new helper: ihold()
[linux-2.6.git] / fs / ext4 / namei.c
1 /*
2  *  linux/fs/ext4/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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39
40 #include "xattr.h"
41 #include "acl.h"
42
43 /*
44  * define how far ahead to read directories while searching them.
45  */
46 #define NAMEI_RA_CHUNKS  2
47 #define NAMEI_RA_BLOCKS  4
48 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
50
51 static struct buffer_head *ext4_append(handle_t *handle,
52                                         struct inode *inode,
53                                         ext4_lblk_t *block, int *err)
54 {
55         struct buffer_head *bh;
56
57         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58
59         bh = ext4_bread(handle, inode, *block, 1, err);
60         if (bh) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT4_I(inode)->i_disksize = inode->i_size;
63                 *err = ext4_journal_get_write_access(handle, bh);
64                 if (*err) {
65                         brelse(bh);
66                         bh = NULL;
67                 }
68         }
69         return bh;
70 }
71
72 #ifndef assert
73 #define assert(test) J_ASSERT(test)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u16 offs;
144         u16 size;
145 };
146
147 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
149 static inline unsigned dx_get_hash(struct dx_entry *entry);
150 static void dx_set_hash(struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count(struct dx_entry *entries);
152 static unsigned dx_get_limit(struct dx_entry *entries);
153 static void dx_set_count(struct dx_entry *entries, unsigned value);
154 static void dx_set_limit(struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit(struct inode *dir);
157 static struct dx_frame *dx_probe(const struct qstr *d_name,
158                                  struct inode *dir,
159                                  struct dx_hash_info *hinfo,
160                                  struct dx_frame *frame,
161                                  int *err);
162 static void dx_release(struct dx_frame *frames);
163 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
164                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
167                 struct dx_map_entry *offsets, int count, unsigned blocksize);
168 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
169 static void dx_insert_block(struct dx_frame *frame,
170                                         u32 hash, ext4_lblk_t block);
171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
172                                  struct dx_frame *frame,
173                                  struct dx_frame *frames,
174                                  __u32 *start_hash);
175 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176                 const struct qstr *d_name,
177                 struct ext4_dir_entry_2 **res_dir,
178                 int *err);
179 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
180                              struct inode *inode);
181
182 /*
183  * p is at least 6 bytes before the end of page
184  */
185 static inline struct ext4_dir_entry_2 *
186 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
187 {
188         return (struct ext4_dir_entry_2 *)((char *)p +
189                 ext4_rec_len_from_disk(p->rec_len, blocksize));
190 }
191
192 /*
193  * Future: use high four bits of block for coalesce-on-delete flags
194  * Mask them off for now.
195  */
196
197 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
198 {
199         return le32_to_cpu(entry->block) & 0x00ffffff;
200 }
201
202 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
203 {
204         entry->block = cpu_to_le32(value);
205 }
206
207 static inline unsigned dx_get_hash(struct dx_entry *entry)
208 {
209         return le32_to_cpu(entry->hash);
210 }
211
212 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
213 {
214         entry->hash = cpu_to_le32(value);
215 }
216
217 static inline unsigned dx_get_count(struct dx_entry *entries)
218 {
219         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
220 }
221
222 static inline unsigned dx_get_limit(struct dx_entry *entries)
223 {
224         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
225 }
226
227 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
228 {
229         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
230 }
231
232 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
233 {
234         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
235 }
236
237 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
238 {
239         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
240                 EXT4_DIR_REC_LEN(2) - infosize;
241         return entry_space / sizeof(struct dx_entry);
242 }
243
244 static inline unsigned dx_node_limit(struct inode *dir)
245 {
246         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
247         return entry_space / sizeof(struct dx_entry);
248 }
249
250 /*
251  * Debug
252  */
253 #ifdef DX_DEBUG
254 static void dx_show_index(char * label, struct dx_entry *entries)
255 {
256         int i, n = dx_get_count (entries);
257         printk(KERN_DEBUG "%s index ", label);
258         for (i = 0; i < n; i++) {
259                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
260                                 0, (unsigned long)dx_get_block(entries + i));
261         }
262         printk("\n");
263 }
264
265 struct stats
266 {
267         unsigned names;
268         unsigned space;
269         unsigned bcount;
270 };
271
272 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
273                                  int size, int show_names)
274 {
275         unsigned names = 0, space = 0;
276         char *base = (char *) de;
277         struct dx_hash_info h = *hinfo;
278
279         printk("names: ");
280         while ((char *) de < base + size)
281         {
282                 if (de->inode)
283                 {
284                         if (show_names)
285                         {
286                                 int len = de->name_len;
287                                 char *name = de->name;
288                                 while (len--) printk("%c", *name++);
289                                 ext4fs_dirhash(de->name, de->name_len, &h);
290                                 printk(":%x.%u ", h.hash,
291                                        ((char *) de - base));
292                         }
293                         space += EXT4_DIR_REC_LEN(de->name_len);
294                         names++;
295                 }
296                 de = ext4_next_entry(de, size);
297         }
298         printk("(%i)\n", names);
299         return (struct stats) { names, space, 1 };
300 }
301
302 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
303                              struct dx_entry *entries, int levels)
304 {
305         unsigned blocksize = dir->i_sb->s_blocksize;
306         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
307         unsigned bcount = 0;
308         struct buffer_head *bh;
309         int err;
310         printk("%i indexed blocks...\n", count);
311         for (i = 0; i < count; i++, entries++)
312         {
313                 ext4_lblk_t block = dx_get_block(entries);
314                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
315                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
316                 struct stats stats;
317                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
318                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
319                 stats = levels?
320                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
321                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
322                 names += stats.names;
323                 space += stats.space;
324                 bcount += stats.bcount;
325                 brelse(bh);
326         }
327         if (bcount)
328                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
329                        levels ? "" : "   ", names, space/bcount,
330                        (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(const struct qstr *d_name, 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 (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
357                 goto fail;
358         root = (struct dx_root *) bh->b_data;
359         if (root->info.hash_version != DX_HASH_TEA &&
360             root->info.hash_version != DX_HASH_HALF_MD4 &&
361             root->info.hash_version != DX_HASH_LEGACY) {
362                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
363                              root->info.hash_version);
364                 brelse(bh);
365                 *err = ERR_BAD_DX_DIR;
366                 goto fail;
367         }
368         hinfo->hash_version = root->info.hash_version;
369         if (hinfo->hash_version <= DX_HASH_TEA)
370                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
371         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
372         if (d_name)
373                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
374         hash = hinfo->hash;
375
376         if (root->info.unused_flags & 1) {
377                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
378                              root->info.unused_flags);
379                 brelse(bh);
380                 *err = ERR_BAD_DX_DIR;
381                 goto fail;
382         }
383
384         if ((indirect = root->info.indirect_levels) > 1) {
385                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
386                              root->info.indirect_levels);
387                 brelse(bh);
388                 *err = ERR_BAD_DX_DIR;
389                 goto fail;
390         }
391
392         entries = (struct dx_entry *) (((char *)&root->info) +
393                                        root->info.info_length);
394
395         if (dx_get_limit(entries) != dx_root_limit(dir,
396                                                    root->info.info_length)) {
397                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
398                 brelse(bh);
399                 *err = ERR_BAD_DX_DIR;
400                 goto fail;
401         }
402
403         dxtrace(printk("Look up %x", hash));
404         while (1)
405         {
406                 count = dx_get_count(entries);
407                 if (!count || count > dx_get_limit(entries)) {
408                         ext4_warning(dir->i_sb,
409                                      "dx entry: no count or count > limit");
410                         brelse(bh);
411                         *err = ERR_BAD_DX_DIR;
412                         goto fail2;
413                 }
414
415                 p = entries + 1;
416                 q = entries + count - 1;
417                 while (p <= q)
418                 {
419                         m = p + (q - p)/2;
420                         dxtrace(printk("."));
421                         if (dx_get_hash(m) > hash)
422                                 q = m - 1;
423                         else
424                                 p = m + 1;
425                 }
426
427                 if (0) // linear search cross check
428                 {
429                         unsigned n = count - 1;
430                         at = entries;
431                         while (n--)
432                         {
433                                 dxtrace(printk(","));
434                                 if (dx_get_hash(++at) > hash)
435                                 {
436                                         at--;
437                                         break;
438                                 }
439                         }
440                         assert (at == p - 1);
441                 }
442
443                 at = p - 1;
444                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
445                 frame->bh = bh;
446                 frame->entries = entries;
447                 frame->at = at;
448                 if (!indirect--) return frame;
449                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
450                         goto fail2;
451                 at = entries = ((struct dx_node *) bh->b_data)->entries;
452                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
453                         ext4_warning(dir->i_sb,
454                                      "dx entry: limit != node limit");
455                         brelse(bh);
456                         *err = ERR_BAD_DX_DIR;
457                         goto fail2;
458                 }
459                 frame++;
460                 frame->bh = NULL;
461         }
462 fail2:
463         while (frame >= frame_in) {
464                 brelse(frame->bh);
465                 frame--;
466         }
467 fail:
468         if (*err == ERR_BAD_DX_DIR)
469                 ext4_warning(dir->i_sb,
470                              "Corrupt dir inode %ld, running e2fsck is "
471                              "recommended.", dir->i_ino);
472         return NULL;
473 }
474
475 static void dx_release (struct dx_frame *frames)
476 {
477         if (frames[0].bh == NULL)
478                 return;
479
480         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
481                 brelse(frames[1].bh);
482         brelse(frames[0].bh);
483 }
484
485 /*
486  * This function increments the frame pointer to search the next leaf
487  * block, and reads in the necessary intervening nodes if the search
488  * should be necessary.  Whether or not the search is necessary is
489  * controlled by the hash parameter.  If the hash value is even, then
490  * the search is only continued if the next block starts with that
491  * hash value.  This is used if we are searching for a specific file.
