check rec_len == 0 case? -Chris */
730                 de = ext3_next_entry(de);
731         }
732         return count;
733 }
734
735 /* Sort map by hash value */
736 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737 {
738         struct dx_map_entry *p, *q, *top = map + count - 1;
739         int more;
740         /* Combsort until bubble sort doesn't suck */
741         while (count > 2)
742         {
743                 count = count*10/13;
744                 if (count - 9 < 2) /* 9, 10 -> 11 */
745                         count = 11;
746                 for (p = top, q = p - count; q >= map; p--, q--)
747                         if (p->hash < q->hash)
748                                 swap(*p, *q);
749         }
750         /* Garden variety bubble sort */
751         do {
752                 more = 0;
753                 q = top;
754                 while (q-- > map)
755                 {
756                         if (q[1].hash >= q[0].hash)
757                                 continue;
758                         swap(*(q+1), *q);
759                         more = 1;
760                 }
761         } while(more);
762 }
763
764 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
765 {
766         struct dx_entry *entries = frame->entries;
767         struct dx_entry *old = frame->at, *new = old + 1;
768         int count = dx_get_count(entries);
769
770         assert(count < dx_get_limit(entries));
771         assert(old < entries + count);
772         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773         dx_set_hash(new, hash);
774         dx_set_block(new, block);
775         dx_set_count(entries, count + 1);
776 }
777
778 static void ext3_update_dx_flag(struct inode *inode)
779 {
780         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
781                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
782                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
783 }
784
785 /*
786  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
787  *
788  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789  * `de != NULL' is guaranteed by caller.
790  */
791 static inline int ext3_match (int len, const char * const name,
792                               struct ext3_dir_entry_2 * de)
793 {
794         if (len != de->name_len)
795                 return 0;
796         if (!de->inode)
797                 return 0;
798         return !memcmp(name, de->name, len);
799 }
800
801 /*
802  * Returns 0 if not found, -1 on failure, and 1 on success
803  */
804 static inline int search_dirblock(struct buffer_head * bh,
805                                   struct inode *dir,
806                                   struct dentry *dentry,
807                                   unsigned long offset,
808                                   struct ext3_dir_entry_2 ** res_dir)
809 {
810         struct ext3_dir_entry_2 * de;
811         char * dlimit;
812         int de_len;
813         const char *name = dentry->d_name.name;
814         int namelen = dentry->d_name.len;
815
816         de = (struct ext3_dir_entry_2 *) bh->b_data;
817         dlimit = bh->b_data + dir->i_sb->s_blocksize;
818         while ((char *) de < dlimit) {
819                 /* this code is executed quadratically often */
820                 /* do minimal checking `by hand' */
821
822                 if ((char *) de + namelen <= dlimit &&
823                     ext3_match (namelen, name, de)) {
824                         /* found a match - just to be sure, do a full check */
825                         if (!ext3_check_dir_entry("ext3_find_entry",
826                                                   dir, de, bh, offset))
827                                 return -1;
828                         *res_dir = de;
829                         return 1;
830                 }
831                 /* prevent looping on a bad block */
832                 de_len = ext3_rec_len_from_disk(de->rec_len);
833                 if (de_len <= 0)
834                         return -1;
835                 offset += de_len;
836                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
837         }
838         return 0;
839 }
840
841
842 /*
843  *      ext3_find_entry()
844  *
845  * finds an entry in the specified directory with the wanted name. It
846  * returns the cache buffer in which the entry was found, and the entry
847  * itself (as a parameter - res_dir). It does NOT read the inode of the
848  * entry - you'll have to do that yourself if you want to.
849  *
850  * The returned buffer_head has ->b_count elevated.  The caller is expected
851  * to brelse() it when appropriate.
852  */
853 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
854                                         struct ext3_dir_entry_2 ** res_dir)
855 {
856         struct super_block * sb;
857         struct buffer_head * bh_use[NAMEI_RA_SIZE];
858         struct buffer_head * bh, *ret = NULL;
859         unsigned long start, block, b;
860         int ra_max = 0;         /* Number of bh's in the readahead
861                                    buffer, bh_use[] */
862         int ra_ptr = 0;         /* Current index into readahead
863                                    buffer */
864         int num = 0;
865         int nblocks, i, err;
866         struct inode *dir = dentry->d_parent->d_inode;
867         int namelen;
868
869         *res_dir = NULL;
870         sb = dir->i_sb;
871         namelen = dentry->d_name.len;
872         if (namelen > EXT3_NAME_LEN)
873                 return NULL;
874         if (is_dx(dir)) {
875                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
876                 /*
877                  * On success, or if the error was file not found,
878                  * return.  Otherwise, fall back to doing a search the
879                  * old fashioned way.
880                  */
881                 if (bh || (err != ERR_BAD_DX_DIR))
882                         return bh;
883                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
884         }
885         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886         start = EXT3_I(dir)->i_dir_start_lookup;
887         if (start >= nblocks)
888                 start = 0;
889         block = start;
890 restart:
891         do {
892                 /*
893                  * We deal with the read-ahead logic here.
894                  */
895                 if (ra_ptr >= ra_max) {
896                         /* Refill the readahead buffer */
897                         ra_ptr = 0;
898                         b = block;
899                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900                                 /*
901                                  * Terminate if we reach the end of the
902                                  * directory and must wrap, or if our
903                                  * search has finished at this block.
904                                  */
905                                 if (b >= nblocks || (num && block == start)) {
906                                         bh_use[ra_max] = NULL;
907                                         break;
908                                 }
909                                 num++;
910                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
911                                 bh_use[ra_max] = bh;
912                                 if (bh)
913                                         ll_rw_block(READ_META, 1, &bh);
914                         }
915                 }
916                 if ((bh = bh_use[ra_ptr++]) == NULL)
917                         goto next;
918                 wait_on_buffer(bh);
919                 if (!buffer_uptodate(bh)) {
920                         /* read error, skip block & hope for the best */
921                         ext3_error(sb, __func__, "reading directory #%lu "
922                                    "offset %lu", dir->i_ino, block);
923                         brelse(bh);
924                         goto next;
925                 }
926                 i = search_dirblock(bh, dir, dentry,
927                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
928                 if (i == 1) {
929                         EXT3_I(dir)->i_dir_start_lookup = block;
930                         ret = bh;
931                         goto cleanup_and_exit;
932                 } else {
933                         brelse(bh);
934                         if (i < 0)
935                                 goto cleanup_and_exit;
936                 }
937         next:
938                 if (++block >= nblocks)
939                         block = 0;
940         } while (block != start);
941
942         /*
943          * If the directory has grown while we were searching, then
944          * search the last part of the directory before giving up.
