Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / 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 "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext4_append(handle_t *handle,
53                                         struct inode *inode,
54                                         ext4_lblk_t *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         bh = ext4_bread(handle, inode, *block, 1, err);
61         if (bh) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT4_I(inode)->i_disksize = inode->i_size;
64                 *err = ext4_journal_get_write_access(handle, bh);
65                 if (*err) {
66                         brelse(bh);
67                         bh = NULL;
68                 }
69         }
70         return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #ifndef swap
78 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79 #endif
80
81 #ifdef DX_DEBUG
82 #define dxtrace(command) command
83 #else
84 #define dxtrace(command)
85 #endif
86
87 struct fake_dirent
88 {
89         __le32 inode;
90         __le16 rec_len;
91         u8 name_len;
92         u8 file_type;
93 };
94
95 struct dx_countlimit
96 {
97         __le16 limit;
98         __le16 count;
99 };
100
101 struct dx_entry
102 {
103         __le32 hash;
104         __le32 block;
105 };
106
107 /*
108  * dx_root_info is laid out so that if it should somehow get overlaid by a
109  * dirent the two low bits of the hash version will be zero.  Therefore, the
110  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
111  */
112
113 struct dx_root
114 {
115         struct fake_dirent dot;
116         char dot_name[4];
117         struct fake_dirent dotdot;
118         char dotdot_name[4];
119         struct dx_root_info
120         {
121                 __le32 reserved_zero;
122                 u8 hash_version;
123                 u8 info_length; /* 8 */
124                 u8 indirect_levels;
125                 u8 unused_flags;
126         }
127         info;
128         struct dx_entry entries[0];
129 };
130
131 struct dx_node
132 {
133         struct fake_dirent fake;
134         struct dx_entry entries[0];
135 };
136
137
138 struct dx_frame
139 {
140         struct buffer_head *bh;
141         struct dx_entry *entries;
142         struct dx_entry *at;
143 };
144
145 struct dx_map_entry
146 {
147         u32 hash;
148         u16 offs;
149         u16 size;
150 };
151
152 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
153 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
154 static inline unsigned dx_get_hash (struct dx_entry *entry);
155 static void dx_set_hash (struct dx_entry *entry, unsigned value);
156 static unsigned dx_get_count (struct dx_entry *entries);
157 static unsigned dx_get_limit (struct dx_entry *entries);
158 static void dx_set_count (struct dx_entry *entries, unsigned value);
159 static void dx_set_limit (struct dx_entry *entries, unsigned value);
160 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
161 static unsigned dx_node_limit (struct inode *dir);
162 static struct dx_frame *dx_probe(struct dentry *dentry,
163                                  struct inode *dir,
164                                  struct dx_hash_info *hinfo,
165                                  struct dx_frame *frame,
166                                  int *err);
167 static void dx_release (struct dx_frame *frames);
168 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
169                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
170 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
171 static struct ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
172                 struct dx_map_entry *offsets, int count);
173 static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
174 static void dx_insert_block(struct dx_frame *frame,
175                                         u32 hash, ext4_lblk_t block);
176 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
177                                  struct dx_frame *frame,
178                                  struct dx_frame *frames,
179                                  __u32 *start_hash);
180 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
181                        struct ext4_dir_entry_2 **res_dir, int *err);
182 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
183                              struct inode *inode);
184
185 /*
186  * Future: use high four bits of block for coalesce-on-delete flags
187  * Mask them off for now.
188  */
189
190 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
191 {
192         return le32_to_cpu(entry->block) & 0x00ffffff;
193 }
194
195 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
196 {
197         entry->block = cpu_to_le32(value);
198 }
199
200 static inline unsigned dx_get_hash (struct dx_entry *entry)
201 {
202         return le32_to_cpu(entry->hash);
203 }
204
205 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
206 {
207         entry->hash = cpu_to_le32(value);
208 }
209
210 static inline unsigned dx_get_count (struct dx_entry *entries)
211 {
212         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
213 }
214
215 static inline unsigned dx_get_limit (struct dx_entry *entries)
216 {
217         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
218 }
219
220 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
221 {
222         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
223 }
224
225 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
226 {
227         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
228 }
229
230 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
231 {
232         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
233                 EXT4_DIR_REC_LEN(2) - infosize;
234         return 0? 20: entry_space / sizeof(struct dx_entry);
235 }
236
237 static inline unsigned dx_node_limit (struct inode *dir)
238 {
239         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
240         return 0? 22: entry_space / sizeof(struct dx_entry);
241 }
242
243 /*
244  * Debug
245  */
246 #ifdef DX_DEBUG
247 static void dx_show_index (char * label, struct dx_entry *entries)
248 {
249         int i, n = dx_get_count (entries);
250         printk("%s index ", label);
251         for (i = 0; i < n; i++) {
252                 printk("%x->%lu ", i? dx_get_hash(entries + i) :
253                                 0, (unsigned long)dx_get_block(entries + i));
254         }
255         printk("\n");
256 }
257
258 struct stats
259 {
260         unsigned names;
261         unsigned space;
262         unsigned bcount;
263 };
264
265 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
266                                  int size, int show_names)
267 {
268         unsigned names = 0, space = 0;
269         char *base = (char *) de;
270         struct dx_hash_info h = *hinfo;
271
272         printk("names: ");
273         while ((char *) de < base + size)
274         {
275                 if (de->inode)
276                 {
277                         if (show_names)
278                         {
279                                 int len = de->name_len;
280                                 char *name = de->name;
281                                 while (len--) printk("%c", *name++);
282                                 ext4fs_dirhash(de->name, de->name_len, &h);
283                                 printk(":%x.%u ", h.hash,
284                                        ((char *) de - base));
285                         }
286                         space += EXT4_DIR_REC_LEN(de->name_len);
287                         names++;
288                 }
289                 de = ext4_next_entry(de);
290         }
291         printk("(%i)\n", names);
292         return (struct stats) { names, space, 1 };
293 }
294
295 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
296                              struct dx_entry *entries, int levels)
297 {
298         unsigned blocksize = dir->i_sb->s_blocksize;
299         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
300         unsigned bcount = 0;
301         struct buffer_head *bh;
302         int err;
303         printk("%i indexed blocks...\n", count);
304         for (i = 0; i < count; i++, entries++)
305         {
306                 ext4_lblk_t block = dx_get_block(entries);
307                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
308                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
309                 struct stats stats;
310                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
311                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
312                 stats = levels?
313                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
314                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
315                 names += stats.names;
316                 space += stats.space;
317                 bcount += stats.bcount;
318                 brelse (bh);
319         }
320         if (bcount)
321                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
322                         names, space/bcount,(space/bcount)*100/blocksize);
323         return (struct stats) { names, space, bcount};
324 }
325 #endif /* DX_DEBUG */
326
327 /*
328  * Probe for a directory leaf block to search.
329  *
330  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
331  * error in the directory index, and the caller should fall back to
332  * searching the directory normally.  The callers of dx_probe **MUST**
333  * check for this error code, and make sure it never gets reflected
334  * back to userspace.