492  *
493  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
494  *
495  * This function returns 1 if the caller should continue to search,
496  * or 0 if it should not.  If there is an error reading one of the
497  * index blocks, it will a negative error code.
498  *
499  * If start_hash is non-null, it will be filled in with the starting
500  * hash of the next page.
501  */
502 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
503                                  struct dx_frame *frame,
504                                  struct dx_frame *frames,
505                                  __u32 *start_hash)
506 {
507         struct dx_frame *p;
508         struct buffer_head *bh;
509         int err, num_frames = 0;
510         __u32 bhash;
511
512         p = frame;
513         /*
514          * Find the next leaf page by incrementing the frame pointer.
515          * If we run out of entries in the interior node, loop around and
516          * increment pointer in the parent node.  When we break out of
517          * this loop, num_frames indicates the number of interior
518          * nodes need to be read.
519          */
520         while (1) {
521                 if (++(p->at) < p->entries + dx_get_count(p->entries))
522                         break;
523                 if (p == frames)
524                         return 0;
525                 num_frames++;
526                 p--;
527         }
528
529         /*
530          * If the hash is 1, then continue only if the next page has a
531          * continuation hash of any value.  This is used for readdir
532          * handling.  Otherwise, check to see if the hash matches the
533          * desired contiuation hash.  If it doesn't, return since
534          * there's no point to read in the successive index pages.
535          */
536         bhash = dx_get_hash(p->at);
537         if (start_hash)
538                 *start_hash = bhash;
539         if ((hash & 1) == 0) {
540                 if ((bhash & ~1) != hash)
541                         return 0;
542         }
543         /*
544          * If the hash is HASH_NB_ALWAYS, we always go to the next
545          * block so no check is necessary
546          */
547         while (num_frames--) {
548                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
549                                       0, &err)))
550                         return err; /* Failure */
551                 p++;
552                 brelse(p->bh);
553                 p->bh = bh;
554                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
555         }
556         return 1;
557 }
558
559
560 /*
561  * This function fills a red-black tree with information from a
562  * directory block.  It returns the number directory entries loaded
563  * into the tree.  If there is an error it is returned in err.
564  */
565 static int htree_dirblock_to_tree(struct file *dir_file,
566                                   struct inode *dir, ext4_lblk_t block,
567                                   struct dx_hash_info *hinfo,
568                                   __u32 start_hash, __u32 start_minor_hash)
569 {
570         struct buffer_head *bh;
571         struct ext4_dir_entry_2 *de, *top;
572         int err, count = 0;
573
574         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
575                                                         (unsigned long)block));
576         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
577                 return err;
578
579         de = (struct ext4_dir_entry_2 *) bh->b_data;
580         top = (struct ext4_dir_entry_2 *) ((char *) de +
581                                            dir->i_sb->s_blocksize -
582                                            EXT4_DIR_REC_LEN(0));
583         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
584                 if (!ext4_check_dir_entry(dir, de, bh,
585                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
586                                                 +((char *)de - bh->b_data))) {
587                         /* On error, skip the f_pos to the next block. */
588                         dir_file->f_pos = (dir_file->f_pos |
589                                         (dir->i_sb->s_blocksize - 1)) + 1;
590                         brelse(bh);
591                         return count;
592                 }
593                 ext4fs_dirhash(de->name, de->name_len, hinfo);
594                 if ((hinfo->hash < start_hash) ||
595                     ((hinfo->hash == start_hash) &&
596                      (hinfo->minor_hash < start_minor_hash)))
597                         continue;
598                 if (de->inode == 0)
599                         continue;
600                 if ((err = ext4_htree_store_dirent(dir_file,
601                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
602                         brelse(bh);
603                         return err;
604                 }
605                 count++;
606         }
607         brelse(bh);
608         return count;
609 }
610
611
612 /*
613  * This function fills a red-black tree with information from a
614  * directory.  We start scanning the directory in hash order, starting
615  * at start_hash and start_minor_hash.
616  *
617  * This function returns the number of entries inserted into the tree,
618  * or a negative error code.
619  */
620 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
621                          __u32 start_minor_hash, __u32 *next_hash)
622 {
623         struct dx_hash_info hinfo;
624         struct ext4_dir_entry_2 *de;
625         struct dx_frame frames[2], *frame;
626         struct inode *dir;
627         ext4_lblk_t block;
628         int count = 0;
629         int ret, err;
630         __u32 hashval;
631
632         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
633                        start_hash, start_minor_hash));
634         dir = dir_file->f_path.dentry->d_inode;
635         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
636                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
637                 if (hinfo.hash_version <= DX_HASH_TEA)
638                         hinfo.hash_version +=
639                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
640                 hinfo.seed = EXT4_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, &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 ext4_dir_entry_2 *) frames[0].bh->b_data;
655                 if ((err = ext4_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 ext4_dir_entry_2 *) frames[0].bh->b_data;
661                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
662                 if ((err = ext4_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 = ext4_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(KERN_DEBUG "Fill tree: returned %d entries, "
695                        "next hash: %x\n", 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 ext4_dir_entry_2 *de, unsigned blocksize,
712                        struct dx_hash_info *hinfo,
713                        struct dx_map_entry *map_tail)
714 {
715         int count = 0;
716         char *base = (char *) de;
717         struct dx_hash_info h = *hinfo;
718
719         while ((char *) de < base + blocksize) {
720                 if (de->name_len && de->inode) {
721                         ext4fs_dirhash(de->name, de->name_len, &h);
722                         map_tail--;
723                         map_tail->hash = h.hash;
724                         map_tail->offs = ((char *) de - base)>>2;
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 = ext4_next_entry(de, blocksize);
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                 count = count*10/13;
743                 if (count - 9 < 2) /* 9, 10 -> 11 */
744                         count = 11;
745                 for (p = top, q = p - count; q >= map; p--, q--)
746                         if (p->hash < q->hash)
747                                 swap(*p, *q);
748         }
749         /* Garden variety bubble sort */
750         do {
751                 more = 0;
752                 q = top;
753                 while (q-- > map) {
754                         if (q[1].hash >= q[0].hash)
755                                 continue;
756                         swap(*(q+1), *q);
757                         more = 1;
758                 }
759         } while(more);
760 }
761
762 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
763 {
764         struct dx_entry *entries = frame->entries;
765         struct dx_entry *old = frame->at, *new = old + 1;
766         int count = dx_get_count(entries);
767
768         assert(count < dx_get_limit(entries));
769         assert(old < entries + count);
770         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
771         dx_set_hash(new, hash);
772         dx_set_block(new, block);
773         dx_set_count(entries, count + 1);
774 }
775
776 static void ext4_update_dx_flag(struct inode *inode)
777 {
778         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
779                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
780                 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
781 }
782
783 /*
784  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
785  *
786  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
787  * `de != NULL' is guaranteed by caller.