945          */
946         block = nblocks;
947         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
948         if (block < nblocks) {
949                 start = 0;
950                 goto restart;
951         }
952
953 cleanup_and_exit:
954         /* Clean up the read-ahead blocks */
955         for (; ra_ptr < ra_max; ra_ptr++)
956                 brelse (bh_use[ra_ptr]);
957         return ret;
958 }
959
960 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
961                        struct ext3_dir_entry_2 **res_dir, int *err)
962 {
963         struct super_block * sb;
964         struct dx_hash_info     hinfo;
965         u32 hash;
966         struct dx_frame frames[2], *frame;
967         struct ext3_dir_entry_2 *de, *top;
968         struct buffer_head *bh;
969         unsigned long block;
970         int retval;
971         int namelen = dentry->d_name.len;
972         const u8 *name = dentry->d_name.name;
973         struct inode *dir = dentry->d_parent->d_inode;
974
975         sb = dir->i_sb;
976         /* NFS may look up ".." - look at dx_root directory block */
977         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
978                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
979                         return NULL;
980         } else {
981                 frame = frames;
982                 frame->bh = NULL;                       /* for dx_release() */
983                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
984                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
985         }
986         hash = hinfo.hash;
987         do {
988                 block = dx_get_block(frame->at);
989                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
990                         goto errout;
991                 de = (struct ext3_dir_entry_2 *) bh->b_data;
992                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
993                                        EXT3_DIR_REC_LEN(0));
994                 for (; de < top; de = ext3_next_entry(de)) {
995                         int off = (block << EXT3_BLOCK_SIZE_BITS(sb))
996                                   + ((char *) de - bh->b_data);
997
998                         if (!ext3_check_dir_entry(__func__, dir, de, bh, off)) {
999                                 brelse(bh);
1000                                 *err = ERR_BAD_DX_DIR;
1001                                 goto errout;
1002                         }
1003
1004                         if (ext3_match(namelen, name, de)) {
1005                                 *res_dir = de;
1006                                 dx_release(frames);
1007                                 return bh;
1008                         }
1009                 }
1010                 brelse (bh);
1011                 /* Check to see if we should continue to search */
1012                 retval = ext3_htree_next_block(dir, hash, frame,
1013                                                frames, NULL);
1014                 if (retval < 0) {
1015                         ext3_warning(sb, __func__,
1016                              "error reading index page in directory #%lu",
1017                              dir->i_ino);
1018                         *err = retval;
1019                         goto errout;
1020                 }
1021         } while (retval == 1);
1022
1023         *err = -ENOENT;
1024 errout:
1025         dxtrace(printk("%s not found\n", name));
1026         dx_release (frames);
1027         return NULL;
1028 }
1029
1030 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1031 {
1032         struct inode * inode;
1033         struct ext3_dir_entry_2 * de;
1034         struct buffer_head * bh;
1035
1036         if (dentry->d_name.len > EXT3_NAME_LEN)
1037                 return ERR_PTR(-ENAMETOOLONG);
1038
1039         bh = ext3_find_entry(dentry, &de);
1040         inode = NULL;
1041         if (bh) {
1042                 unsigned long ino = le32_to_cpu(de->inode);
1043                 brelse (bh);
1044                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1045                         ext3_error(dir->i_sb, "ext3_lookup",
1046                                    "bad inode number: %lu", ino);
1047                         return ERR_PTR(-EIO);
1048                 }
1049                 inode = ext3_iget(dir->i_sb, ino);
1050                 if (IS_ERR(inode))
1051                         return ERR_CAST(inode);
1052         }
1053         return d_splice_alias(inode, dentry);
1054 }
1055
1056
1057 struct dentry *ext3_get_parent(struct dentry *child)
1058 {
1059         unsigned long ino;
1060         struct dentry *parent;
1061         struct inode *inode;
1062         struct dentry dotdot;
1063         struct ext3_dir_entry_2 * de;
1064         struct buffer_head *bh;
1065
1066         dotdot.d_name.name = "..";
1067         dotdot.d_name.len = 2;
1068         dotdot.d_parent = child; /* confusing, isn't it! */
1069
1070         bh = ext3_find_entry(&dotdot, &de);
1071         inode = NULL;
1072         if (!bh)
1073                 return ERR_PTR(-ENOENT);
1074         ino = le32_to_cpu(de->inode);
1075         brelse(bh);
1076
1077         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1078                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1079                            "bad inode number: %lu", ino);
1080                 return ERR_PTR(-EIO);
1081         }
1082
1083         inode = ext3_iget(child->d_inode->i_sb, ino);
1084         if (IS_ERR(inode))
1085                 return ERR_CAST(inode);
1086
1087         parent = d_alloc_anon(inode);
1088         if (!parent) {
1089                 iput(inode);
1090                 parent = ERR_PTR(-ENOMEM);
1091         }
1092         return parent;
1093 }
1094
1095 #define S_SHIFT 12
1096 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1097         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1098         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1099         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1100         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1101         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1102         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1103         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1104 };
1105
1106 static inline void ext3_set_de_type(struct super_block *sb,
1107                                 struct ext3_dir_entry_2 *de,
1108                                 umode_t mode) {
1109         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1110                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1111 }
1112
1113 /*
1114  * Move count entries from end of map between two memory locations.
1115  * Returns pointer to last entry moved.
1116  */
1117 static struct ext3_dir_entry_2 *
1118 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1119 {
1120         unsigned rec_len = 0;
1121
1122         while (count--) {
1123                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1124                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1125                 memcpy (to, de, rec_len);
1126                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1127                                 ext3_rec_len_to_disk(rec_len);
1128                 de->inode = 0;
1129                 map++;
1130                 to += rec_len;
1131         }
1132         return (struct ext3_dir_entry_2 *) (to - rec_len);
1133 }
1134
1135 /*
1136  * Compact each dir entry in the range to the minimal rec_len.
1137  * Returns pointer to last entry in range.
1138  */
1139 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1140 {
1141         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1142         unsigned rec_len = 0;
1143
1144         prev = to = de;
1145         while ((char*)de < base + size) {
1146                 next = ext3_next_entry(de);
1147                 if (de->inode && de->name_len) {
1148                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1149                         if (de > to)
1150                                 memmove(to, de, rec_len);
1151                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1152                         prev = to;
1153                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1154                 }
1155                 de = next;
1156         }
1157         return prev;
1158 }
1159
1160 /*
1161  * Split a full leaf block to make room for a new dir entry.
1162  * Allocate a new block, and move entries so that they are approx. equally full.
1163  * Returns pointer to de in block into which the new entry will be inserted.