335  */
336 static struct dx_frame *
337 dx_probe(struct dentry *dentry, struct inode *dir,
338          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
339 {
340         unsigned count, indirect;
341         struct dx_entry *at, *entries, *p, *q, *m;
342         struct dx_root *root;
343         struct buffer_head *bh;
344         struct dx_frame *frame = frame_in;
345         u32 hash;
346
347         frame->bh = NULL;
348         if (dentry)
349                 dir = dentry->d_parent->d_inode;
350         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
351                 goto fail;
352         root = (struct dx_root *) bh->b_data;
353         if (root->info.hash_version != DX_HASH_TEA &&
354             root->info.hash_version != DX_HASH_HALF_MD4 &&
355             root->info.hash_version != DX_HASH_LEGACY) {
356                 ext4_warning(dir->i_sb, __func__,
357                              "Unrecognised inode hash code %d",
358                              root->info.hash_version);
359                 brelse(bh);
360                 *err = ERR_BAD_DX_DIR;
361                 goto fail;
362         }
363         hinfo->hash_version = root->info.hash_version;
364         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
365         if (dentry)
366                 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
367         hash = hinfo->hash;
368
369         if (root->info.unused_flags & 1) {
370                 ext4_warning(dir->i_sb, __func__,
371                              "Unimplemented inode hash flags: %#06x",
372                              root->info.unused_flags);
373                 brelse(bh);
374                 *err = ERR_BAD_DX_DIR;
375                 goto fail;
376         }
377
378         if ((indirect = root->info.indirect_levels) > 1) {
379                 ext4_warning(dir->i_sb, __func__,
380                              "Unimplemented inode hash depth: %#06x",
381                              root->info.indirect_levels);
382                 brelse(bh);
383                 *err = ERR_BAD_DX_DIR;
384                 goto fail;
385         }
386
387         entries = (struct dx_entry *) (((char *)&root->info) +
388                                        root->info.info_length);
389
390         if (dx_get_limit(entries) != dx_root_limit(dir,
391                                                    root->info.info_length)) {
392                 ext4_warning(dir->i_sb, __func__,
393                              "dx entry: limit != root limit");
394                 brelse(bh);
395                 *err = ERR_BAD_DX_DIR;
396                 goto fail;
397         }
398
399         dxtrace (printk("Look up %x", hash));
400         while (1)
401         {
402                 count = dx_get_count(entries);
403                 if (!count || count > dx_get_limit(entries)) {
404                         ext4_warning(dir->i_sb, __func__,
405                                      "dx entry: no count or count > limit");
406                         brelse(bh);
407                         *err = ERR_BAD_DX_DIR;
408                         goto fail2;
409                 }
410
411                 p = entries + 1;
412                 q = entries + count - 1;
413                 while (p <= q)
414                 {
415                         m = p + (q - p)/2;
416                         dxtrace(printk("."));
417                         if (dx_get_hash(m) > hash)
418                                 q = m - 1;
419                         else
420                                 p = m + 1;
421                 }
422
423                 if (0) // linear search cross check
424                 {
425                         unsigned n = count - 1;
426                         at = entries;
427                         while (n--)
428                         {
429                                 dxtrace(printk(","));
430                                 if (dx_get_hash(++at) > hash)
431                                 {
432                                         at--;
433                                         break;
434                                 }
435                         }
436                         assert (at == p - 1);
437                 }
438
439                 at = p - 1;
440                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
441                 frame->bh = bh;
442                 frame->entries = entries;
443                 frame->at = at;
444                 if (!indirect--) return frame;
445                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
446                         goto fail2;
447                 at = entries = ((struct dx_node *) bh->b_data)->entries;
448                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
449                         ext4_warning(dir->i_sb, __func__,
450                                      "dx entry: limit != node limit");
451                         brelse(bh);
452                         *err = ERR_BAD_DX_DIR;
453                         goto fail2;
454                 }
455                 frame++;
456                 frame->bh = NULL;
457         }
458 fail2:
459         while (frame >= frame_in) {
460                 brelse(frame->bh);
461                 frame--;
462         }
463 fail:
464         if (*err == ERR_BAD_DX_DIR)
465                 ext4_warning(dir->i_sb, __func__,
466                              "Corrupt dir inode %ld, running e2fsck is "
467                              "recommended.", dir->i_ino);
468         return NULL;
469 }
470
471 static void dx_release (struct dx_frame *frames)
472 {
473         if (frames[0].bh == NULL)
474                 return;
475
476         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
477                 brelse(frames[1].bh);
478         brelse(frames[0].bh);
479 }
480
481 /*
482  * This function increments the frame pointer to search the next leaf
483  * block, and reads in the necessary intervening nodes if the search
484  * should be necessary.  Whether or not the search is necessary is
485  * controlled by the hash parameter.  If the hash value is even, then
486  * the search is only continued if the next block starts with that
487  * hash value.  This is used if we are searching for a specific file.
488  *
489  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
490  *
491  * This function returns 1 if the caller should continue to search,
492  * or 0 if it should not.  If there is an error reading one of the
493  * index blocks, it will a negative error code.
494  *
495  * If start_hash is non-null, it will be filled in with the starting
496  * hash of the next page.
497  */
498 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
499                                  struct dx_frame *frame,
500                                  struct dx_frame *frames,
501                                  __u32 *start_hash)
502 {
503         struct dx_frame *p;
504         struct buffer_head *bh;
505         int err, num_frames = 0;
506         __u32 bhash;
507
508         p = frame;
509         /*
510          * Find the next leaf page by incrementing the frame pointer.
511          * If we run out of entries in the interior node, loop around and
512          * increment pointer in the parent node.  When we break out of
513          * this loop, num_frames indicates the number of interior
514          * nodes need to be read.
515          */
516         while (1) {
517                 if (++(p->at) < p->entries + dx_get_count(p->entries))
518                         break;
519                 if (p == frames)
520                         return 0;
521                 num_frames++;
522                 p--;
523         }
524
525         /*
526          * If the hash is 1, then continue only if the next page has a
527          * continuation hash of any value.  This is used for readdir
528          * handling.  Otherwise, check to see if the hash matches the
529          * desired contiuation hash.  If it doesn't, return since
530          * there's no point to read in the successive index pages.
531          */
532         bhash = dx_get_hash(p->at);
533         if (start_hash)
534                 *start_hash = bhash;
535         if ((hash & 1) == 0) {
536                 if ((bhash & ~1) != hash)
537                         return 0;
538         }
539         /*
540          * If the hash is HASH_NB_ALWAYS, we always go to the next
541          * block so no check is necessary
542          */
543         while (num_frames--) {
544                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
545                                       0, &err)))
546                         return err; /* Failure */
547                 p++;
548                 brelse (p->bh);
549                 p->bh = bh;
550                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
551         }
552         return 1;
553 }
554
555
556 /*
557  * p is at least 6 bytes before the end of page
558  */
559 static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
560 {
561         return (struct ext4_dir_entry_2 *)((char *)p +
562                 ext4_rec_len_from_disk(p->rec_len));
563 }
564
565 /*
566  * This function fills a red-black tree with information from a
567  * directory block.  It returns the number directory entries loaded
568  * into the tree.  If there is an error it is returned in err.
569  */
570 static int htree_dirblock_to_tree(struct file *dir_file,
571                                   struct inode *dir, ext4_lblk_t block,
572                                   struct dx_hash_info *hinfo,
573                                   __u32 start_hash, __u32 start_minor_hash)
574 {
575         struct buffer_head *bh;
576         struct ext4_dir_entry_2 *de, *top;
577         int err, count = 0;
578
579         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
580                                                         (unsigned long)block));
581         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
582                 return err;
583
584         de = (struct ext4_dir_entry_2 *) bh->b_data;
585         top = (struct ext4_dir_entry_2 *) ((char *) de +
586                                            dir->i_sb->s_blocksize -
587                                            EXT4_DIR_REC_LEN(0));
588         for (; de < top; de = ext4_next_entry(de)) {
589                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
590                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
591                                                 +((char *)de - bh->b_data))) {
592                         /* On error, skip the f_pos to the next block. */
593                         dir_file->f_pos = (dir_file->f_pos |
594                                         (dir->i_sb->s_blocksize - 1)) + 1;
595                         brelse (bh);
596                         return count;
597                 }
598                 ext4fs_dirhash(de->name, de->name_len, hinfo);
599                 if ((hinfo->hash < start_hash) ||
600                     ((hinfo->hash == start_hash) &&
601                      (hinfo->minor_hash < start_minor_hash)))
602                         continue;
603                 if (de->inode == 0)
604                         continue;
605                 if ((err = ext4_htree_store_dirent(dir_file,
606                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
607                         brelse(bh);
608                         return err;
609                 }
610                 count++;
611         }
612         brelse(bh);
613         return count;
614 }
615
616
617 /*
618  * This function fills a red-black tree with information from a
619  * directory.  We start scanning the directory in hash order, starting
620  * at start_hash and start_minor_hash.