788  */
789 static inline int ext4_match (int len, const char * const name,
790                               struct ext4_dir_entry_2 * de)
791 {
792         if (len != de->name_len)
793                 return 0;
794         if (!de->inode)
795                 return 0;
796         return !memcmp(name, de->name, len);
797 }
798
799 /*
800  * Returns 0 if not found, -1 on failure, and 1 on success
801  */
802 static inline int search_dirblock(struct buffer_head *bh,
803                                   struct inode *dir,
804                                   const struct qstr *d_name,
805                                   unsigned int offset,
806                                   struct ext4_dir_entry_2 ** res_dir)
807 {
808         struct ext4_dir_entry_2 * de;
809         char * dlimit;
810         int de_len;
811         const char *name = d_name->name;
812         int namelen = d_name->len;
813
814         de = (struct ext4_dir_entry_2 *) bh->b_data;
815         dlimit = bh->b_data + dir->i_sb->s_blocksize;
816         while ((char *) de < dlimit) {
817                 /* this code is executed quadratically often */
818                 /* do minimal checking `by hand' */
819
820                 if ((char *) de + namelen <= dlimit &&
821                     ext4_match (namelen, name, de)) {
822                         /* found a match - just to be sure, do a full check */
823                         if (!ext4_check_dir_entry(dir, de, bh, offset))
824                                 return -1;
825                         *res_dir = de;
826                         return 1;
827                 }
828                 /* prevent looping on a bad block */
829                 de_len = ext4_rec_len_from_disk(de->rec_len,
830                                                 dir->i_sb->s_blocksize);
831                 if (de_len <= 0)
832                         return -1;
833                 offset += de_len;
834                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
835         }
836         return 0;
837 }
838
839
840 /*
841  *      ext4_find_entry()
842  *
843  * finds an entry in the specified directory with the wanted name. It
844  * returns the cache buffer in which the entry was found, and the entry
845  * itself (as a parameter - res_dir). It does NOT read the inode of the
846  * entry - you'll have to do that yourself if you want to.
847  *
848  * The returned buffer_head has ->b_count elevated.  The caller is expected
849  * to brelse() it when appropriate.
850  */
851 static struct buffer_head * ext4_find_entry (struct inode *dir,
852                                         const struct qstr *d_name,
853                                         struct ext4_dir_entry_2 ** res_dir)
854 {
855         struct super_block *sb;
856         struct buffer_head *bh_use[NAMEI_RA_SIZE];
857         struct buffer_head *bh, *ret = NULL;
858         ext4_lblk_t start, block, b;
859         int ra_max = 0;         /* Number of bh's in the readahead
860                                    buffer, bh_use[] */
861         int ra_ptr = 0;         /* Current index into readahead
862                                    buffer */
863         int num = 0;
864         ext4_lblk_t  nblocks;
865         int i, err;
866         int namelen;
867
868         *res_dir = NULL;
869         sb = dir->i_sb;
870         namelen = d_name->len;
871         if (namelen > EXT4_NAME_LEN)
872                 return NULL;
873         if (is_dx(dir)) {
874                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
875                 /*
876                  * On success, or if the error was file not found,
877                  * return.  Otherwise, fall back to doing a search the
878                  * old fashioned way.
879                  */
880                 if (bh || (err != ERR_BAD_DX_DIR))
881                         return bh;
882                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
883                                "falling back\n"));
884         }
885         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
886         start = EXT4_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 = ext4_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                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
922                                          (unsigned long) block);
923                         brelse(bh);
924                         goto next;
925                 }
926                 i = search_dirblock(bh, dir, d_name,
927                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
928                 if (i == 1) {
929                         EXT4_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 >> EXT4_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 * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
961                        struct ext4_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 ext4_dir_entry_2 *de, *top;
968         struct buffer_head *bh;
969         ext4_lblk_t block;
970         int retval;
971         int namelen = d_name->len;
972         const u8 *name = d_name->name;
973
974         sb = dir->i_sb;
975         /* NFS may look up ".." - look at dx_root directory block */
976         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
977                 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
978                         return NULL;
979         } else {
980                 frame = frames;
981                 frame->bh = NULL;                       /* for dx_release() */
982                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
983                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
984         }
985         hash = hinfo.hash;
986         do {
987                 block = dx_get_block(frame->at);
988                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
989                         goto errout;
990                 de = (struct ext4_dir_entry_2 *) bh->b_data;
991                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
992                                        EXT4_DIR_REC_LEN(0));
993                 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
994                         int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
995                                   + ((char *) de - bh->b_data);
996
997                         if (!ext4_check_dir_entry(dir, de, bh, off)) {
998                                 brelse(bh);
999                                 *err = ERR_BAD_DX_DIR;
1000                                 goto errout;
1001                         }
1002
1003                         if (ext4_match(namelen, name, de)) {
1004                                 *res_dir = de;
1005                                 dx_release(frames);
1006                                 return bh;
1007                         }
1008                 }
1009                 brelse(bh);
1010                 /* Check to see if we should continue to search */
1011                 retval = ext4_htree_next_block(dir, hash, frame,
1012                                                frames, NULL);
1013                 if (retval < 0) {
1014                         ext4_warning(sb,
1015                              "error reading index page in directory #%lu",
1016                              dir->i_ino);
1017                         *err = retval;
1018                         goto errout;
1019                 }
1020         } while (retval == 1);
1021
1022         *err = -ENOENT;
1023 errout:
1024         dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1025         dx_release (frames);
1026         return NULL;
1027 }
1028
1029 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1030 {
1031         struct inode *inode;
1032         struct ext4_dir_entry_2 *de;
1033         struct buffer_head *bh;
1034
1035         if (dentry->d_name.len > EXT4_NAME_LEN)
1036                 return ERR_PTR(-ENAMETOOLONG);
1037
1038         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1039         inode = NULL;
1040         if (bh) {
1041                 __u32 ino = le32_to_cpu(de->inode);
1042                 brelse(bh);
1043                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1044                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1045                         return ERR_PTR(-EIO);
1046                 }
1047                 inode = ext4_iget(dir->i_sb, ino);
1048                 if (unlikely(IS_ERR(inode))) {
1049                         if (PTR_ERR(inode) == -ESTALE) {
1050                                 EXT4_ERROR_INODE(dir,
1051                                                  "deleted inode referenced: %u",
1052                                                  ino);
1053                                 return ERR_PTR(-EIO);
1054                         } else {
1055                                 return ERR_CAST(inode);
1056                         }
1057                 }
1058         }
1059         return d_splice_alias(inode, dentry);
1060 }
1061
1062
1063 struct dentry *ext4_get_parent(struct dentry *child)
1064 {
1065         __u32 ino;
1066         static const struct qstr dotdot = {
1067                 .name = "..",
1068                 .len = 2,
1069         };
1070         struct ext4_dir_entry_2 * de;
1071         struct buffer_head *bh;
1072
1073         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1074         if (!bh)
1075                 return ERR_PTR(-ENOENT);
1076         ino = le32_to_cpu(de->inode);
1077         brelse(bh);
1078
1079         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1080                 EXT4_ERROR_INODE(child->d_inode,
1081                                  "bad parent inode number: %u", ino);
1082                 return ERR_PTR(-EIO);
1083         }
1084
1085         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1086 }
1087
1088 #define S_SHIFT 12
1089 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1090         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1091         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1092         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1093         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1094         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1095         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1096         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1097 };
1098
1099 static inline void ext4_set_de_type(struct super_block *sb,
1100                                 struct ext4_dir_entry_2 *de,
1101                                 umode_t mode) {
1102         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1103                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1104 }
1105
1106 /*
1107  * Move count entries from end of map between two memory locations.
1108  * Returns pointer to last entry moved.
1109  */
1110 static struct ext4_dir_entry_2 *
1111 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1112                 unsigned blocksize)
1113 {
1114         unsigned rec_len = 0;
1115
1116         while (count--) {
1117                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1118                                                 (from + (map->offs<<2));
1119                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1120                 memcpy (to, de, rec_len);
1121                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1122                                 ext4_rec_len_to_disk(rec_len, blocksize);
1123                 de->inode = 0;
1124                 map++;
1125                 to += rec_len;
1126         }
1127         return (struct ext4_dir_entry_2 *) (to - rec_len);
1128 }
1129
1130 /*
1131  * Compact each dir entry in the range to the minimal rec_len.
1132  * Returns pointer to last entry in range.
1133  */
1134 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1135 {
1136         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1137         unsigned rec_len = 0;
1138
1139         prev = to = de;
1140         while ((char*)de < base + blocksize) {
1141                 next = ext4_next_entry(de, blocksize);
1142                 if (de->inode && de->name_len) {
1143                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1144                         if (de > to)
1145                                 memmove(to, de, rec_len);
1146                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1147                         prev = to;
1148                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1149                 }
1150                 de = next;
1151         }
1152         return prev;
1153 }
1154
1155 /*
1156  * Split a full leaf block to make room for a new dir entry.
1157  * Allocate a new block, and move entries so that they are approx. equally full.
1158  * Returns pointer to de in block into which the new entry will be inserted.