1164  */
1165 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1166                         struct buffer_head **bh,struct dx_frame *frame,
1167                         struct dx_hash_info *hinfo, int *error)
1168 {
1169         unsigned blocksize = dir->i_sb->s_blocksize;
1170         unsigned count, continued;
1171         struct buffer_head *bh2;
1172         u32 newblock;
1173         u32 hash2;
1174         struct dx_map_entry *map;
1175         char *data1 = (*bh)->b_data, *data2;
1176         unsigned split, move, size, i;
1177         struct ext3_dir_entry_2 *de = NULL, *de2;
1178         int     err = 0;
1179
1180         bh2 = ext3_append (handle, dir, &newblock, &err);
1181         if (!(bh2)) {
1182                 brelse(*bh);
1183                 *bh = NULL;
1184                 goto errout;
1185         }
1186
1187         BUFFER_TRACE(*bh, "get_write_access");
1188         err = ext3_journal_get_write_access(handle, *bh);
1189         if (err)
1190                 goto journal_error;
1191
1192         BUFFER_TRACE(frame->bh, "get_write_access");
1193         err = ext3_journal_get_write_access(handle, frame->bh);
1194         if (err)
1195                 goto journal_error;
1196
1197         data2 = bh2->b_data;
1198
1199         /* create map in the end of data2 block */
1200         map = (struct dx_map_entry *) (data2 + blocksize);
1201         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1202                              blocksize, hinfo, map);
1203         map -= count;
1204         dx_sort_map (map, count);
1205         /* Split the existing block in the middle, size-wise */
1206         size = 0;
1207         move = 0;
1208         for (i = count-1; i >= 0; i--) {
1209                 /* is more than half of this entry in 2nd half of the block? */
1210                 if (size + map[i].size/2 > blocksize/2)
1211                         break;
1212                 size += map[i].size;
1213                 move++;
1214         }
1215         /* map index at which we will split */
1216         split = count - move;
1217         hash2 = map[split].hash;
1218         continued = hash2 == map[split - 1].hash;
1219         dxtrace(printk("Split block %i at %x, %i/%i\n",
1220                 dx_get_block(frame->at), hash2, split, count-split));
1221
1222         /* Fancy dance to stay within two buffers */
1223         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1224         de = dx_pack_dirents(data1,blocksize);
1225         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1226         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1227         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1228         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1229
1230         /* Which block gets the new entry? */
1231         if (hinfo->hash >= hash2)
1232         {
1233                 swap(*bh, bh2);
1234                 de = de2;
1235         }
1236         dx_insert_block (frame, hash2 + continued, newblock);
1237         err = ext3_journal_dirty_metadata (handle, bh2);
1238         if (err)
1239                 goto journal_error;
1240         err = ext3_journal_dirty_metadata (handle, frame->bh);
1241         if (err)
1242                 goto journal_error;
1243         brelse (bh2);
1244         dxtrace(dx_show_index ("frame", frame->entries));
1245         return de;
1246
1247 journal_error:
1248         brelse(*bh);
1249         brelse(bh2);
1250         *bh = NULL;
1251         ext3_std_error(dir->i_sb, err);
1252 errout:
1253         *error = err;
1254         return NULL;
1255 }
1256
1257
1258 /*
1259  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1260  * it points to a directory entry which is guaranteed to be large
1261  * enough for new directory entry.  If de is NULL, then
1262  * add_dirent_to_buf will attempt search the directory block for
1263  * space.  It will return -ENOSPC if no space is available, and -EIO
1264  * and -EEXIST if directory entry already exists.
1265  *
1266  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1267  * all other cases bh is released.
1268  */
1269 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1270                              struct inode *inode, struct ext3_dir_entry_2 *de,
1271                              struct buffer_head * bh)
1272 {
1273         struct inode    *dir = dentry->d_parent->d_inode;
1274         const char      *name = dentry->d_name.name;
1275         int             namelen = dentry->d_name.len;
1276         unsigned long   offset = 0;
1277         unsigned short  reclen;
1278         int             nlen, rlen, err;
1279         char            *top;
1280
1281         reclen = EXT3_DIR_REC_LEN(namelen);
1282         if (!de) {
1283                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1284                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1285                 while ((char *) de <= top) {
1286                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1287                                                   bh, offset)) {
1288                                 brelse (bh);
1289                                 return -EIO;
1290                         }
1291                         if (ext3_match (namelen, name, de)) {
1292                                 brelse (bh);
1293                                 return -EEXIST;
1294                         }
1295                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1296                         rlen = ext3_rec_len_from_disk(de->rec_len);
1297                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1298                                 break;
1299                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1300                         offset += rlen;
1301                 }
1302                 if ((char *) de > top)
1303                         return -ENOSPC;
1304         }
1305         BUFFER_TRACE(bh, "get_write_access");
1306         err = ext3_journal_get_write_access(handle, bh);
1307         if (err) {
1308                 ext3_std_error(dir->i_sb, err);
1309                 brelse(bh);
1310                 return err;
1311         }
1312
1313         /* By now the buffer is marked for journaling */
1314         nlen = EXT3_DIR_REC_LEN(de->name_len);
1315         rlen = ext3_rec_len_from_disk(de->rec_len);
1316         if (de->inode) {
1317                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1318                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1319                 de->rec_len = ext3_rec_len_to_disk(nlen);
1320                 de = de1;
1321         }
1322         de->file_type = EXT3_FT_UNKNOWN;
1323         if (inode) {
1324                 de->inode = cpu_to_le32(inode->i_ino);
1325                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1326         } else
1327                 de->inode = 0;
1328         de->name_len = namelen;
1329         memcpy (de->name, name, namelen);
1330         /*
1331          * XXX shouldn't update any times until successful
1332          * completion of syscall, but too many callers depend
1333          * on this.
1334          *
1335          * XXX similarly, too many callers depend on
1336          * ext3_new_inode() setting the times, but error
1337          * recovery deletes the inode, so the worst that can
1338          * happen is that the times are slightly out of date
1339          * and/or different from the directory change time.
1340          */
1341         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1342         ext3_update_dx_flag(dir);
1343         dir->i_version++;
1344         ext3_mark_inode_dirty(handle, dir);
1345         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1346         err = ext3_journal_dirty_metadata(handle, bh);
1347         if (err)
1348                 ext3_std_error(dir->i_sb, err);
1349         brelse(bh);
1350         return 0;
1351 }
1352
1353 /*
1354  * This converts a one block unindexed directory to a 3 block indexed
1355  * directory, and adds the dentry to the indexed directory.