621  *
622  * This function returns the number of entries inserted into the tree,
623  * or a negative error code.
624  */
625 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
626                          __u32 start_minor_hash, __u32 *next_hash)
627 {
628         struct dx_hash_info hinfo;
629         struct ext4_dir_entry_2 *de;
630         struct dx_frame frames[2], *frame;
631         struct inode *dir;
632         ext4_lblk_t block;
633         int count = 0;
634         int ret, err;
635         __u32 hashval;
636
637         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
638                        start_minor_hash));
639         dir = dir_file->f_path.dentry->d_inode;
640         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
641                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
642                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
643                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
644                                                start_hash, start_minor_hash);
645                 *next_hash = ~0;
646                 return count;
647         }
648         hinfo.hash = start_hash;
649         hinfo.minor_hash = 0;
650         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
651         if (!frame)
652                 return err;
653
654         /* Add '.' and '..' from the htree header */
655         if (!start_hash && !start_minor_hash) {
656                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
657                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
658                         goto errout;
659                 count++;
660         }
661         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
662                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
663                 de = ext4_next_entry(de);
664                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
665                         goto errout;
666                 count++;
667         }
668
669         while (1) {
670                 block = dx_get_block(frame->at);
671                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
672                                              start_hash, start_minor_hash);
673                 if (ret < 0) {
674                         err = ret;
675                         goto errout;
676                 }
677                 count += ret;
678                 hashval = ~0;
679                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
680                                             frame, frames, &hashval);
681                 *next_hash = hashval;
682                 if (ret < 0) {
683                         err = ret;
684                         goto errout;
685                 }
686                 /*
687                  * Stop if:  (a) there are no more entries, or
688                  * (b) we have inserted at least one entry and the
689                  * next hash value is not a continuation
690                  */
691                 if ((ret == 0) ||
692                     (count && ((hashval & 1) == 0)))
693                         break;
694         }
695         dx_release(frames);
696         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
697                        count, *next_hash));
698         return count;
699 errout:
700         dx_release(frames);
701         return (err);
702 }
703
704
705 /*
706  * Directory block splitting, compacting
707  */
708
709 /*
710  * Create map of hash values, offsets, and sizes, stored at end of block.
711  * Returns number of entries mapped.
712  */
713 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
714                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
715 {
716         int count = 0;
717         char *base = (char *) de;
718         struct dx_hash_info h = *hinfo;
719
720         while ((char *) de < base + size)
721         {
722                 if (de->name_len && de->inode) {
723                         ext4fs_dirhash(de->name, de->name_len, &h);
724                         map_tail--;
725                         map_tail->hash = h.hash;
726                         map_tail->offs = (u16) ((char *) de - base);
727                         map_tail->size = le16_to_cpu(de->rec_len);
728                         count++;
729                         cond_resched();
730                 }
731                 /* XXX: do we need to check rec_len == 0 case? -Chris */
732                 de = ext4_next_entry(de);
733         }
734         return count;
735 }
736
737 /* Sort map by hash value */
738 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
739 {
740         struct dx_map_entry *p, *q, *top = map + count - 1;
741         int more;
742         /* Combsort until bubble sort doesn't suck */
743         while (count > 2) {
744                 count = count*10/13;
745                 if (count - 9 < 2) /* 9, 10 -> 11 */
746                         count = 11;
747                 for (p = top, q = p - count; q >= map; p--, q--)
748                         if (p->hash < q->hash)
749                                 swap(*p, *q);
750         }
751         /* Garden variety bubble sort */
752         do {
753                 more = 0;
754                 q = top;
755                 while (q-- > map) {
756                         if (q[1].hash >= q[0].hash)
757                                 continue;
758                         swap(*(q+1), *q);
759                         more = 1;
760                 }
761         } while(more);
762 }
763
764 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
765 {
766         struct dx_entry *entries = frame->entries;
767         struct dx_entry *old = frame->at, *new = old + 1;
768         int count = dx_get_count(entries);
769
770         assert(count < dx_get_limit(entries));
771         assert(old < entries + count);
772         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773         dx_set_hash(new, hash);
774         dx_set_block(new, block);
775         dx_set_count(entries, count + 1);
776 }
777
778 static void ext4_update_dx_flag(struct inode *inode)
779 {
780         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
781                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
782                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
783 }
784
785 /*
786  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
787  *
788  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
789  * `de != NULL' is guaranteed by caller.
790  */
791 static inline int ext4_match (int len, const char * const name,
792                               struct ext4_dir_entry_2 * de)
793 {
794         if (len != de->name_len)
795                 return 0;
796         if (!de->inode)
797                 return 0;
798         return !memcmp(name, de->name, len);
799 }
800
801 /*
802  * Returns 0 if not found, -1 on failure, and 1 on success
803  */
804 static inline int search_dirblock(struct buffer_head * bh,
805                                   struct inode *dir,
806                                   struct dentry *dentry,
807                                   unsigned long offset,
808                                   struct ext4_dir_entry_2 ** res_dir)
809 {
810         struct ext4_dir_entry_2 * de;
811         char * dlimit;
812         int de_len;
813         const char *name = dentry->d_name.name;
814         int namelen = dentry->d_name.len;
815
816         de = (struct ext4_dir_entry_2 *) bh->b_data;
817         dlimit = bh->b_data + dir->i_sb->s_blocksize;
818         while ((char *) de < dlimit) {
819                 /* this code is executed quadratically often */
820                 /* do minimal checking `by hand' */
821
822                 if ((char *) de + namelen <= dlimit &&
823                     ext4_match (namelen, name, de)) {
824                         /* found a match - just to be sure, do a full check */
825                         if (!ext4_check_dir_entry("ext4_find_entry",
826                                                   dir, de, bh, offset))
827                                 return -1;
828                         *res_dir = de;
829                         return 1;
830                 }
831                 /* prevent looping on a bad block */
832                 de_len = ext4_rec_len_from_disk(de->rec_len);
833                 if (de_len <= 0)
834                         return -1;
835                 offset += de_len;
836                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
837         }
838         return 0;
839 }
840
841
842 /*
843  *      ext4_find_entry()
844  *
845  * finds an entry in the specified directory with the wanted name. It
846  * returns the cache buffer in which the entry was found, and the entry
847  * itself (as a parameter - res_dir). It does NOT read the inode of the
848  * entry - you'll have to do that yourself if you want to.
849  *
850  * The returned buffer_head has ->b_count elevated.  The caller is expected
851  * to brelse() it when appropriate.
852  */
853 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
854                                         struct ext4_dir_entry_2 ** res_dir)
855 {
856         struct super_block * sb;
857         struct buffer_head * bh_use[NAMEI_RA_SIZE];
858         struct buffer_head * bh, *ret = NULL;
859         ext4_lblk_t start, block, b;
860         int ra_max = 0;         /* Number of bh's in the readahead
861                                    buffer, bh_use[] */
862         int ra_ptr = 0;         /* Current index into readahead
863                                    buffer */
864         int num = 0;
865         ext4_lblk_t  nblocks;
866         int i, err;
867         struct inode *dir = dentry->d_parent->d_inode;
868         int namelen;
869
870         *res_dir = NULL;
871         sb = dir->i_sb;
872         namelen = dentry->d_name.len;
873         if (namelen > EXT4_NAME_LEN)
874                 return NULL;
875         if (is_dx(dir)) {
876                 bh = ext4_dx_find_entry(dentry, res_dir, &err);
877                 /*
878                  * On success, or if the error was file not found,
879                  * return.  Otherwise, fall back to doing a search the
880                  * old fashioned way.
881                  */
882                 if (bh || (err != ERR_BAD_DX_DIR))
883                         return bh;
884                 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
885         }
886         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
887         start = EXT4_I(dir)->i_dir_start_lookup;
888         if (start >= nblocks)
889                 start = 0;
890         block = start;
891 restart:
892         do {
893                 /*
894                  * We deal with the read-ahead logic here.