1159  */
1160 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1161                         struct buffer_head **bh,struct dx_frame *frame,
1162                         struct dx_hash_info *hinfo, int *error)
1163 {
1164         unsigned blocksize = dir->i_sb->s_blocksize;
1165         unsigned count, continued;
1166         struct buffer_head *bh2;
1167         ext4_lblk_t newblock;
1168         u32 hash2;
1169         struct dx_map_entry *map;
1170         char *data1 = (*bh)->b_data, *data2;
1171         unsigned split, move, size;
1172         struct ext4_dir_entry_2 *de = NULL, *de2;
1173         int     err = 0, i;
1174
1175         bh2 = ext4_append (handle, dir, &newblock, &err);
1176         if (!(bh2)) {
1177                 brelse(*bh);
1178                 *bh = NULL;
1179                 goto errout;
1180         }
1181
1182         BUFFER_TRACE(*bh, "get_write_access");
1183         err = ext4_journal_get_write_access(handle, *bh);
1184         if (err)
1185                 goto journal_error;
1186
1187         BUFFER_TRACE(frame->bh, "get_write_access");
1188         err = ext4_journal_get_write_access(handle, frame->bh);
1189         if (err)
1190                 goto journal_error;
1191
1192         data2 = bh2->b_data;
1193
1194         /* create map in the end of data2 block */
1195         map = (struct dx_map_entry *) (data2 + blocksize);
1196         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1197                              blocksize, hinfo, map);
1198         map -= count;
1199         dx_sort_map(map, count);
1200         /* Split the existing block in the middle, size-wise */
1201         size = 0;
1202         move = 0;
1203         for (i = count-1; i >= 0; i--) {
1204                 /* is more than half of this entry in 2nd half of the block? */
1205                 if (size + map[i].size/2 > blocksize/2)
1206                         break;
1207                 size += map[i].size;
1208                 move++;
1209         }
1210         /* map index at which we will split */
1211         split = count - move;
1212         hash2 = map[split].hash;
1213         continued = hash2 == map[split - 1].hash;
1214         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1215                         (unsigned long)dx_get_block(frame->at),
1216                                         hash2, split, count-split));
1217
1218         /* Fancy dance to stay within two buffers */
1219         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1220         de = dx_pack_dirents(data1, blocksize);
1221         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1222                                            blocksize);
1223         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1224                                             blocksize);
1225         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1226         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1227
1228         /* Which block gets the new entry? */
1229         if (hinfo->hash >= hash2)
1230         {
1231                 swap(*bh, bh2);
1232                 de = de2;
1233         }
1234         dx_insert_block(frame, hash2 + continued, newblock);
1235         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1236         if (err)
1237                 goto journal_error;
1238         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1239         if (err)
1240                 goto journal_error;
1241         brelse(bh2);
1242         dxtrace(dx_show_index("frame", frame->entries));
1243         return de;
1244
1245 journal_error:
1246         brelse(*bh);
1247         brelse(bh2);
1248         *bh = NULL;
1249         ext4_std_error(dir->i_sb, err);
1250 errout:
1251         *error = err;
1252         return NULL;
1253 }
1254
1255 /*
1256  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1257  * it points to a directory entry which is guaranteed to be large
1258  * enough for new directory entry.  If de is NULL, then
1259  * add_dirent_to_buf will attempt search the directory block for
1260  * space.  It will return -ENOSPC if no space is available, and -EIO
1261  * and -EEXIST if directory entry already exists.
1262  */
1263 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1264                              struct inode *inode, struct ext4_dir_entry_2 *de,
1265                              struct buffer_head *bh)
1266 {
1267         struct inode    *dir = dentry->d_parent->d_inode;
1268         const char      *name = dentry->d_name.name;
1269         int             namelen = dentry->d_name.len;
1270         unsigned int    offset = 0;
1271         unsigned int    blocksize = dir->i_sb->s_blocksize;
1272         unsigned short  reclen;
1273         int             nlen, rlen, err;
1274         char            *top;
1275
1276         reclen = EXT4_DIR_REC_LEN(namelen);
1277         if (!de) {
1278                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1279                 top = bh->b_data + blocksize - reclen;
1280                 while ((char *) de <= top) {
1281                         if (!ext4_check_dir_entry(dir, de, bh, offset))
1282                                 return -EIO;
1283                         if (ext4_match(namelen, name, de))
1284                                 return -EEXIST;
1285                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1286                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1287                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1288                                 break;
1289                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1290                         offset += rlen;
1291                 }
1292                 if ((char *) de > top)
1293                         return -ENOSPC;
1294         }
1295         BUFFER_TRACE(bh, "get_write_access");
1296         err = ext4_journal_get_write_access(handle, bh);
1297         if (err) {
1298                 ext4_std_error(dir->i_sb, err);
1299                 return err;
1300         }
1301
1302         /* By now the buffer is marked for journaling */
1303         nlen = EXT4_DIR_REC_LEN(de->name_len);
1304         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1305         if (de->inode) {
1306                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1307                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1308                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1309                 de = de1;
1310         }
1311         de->file_type = EXT4_FT_UNKNOWN;
1312         if (inode) {
1313                 de->inode = cpu_to_le32(inode->i_ino);
1314                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1315         } else
1316                 de->inode = 0;
1317         de->name_len = namelen;
1318         memcpy(de->name, name, namelen);
1319         /*
1320          * XXX shouldn't update any times until successful
1321          * completion of syscall, but too many callers depend
1322          * on this.
1323          *
1324          * XXX similarly, too many callers depend on
1325          * ext4_new_inode() setting the times, but error
1326          * recovery deletes the inode, so the worst that can
1327          * happen is that the times are slightly out of date
1328          * and/or different from the directory change time.
1329          */
1330         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1331         ext4_update_dx_flag(dir);
1332         dir->i_version++;
1333         ext4_mark_inode_dirty(handle, dir);
1334         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1335         err = ext4_handle_dirty_metadata(handle, dir, bh);
1336         if (err)
1337                 ext4_std_error(dir->i_sb, err);
1338         return 0;
1339 }
1340
1341 /*
1342  * This converts a one block unindexed directory to a 3 block indexed
1343  * directory, and adds the dentry to the indexed directory.
1344  */
1345 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1346                             struct inode *inode, struct buffer_head *bh)
1347 {
1348         struct inode    *dir = dentry->d_parent->d_inode;
1349         const char      *name = dentry->d_name.name;
1350         int             namelen = dentry->d_name.len;
1351         struct buffer_head *bh2;
1352         struct dx_root  *root;
1353         struct dx_frame frames[2], *frame;
1354         struct dx_entry *entries;
1355         struct ext4_dir_entry_2 *de, *de2;
1356         char            *data1, *top;
1357         unsigned        len;
1358         int             retval;
1359         unsigned        blocksize;
1360         struct dx_hash_info hinfo;
1361         ext4_lblk_t  block;
1362         struct fake_dirent *fde;
1363
1364         blocksize =  dir->i_sb->s_blocksize;
1365         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1366         retval = ext4_journal_get_write_access(handle, bh);
1367         if (retval) {
1368                 ext4_std_error(dir->i_sb, retval);
1369                 brelse(bh);
1370                 return retval;
1371         }
1372         root = (struct dx_root *) bh->b_data;
1373
1374         /* The 0th block becomes the root, move the dirents out */
1375         fde = &root->dotdot;
1376         de = (struct ext4_dir_entry_2 *)((char *)fde +
1377                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1378         if ((char *) de >= (((char *) root) + blocksize)) {
1379                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1380                 brelse(bh);
1381                 return -EIO;
1382         }
1383         len = ((char *) root) + blocksize - (char *) de;
1384
1385         /* Allocate new block for the 0th block's dirents */
1386         bh2 = ext4_append(handle, dir, &block, &retval);
1387         if (!(bh2)) {
1388                 brelse(bh);
1389                 return retval;
1390         }
1391         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1392         data1 = bh2->b_data;
1393
1394         memcpy (data1, de, len);
1395         de = (struct ext4_dir_entry_2 *) data1;
1396         top = data1 + len;
1397         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1398                 de = de2;
1399         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1400                                            blocksize);
1401         /* Initialize the root; the dot dirents already exist */
1402         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1403         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1404                                            blocksize);
1405         memset (&root->info, 0, sizeof(root->info));
1406         root->info.info_length = sizeof(root->info);
1407         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1408         entries = root->entries;
1409         dx_set_block(entries, 1);
1410         dx_set_count(entries, 1);
1411         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1412
1413         /* Initialize as for dx_probe */
1414         hinfo.hash_version = root->info.hash_version;
1415         if (hinfo.hash_version <= DX_HASH_TEA)
1416                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1417         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1418         ext4fs_dirhash(name, namelen, &hinfo);
1419         frame = frames;
1420         frame->entries = entries;
1421         frame->at = entries;
1422         frame->bh = bh;
1423         bh = bh2;
1424         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1425         dx_release (frames);
1426         if (!(de))
1427                 return retval;
1428
1429         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1430         brelse(bh);
1431         return retval;
1432 }
1433
1434 /*
1435  *      ext4_add_entry()
1436  *
1437  * adds a file entry to the specified directory, using the same
1438  * semantics as ext4_find_entry(). It returns NULL if it failed.
1439  *
1440  * NOTE!! The inode part of 'de' is left at 0 - which means you
1441  * may not sleep between calling this and putting something into
1442  * the entry, as someone else might have used it while you slept.