1356  */
1357 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1358                             struct inode *inode, struct buffer_head *bh)
1359 {
1360         struct inode    *dir = dentry->d_parent->d_inode;
1361         const char      *name = dentry->d_name.name;
1362         int             namelen = dentry->d_name.len;
1363         struct buffer_head *bh2;
1364         struct dx_root  *root;
1365         struct dx_frame frames[2], *frame;
1366         struct dx_entry *entries;
1367         struct ext3_dir_entry_2 *de, *de2;
1368         char            *data1, *top;
1369         unsigned        len;
1370         int             retval;
1371         unsigned        blocksize;
1372         struct dx_hash_info hinfo;
1373         u32             block;
1374         struct fake_dirent *fde;
1375
1376         blocksize =  dir->i_sb->s_blocksize;
1377         dxtrace(printk("Creating index\n"));
1378         retval = ext3_journal_get_write_access(handle, bh);
1379         if (retval) {
1380                 ext3_std_error(dir->i_sb, retval);
1381                 brelse(bh);
1382                 return retval;
1383         }
1384         root = (struct dx_root *) bh->b_data;
1385
1386         bh2 = ext3_append (handle, dir, &block, &retval);
1387         if (!(bh2)) {
1388                 brelse(bh);
1389                 return retval;
1390         }
1391         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1392         data1 = bh2->b_data;
1393
1394         /* The 0th block becomes the root, move the dirents out */
1395         fde = &root->dotdot;
1396         de = (struct ext3_dir_entry_2 *)((char *)fde +
1397                         ext3_rec_len_from_disk(fde->rec_len));
1398         len = ((char *) root) + blocksize - (char *) de;
1399         memcpy (data1, de, len);
1400         de = (struct ext3_dir_entry_2 *) data1;
1401         top = data1 + len;
1402         while ((char *)(de2 = ext3_next_entry(de)) < top)
1403                 de = de2;
1404         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1405         /* Initialize the root; the dot dirents already exist */
1406         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1407         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1408         memset (&root->info, 0, sizeof(root->info));
1409         root->info.info_length = sizeof(root->info);
1410         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1411         entries = root->entries;
1412         dx_set_block (entries, 1);
1413         dx_set_count (entries, 1);
1414         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1415
1416         /* Initialize as for dx_probe */
1417         hinfo.hash_version = root->info.hash_version;
1418         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1419         ext3fs_dirhash(name, namelen, &hinfo);
1420         frame = frames;
1421         frame->entries = entries;
1422         frame->at = entries;
1423         frame->bh = bh;
1424         bh = bh2;
1425         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1426         dx_release (frames);
1427         if (!(de))
1428                 return retval;
1429
1430         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1431 }
1432
1433 /*
1434  *      ext3_add_entry()
1435  *
1436  * adds a file entry to the specified directory, using the same
1437  * semantics as ext3_find_entry(). It returns NULL if it failed.
1438  *
1439  * NOTE!! The inode part of 'de' is left at 0 - which means you
1440  * may not sleep between calling this and putting something into
1441  * the entry, as someone else might have used it while you slept.
1442  */
1443 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1444         struct inode *inode)
1445 {
1446         struct inode *dir = dentry->d_parent->d_inode;
1447         unsigned long offset;
1448         struct buffer_head * bh;
1449         struct ext3_dir_entry_2 *de;
1450         struct super_block * sb;
1451         int     retval;
1452         int     dx_fallback=0;
1453         unsigned blocksize;
1454         u32 block, blocks;
1455
1456         sb = dir->i_sb;
1457         blocksize = sb->s_blocksize;
1458         if (!dentry->d_name.len)
1459                 return -EINVAL;
1460         if (is_dx(dir)) {
1461                 retval = ext3_dx_add_entry(handle, dentry, inode);
1462                 if (!retval || (retval != ERR_BAD_DX_DIR))
1463                         return retval;
1464                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1465                 dx_fallback++;
1466                 ext3_mark_inode_dirty(handle, dir);
1467         }
1468         blocks = dir->i_size >> sb->s_blocksize_bits;
1469         for (block = 0, offset = 0; block < blocks; block++) {
1470                 bh = ext3_bread(handle, dir, block, 0, &retval);
1471                 if(!bh)
1472                         return retval;
1473                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1474                 if (retval != -ENOSPC)
1475                         return retval;
1476
1477                 if (blocks == 1 && !dx_fallback &&
1478                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1479                         return make_indexed_dir(handle, dentry, inode, bh);
1480                 brelse(bh);
1481         }
1482         bh = ext3_append(handle, dir, &block, &retval);
1483         if (!bh)
1484                 return retval;
1485         de = (struct ext3_dir_entry_2 *) bh->b_data;
1486         de->inode = 0;
1487         de->rec_len = ext3_rec_len_to_disk(blocksize);
1488         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1489 }
1490
1491 /*
1492  * Returns 0 for success, or a negative error value
1493  */
1494 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1495                              struct inode *inode)
1496 {
1497         struct dx_frame frames[2], *frame;
1498         struct dx_entry *entries, *at;
1499         struct dx_hash_info hinfo;
1500         struct buffer_head * bh;
1501         struct inode *dir = dentry->d_parent->d_inode;
1502         struct super_block * sb = dir->i_sb;
1503         struct ext3_dir_entry_2 *de;
1504         int err;
1505
1506         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1507         if (!frame)
1508                 return err;
1509         entries = frame->entries;
1510         at = frame->at;
1511
1512         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1513                 goto cleanup;
1514
1515         BUFFER_TRACE(bh, "get_write_access");
1516         err = ext3_journal_get_write_access(handle, bh);
1517         if (err)
1518                 goto journal_error;
1519
1520         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1521         if (err != -ENOSPC) {
1522                 bh = NULL;
1523                 goto cleanup;
1524         }
1525
1526         /* Block full, should compress but for now just split */
1527         dxtrace(printk("using %u of %u node entries\n",
1528                        dx_get_count(entries), dx_get_limit(entries)));
1529         /* Need to split index? */
1530         if (dx_get_count(entries) == dx_get_limit(entries)) {
1531                 u32 newblock;
1532                 unsigned icount = dx_get_count(entries);
1533                 int levels = frame - frames;
1534                 struct dx_entry *entries2;
1535                 struct dx_node *node2;
1536                 struct buffer_head *bh2;
1537
1538                 if (levels && (dx_get_count(frames->entries) ==
1539                                dx_get_limit(frames->entries))) {
1540                         ext3_warning(sb, __func__,
1541                                      "Directory index full!");
1542                         err = -ENOSPC;
1543                         goto cleanup;
1544                 }
1545                 bh2 = ext3_append (handle, dir, &newblock, &err);
1546                 if (!(bh2))
1547                         goto cleanup;
1548                 node2 = (struct dx_node *)(bh2->b_data);
1549                 entries2 = node2->entries;
1550                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1551                 node2->fake.inode = 0;
1552                 BUFFER_TRACE(frame->bh, "get_write_access");
1553                 err = ext3_journal_get_write_access(handle, frame->bh);
1554                 if (err)
1555                         goto journal_error;
1556                 if (levels) {
1557                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1558                         unsigned hash2 = dx_get_hash(entries + icount1);
1559                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1560
1561                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1562                         err = ext3_journal_get_write_access(handle,