895                  */
896                 if (ra_ptr >= ra_max) {
897                         /* Refill the readahead buffer */
898                         ra_ptr = 0;
899                         b = block;
900                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
901                                 /*
902                                  * Terminate if we reach the end of the
903                                  * directory and must wrap, or if our
904                                  * search has finished at this block.
905                                  */
906                                 if (b >= nblocks || (num && block == start)) {
907                                         bh_use[ra_max] = NULL;
908                                         break;
909                                 }
910                                 num++;
911                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
912                                 bh_use[ra_max] = bh;
913                                 if (bh)
914                                         ll_rw_block(READ_META, 1, &bh);
915                         }
916                 }
917                 if ((bh = bh_use[ra_ptr++]) == NULL)
918                         goto next;
919                 wait_on_buffer(bh);
920                 if (!buffer_uptodate(bh)) {
921                         /* read error, skip block & hope for the best */
922                         ext4_error(sb, __func__, "reading directory #%lu "
923                                    "offset %lu", dir->i_ino,
924                                    (unsigned long)block);
925                         brelse(bh);
926                         goto next;
927                 }
928                 i = search_dirblock(bh, dir, dentry,
929                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
930                 if (i == 1) {
931                         EXT4_I(dir)->i_dir_start_lookup = block;
932                         ret = bh;
933                         goto cleanup_and_exit;
934                 } else {
935                         brelse(bh);
936                         if (i < 0)
937                                 goto cleanup_and_exit;
938                 }
939         next:
940                 if (++block >= nblocks)
941                         block = 0;
942         } while (block != start);
943
944         /*
945          * If the directory has grown while we were searching, then
946          * search the last part of the directory before giving up.
947          */
948         block = nblocks;
949         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
950         if (block < nblocks) {
951                 start = 0;
952                 goto restart;
953         }
954
955 cleanup_and_exit:
956         /* Clean up the read-ahead blocks */
957         for (; ra_ptr < ra_max; ra_ptr++)
958                 brelse (bh_use[ra_ptr]);
959         return ret;
960 }
961
962 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
963                        struct ext4_dir_entry_2 **res_dir, int *err)
964 {
965         struct super_block * sb;
966         struct dx_hash_info     hinfo;
967         u32 hash;
968         struct dx_frame frames[2], *frame;
969         struct ext4_dir_entry_2 *de, *top;
970         struct buffer_head *bh;
971         ext4_lblk_t block;
972         int retval;
973         int namelen = dentry->d_name.len;
974         const u8 *name = dentry->d_name.name;
975         struct inode *dir = dentry->d_parent->d_inode;
976
977         sb = dir->i_sb;
978         /* NFS may look up ".." - look at dx_root directory block */
979         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
980                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
981                         return NULL;
982         } else {
983                 frame = frames;
984                 frame->bh = NULL;                       /* for dx_release() */
985                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
986                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
987         }
988         hash = hinfo.hash;
989         do {
990                 block = dx_get_block(frame->at);
991                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
992                         goto errout;
993                 de = (struct ext4_dir_entry_2 *) bh->b_data;
994                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
995                                        EXT4_DIR_REC_LEN(0));
996                 for (; de < top; de = ext4_next_entry(de))
997                 if (ext4_match (namelen, name, de)) {
998                         if (!ext4_check_dir_entry("ext4_find_entry",
999                                                   dir, de, bh,
1000                                   (block<<EXT4_BLOCK_SIZE_BITS(sb))
1001                                           +((char *)de - bh->b_data))) {
1002                                 brelse (bh);
1003                                 *err = ERR_BAD_DX_DIR;
1004                                 goto errout;
1005                         }
1006                         *res_dir = de;
1007                         dx_release (frames);
1008                         return bh;
1009                 }
1010                 brelse (bh);
1011                 /* Check to see if we should continue to search */
1012                 retval = ext4_htree_next_block(dir, hash, frame,
1013                                                frames, NULL);
1014                 if (retval < 0) {
1015                         ext4_warning(sb, __func__,
1016                              "error reading index page in directory #%lu",
1017                              dir->i_ino);
1018                         *err = retval;
1019                         goto errout;
1020                 }
1021         } while (retval == 1);
1022
1023         *err = -ENOENT;
1024 errout:
1025         dxtrace(printk("%s not found\n", name));
1026         dx_release (frames);
1027         return NULL;
1028 }
1029
1030 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1031 {
1032         struct inode * inode;
1033         struct ext4_dir_entry_2 * de;
1034         struct buffer_head * bh;
1035
1036         if (dentry->d_name.len > EXT4_NAME_LEN)
1037                 return ERR_PTR(-ENAMETOOLONG);
1038
1039         bh = ext4_find_entry(dentry, &de);
1040         inode = NULL;
1041         if (bh) {
1042                 unsigned long ino = le32_to_cpu(de->inode);
1043                 brelse (bh);
1044                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1045                         ext4_error(dir->i_sb, "ext4_lookup",
1046                                    "bad inode number: %lu", ino);
1047                         return ERR_PTR(-EIO);
1048                 }
1049                 inode = ext4_iget(dir->i_sb, ino);
1050                 if (IS_ERR(inode))
1051                         return ERR_CAST(inode);
1052         }
1053         return d_splice_alias(inode, dentry);
1054 }
1055
1056
1057 struct dentry *ext4_get_parent(struct dentry *child)
1058 {
1059         unsigned long ino;
1060         struct dentry *parent;
1061         struct inode *inode;
1062         struct dentry dotdot;
1063         struct ext4_dir_entry_2 * de;
1064         struct buffer_head *bh;
1065
1066         dotdot.d_name.name = "..";
1067         dotdot.d_name.len = 2;
1068         dotdot.d_parent = child; /* confusing, isn't it! */
1069
1070         bh = ext4_find_entry(&dotdot, &de);
1071         inode = NULL;
1072         if (!bh)
1073                 return ERR_PTR(-ENOENT);
1074         ino = le32_to_cpu(de->inode);
1075         brelse(bh);
1076
1077         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1078                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1079                            "bad inode number: %lu", ino);
1080                 return ERR_PTR(-EIO);
1081         }
1082
1083         inode = ext4_iget(child->d_inode->i_sb, ino);
1084         if (IS_ERR(inode))
1085                 return ERR_CAST(inode);
1086
1087         parent = d_alloc_anon(inode);
1088         if (!parent) {
1089                 iput(inode);
1090                 parent = ERR_PTR(-ENOMEM);
1091         }
1092         return parent;
1093 }
1094
1095 #define S_SHIFT 12
1096 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1097         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1098         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1099         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1100         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1101         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1102         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1103         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1104 };
1105
1106 static inline void ext4_set_de_type(struct super_block *sb,
1107                                 struct ext4_dir_entry_2 *de,
1108                                 umode_t mode) {
1109         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1110                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1111 }
1112
1113 /*
1114  * Move count entries from end of map between two memory locations.
1115  * Returns pointer to last entry moved.
1116  */
1117 static struct ext4_dir_entry_2 *
1118 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1119 {
1120         unsigned rec_len = 0;
1121
1122         while (count--) {
1123                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1124                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1125                 memcpy (to, de, rec_len);
1126                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1127                                 ext4_rec_len_to_disk(rec_len);
1128                 de->inode = 0;
1129                 map++;
1130                 to += rec_len;
1131         }
1132         return (struct ext4_dir_entry_2 *) (to - rec_len);
1133 }
1134
1135 /*
1136  * Compact each dir entry in the range to the minimal rec_len.
1137  * Returns pointer to last entry in range.
1138  */
1139 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1140 {
1141         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1142         unsigned rec_len = 0;
1143
1144         prev = to = de;
1145         while ((char*)de < base + size) {
1146                 next = ext4_next_entry(de);
1147                 if (de->inode && de->name_len) {
1148                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1149                         if (de > to)
1150                                 memmove(to, de, rec_len);
1151                         to->rec_len = ext4_rec_len_to_disk(rec_len);
1152                         prev = to;
1153                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1154                 }
1155                 de = next;
1156         }
1157         return prev;
1158 }
1159
1160 /*
1161  * Split a full leaf block to make room for a new dir entry.