1443  */
1444 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1445                           struct inode *inode)
1446 {
1447         struct inode *dir = dentry->d_parent->d_inode;
1448         struct buffer_head *bh;
1449         struct ext4_dir_entry_2 *de;
1450         struct super_block *sb;
1451         int     retval;
1452         int     dx_fallback=0;
1453         unsigned blocksize;
1454         ext4_lblk_t 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 = ext4_dx_add_entry(handle, dentry, inode);
1462                 if (!retval || (retval != ERR_BAD_DX_DIR))
1463                         return retval;
1464                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1465                 dx_fallback++;
1466                 ext4_mark_inode_dirty(handle, dir);
1467         }
1468         blocks = dir->i_size >> sb->s_blocksize_bits;
1469         for (block = 0; block < blocks; block++) {
1470                 bh = ext4_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                         brelse(bh);
1476                         return retval;
1477                 }
1478
1479                 if (blocks == 1 && !dx_fallback &&
1480                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1481                         return make_indexed_dir(handle, dentry, inode, bh);
1482                 brelse(bh);
1483         }
1484         bh = ext4_append(handle, dir, &block, &retval);
1485         if (!bh)
1486                 return retval;
1487         de = (struct ext4_dir_entry_2 *) bh->b_data;
1488         de->inode = 0;
1489         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1490         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1491         brelse(bh);
1492         if (retval == 0)
1493                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1494         return retval;
1495 }
1496
1497 /*
1498  * Returns 0 for success, or a negative error value
1499  */
1500 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1501                              struct inode *inode)
1502 {
1503         struct dx_frame frames[2], *frame;
1504         struct dx_entry *entries, *at;
1505         struct dx_hash_info hinfo;
1506         struct buffer_head *bh;
1507         struct inode *dir = dentry->d_parent->d_inode;
1508         struct super_block *sb = dir->i_sb;
1509         struct ext4_dir_entry_2 *de;
1510         int err;
1511
1512         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1513         if (!frame)
1514                 return err;
1515         entries = frame->entries;
1516         at = frame->at;
1517
1518         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1519                 goto cleanup;
1520
1521         BUFFER_TRACE(bh, "get_write_access");
1522         err = ext4_journal_get_write_access(handle, bh);
1523         if (err)
1524                 goto journal_error;
1525
1526         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1527         if (err != -ENOSPC)
1528                 goto cleanup;
1529
1530         /* Block full, should compress but for now just split */
1531         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1532                        dx_get_count(entries), dx_get_limit(entries)));
1533         /* Need to split index? */
1534         if (dx_get_count(entries) == dx_get_limit(entries)) {
1535                 ext4_lblk_t newblock;
1536                 unsigned icount = dx_get_count(entries);
1537                 int levels = frame - frames;
1538                 struct dx_entry *entries2;
1539                 struct dx_node *node2;
1540                 struct buffer_head *bh2;
1541
1542                 if (levels && (dx_get_count(frames->entries) ==
1543                                dx_get_limit(frames->entries))) {
1544                         ext4_warning(sb, "Directory index full!");
1545                         err = -ENOSPC;
1546                         goto cleanup;
1547                 }
1548                 bh2 = ext4_append (handle, dir, &newblock, &err);
1549                 if (!(bh2))
1550                         goto cleanup;
1551                 node2 = (struct dx_node *)(bh2->b_data);
1552                 entries2 = node2->entries;
1553                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1554                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1555                                                            sb->s_blocksize);
1556                 BUFFER_TRACE(frame->bh, "get_write_access");
1557                 err = ext4_journal_get_write_access(handle, frame->bh);
1558                 if (err)
1559                         goto journal_error;
1560                 if (levels) {
1561                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1562                         unsigned hash2 = dx_get_hash(entries + icount1);
1563                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1564                                        icount1, icount2));
1565
1566                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1567                         err = ext4_journal_get_write_access(handle,
1568                                                              frames[0].bh);
1569                         if (err)
1570                                 goto journal_error;
1571
1572                         memcpy((char *) entries2, (char *) (entries + icount1),
1573                                icount2 * sizeof(struct dx_entry));
1574                         dx_set_count(entries, icount1);
1575                         dx_set_count(entries2, icount2);
1576                         dx_set_limit(entries2, dx_node_limit(dir));
1577
1578                         /* Which index block gets the new entry? */
1579                         if (at - entries >= icount1) {
1580                                 frame->at = at = at - entries - icount1 + entries2;
1581                                 frame->entries = entries = entries2;
1582                                 swap(frame->bh, bh2);
1583                         }
1584                         dx_insert_block(frames + 0, hash2, newblock);
1585                         dxtrace(dx_show_index("node", frames[1].entries));
1586                         dxtrace(dx_show_index("node",
1587                                ((struct dx_node *) bh2->b_data)->entries));
1588                         err = ext4_handle_dirty_metadata(handle, inode, bh2);
1589                         if (err)
1590                                 goto journal_error;
1591                         brelse (bh2);
1592                 } else {
1593                         dxtrace(printk(KERN_DEBUG
1594                                        "Creating second level index...\n"));
1595                         memcpy((char *) entries2, (char *) entries,
1596                                icount * sizeof(struct dx_entry));
1597                         dx_set_limit(entries2, dx_node_limit(dir));
1598
1599                         /* Set up root */
1600                         dx_set_count(entries, 1);
1601                         dx_set_block(entries + 0, newblock);
1602                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1603
1604                         /* Add new access path frame */
1605                         frame = frames + 1;
1606                         frame->at = at = at - entries + entries2;
1607                         frame->entries = entries = entries2;
1608                         frame->bh = bh2;
1609                         err = ext4_journal_get_write_access(handle,
1610                                                              frame->bh);
1611                         if (err)
1612                                 goto journal_error;
1613                 }
1614                 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1615         }
1616         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1617         if (!de)
1618                 goto cleanup;
1619         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1620         goto cleanup;
1621
1622 journal_error:
1623         ext4_std_error(dir->i_sb, err);
1624 cleanup:
1625         if (bh)
1626                 brelse(bh);
1627         dx_release(frames);
1628         return err;
1629 }
1630
1631 /*
1632  * ext4_delete_entry deletes a directory entry by merging it with the
1633  * previous entry
1634  */
1635 static int ext4_delete_entry(handle_t *handle,
1636                              struct inode *dir,
1637                              struct ext4_dir_entry_2 *de_del,
1638                              struct buffer_head *bh)
1639 {
1640         struct ext4_dir_entry_2 *de, *pde;
1641         unsigned int blocksize = dir->i_sb->s_blocksize;
1642         int i;
1643
1644         i = 0;
1645         pde = NULL;
1646         de = (struct ext4_dir_entry_2 *) bh->b_data;
1647         while (i < bh->b_size) {
1648                 if (!ext4_check_dir_entry(dir, de, bh, i))
1649                         return -EIO;
1650                 if (de == de_del)  {
1651                         BUFFER_TRACE(bh, "get_write_access");
1652                         ext4_journal_get_write_access(handle, bh);
1653                         if (pde)
1654                                 pde->rec_len = ext4_rec_len_to_disk(
1655                                         ext4_rec_len_from_disk(pde->rec_len,
1656                                                                blocksize) +
1657                                         ext4_rec_len_from_disk(de->rec_len,
1658                                                                blocksize),
1659                                         blocksize);
1660                         else
1661                                 de->inode = 0;
1662                         dir->i_version++;
1663                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1664                         ext4_handle_dirty_metadata(handle, dir, bh);
1665                         return 0;
1666                 }
1667                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1668                 pde = de;
1669                 de = ext4_next_entry(de, blocksize);
1670         }
1671         return -ENOENT;
1672 }
1673
1674 /*
1675  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1676  * since this indicates that nlinks count was previously 1.
1677  */
1678 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1679 {
1680         inc_nlink(inode);
1681         if (is_dx(inode) && inode->i_nlink > 1) {
1682                 /* limit is 16-bit i_links_count */
1683                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1684                         inode->i_nlink = 1;
1685                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1686                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1687                 }
1688         }
1689 }
1690
1691 /*
1692  * If a directory had nlink == 1, then we should let it be 1. This indicates
1693  * directory has >EXT4_LINK_MAX subdirs.
1694  */
1695 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1696 {
1697         drop_nlink(inode);
1698         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1699                 inc_nlink(inode);
1700 }
1701
1702
1703 static int ext4_add_nondir(handle_t *handle,
1704                 struct dentry *dentry, struct inode *inode)
1705 {
1706         int err = ext4_add_entry(handle, dentry, inode);
1707         if (!err) {
1708                 ext4_mark_inode_dirty(handle, inode);
1709                 d_instantiate(dentry, inode);
1710                 unlock_new_inode(inode);
1711                 return 0;
1712         }
1713         drop_nlink(inode);
1714         unlock_new_inode(inode);
1715         iput(inode);
1716         return err;
1717 }
1718
1719 /*
1720  * By the time this is called, we already have created
1721  * the directory cache entry for the new file, but it
1722  * is so far negative - it has no inode.