1563                                                              frames[0].bh);
1564                         if (err)
1565                                 goto journal_error;
1566
1567                         memcpy ((char *) entries2, (char *) (entries + icount1),
1568                                 icount2 * sizeof(struct dx_entry));
1569                         dx_set_count (entries, icount1);
1570                         dx_set_count (entries2, icount2);
1571                         dx_set_limit (entries2, dx_node_limit(dir));
1572
1573                         /* Which index block gets the new entry? */
1574                         if (at - entries >= icount1) {
1575                                 frame->at = at = at - entries - icount1 + entries2;
1576                                 frame->entries = entries = entries2;
1577                                 swap(frame->bh, bh2);
1578                         }
1579                         dx_insert_block (frames + 0, hash2, newblock);
1580                         dxtrace(dx_show_index ("node", frames[1].entries));
1581                         dxtrace(dx_show_index ("node",
1582                                ((struct dx_node *) bh2->b_data)->entries));
1583                         err = ext3_journal_dirty_metadata(handle, bh2);
1584                         if (err)
1585                                 goto journal_error;
1586                         brelse (bh2);
1587                 } else {
1588                         dxtrace(printk("Creating second level index...\n"));
1589                         memcpy((char *) entries2, (char *) entries,
1590                                icount * sizeof(struct dx_entry));
1591                         dx_set_limit(entries2, dx_node_limit(dir));
1592
1593                         /* Set up root */
1594                         dx_set_count(entries, 1);
1595                         dx_set_block(entries + 0, newblock);
1596                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1597
1598                         /* Add new access path frame */
1599                         frame = frames + 1;
1600                         frame->at = at = at - entries + entries2;
1601                         frame->entries = entries = entries2;
1602                         frame->bh = bh2;
1603                         err = ext3_journal_get_write_access(handle,
1604                                                              frame->bh);
1605                         if (err)
1606                                 goto journal_error;
1607                 }
1608                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1609         }
1610         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1611         if (!de)
1612                 goto cleanup;
1613         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1614         bh = NULL;
1615         goto cleanup;
1616
1617 journal_error:
1618         ext3_std_error(dir->i_sb, err);
1619 cleanup:
1620         if (bh)
1621                 brelse(bh);
1622         dx_release(frames);
1623         return err;
1624 }
1625
1626 /*
1627  * ext3_delete_entry deletes a directory entry by merging it with the
1628  * previous entry
1629  */
1630 static int ext3_delete_entry (handle_t *handle,
1631                               struct inode * dir,
1632                               struct ext3_dir_entry_2 * de_del,
1633                               struct buffer_head * bh)
1634 {
1635         struct ext3_dir_entry_2 * de, * pde;
1636         int i;
1637
1638         i = 0;
1639         pde = NULL;
1640         de = (struct ext3_dir_entry_2 *) bh->b_data;
1641         while (i < bh->b_size) {
1642                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1643                         return -EIO;
1644                 if (de == de_del)  {
1645                         BUFFER_TRACE(bh, "get_write_access");
1646                         ext3_journal_get_write_access(handle, bh);
1647                         if (pde)
1648                                 pde->rec_len = ext3_rec_len_to_disk(
1649                                         ext3_rec_len_from_disk(pde->rec_len) +
1650                                         ext3_rec_len_from_disk(de->rec_len));
1651                         else
1652                                 de->inode = 0;
1653                         dir->i_version++;
1654                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1655                         ext3_journal_dirty_metadata(handle, bh);
1656                         return 0;
1657                 }
1658                 i += ext3_rec_len_from_disk(de->rec_len);
1659                 pde = de;
1660                 de = ext3_next_entry(de);
1661         }
1662         return -ENOENT;
1663 }
1664
1665 static int ext3_add_nondir(handle_t *handle,
1666                 struct dentry *dentry, struct inode *inode)
1667 {
1668         int err = ext3_add_entry(handle, dentry, inode);
1669         if (!err) {
1670                 ext3_mark_inode_dirty(handle, inode);
1671                 d_instantiate(dentry, inode);
1672                 return 0;
1673         }
1674         drop_nlink(inode);
1675         iput(inode);
1676         return err;
1677 }
1678
1679 /*
1680  * By the time this is called, we already have created
1681  * the directory cache entry for the new file, but it
1682  * is so far negative - it has no inode.
1683  *
1684  * If the create succeeds, we fill in the inode information
1685  * with d_instantiate().
1686  */
1687 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1688                 struct nameidata *nd)
1689 {
1690         handle_t *handle;
1691         struct inode * inode;
1692         int err, retries = 0;
1693
1694 retry:
1695         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1696                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1697                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1698         if (IS_ERR(handle))
1699                 return PTR_ERR(handle);
1700
1701         if (IS_DIRSYNC(dir))
1702                 handle->h_sync = 1;
1703
1704         inode = ext3_new_inode (handle, dir, mode);
1705         err = PTR_ERR(inode);
1706         if (!IS_ERR(inode)) {
1707                 inode->i_op = &ext3_file_inode_operations;
1708                 inode->i_fop = &ext3_file_operations;
1709                 ext3_set_aops(inode);
1710                 err = ext3_add_nondir(handle, dentry, inode);
1711         }
1712         ext3_journal_stop(handle);
1713         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1714                 goto retry;
1715         return err;
1716 }
1717
1718 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1719                         int mode, dev_t rdev)
1720 {
1721         handle_t *handle;
1722         struct inode *inode;
1723         int err, retries = 0;
1724
1725         if (!new_valid_dev(rdev))
1726                 return -EINVAL;
1727
1728 retry:
1729         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1730                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1731                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1732         if (IS_ERR(handle))
1733                 return PTR_ERR(handle);
1734
1735         if (IS_DIRSYNC(dir))
1736                 handle->h_sync = 1;
1737
1738         inode = ext3_new_inode (handle, dir, mode);
1739         err = PTR_ERR(inode);
1740         if (!IS_ERR(inode)) {
1741                 init_special_inode(inode, inode->i_mode, rdev);
1742 #ifdef CONFIG_EXT3_FS_XATTR
1743                 inode->i_op = &ext3_special_inode_operations;
1744 #endif
1745                 err = ext3_add_nondir(handle, dentry, inode);
1746         }
1747         ext3_journal_stop(handle);
1748         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1749                 goto retry;
1750         return err;
1751 }
1752
1753 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1754 {
1755         handle_t *handle;
1756         struct inode * inode;
1757         struct buffer_head * dir_block;
1758         struct ext3_dir_entry_2 * de;
1759         int err, retries = 0;
1760
1761         if (dir->i_nlink >= EXT3_LINK_MAX)
1762                 return -EMLINK;
1763
1764 retry:
1765         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1766                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1767                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1768         if (IS_ERR(handle))
1769                 return PTR_ERR(handle);
1770
1771         if (IS_DIRSYNC(dir))
1772                 handle->h_sync = 1;
1773
1774         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1775         err = PTR_ERR(inode);
1776         if (IS_ERR(inode))