1162  * Allocate a new block, and move entries so that they are approx. equally full.
1163  * Returns pointer to de in block into which the new entry will be inserted.
1164  */
1165 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1166                         struct buffer_head **bh,struct dx_frame *frame,
1167                         struct dx_hash_info *hinfo, int *error)
1168 {
1169         unsigned blocksize = dir->i_sb->s_blocksize;
1170         unsigned count, continued;
1171         struct buffer_head *bh2;
1172         ext4_lblk_t newblock;
1173         u32 hash2;
1174         struct dx_map_entry *map;
1175         char *data1 = (*bh)->b_data, *data2;
1176         unsigned split, move, size, i;
1177         struct ext4_dir_entry_2 *de = NULL, *de2;
1178         int     err = 0;
1179
1180         bh2 = ext4_append (handle, dir, &newblock, &err);
1181         if (!(bh2)) {
1182                 brelse(*bh);
1183                 *bh = NULL;
1184                 goto errout;
1185         }
1186
1187         BUFFER_TRACE(*bh, "get_write_access");
1188         err = ext4_journal_get_write_access(handle, *bh);
1189         if (err)
1190                 goto journal_error;
1191
1192         BUFFER_TRACE(frame->bh, "get_write_access");
1193         err = ext4_journal_get_write_access(handle, frame->bh);
1194         if (err)
1195                 goto journal_error;
1196
1197         data2 = bh2->b_data;
1198
1199         /* create map in the end of data2 block */
1200         map = (struct dx_map_entry *) (data2 + blocksize);
1201         count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1202                              blocksize, hinfo, map);
1203         map -= count;
1204         dx_sort_map (map, count);
1205         /* Split the existing block in the middle, size-wise */
1206         size = 0;
1207         move = 0;
1208         for (i = count-1; i >= 0; i--) {
1209                 /* is more than half of this entry in 2nd half of the block? */
1210                 if (size + map[i].size/2 > blocksize/2)
1211                         break;
1212                 size += map[i].size;
1213                 move++;
1214         }
1215         /* map index at which we will split */
1216         split = count - move;
1217         hash2 = map[split].hash;
1218         continued = hash2 == map[split - 1].hash;
1219         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1220                         (unsigned long)dx_get_block(frame->at),
1221                                         hash2, split, count-split));
1222
1223         /* Fancy dance to stay within two buffers */
1224         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1225         de = dx_pack_dirents(data1,blocksize);
1226         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1227         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
1228         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1229         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1230
1231         /* Which block gets the new entry? */
1232         if (hinfo->hash >= hash2)
1233         {
1234                 swap(*bh, bh2);
1235                 de = de2;
1236         }
1237         dx_insert_block (frame, hash2 + continued, newblock);
1238         err = ext4_journal_dirty_metadata (handle, bh2);
1239         if (err)
1240                 goto journal_error;
1241         err = ext4_journal_dirty_metadata (handle, frame->bh);
1242         if (err)
1243                 goto journal_error;
1244         brelse (bh2);
1245         dxtrace(dx_show_index ("frame", frame->entries));
1246         return de;
1247
1248 journal_error:
1249         brelse(*bh);
1250         brelse(bh2);
1251         *bh = NULL;
1252         ext4_std_error(dir->i_sb, err);
1253 errout:
1254         *error = err;
1255         return NULL;
1256 }
1257
1258 /*
1259  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1260  * it points to a directory entry which is guaranteed to be large
1261  * enough for new directory entry.  If de is NULL, then
1262  * add_dirent_to_buf will attempt search the directory block for
1263  * space.  It will return -ENOSPC if no space is available, and -EIO
1264  * and -EEXIST if directory entry already exists.
1265  *
1266  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1267  * all other cases bh is released.
1268  */
1269 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1270                              struct inode *inode, struct ext4_dir_entry_2 *de,
1271                              struct buffer_head * bh)
1272 {
1273         struct inode    *dir = dentry->d_parent->d_inode;
1274         const char      *name = dentry->d_name.name;
1275         int             namelen = dentry->d_name.len;
1276         unsigned long   offset = 0;
1277         unsigned short  reclen;
1278         int             nlen, rlen, err;
1279         char            *top;
1280
1281         reclen = EXT4_DIR_REC_LEN(namelen);
1282         if (!de) {
1283                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1284                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1285                 while ((char *) de <= top) {
1286                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1287                                                   bh, offset)) {
1288                                 brelse (bh);
1289                                 return -EIO;
1290                         }
1291                         if (ext4_match (namelen, name, de)) {
1292                                 brelse (bh);
1293                                 return -EEXIST;
1294                         }
1295                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1296                         rlen = ext4_rec_len_from_disk(de->rec_len);
1297                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1298                                 break;
1299                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1300                         offset += rlen;
1301                 }
1302                 if ((char *) de > top)
1303                         return -ENOSPC;
1304         }
1305         BUFFER_TRACE(bh, "get_write_access");
1306         err = ext4_journal_get_write_access(handle, bh);
1307         if (err) {
1308                 ext4_std_error(dir->i_sb, err);
1309                 brelse(bh);
1310                 return err;
1311         }
1312
1313         /* By now the buffer is marked for journaling */
1314         nlen = EXT4_DIR_REC_LEN(de->name_len);
1315         rlen = ext4_rec_len_from_disk(de->rec_len);
1316         if (de->inode) {
1317                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1318                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1319                 de->rec_len = ext4_rec_len_to_disk(nlen);
1320                 de = de1;
1321         }
1322         de->file_type = EXT4_FT_UNKNOWN;
1323         if (inode) {
1324                 de->inode = cpu_to_le32(inode->i_ino);
1325                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1326         } else
1327                 de->inode = 0;
1328         de->name_len = namelen;
1329         memcpy (de->name, name, namelen);
1330         /*
1331          * XXX shouldn't update any times until successful
1332          * completion of syscall, but too many callers depend
1333          * on this.
1334          *
1335          * XXX similarly, too many callers depend on
1336          * ext4_new_inode() setting the times, but error
1337          * recovery deletes the inode, so the worst that can
1338          * happen is that the times are slightly out of date
1339          * and/or different from the directory change time.
1340          */
1341         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1342         ext4_update_dx_flag(dir);
1343         dir->i_version++;
1344         ext4_mark_inode_dirty(handle, dir);
1345         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1346         err = ext4_journal_dirty_metadata(handle, bh);
1347         if (err)
1348                 ext4_std_error(dir->i_sb, err);
1349         brelse(bh);
1350         return 0;
1351 }
1352
1353 /*
1354  * This converts a one block unindexed directory to a 3 block indexed
1355  * directory, and adds the dentry to the indexed directory.