1723  *
1724  * If the create succeeds, we fill in the inode information
1725  * with d_instantiate().
1726  */
1727 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1728                        struct nameidata *nd)
1729 {
1730         handle_t *handle;
1731         struct inode *inode;
1732         int err, retries = 0;
1733
1734         dquot_initialize(dir);
1735
1736 retry:
1737         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1738                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1739                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1740         if (IS_ERR(handle))
1741                 return PTR_ERR(handle);
1742
1743         if (IS_DIRSYNC(dir))
1744                 ext4_handle_sync(handle);
1745
1746         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1747         err = PTR_ERR(inode);
1748         if (!IS_ERR(inode)) {
1749                 inode->i_op = &ext4_file_inode_operations;
1750                 inode->i_fop = &ext4_file_operations;
1751                 ext4_set_aops(inode);
1752                 err = ext4_add_nondir(handle, dentry, inode);
1753         }
1754         ext4_journal_stop(handle);
1755         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1756                 goto retry;
1757         return err;
1758 }
1759
1760 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1761                       int mode, dev_t rdev)
1762 {
1763         handle_t *handle;
1764         struct inode *inode;
1765         int err, retries = 0;
1766
1767         if (!new_valid_dev(rdev))
1768                 return -EINVAL;
1769
1770         dquot_initialize(dir);
1771
1772 retry:
1773         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1774                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1775                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1776         if (IS_ERR(handle))
1777                 return PTR_ERR(handle);
1778
1779         if (IS_DIRSYNC(dir))
1780                 ext4_handle_sync(handle);
1781
1782         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1783         err = PTR_ERR(inode);
1784         if (!IS_ERR(inode)) {
1785                 init_special_inode(inode, inode->i_mode, rdev);
1786 #ifdef CONFIG_EXT4_FS_XATTR
1787                 inode->i_op = &ext4_special_inode_operations;
1788 #endif
1789                 err = ext4_add_nondir(handle, dentry, inode);
1790         }
1791         ext4_journal_stop(handle);
1792         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1793                 goto retry;
1794         return err;
1795 }
1796
1797 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1798 {
1799         handle_t *handle;
1800         struct inode *inode;
1801         struct buffer_head *dir_block;
1802         struct ext4_dir_entry_2 *de;
1803         unsigned int blocksize = dir->i_sb->s_blocksize;
1804         int err, retries = 0;
1805
1806         if (EXT4_DIR_LINK_MAX(dir))
1807                 return -EMLINK;
1808
1809         dquot_initialize(dir);
1810
1811 retry:
1812         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1813                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1814                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1815         if (IS_ERR(handle))
1816                 return PTR_ERR(handle);
1817
1818         if (IS_DIRSYNC(dir))
1819                 ext4_handle_sync(handle);
1820
1821         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1822                                &dentry->d_name, 0);
1823         err = PTR_ERR(inode);
1824         if (IS_ERR(inode))
1825                 goto out_stop;
1826
1827         inode->i_op = &ext4_dir_inode_operations;
1828         inode->i_fop = &ext4_dir_operations;
1829         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1830         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1831         if (!dir_block)
1832                 goto out_clear_inode;
1833         BUFFER_TRACE(dir_block, "get_write_access");
1834         ext4_journal_get_write_access(handle, dir_block);
1835         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1836         de->inode = cpu_to_le32(inode->i_ino);
1837         de->name_len = 1;
1838         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1839                                            blocksize);
1840         strcpy(de->name, ".");
1841         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1842         de = ext4_next_entry(de, blocksize);
1843         de->inode = cpu_to_le32(dir->i_ino);
1844         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1845                                            blocksize);
1846         de->name_len = 2;
1847         strcpy(de->name, "..");
1848         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1849         inode->i_nlink = 2;
1850         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1851         ext4_handle_dirty_metadata(handle, dir, dir_block);
1852         brelse(dir_block);
1853         ext4_mark_inode_dirty(handle, inode);
1854         err = ext4_add_entry(handle, dentry, inode);
1855         if (err) {
1856 out_clear_inode:
1857                 clear_nlink(inode);
1858                 unlock_new_inode(inode);
1859                 ext4_mark_inode_dirty(handle, inode);
1860                 iput(inode);
1861                 goto out_stop;
1862         }
1863         ext4_inc_count(handle, dir);
1864         ext4_update_dx_flag(dir);
1865         ext4_mark_inode_dirty(handle, dir);
1866         d_instantiate(dentry, inode);
1867         unlock_new_inode(inode);
1868 out_stop:
1869         ext4_journal_stop(handle);
1870         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1871                 goto retry;
1872         return err;
1873 }
1874
1875 /*
1876  * routine to check that the specified directory is empty (for rmdir)
1877  */
1878 static int empty_dir(struct inode *inode)
1879 {
1880         unsigned int offset;
1881         struct buffer_head *bh;
1882         struct ext4_dir_entry_2 *de, *de1;
1883         struct super_block *sb;
1884         int err = 0;
1885
1886         sb = inode->i_sb;
1887         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1888             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1889                 if (err)
1890                         EXT4_ERROR_INODE(inode,
1891                                 "error %d reading directory lblock 0", err);
1892                 else
1893                         ext4_warning(inode->i_sb,
1894                                      "bad directory (dir #%lu) - no data block",
1895                                      inode->i_ino);
1896                 return 1;
1897         }
1898         de = (struct ext4_dir_entry_2 *) bh->b_data;
1899         de1 = ext4_next_entry(de, sb->s_blocksize);
1900         if (le32_to_cpu(de->inode) != inode->i_ino ||
1901                         !le32_to_cpu(de1->inode) ||
1902                         strcmp(".", de->name) ||
1903                         strcmp("..", de1->name)) {
1904                 ext4_warning(inode->i_sb,
1905                              "bad directory (dir #%lu) - no `.' or `..'",
1906                              inode->i_ino);
1907                 brelse(bh);
1908                 return 1;
1909         }
1910         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1911                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1912         de = ext4_next_entry(de1, sb->s_blocksize);
1913         while (offset < inode->i_size) {
1914                 if (!bh ||
1915                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1916                         unsigned int lblock;
1917                         err = 0;
1918                         brelse(bh);
1919                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1920                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
1921                         if (!bh) {
1922                                 if (err)
1923                                         EXT4_ERROR_INODE(inode,
1924                                                 "error %d reading directory "
1925                                                 "lblock %u", err, lblock);
1926                                 offset += sb->s_blocksize;
1927                                 continue;
1928                         }
1929                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1930                 }
1931                 if (!ext4_check_dir_entry(inode, de, bh, offset)) {
1932                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1933                                                          sb->s_blocksize);
1934                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1935                         continue;
1936                 }
1937                 if (le32_to_cpu(de->inode)) {
1938                         brelse(bh);
1939                         return 0;
1940                 }
1941                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1942                 de = ext4_next_entry(de, sb->s_blocksize);
1943         }
1944         brelse(bh);
1945         return 1;
1946 }
1947
1948 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1949  * such inodes, starting at the superblock, in case we crash before the
1950  * file is closed/deleted, or in case the inode truncate spans multiple
1951  * transactions and the last transaction is not recovered after a crash.
1952  *
1953  * At filesystem recovery time, we walk this list deleting unlinked
1954  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1955  */
1956 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1957 {
1958         struct super_block *sb = inode->i_sb;
1959         struct ext4_iloc iloc;
1960         int err = 0, rc;
1961
1962         if (!ext4_handle_valid(handle))
1963                 return 0;
1964
1965         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1966         if (!list_empty(&EXT4_I(inode)->i_orphan))
1967                 goto out_unlock;
1968
1969         /* Orphan handling is only valid for files with data blocks
1970          * being truncated, or files being unlinked. */
1971
1972         /* @@@ FIXME: Observation from aviro:
1973          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1974          * here (on s_orphan_lock), so race with ext4_link() which might bump
1975          * ->i_nlink. For, say it, character device. Not a regular file,
1976          * not a directory, not a symlink and ->i_nlink > 0.
1977          *
1978          * tytso, 4/25/2009: I'm not sure how that could happen;
1979          * shouldn't the fs core protect us from these sort of
1980          * unlink()/link() races?
1981          */
1982         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1983                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1984
1985         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1986         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1987         if (err)
1988                 goto out_unlock;
1989
1990         err = ext4_reserve_inode_write(handle, inode, &iloc);
1991         if (err)
1992                 goto out_unlock;
1993         /*
1994          * Due to previous errors inode may be already a part of on-disk
1995          * orphan list. If so skip on-disk list modification.