1777                 goto out_stop;
1778
1779         inode->i_op = &ext3_dir_inode_operations;
1780         inode->i_fop = &ext3_dir_operations;
1781         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1782         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1783         if (!dir_block) {
1784                 drop_nlink(inode); /* is this nlink == 0? */
1785                 ext3_mark_inode_dirty(handle, inode);
1786                 iput (inode);
1787                 goto out_stop;
1788         }
1789         BUFFER_TRACE(dir_block, "get_write_access");
1790         ext3_journal_get_write_access(handle, dir_block);
1791         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1792         de->inode = cpu_to_le32(inode->i_ino);
1793         de->name_len = 1;
1794         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1795         strcpy (de->name, ".");
1796         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1797         de = ext3_next_entry(de);
1798         de->inode = cpu_to_le32(dir->i_ino);
1799         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1800                                         EXT3_DIR_REC_LEN(1));
1801         de->name_len = 2;
1802         strcpy (de->name, "..");
1803         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1804         inode->i_nlink = 2;
1805         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1806         ext3_journal_dirty_metadata(handle, dir_block);
1807         brelse (dir_block);
1808         ext3_mark_inode_dirty(handle, inode);
1809         err = ext3_add_entry (handle, dentry, inode);
1810         if (err) {
1811                 inode->i_nlink = 0;
1812                 ext3_mark_inode_dirty(handle, inode);
1813                 iput (inode);
1814                 goto out_stop;
1815         }
1816         inc_nlink(dir);
1817         ext3_update_dx_flag(dir);
1818         ext3_mark_inode_dirty(handle, dir);
1819         d_instantiate(dentry, inode);
1820 out_stop:
1821         ext3_journal_stop(handle);
1822         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1823                 goto retry;
1824         return err;
1825 }
1826
1827 /*
1828  * routine to check that the specified directory is empty (for rmdir)
1829  */
1830 static int empty_dir (struct inode * inode)
1831 {
1832         unsigned long offset;
1833         struct buffer_head * bh;
1834         struct ext3_dir_entry_2 * de, * de1;
1835         struct super_block * sb;
1836         int err = 0;
1837
1838         sb = inode->i_sb;
1839         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1840             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1841                 if (err)
1842                         ext3_error(inode->i_sb, __func__,
1843                                    "error %d reading directory #%lu offset 0",
1844                                    err, inode->i_ino);
1845                 else
1846                         ext3_warning(inode->i_sb, __func__,
1847                                      "bad directory (dir #%lu) - no data block",
1848                                      inode->i_ino);
1849                 return 1;
1850         }
1851         de = (struct ext3_dir_entry_2 *) bh->b_data;
1852         de1 = ext3_next_entry(de);
1853         if (le32_to_cpu(de->inode) != inode->i_ino ||
1854                         !le32_to_cpu(de1->inode) ||
1855                         strcmp (".", de->name) ||
1856                         strcmp ("..", de1->name)) {
1857                 ext3_warning (inode->i_sb, "empty_dir",
1858                               "bad directory (dir #%lu) - no `.' or `..'",
1859                               inode->i_ino);
1860                 brelse (bh);
1861                 return 1;
1862         }
1863         offset = ext3_rec_len_from_disk(de->rec_len) +
1864                         ext3_rec_len_from_disk(de1->rec_len);
1865         de = ext3_next_entry(de1);
1866         while (offset < inode->i_size ) {
1867                 if (!bh ||
1868                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1869                         err = 0;
1870                         brelse (bh);
1871                         bh = ext3_bread (NULL, inode,
1872                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1873                         if (!bh) {
1874                                 if (err)
1875                                         ext3_error(sb, __func__,
1876                                                    "error %d reading directory"
1877                                                    " #%lu offset %lu",
1878                                                    err, inode->i_ino, offset);
1879                                 offset += sb->s_blocksize;
1880                                 continue;
1881                         }
1882                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1883                 }
1884                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1885                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1886                                                          sb->s_blocksize);
1887                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1888                         continue;
1889                 }
1890                 if (le32_to_cpu(de->inode)) {
1891                         brelse (bh);
1892                         return 0;
1893                 }
1894                 offset += ext3_rec_len_from_disk(de->rec_len);
1895                 de = ext3_next_entry(de);
1896         }
1897         brelse (bh);
1898         return 1;
1899 }
1900
1901 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1902  * such inodes, starting at the superblock, in case we crash before the
1903  * file is closed/deleted, or in case the inode truncate spans multiple
1904  * transactions and the last transaction is not recovered after a crash.
1905  *
1906  * At filesystem recovery time, we walk this list deleting unlinked
1907  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1908  */
1909 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1910 {
1911         struct super_block *sb = inode->i_sb;
1912         struct ext3_iloc iloc;
1913         int err = 0, rc;
1914
1915         lock_super(sb);
1916         if (!list_empty(&EXT3_I(inode)->i_orphan))
1917                 goto out_unlock;
1918
1919         /* Orphan handling is only valid for files with data blocks
1920          * being truncated, or files being unlinked. */
1921
1922         /* @@@ FIXME: Observation from aviro:
1923          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1924          * here (on lock_super()), so race with ext3_link() which might bump
1925          * ->i_nlink. For, say it, character device. Not a regular file,
1926          * not a directory, not a symlink and ->i_nlink > 0.
1927          */
1928         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1929                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1930
1931         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1932         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1933         if (err)
1934                 goto out_unlock;
1935
1936         err = ext3_reserve_inode_write(handle, inode, &iloc);
1937         if (err)
1938                 goto out_unlock;
1939
1940         /* Insert this inode at the head of the on-disk orphan list... */
1941         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1942         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1943         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1944         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1945         if (!err)
1946                 err = rc;
1947
1948         /* Only add to the head of the in-memory list if all the
1949          * previous operations succeeded.  If the orphan_add is going to
1950          * fail (possibly taking the journal offline), we can't risk
1951          * leaving the inode on the orphan list: stray orphan-list
1952          * entries can cause panics at unmount time.
1953          *
1954          * This is safe: on error we're going to ignore the orphan list
1955          * anyway on the next recovery. */
1956         if (!err)
1957                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1958
1959         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1960         jbd_debug(4, "orphan inode %lu will point to %d\n",
1961                         inode->i_ino, NEXT_ORPHAN(inode));
1962 out_unlock:
1963         unlock_super(sb);
1964         ext3_std_error(inode->i_sb, err);
1965         return err;
1966 }
1967
1968 /*
1969  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1970  * of such inodes stored on disk, because it is finally being cleaned up.