1356  */
1357 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1358                             struct inode *inode, struct buffer_head *bh)
1359 {
1360         struct inode    *dir = dentry->d_parent->d_inode;
1361         const char      *name = dentry->d_name.name;
1362         int             namelen = dentry->d_name.len;
1363         struct buffer_head *bh2;
1364         struct dx_root  *root;
1365         struct dx_frame frames[2], *frame;
1366         struct dx_entry *entries;
1367         struct ext4_dir_entry_2 *de, *de2;
1368         char            *data1, *top;
1369         unsigned        len;
1370         int             retval;
1371         unsigned        blocksize;
1372         struct dx_hash_info hinfo;
1373         ext4_lblk_t  block;
1374         struct fake_dirent *fde;
1375
1376         blocksize =  dir->i_sb->s_blocksize;
1377         dxtrace(printk("Creating index\n"));
1378         retval = ext4_journal_get_write_access(handle, bh);
1379         if (retval) {
1380                 ext4_std_error(dir->i_sb, retval);
1381                 brelse(bh);
1382                 return retval;
1383         }
1384         root = (struct dx_root *) bh->b_data;
1385
1386         bh2 = ext4_append (handle, dir, &block, &retval);
1387         if (!(bh2)) {
1388                 brelse(bh);
1389                 return retval;
1390         }
1391         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1392         data1 = bh2->b_data;
1393
1394         /* The 0th block becomes the root, move the dirents out */
1395         fde = &root->dotdot;
1396         de = (struct ext4_dir_entry_2 *)((char *)fde +
1397                 ext4_rec_len_from_disk(fde->rec_len));
1398         len = ((char *) root) + blocksize - (char *) de;
1399         memcpy (data1, de, len);
1400         de = (struct ext4_dir_entry_2 *) data1;
1401         top = data1 + len;
1402         while ((char *)(de2 = ext4_next_entry(de)) < top)
1403                 de = de2;
1404         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1405         /* Initialize the root; the dot dirents already exist */
1406         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1407         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
1408         memset (&root->info, 0, sizeof(root->info));
1409         root->info.info_length = sizeof(root->info);
1410         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1411         entries = root->entries;
1412         dx_set_block (entries, 1);
1413         dx_set_count (entries, 1);
1414         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1415
1416         /* Initialize as for dx_probe */
1417         hinfo.hash_version = root->info.hash_version;
1418         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1419         ext4fs_dirhash(name, namelen, &hinfo);
1420         frame = frames;
1421         frame->entries = entries;
1422         frame->at = entries;
1423         frame->bh = bh;
1424         bh = bh2;
1425         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1426         dx_release (frames);
1427         if (!(de))
1428                 return retval;
1429
1430         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1431 }
1432
1433 /*
1434  *      ext4_add_entry()
1435  *
1436  * adds a file entry to the specified directory, using the same
1437  * semantics as ext4_find_entry(). It returns NULL if it failed.
1438  *
1439  * NOTE!! The inode part of 'de' is left at 0 - which means you
1440  * may not sleep between calling this and putting something into
1441  * the entry, as someone else might have used it while you slept.
1442  */
1443 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1444         struct inode *inode)
1445 {
1446         struct inode *dir = dentry->d_parent->d_inode;
1447         unsigned long offset;
1448         struct buffer_head * bh;
1449         struct 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_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1465                 dx_fallback++;
1466                 ext4_mark_inode_dirty(handle, dir);
1467         }
1468         blocks = dir->i_size >> sb->s_blocksize_bits;
1469         for (block = 0, offset = 0; block < blocks; block++) {
1470                 bh = 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                         return retval;
1476
1477                 if (blocks == 1 && !dx_fallback &&
1478                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1479                         return make_indexed_dir(handle, dentry, inode, bh);
1480                 brelse(bh);
1481         }
1482         bh = ext4_append(handle, dir, &block, &retval);
1483         if (!bh)
1484                 return retval;
1485         de = (struct ext4_dir_entry_2 *) bh->b_data;
1486         de->inode = 0;
1487         de->rec_len = ext4_rec_len_to_disk(blocksize);
1488         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1489 }
1490
1491 /*
1492  * Returns 0 for success, or a negative error value
1493  */
1494 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1495                              struct inode *inode)
1496 {
1497         struct dx_frame frames[2], *frame;
1498         struct dx_entry *entries, *at;
1499         struct dx_hash_info hinfo;
1500         struct buffer_head * bh;
1501         struct inode *dir = dentry->d_parent->d_inode;
1502         struct super_block * sb = dir->i_sb;
1503         struct ext4_dir_entry_2 *de;
1504         int err;
1505
1506         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1507         if (!frame)
1508                 return err;
1509         entries = frame->entries;
1510         at = frame->at;
1511
1512         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1513                 goto cleanup;
1514
1515         BUFFER_TRACE(bh, "get_write_access");
1516         err = ext4_journal_get_write_access(handle, bh);
1517         if (err)
1518                 goto journal_error;
1519
1520         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1521         if (err != -ENOSPC) {
1522                 bh = NULL;
1523                 goto cleanup;
1524         }
1525
1526         /* Block full, should compress but for now just split */
1527         dxtrace(printk("using %u of %u node entries\n",
1528                        dx_get_count(entries), dx_get_limit(entries)));
1529         /* Need to split index? */
1530         if (dx_get_count(entries) == dx_get_limit(entries)) {
1531                 ext4_lblk_t newblock;
1532                 unsigned icount = dx_get_count(entries);
1533                 int levels = frame - frames;
1534                 struct dx_entry *entries2;
1535                 struct dx_node *node2;
1536                 struct buffer_head *bh2;
1537
1538                 if (levels && (dx_get_count(frames->entries) ==
1539                                dx_get_limit(frames->entries))) {
1540                         ext4_warning(sb, __func__,
1541                                      "Directory index full!");
1542                         err = -ENOSPC;
1543                         goto cleanup;
1544                 }
1545                 bh2 = ext4_append (handle, dir, &newblock, &err);
1546                 if (!(bh2))
1547                         goto cleanup;
1548                 node2 = (struct dx_node *)(bh2->b_data);
1549                 entries2 = node2->entries;
1550                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
1551                 node2->fake.inode = 0;
1552                 BUFFER_TRACE(frame->bh, "get_write_access");
1553                 err = ext4_journal_get_write_access(handle, frame->bh);
1554                 if (err)
1555                         goto journal_error;
1556                 if (levels) {
1557                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1558                         unsigned hash2 = dx_get_hash(entries + icount1);
1559                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1560
1561                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1562                         err = ext4_journal_get_write_access(handle,
1563                                                              frames[0].bh);
1564                         if (err)
1565                                 goto journal_error;
1566
1567                         memcpy ((char *) entries2, (char *) (entries + icount1),
1568                                 icount2 * sizeof(struct dx_entry));
1569                         dx_set_count (entries, icount1);
1570                         dx_set_count (entries2, icount2);
1571                         dx_set_limit (entries2, dx_node_limit(dir));
1572
1573                         /* Which index block gets the new entry? */
1574                         if (at - entries >= icount1) {
1575                                 frame->at = at = at - entries - icount1 + entries2;
1576                                 frame->entries = entries = entries2;
1577                                 swap(frame->bh, bh2);
1578                         }
1579                         dx_insert_block (frames + 0, hash2, newblock);
1580                         dxtrace(dx_show_index ("node", frames[1].entries));
1581                         dxtrace(dx_show_index ("node",
1582                                ((struct dx_node *) bh2->b_data)->entries));
1583                         err = ext4_journal_dirty_metadata(handle, bh2);
1584                         if (err)
1585                                 goto journal_error;
1586                         brelse (bh2);
1587                 } else {
1588                         dxtrace(printk("Creating second level index...\n"));
1589                         memcpy((char *) entries2, (char *) entries,
1590                                icount * sizeof(struct dx_entry));
1591                         dx_set_limit(entries2, dx_node_limit(dir));
1592
1593                         /* Set up root */
1594                         dx_set_count(entries, 1);
1595                         dx_set_block(entries + 0, newblock);
1596                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1597
1598                         /* Add new access path frame */
1599                         frame = frames + 1;
1600                         frame->at = at = at - entries + entries2;
1601                         frame->entries = entries = entries2;
1602                         frame->bh = bh2;
1603                         err = ext4_journal_get_write_access(handle,
1604                                                              frame->bh);
1605                         if (err)
1606                                 goto journal_error;
1607                 }
1608                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1609         }
1610         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1611         if (!de)
1612                 goto cleanup;
1613         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1614         bh = NULL;
1615         goto cleanup;
1616
1617 journal_error:
1618         ext4_std_error(dir->i_sb, err);
1619 cleanup:
1620         if (bh)
1621                 brelse(bh);
1622         dx_release(frames);
1623         return err;
1624 }
1625
1626 /*
1627  * ext4_delete_entry deletes a directory entry by merging it with the
1628  * previous entry
1629  */
1630 static int ext4_delete_entry (handle_t *handle,
1631                               struct inode * dir,
1632                               struct ext4_dir_entry_2 * de_del,
1633                               struct buffer_head * bh)
1634 {
1635         struct ext4_dir_entry_2 * de, * pde;
1636         int i;
1637
1638         i = 0;
1639         pde = NULL;
1640         de = (struct ext4_dir_entry_2 *) bh->b_data;
1641         while (i < bh->b_size) {
1642                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1643                         return -EIO;
1644                 if (de == de_del)  {
1645                         BUFFER_TRACE(bh, "get_write_access");
1646                         ext4_journal_get_write_access(handle, bh);
1647                         if (pde)
1648                                 pde->rec_len = ext4_rec_len_to_disk(
1649                                         ext4_rec_len_from_disk(pde->rec_len) +
1650                                         ext4_rec_len_from_disk(de->rec_len));
1651                         else
1652                                 de->inode = 0;
1653                         dir->i_version++;
1654                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1655                         ext4_journal_dirty_metadata(handle, bh);
1656                         return 0;
1657                 }
1658                 i += ext4_rec_len_from_disk(de->rec_len);
1659                 pde = de;
1660                 de = ext4_next_entry(de);
1661         }
1662         return -ENOENT;
1663 }
1664
1665 /*
1666  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1667  * since this indicates that nlinks count was previously 1.