1996          */
1997         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
1998                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
1999                         goto mem_insert;
2000
2001         /* Insert this inode at the head of the on-disk orphan list... */
2002         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2003         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2004         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2005         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2006         if (!err)
2007                 err = rc;
2008
2009         /* Only add to the head of the in-memory list if all the
2010          * previous operations succeeded.  If the orphan_add is going to
2011          * fail (possibly taking the journal offline), we can't risk
2012          * leaving the inode on the orphan list: stray orphan-list
2013          * entries can cause panics at unmount time.
2014          *
2015          * This is safe: on error we're going to ignore the orphan list
2016          * anyway on the next recovery. */
2017 mem_insert:
2018         if (!err)
2019                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2020
2021         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2022         jbd_debug(4, "orphan inode %lu will point to %d\n",
2023                         inode->i_ino, NEXT_ORPHAN(inode));
2024 out_unlock:
2025         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2026         ext4_std_error(inode->i_sb, err);
2027         return err;
2028 }
2029
2030 /*
2031  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2032  * of such inodes stored on disk, because it is finally being cleaned up.
2033  */
2034 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2035 {
2036         struct list_head *prev;
2037         struct ext4_inode_info *ei = EXT4_I(inode);
2038         struct ext4_sb_info *sbi;
2039         __u32 ino_next;
2040         struct ext4_iloc iloc;
2041         int err = 0;
2042
2043         /* ext4_handle_valid() assumes a valid handle_t pointer */
2044         if (handle && !ext4_handle_valid(handle))
2045                 return 0;
2046
2047         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2048         if (list_empty(&ei->i_orphan))
2049                 goto out;
2050
2051         ino_next = NEXT_ORPHAN(inode);
2052         prev = ei->i_orphan.prev;
2053         sbi = EXT4_SB(inode->i_sb);
2054
2055         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2056
2057         list_del_init(&ei->i_orphan);
2058
2059         /* If we're on an error path, we may not have a valid
2060          * transaction handle with which to update the orphan list on
2061          * disk, but we still need to remove the inode from the linked
2062          * list in memory. */
2063         if (sbi->s_journal && !handle)
2064                 goto out;
2065
2066         err = ext4_reserve_inode_write(handle, inode, &iloc);
2067         if (err)
2068                 goto out_err;
2069
2070         if (prev == &sbi->s_orphan) {
2071                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2072                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2073                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2074                 if (err)
2075                         goto out_brelse;
2076                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2077                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2078         } else {
2079                 struct ext4_iloc iloc2;
2080                 struct inode *i_prev =
2081                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2082
2083                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2084                           i_prev->i_ino, ino_next);
2085                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2086                 if (err)
2087                         goto out_brelse;
2088                 NEXT_ORPHAN(i_prev) = ino_next;
2089                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2090         }
2091         if (err)
2092                 goto out_brelse;
2093         NEXT_ORPHAN(inode) = 0;
2094         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2095
2096 out_err:
2097         ext4_std_error(inode->i_sb, err);
2098 out:
2099         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2100         return err;
2101
2102 out_brelse:
2103         brelse(iloc.bh);
2104         goto out_err;
2105 }
2106
2107 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2108 {
2109         int retval;
2110         struct inode *inode;
2111         struct buffer_head *bh;
2112         struct ext4_dir_entry_2 *de;
2113         handle_t *handle;
2114
2115         /* Initialize quotas before so that eventual writes go in
2116          * separate transaction */
2117         dquot_initialize(dir);
2118         dquot_initialize(dentry->d_inode);
2119
2120         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2121         if (IS_ERR(handle))
2122                 return PTR_ERR(handle);
2123
2124         retval = -ENOENT;
2125         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2126         if (!bh)
2127                 goto end_rmdir;
2128
2129         if (IS_DIRSYNC(dir))
2130                 ext4_handle_sync(handle);
2131
2132         inode = dentry->d_inode;
2133
2134         retval = -EIO;
2135         if (le32_to_cpu(de->inode) != inode->i_ino)
2136                 goto end_rmdir;
2137
2138         retval = -ENOTEMPTY;
2139         if (!empty_dir(inode))
2140                 goto end_rmdir;
2141
2142         retval = ext4_delete_entry(handle, dir, de, bh);
2143         if (retval)
2144                 goto end_rmdir;
2145         if (!EXT4_DIR_LINK_EMPTY(inode))
2146                 ext4_warning(inode->i_sb,
2147                              "empty directory has too many links (%d)",
2148                              inode->i_nlink);
2149         inode->i_version++;
2150         clear_nlink(inode);
2151         /* There's no need to set i_disksize: the fact that i_nlink is
2152          * zero will ensure that the right thing happens during any
2153          * recovery. */
2154         inode->i_size = 0;
2155         ext4_orphan_add(handle, inode);
2156         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2157         ext4_mark_inode_dirty(handle, inode);
2158         ext4_dec_count(handle, dir);
2159         ext4_update_dx_flag(dir);
2160         ext4_mark_inode_dirty(handle, dir);
2161
2162 end_rmdir:
2163         ext4_journal_stop(handle);
2164         brelse(bh);
2165         return retval;
2166 }
2167
2168 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2169 {
2170         int retval;
2171         struct inode *inode;
2172         struct buffer_head *bh;
2173         struct ext4_dir_entry_2 *de;
2174         handle_t *handle;
2175
2176         /* Initialize quotas before so that eventual writes go
2177          * in separate transaction */
2178         dquot_initialize(dir);
2179         dquot_initialize(dentry->d_inode);
2180
2181         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2182         if (IS_ERR(handle))
2183                 return PTR_ERR(handle);
2184
2185         if (IS_DIRSYNC(dir))
2186                 ext4_handle_sync(handle);
2187
2188         retval = -ENOENT;
2189         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2190         if (!bh)
2191                 goto end_unlink;
2192
2193         inode = dentry->d_inode;
2194
2195         retval = -EIO;
2196         if (le32_to_cpu(de->inode) != inode->i_ino)
2197                 goto end_unlink;
2198
2199         if (!inode->i_nlink) {
2200                 ext4_warning(inode->i_sb,
2201                              "Deleting nonexistent file (%lu), %d",
2202                              inode->i_ino, inode->i_nlink);
2203                 inode->i_nlink = 1;
2204         }
2205         retval = ext4_delete_entry(handle, dir, de, bh);
2206         if (retval)
2207                 goto end_unlink;
2208         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2209         ext4_update_dx_flag(dir);
2210         ext4_mark_inode_dirty(handle, dir);
2211         drop_nlink(inode);
2212         if (!inode->i_nlink)
2213                 ext4_orphan_add(handle, inode);
2214         inode->i_ctime = ext4_current_time(inode);
2215         ext4_mark_inode_dirty(handle, inode);
2216         retval = 0;
2217
2218 end_unlink:
2219         ext4_journal_stop(handle);
2220         brelse(bh);
2221         return retval;
2222 }
2223
2224 static int ext4_symlink(struct inode *dir,
2225                         struct dentry *dentry, const char *symname)
2226 {
2227         handle_t *handle;
2228         struct inode *inode;
2229         int l, err, retries = 0;
2230
2231         l = strlen(symname)+1;
2232         if (l > dir->i_sb->s_blocksize)
2233                 return -ENAMETOOLONG;
2234
2235         dquot_initialize(dir);
2236
2237 retry:
2238         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2239                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2240                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2241         if (IS_ERR(handle))
2242                 return PTR_ERR(handle);
2243
2244         if (IS_DIRSYNC(dir))
2245                 ext4_handle_sync(handle);
2246
2247         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2248                                &dentry->d_name, 0);
2249         err = PTR_ERR(inode);
2250         if (IS_ERR(inode))
2251                 goto out_stop;
2252
2253         if (l > sizeof(EXT4_I(inode)->i_data)) {
2254                 inode->i_op = &ext4_symlink_inode_operations;
2255                 ext4_set_aops(inode);
2256                 /*
2257                  * page_symlink() calls into ext4_prepare/commit_write.
2258                  * We have a transaction open.  All is sweetness.  It also sets
2259                  * i_size in generic_commit_write().
2260                  */
2261                 err = __page_symlink(inode, symname, l, 1);
2262                 if (err) {
2263                         clear_nlink(inode);
2264                         unlock_new_inode(inode);
2265                         ext4_mark_inode_dirty(handle, inode);
2266                         iput(inode);
2267                         goto out_stop;
2268                 }
2269         } else {
2270                 /* clear the extent format for fast symlink */
2271                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2272                 inode->i_op = &ext4_fast_symlink_inode_operations;
2273                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2274                 inode->i_size = l-1;
2275         }
2276         EXT4_I(inode)->i_disksize = inode->i_size;
2277         err = ext4_add_nondir(handle, dentry, inode);
2278 out_stop:
2279         ext4_journal_stop(handle);
2280         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2281                 goto retry;
2282         return err;
2283 }
2284
2285 static int ext4_link(struct dentry *old_dentry,
2286                      struct inode *dir, struct dentry *dentry)
2287 {
2288         handle_t *handle;
2289         struct inode *inode = old_dentry->d_inode;
2290         int err, retries = 0;
2291
2292         if (inode->i_nlink >= EXT4_LINK_MAX)
2293                 return -EMLINK;
2294
2295         dquot_initialize(dir);
2296
2297         /*
2298          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2299          * otherwise has the potential to corrupt the orphan inode list.