1971  */
1972 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1973 {
1974         struct list_head *prev;
1975         struct ext3_inode_info *ei = EXT3_I(inode);
1976         struct ext3_sb_info *sbi;
1977         unsigned long ino_next;
1978         struct ext3_iloc iloc;
1979         int err = 0;
1980
1981         lock_super(inode->i_sb);
1982         if (list_empty(&ei->i_orphan)) {
1983                 unlock_super(inode->i_sb);
1984                 return 0;
1985         }
1986
1987         ino_next = NEXT_ORPHAN(inode);
1988         prev = ei->i_orphan.prev;
1989         sbi = EXT3_SB(inode->i_sb);
1990
1991         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1992
1993         list_del_init(&ei->i_orphan);
1994
1995         /* If we're on an error path, we may not have a valid
1996          * transaction handle with which to update the orphan list on
1997          * disk, but we still need to remove the inode from the linked
1998          * list in memory. */
1999         if (!handle)
2000                 goto out;
2001
2002         err = ext3_reserve_inode_write(handle, inode, &iloc);
2003         if (err)
2004                 goto out_err;
2005
2006         if (prev == &sbi->s_orphan) {
2007                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2008                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2009                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2010                 if (err)
2011                         goto out_brelse;
2012                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2013                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2014         } else {
2015                 struct ext3_iloc iloc2;
2016                 struct inode *i_prev =
2017                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2018
2019                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2020                           i_prev->i_ino, ino_next);
2021                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2022                 if (err)
2023                         goto out_brelse;
2024                 NEXT_ORPHAN(i_prev) = ino_next;
2025                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2026         }
2027         if (err)
2028                 goto out_brelse;
2029         NEXT_ORPHAN(inode) = 0;
2030         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2031
2032 out_err:
2033         ext3_std_error(inode->i_sb, err);
2034 out:
2035         unlock_super(inode->i_sb);
2036         return err;
2037
2038 out_brelse:
2039         brelse(iloc.bh);
2040         goto out_err;
2041 }
2042
2043 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2044 {
2045         int retval;
2046         struct inode * inode;
2047         struct buffer_head * bh;
2048         struct ext3_dir_entry_2 * de;
2049         handle_t *handle;
2050
2051         /* Initialize quotas before so that eventual writes go in
2052          * separate transaction */
2053         DQUOT_INIT(dentry->d_inode);
2054         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2055         if (IS_ERR(handle))
2056                 return PTR_ERR(handle);
2057
2058         retval = -ENOENT;
2059         bh = ext3_find_entry (dentry, &de);
2060         if (!bh)
2061                 goto end_rmdir;
2062
2063         if (IS_DIRSYNC(dir))
2064                 handle->h_sync = 1;
2065
2066         inode = dentry->d_inode;
2067
2068         retval = -EIO;
2069         if (le32_to_cpu(de->inode) != inode->i_ino)
2070                 goto end_rmdir;
2071
2072         retval = -ENOTEMPTY;
2073         if (!empty_dir (inode))
2074                 goto end_rmdir;
2075
2076         retval = ext3_delete_entry(handle, dir, de, bh);
2077         if (retval)
2078                 goto end_rmdir;
2079         if (inode->i_nlink != 2)
2080                 ext3_warning (inode->i_sb, "ext3_rmdir",
2081                               "empty directory has nlink!=2 (%d)",
2082                               inode->i_nlink);
2083         inode->i_version++;
2084         clear_nlink(inode);
2085         /* There's no need to set i_disksize: the fact that i_nlink is
2086          * zero will ensure that the right thing happens during any
2087          * recovery. */
2088         inode->i_size = 0;
2089         ext3_orphan_add(handle, inode);
2090         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2091         ext3_mark_inode_dirty(handle, inode);
2092         drop_nlink(dir);
2093         ext3_update_dx_flag(dir);
2094         ext3_mark_inode_dirty(handle, dir);
2095
2096 end_rmdir:
2097         ext3_journal_stop(handle);
2098         brelse (bh);
2099         return retval;
2100 }
2101
2102 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2103 {
2104         int retval;
2105         struct inode * inode;
2106         struct buffer_head * bh;
2107         struct ext3_dir_entry_2 * de;
2108         handle_t *handle;
2109
2110         /* Initialize quotas before so that eventual writes go
2111          * in separate transaction */
2112         DQUOT_INIT(dentry->d_inode);
2113         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2114         if (IS_ERR(handle))
2115                 return PTR_ERR(handle);
2116
2117         if (IS_DIRSYNC(dir))
2118                 handle->h_sync = 1;
2119
2120         retval = -ENOENT;
2121         bh = ext3_find_entry (dentry, &de);
2122         if (!bh)
2123                 goto end_unlink;
2124
2125         inode = dentry->d_inode;
2126
2127         retval = -EIO;
2128         if (le32_to_cpu(de->inode) != inode->i_ino)
2129                 goto end_unlink;
2130
2131         if (!inode->i_nlink) {
2132                 ext3_warning (inode->i_sb, "ext3_unlink",
2133                               "Deleting nonexistent file (%lu), %d",
2134                               inode->i_ino, inode->i_nlink);
2135                 inode->i_nlink = 1;
2136         }
2137         retval = ext3_delete_entry(handle, dir, de, bh);
2138         if (retval)
2139                 goto end_unlink;
2140         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2141         ext3_update_dx_flag(dir);
2142         ext3_mark_inode_dirty(handle, dir);
2143         drop_nlink(inode);
2144         if (!inode->i_nlink)
2145                 ext3_orphan_add(handle, inode);
2146         inode->i_ctime = dir->i_ctime;
2147         ext3_mark_inode_dirty(handle, inode);
2148         retval = 0;
2149
2150 end_unlink:
2151         ext3_journal_stop(handle);
2152         brelse (bh);
2153         return retval;
2154 }
2155
2156 static int ext3_symlink (struct inode * dir,
2157                 struct dentry *dentry, const char * symname)
2158 {
2159         handle_t *handle;
2160         struct inode * inode;
2161         int l, err, retries = 0;
2162
2163         l = strlen(symname)+1;
2164         if (l > dir->i_sb->s_blocksize)
2165                 return -ENAMETOOLONG;
2166
2167 retry:
2168         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2169                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2170                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2171         if (IS_ERR(handle))
2172                 return PTR_ERR(handle);
2173
2174         if (IS_DIRSYNC(dir))
2175                 handle->h_sync = 1;
2176
2177         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2178         err = PTR_ERR(inode);
2179         if (IS_ERR(inode))
2180                 goto out_stop;
2181
2182         if (l > sizeof (EXT3_I(inode)->i_data)) {
2183                 inode->i_op = &ext3_symlink_inode_operations;
2184                 ext3_set_aops(inode);
2185                 /*
2186                  * page_symlink() calls into ext3_prepare/commit_write.
2187                  * We have a transaction open.  All is sweetness.  It also sets
2188                  * i_size in generic_commit_write().
2189                  */
2190                 err = __page_symlink(inode, symname, l,
2191                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2192                 if (err) {
2193                         drop_nlink(inode);
2194                         ext3_mark_inode_dirty(handle, inode);
2195                         iput (inode);
2196                         goto out_stop;
2197                 }
2198         } else {
2199                 inode->i_op = &ext3_fast_symlink_inode_operations;
2200                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2201                 inode->i_size = l-1;
2202         }
2203         EXT3_I(inode)->i_disksize = inode->i_size;
2204         err = ext3_add_nondir(handle, dentry, inode);
2205 out_stop:
2206         ext3_journal_stop(handle);
2207         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2208                 goto retry;
2209         return err;
2210 }
2211
2212 static int ext3_link (struct dentry * old_dentry,
2213                 struct inode * dir, struct dentry *dentry)
2214 {
2215         handle_t *handle;
2216         struct inode *inode = old_dentry->d_inode;
2217         int err, retries = 0;
2218
2219         if (inode->i_nlink >= EXT3_LINK_MAX)
2220                 return -EMLINK;
2221         /*
2222          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2223          * otherwise has the potential to corrupt the orphan inode list.
2224          */
2225         if (inode->i_nlink == 0)
2226                 return -ENOENT;
2227
2228 retry:
2229         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2230                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2231         if (IS_ERR(handle))
2232                 return PTR_ERR(handle);
2233
2234         if (IS_DIRSYNC(dir))
2235                 handle->h_sync = 1;
2236
2237         inode->i_ctime = CURRENT_TIME_SEC;
2238         inc_nlink(inode);
2239         atomic_inc(&inode->i_count);
2240
2241         err = ext3_add_nondir(handle, dentry, inode);
2242         ext3_journal_stop(handle);
2243         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2244                 goto retry;
2245         return err;
2246 }
2247
2248 #define PARENT_INO(buffer) \
2249         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2250
2251 /*
2252  * Anybody can rename anything with this: the permission checks are left to the
2253  * higher-level routines.