1668  */
1669 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1670 {
1671         inc_nlink(inode);
1672         if (is_dx(inode) && inode->i_nlink > 1) {
1673                 /* limit is 16-bit i_links_count */
1674                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1675                         inode->i_nlink = 1;
1676                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1677                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1678                 }
1679         }
1680 }
1681
1682 /*
1683  * If a directory had nlink == 1, then we should let it be 1. This indicates
1684  * directory has >EXT4_LINK_MAX subdirs.
1685  */
1686 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1687 {
1688         drop_nlink(inode);
1689         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1690                 inc_nlink(inode);
1691 }
1692
1693
1694 static int ext4_add_nondir(handle_t *handle,
1695                 struct dentry *dentry, struct inode *inode)
1696 {
1697         int err = ext4_add_entry(handle, dentry, inode);
1698         if (!err) {
1699                 ext4_mark_inode_dirty(handle, inode);
1700                 d_instantiate(dentry, inode);
1701                 return 0;
1702         }
1703         drop_nlink(inode);
1704         iput(inode);
1705         return err;
1706 }
1707
1708 /*
1709  * By the time this is called, we already have created
1710  * the directory cache entry for the new file, but it
1711  * is so far negative - it has no inode.
1712  *
1713  * If the create succeeds, we fill in the inode information
1714  * with d_instantiate().
1715  */
1716 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1717                 struct nameidata *nd)
1718 {
1719         handle_t *handle;
1720         struct inode * inode;
1721         int err, retries = 0;
1722
1723 retry:
1724         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1725                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1726                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1727         if (IS_ERR(handle))
1728                 return PTR_ERR(handle);
1729
1730         if (IS_DIRSYNC(dir))
1731                 handle->h_sync = 1;
1732
1733         inode = ext4_new_inode (handle, dir, mode);
1734         err = PTR_ERR(inode);
1735         if (!IS_ERR(inode)) {
1736                 inode->i_op = &ext4_file_inode_operations;
1737                 inode->i_fop = &ext4_file_operations;
1738                 ext4_set_aops(inode);
1739                 err = ext4_add_nondir(handle, dentry, inode);
1740         }
1741         ext4_journal_stop(handle);
1742         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1743                 goto retry;
1744         return err;
1745 }
1746
1747 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1748                         int mode, dev_t rdev)
1749 {
1750         handle_t *handle;
1751         struct inode *inode;
1752         int err, retries = 0;
1753
1754         if (!new_valid_dev(rdev))
1755                 return -EINVAL;
1756
1757 retry:
1758         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1759                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1760                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1761         if (IS_ERR(handle))
1762                 return PTR_ERR(handle);
1763
1764         if (IS_DIRSYNC(dir))
1765                 handle->h_sync = 1;
1766
1767         inode = ext4_new_inode (handle, dir, mode);
1768         err = PTR_ERR(inode);
1769         if (!IS_ERR(inode)) {
1770                 init_special_inode(inode, inode->i_mode, rdev);
1771 #ifdef CONFIG_EXT4DEV_FS_XATTR
1772                 inode->i_op = &ext4_special_inode_operations;
1773 #endif
1774                 err = ext4_add_nondir(handle, dentry, inode);
1775         }
1776         ext4_journal_stop(handle);
1777         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1778                 goto retry;
1779         return err;
1780 }
1781
1782 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1783 {
1784         handle_t *handle;
1785         struct inode * inode;
1786         struct buffer_head * dir_block;
1787         struct ext4_dir_entry_2 * de;
1788         int err, retries = 0;
1789
1790         if (EXT4_DIR_LINK_MAX(dir))
1791                 return -EMLINK;
1792
1793 retry:
1794         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1795                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1796                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1797         if (IS_ERR(handle))
1798                 return PTR_ERR(handle);
1799
1800         if (IS_DIRSYNC(dir))
1801                 handle->h_sync = 1;
1802
1803         inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1804         err = PTR_ERR(inode);
1805         if (IS_ERR(inode))
1806                 goto out_stop;
1807
1808         inode->i_op = &ext4_dir_inode_operations;
1809         inode->i_fop = &ext4_dir_operations;
1810         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1811         dir_block = ext4_bread (handle, inode, 0, 1, &err);
1812         if (!dir_block)
1813                 goto out_clear_inode;
1814         BUFFER_TRACE(dir_block, "get_write_access");
1815         ext4_journal_get_write_access(handle, dir_block);
1816         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1817         de->inode = cpu_to_le32(inode->i_ino);
1818         de->name_len = 1;
1819         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
1820         strcpy (de->name, ".");
1821         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1822         de = ext4_next_entry(de);
1823         de->inode = cpu_to_le32(dir->i_ino);
1824         de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1825                                                 EXT4_DIR_REC_LEN(1));
1826         de->name_len = 2;
1827         strcpy (de->name, "..");
1828         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1829         inode->i_nlink = 2;
1830         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1831         ext4_journal_dirty_metadata(handle, dir_block);
1832         brelse (dir_block);
1833         ext4_mark_inode_dirty(handle, inode);
1834         err = ext4_add_entry (handle, dentry, inode);
1835         if (err) {
1836 out_clear_inode:
1837                 clear_nlink(inode);
1838                 ext4_mark_inode_dirty(handle, inode);
1839                 iput (inode);
1840                 goto out_stop;
1841         }
1842         ext4_inc_count(handle, dir);
1843         ext4_update_dx_flag(dir);
1844         ext4_mark_inode_dirty(handle, dir);
1845         d_instantiate(dentry, inode);
1846 out_stop:
1847         ext4_journal_stop(handle);
1848         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1849                 goto retry;
1850         return err;
1851 }
1852
1853 /*
1854  * routine to check that the specified directory is empty (for rmdir)
1855  */
1856 static int empty_dir (struct inode * inode)
1857 {
1858         unsigned long offset;
1859         struct buffer_head * bh;
1860         struct ext4_dir_entry_2 * de, * de1;
1861         struct super_block * sb;
1862         int err = 0;
1863
1864         sb = inode->i_sb;
1865         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1866             !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1867                 if (err)
1868                         ext4_error(inode->i_sb, __func__,
1869                                    "error %d reading directory #%lu offset 0",
1870                                    err, inode->i_ino);
1871                 else
1872                         ext4_warning(inode->i_sb, __func__,
1873                                      "bad directory (dir #%lu) - no data block",
1874                                      inode->i_ino);
1875                 return 1;
1876         }
1877         de = (struct ext4_dir_entry_2 *) bh->b_data;
1878         de1 = ext4_next_entry(de);
1879         if (le32_to_cpu(de->inode) != inode->i_ino ||
1880                         !le32_to_cpu(de1->inode) ||
1881                         strcmp (".", de->name) ||
1882                         strcmp ("..", de1->name)) {
1883                 ext4_warning (inode->i_sb, "empty_dir",
1884                               "bad directory (dir #%lu) - no `.' or `..'",
1885                               inode->i_ino);
1886                 brelse (bh);
1887                 return 1;
1888         }
1889         offset = ext4_rec_len_from_disk(de->rec_len) +
1890                  ext4_rec_len_from_disk(de1->rec_len);
1891         de = ext4_next_entry(de1);
1892         while (offset < inode->i_size ) {
1893                 if (!bh ||
1894                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1895                         err = 0;
1896                         brelse (bh);
1897                         bh = ext4_bread (NULL, inode,
1898                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1899                         if (!bh) {
1900                                 if (err)
1901                                         ext4_error(sb, __func__,
1902                                                    "error %d reading directory"
1903                                                    " #%lu offset %lu",
1904                                                    err, inode->i_ino, offset);
1905                                 offset += sb->s_blocksize;
1906                                 continue;
1907                         }
1908                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1909                 }
1910                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1911                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1912                                                          sb->s_blocksize);
1913                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1914                         continue;
1915                 }
1916                 if (le32_to_cpu(de->inode)) {
1917                         brelse (bh);
1918                         return 0;
1919                 }
1920                 offset += ext4_rec_len_from_disk(de->rec_len);
1921                 de = ext4_next_entry(de);
1922         }
1923         brelse (bh);
1924         return 1;
1925 }
1926
1927 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1928  * such inodes, starting at the superblock, in case we crash before the
1929  * file is closed/deleted, or in case the inode truncate spans multiple
1930  * transactions and the last transaction is not recovered after a crash.