2300          */
2301         if (inode->i_nlink == 0)
2302                 return -ENOENT;
2303
2304 retry:
2305         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2306                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2307         if (IS_ERR(handle))
2308                 return PTR_ERR(handle);
2309
2310         if (IS_DIRSYNC(dir))
2311                 ext4_handle_sync(handle);
2312
2313         inode->i_ctime = ext4_current_time(inode);
2314         ext4_inc_count(handle, inode);
2315         ihold(inode);
2316
2317         err = ext4_add_entry(handle, dentry, inode);
2318         if (!err) {
2319                 ext4_mark_inode_dirty(handle, inode);
2320                 d_instantiate(dentry, inode);
2321         } else {
2322                 drop_nlink(inode);
2323                 iput(inode);
2324         }
2325         ext4_journal_stop(handle);
2326         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2327                 goto retry;
2328         return err;
2329 }
2330
2331 #define PARENT_INO(buffer, size) \
2332         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2333
2334 /*
2335  * Anybody can rename anything with this: the permission checks are left to the
2336  * higher-level routines.
2337  */
2338 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2339                        struct inode *new_dir, struct dentry *new_dentry)
2340 {
2341         handle_t *handle;
2342         struct inode *old_inode, *new_inode;
2343         struct buffer_head *old_bh, *new_bh, *dir_bh;
2344         struct ext4_dir_entry_2 *old_de, *new_de;
2345         int retval, force_da_alloc = 0;
2346
2347         dquot_initialize(old_dir);
2348         dquot_initialize(new_dir);
2349
2350         old_bh = new_bh = dir_bh = NULL;
2351
2352         /* Initialize quotas before so that eventual writes go
2353          * in separate transaction */
2354         if (new_dentry->d_inode)
2355                 dquot_initialize(new_dentry->d_inode);
2356         handle = ext4_journal_start(old_dir, 2 *
2357                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2358                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2359         if (IS_ERR(handle))
2360                 return PTR_ERR(handle);
2361
2362         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2363                 ext4_handle_sync(handle);
2364
2365         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2366         /*
2367          *  Check for inode number is _not_ due to possible IO errors.
2368          *  We might rmdir the source, keep it as pwd of some process
2369          *  and merrily kill the link to whatever was created under the
2370          *  same name. Goodbye sticky bit ;-<
2371          */
2372         old_inode = old_dentry->d_inode;
2373         retval = -ENOENT;
2374         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2375                 goto end_rename;
2376
2377         new_inode = new_dentry->d_inode;
2378         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2379         if (new_bh) {
2380                 if (!new_inode) {
2381                         brelse(new_bh);
2382                         new_bh = NULL;
2383                 }
2384         }
2385         if (S_ISDIR(old_inode->i_mode)) {
2386                 if (new_inode) {
2387                         retval = -ENOTEMPTY;
2388                         if (!empty_dir(new_inode))
2389                                 goto end_rename;
2390                 }
2391                 retval = -EIO;
2392                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2393                 if (!dir_bh)
2394                         goto end_rename;
2395                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2396                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2397                         goto end_rename;
2398                 retval = -EMLINK;
2399                 if (!new_inode && new_dir != old_dir &&
2400                     EXT4_DIR_LINK_MAX(new_dir))
2401                         goto end_rename;
2402         }
2403         if (!new_bh) {
2404                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2405                 if (retval)
2406                         goto end_rename;
2407         } else {
2408                 BUFFER_TRACE(new_bh, "get write access");
2409                 ext4_journal_get_write_access(handle, new_bh);
2410                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2411                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2412                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2413                         new_de->file_type = old_de->file_type;
2414                 new_dir->i_version++;
2415                 new_dir->i_ctime = new_dir->i_mtime =
2416                                         ext4_current_time(new_dir);
2417                 ext4_mark_inode_dirty(handle, new_dir);
2418                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2419                 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2420                 brelse(new_bh);
2421                 new_bh = NULL;
2422         }
2423
2424         /*
2425          * Like most other Unix systems, set the ctime for inodes on a
2426          * rename.
2427          */
2428         old_inode->i_ctime = ext4_current_time(old_inode);
2429         ext4_mark_inode_dirty(handle, old_inode);
2430
2431         /*
2432          * ok, that's it
2433          */
2434         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2435             old_de->name_len != old_dentry->d_name.len ||
2436             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2437             (retval = ext4_delete_entry(handle, old_dir,
2438                                         old_de, old_bh)) == -ENOENT) {
2439                 /* old_de could have moved from under us during htree split, so
2440                  * make sure that we are deleting the right entry.  We might
2441                  * also be pointing to a stale entry in the unused part of
2442                  * old_bh so just checking inum and the name isn't enough. */
2443                 struct buffer_head *old_bh2;
2444                 struct ext4_dir_entry_2 *old_de2;
2445
2446                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2447                 if (old_bh2) {
2448                         retval = ext4_delete_entry(handle, old_dir,
2449                                                    old_de2, old_bh2);
2450                         brelse(old_bh2);
2451                 }
2452         }
2453         if (retval) {
2454                 ext4_warning(old_dir->i_sb,
2455                                 "Deleting old file (%lu), %d, error=%d",
2456                                 old_dir->i_ino, old_dir->i_nlink, retval);
2457         }
2458
2459         if (new_inode) {
2460                 ext4_dec_count(handle, new_inode);
2461                 new_inode->i_ctime = ext4_current_time(new_inode);
2462         }
2463         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2464         ext4_update_dx_flag(old_dir);
2465         if (dir_bh) {
2466                 BUFFER_TRACE(dir_bh, "get_write_access");
2467                 ext4_journal_get_write_access(handle, dir_bh);
2468                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2469                                                 cpu_to_le32(new_dir->i_ino);
2470                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2471                 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2472                 ext4_dec_count(handle, old_dir);
2473                 if (new_inode) {
2474                         /* checked empty_dir above, can't have another parent,
2475                          * ext4_dec_count() won't work for many-linked dirs */
2476                         new_inode->i_nlink = 0;
2477                 } else {
2478                         ext4_inc_count(handle, new_dir);
2479                         ext4_update_dx_flag(new_dir);
2480                         ext4_mark_inode_dirty(handle, new_dir);
2481                 }
2482         }
2483         ext4_mark_inode_dirty(handle, old_dir);
2484         if (new_inode) {
2485                 ext4_mark_inode_dirty(handle, new_inode);
2486                 if (!new_inode->i_nlink)
2487                         ext4_orphan_add(handle, new_inode);
2488                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2489                         force_da_alloc = 1;
2490         }
2491         retval = 0;
2492
2493 end_rename:
2494         brelse(dir_bh);
2495         brelse(old_bh);
2496         brelse(new_bh);
2497         ext4_journal_stop(handle);
2498         if (retval == 0 && force_da_alloc)
2499                 ext4_alloc_da_blocks(old_inode);
2500         return retval;
2501 }
2502
2503 /*
2504  * directories can handle most operations...
2505  */
2506 const struct inode_operations ext4_dir_inode_operations = {
2507         .create         = ext4_create,
2508         .lookup         = ext4_lookup,
2509         .link           = ext4_link,
2510         .unlink         = ext4_unlink,
2511         .symlink        = ext4_symlink,
2512         .mkdir          = ext4_mkdir,
2513         .rmdir          = ext4_rmdir,
2514         .mknod          = ext4_mknod,
2515         .rename         = ext4_rename,
2516         .setattr        = ext4_setattr,
2517 #ifdef CONFIG_EXT4_FS_XATTR
2518         .setxattr       = generic_setxattr,
2519         .getxattr       = generic_getxattr,
2520         .listxattr      = ext4_listxattr,
2521         .removexattr    = generic_removexattr,
2522 #endif
2523         .check_acl      = ext4_check_acl,
2524         .fiemap         = ext4_fiemap,
2525 };
2526
2527 const struct inode_operations ext4_special_inode_operations = {
2528         .setattr        = ext4_setattr,
2529 #ifdef CONFIG_EXT4_FS_XATTR
2530         .setxattr       = generic_setxattr,
2531         .getxattr       = generic_getxattr,
2532         .listxattr      = ext4_listxattr,
2533         .removexattr    = generic_removexattr,
2534 #endif
2535         .check_acl      = ext4_check_acl,
2536 };