2254  */
2255 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2256                            struct inode * new_dir,struct dentry *new_dentry)
2257 {
2258         handle_t *handle;
2259         struct inode * old_inode, * new_inode;
2260         struct buffer_head * old_bh, * new_bh, * dir_bh;
2261         struct ext3_dir_entry_2 * old_de, * new_de;
2262         int retval;
2263
2264         old_bh = new_bh = dir_bh = NULL;
2265
2266         /* Initialize quotas before so that eventual writes go
2267          * in separate transaction */
2268         if (new_dentry->d_inode)
2269                 DQUOT_INIT(new_dentry->d_inode);
2270         handle = ext3_journal_start(old_dir, 2 *
2271                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2272                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2273         if (IS_ERR(handle))
2274                 return PTR_ERR(handle);
2275
2276         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2277                 handle->h_sync = 1;
2278
2279         old_bh = ext3_find_entry (old_dentry, &old_de);
2280         /*
2281          *  Check for inode number is _not_ due to possible IO errors.
2282          *  We might rmdir the source, keep it as pwd of some process
2283          *  and merrily kill the link to whatever was created under the
2284          *  same name. Goodbye sticky bit ;-<
2285          */
2286         old_inode = old_dentry->d_inode;
2287         retval = -ENOENT;
2288         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2289                 goto end_rename;
2290
2291         new_inode = new_dentry->d_inode;
2292         new_bh = ext3_find_entry (new_dentry, &new_de);
2293         if (new_bh) {
2294                 if (!new_inode) {
2295                         brelse (new_bh);
2296                         new_bh = NULL;
2297                 }
2298         }
2299         if (S_ISDIR(old_inode->i_mode)) {
2300                 if (new_inode) {
2301                         retval = -ENOTEMPTY;
2302                         if (!empty_dir (new_inode))
2303                                 goto end_rename;
2304                 }
2305                 retval = -EIO;
2306                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2307                 if (!dir_bh)
2308                         goto end_rename;
2309                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2310                         goto end_rename;
2311                 retval = -EMLINK;
2312                 if (!new_inode && new_dir!=old_dir &&
2313                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2314                         goto end_rename;
2315         }
2316         if (!new_bh) {
2317                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2318                 if (retval)
2319                         goto end_rename;
2320         } else {
2321                 BUFFER_TRACE(new_bh, "get write access");
2322                 ext3_journal_get_write_access(handle, new_bh);
2323                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2324                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2325                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2326                         new_de->file_type = old_de->file_type;
2327                 new_dir->i_version++;
2328                 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2329                 ext3_mark_inode_dirty(handle, new_dir);
2330                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2331                 ext3_journal_dirty_metadata(handle, new_bh);
2332                 brelse(new_bh);
2333                 new_bh = NULL;
2334         }
2335
2336         /*
2337          * Like most other Unix systems, set the ctime for inodes on a
2338          * rename.
2339          */
2340         old_inode->i_ctime = CURRENT_TIME_SEC;
2341         ext3_mark_inode_dirty(handle, old_inode);
2342
2343         /*
2344          * ok, that's it
2345          */
2346         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2347             old_de->name_len != old_dentry->d_name.len ||
2348             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2349             (retval = ext3_delete_entry(handle, old_dir,
2350                                         old_de, old_bh)) == -ENOENT) {
2351                 /* old_de could have moved from under us during htree split, so
2352                  * make sure that we are deleting the right entry.  We might
2353                  * also be pointing to a stale entry in the unused part of
2354                  * old_bh so just checking inum and the name isn't enough. */
2355                 struct buffer_head *old_bh2;
2356                 struct ext3_dir_entry_2 *old_de2;
2357
2358                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2359                 if (old_bh2) {
2360                         retval = ext3_delete_entry(handle, old_dir,
2361                                                    old_de2, old_bh2);
2362                         brelse(old_bh2);
2363                 }
2364         }
2365         if (retval) {
2366                 ext3_warning(old_dir->i_sb, "ext3_rename",
2367                                 "Deleting old file (%lu), %d, error=%d",
2368                                 old_dir->i_ino, old_dir->i_nlink, retval);
2369         }
2370
2371         if (new_inode) {
2372                 drop_nlink(new_inode);
2373                 new_inode->i_ctime = CURRENT_TIME_SEC;
2374         }
2375         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2376         ext3_update_dx_flag(old_dir);
2377         if (dir_bh) {
2378                 BUFFER_TRACE(dir_bh, "get_write_access");
2379                 ext3_journal_get_write_access(handle, dir_bh);
2380                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2381                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2382                 ext3_journal_dirty_metadata(handle, dir_bh);
2383                 drop_nlink(old_dir);
2384                 if (new_inode) {
2385                         drop_nlink(new_inode);
2386                 } else {
2387                         inc_nlink(new_dir);
2388                         ext3_update_dx_flag(new_dir);
2389                         ext3_mark_inode_dirty(handle, new_dir);
2390                 }
2391         }
2392         ext3_mark_inode_dirty(handle, old_dir);
2393         if (new_inode) {
2394                 ext3_mark_inode_dirty(handle, new_inode);
2395                 if (!new_inode->i_nlink)
2396                         ext3_orphan_add(handle, new_inode);
2397         }
2398         retval = 0;
2399
2400 end_rename:
2401         brelse (dir_bh);
2402         brelse (old_bh);
2403         brelse (new_bh);
2404         ext3_journal_stop(handle);
2405         return retval;
2406 }
2407
2408 /*
2409  * directories can handle most operations...
2410  */
2411 const struct inode_operations ext3_dir_inode_operations = {
2412         .create         = ext3_create,
2413         .lookup         = ext3_lookup,
2414         .link           = ext3_link,
2415         .unlink         = ext3_unlink,
2416         .symlink        = ext3_symlink,
2417         .mkdir          = ext3_mkdir,
2418         .rmdir          = ext3_rmdir,
2419         .mknod          = ext3_mknod,
2420         .rename         = ext3_rename,
2421         .setattr        = ext3_setattr,
2422 #ifdef CONFIG_EXT3_FS_XATTR
2423         .setxattr       = generic_setxattr,
2424         .getxattr       = generic_getxattr,
2425         .listxattr      = ext3_listxattr,
2426         .removexattr    = generic_removexattr,
2427 #endif
2428         .permission     = ext3_permission,
2429 };
2430
2431 const struct inode_operations ext3_special_inode_operations = {
2432         .setattr        = ext3_setattr,
2433 #ifdef CONFIG_EXT3_FS_XATTR
2434         .setxattr       = generic_setxattr,
2435         .getxattr       = generic_getxattr,
2436         .listxattr      = ext3_listxattr,
2437         .removexattr    = generic_removexattr,
2438 #endif
2439         .permission     = ext3_permission,
2440 };