1931  *
1932  * At filesystem recovery time, we walk this list deleting unlinked
1933  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1934  */
1935 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1936 {
1937         struct super_block *sb = inode->i_sb;
1938         struct ext4_iloc iloc;
1939         int err = 0, rc;
1940
1941         lock_super(sb);
1942         if (!list_empty(&EXT4_I(inode)->i_orphan))
1943                 goto out_unlock;
1944
1945         /* Orphan handling is only valid for files with data blocks
1946          * being truncated, or files being unlinked. */
1947
1948         /* @@@ FIXME: Observation from aviro:
1949          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1950          * here (on lock_super()), so race with ext4_link() which might bump
1951          * ->i_nlink. For, say it, character device. Not a regular file,
1952          * not a directory, not a symlink and ->i_nlink > 0.
1953          */
1954         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1955                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1956
1957         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1958         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1959         if (err)
1960                 goto out_unlock;
1961
1962         err = ext4_reserve_inode_write(handle, inode, &iloc);
1963         if (err)
1964                 goto out_unlock;
1965
1966         /* Insert this inode at the head of the on-disk orphan list... */
1967         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1968         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1969         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1970         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1971         if (!err)
1972                 err = rc;
1973
1974         /* Only add to the head of the in-memory list if all the
1975          * previous operations succeeded.  If the orphan_add is going to
1976          * fail (possibly taking the journal offline), we can't risk
1977          * leaving the inode on the orphan list: stray orphan-list
1978          * entries can cause panics at unmount time.
1979          *
1980          * This is safe: on error we're going to ignore the orphan list
1981          * anyway on the next recovery. */
1982         if (!err)
1983                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1984
1985         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1986         jbd_debug(4, "orphan inode %lu will point to %d\n",
1987                         inode->i_ino, NEXT_ORPHAN(inode));
1988 out_unlock:
1989         unlock_super(sb);
1990         ext4_std_error(inode->i_sb, err);
1991         return err;
1992 }
1993
1994 /*
1995  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1996  * of such inodes stored on disk, because it is finally being cleaned up.
1997  */
1998 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1999 {
2000         struct list_head *prev;
2001         struct ext4_inode_info *ei = EXT4_I(inode);
2002         struct ext4_sb_info *sbi;
2003         unsigned long ino_next;
2004         struct ext4_iloc iloc;
2005         int err = 0;
2006
2007         lock_super(inode->i_sb);
2008         if (list_empty(&ei->i_orphan)) {
2009                 unlock_super(inode->i_sb);
2010                 return 0;
2011         }
2012
2013         ino_next = NEXT_ORPHAN(inode);
2014         prev = ei->i_orphan.prev;
2015         sbi = EXT4_SB(inode->i_sb);
2016
2017         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2018
2019         list_del_init(&ei->i_orphan);
2020
2021         /* If we're on an error path, we may not have a valid
2022          * transaction handle with which to update the orphan list on
2023          * disk, but we still need to remove the inode from the linked
2024          * list in memory. */
2025         if (!handle)
2026                 goto out;
2027
2028         err = ext4_reserve_inode_write(handle, inode, &iloc);
2029         if (err)
2030                 goto out_err;
2031
2032         if (prev == &sbi->s_orphan) {
2033                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2034                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2035                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2036                 if (err)
2037                         goto out_brelse;
2038                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2039                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2040         } else {
2041                 struct ext4_iloc iloc2;
2042                 struct inode *i_prev =
2043                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2044
2045                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2046                           i_prev->i_ino, ino_next);
2047                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2048                 if (err)
2049                         goto out_brelse;
2050                 NEXT_ORPHAN(i_prev) = ino_next;
2051                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2052         }
2053         if (err)
2054                 goto out_brelse;
2055         NEXT_ORPHAN(inode) = 0;
2056         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2057
2058 out_err:
2059         ext4_std_error(inode->i_sb, err);
2060 out:
2061         unlock_super(inode->i_sb);
2062         return err;
2063
2064 out_brelse:
2065         brelse(iloc.bh);
2066         goto out_err;
2067 }
2068
2069 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2070 {
2071         int retval;
2072         struct inode * inode;
2073         struct buffer_head * bh;
2074         struct ext4_dir_entry_2 * de;
2075         handle_t *handle;
2076
2077         /* Initialize quotas before so that eventual writes go in
2078          * separate transaction */
2079         DQUOT_INIT(dentry->d_inode);
2080         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2081         if (IS_ERR(handle))
2082                 return PTR_ERR(handle);
2083
2084         retval = -ENOENT;
2085         bh = ext4_find_entry (dentry, &de);
2086         if (!bh)
2087                 goto end_rmdir;
2088
2089         if (IS_DIRSYNC(dir))
2090                 handle->h_sync = 1;
2091
2092         inode = dentry->d_inode;
2093
2094         retval = -EIO;
2095         if (le32_to_cpu(de->inode) != inode->i_ino)
2096                 goto end_rmdir;
2097
2098         retval = -ENOTEMPTY;
2099         if (!empty_dir (inode))
2100                 goto end_rmdir;
2101
2102         retval = ext4_delete_entry(handle, dir, de, bh);
2103         if (retval)
2104                 goto end_rmdir;
2105         if (!EXT4_DIR_LINK_EMPTY(inode))
2106                 ext4_warning (inode->i_sb, "ext4_rmdir",
2107                               "empty directory has too many links (%d)",
2108                               inode->i_nlink);
2109         inode->i_version++;
2110         clear_nlink(inode);
2111         /* There's no need to set i_disksize: the fact that i_nlink is
2112          * zero will ensure that the right thing happens during any
2113          * recovery. */
2114         inode->i_size = 0;
2115         ext4_orphan_add(handle, inode);
2116         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2117         ext4_mark_inode_dirty(handle, inode);
2118         ext4_dec_count(handle, dir);
2119         ext4_update_dx_flag(dir);
2120         ext4_mark_inode_dirty(handle, dir);
2121
2122 end_rmdir:
2123         ext4_journal_stop(handle);
2124         brelse (bh);
2125         return retval;
2126 }
2127
2128 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2129 {
2130         int retval;
2131         struct inode * inode;
2132         struct buffer_head * bh;
2133         struct ext4_dir_entry_2 * de;
2134         handle_t *handle;
2135
2136         /* Initialize quotas before so that eventual writes go
2137          * in separate transaction */
2138         DQUOT_INIT(dentry->d_inode);
2139         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2140         if (IS_ERR(handle))
2141                 return PTR_ERR(handle);
2142
2143         if (IS_DIRSYNC(dir))
2144                 handle->h_sync = 1;
2145
2146         retval = -ENOENT;
2147         bh = ext4_find_entry (dentry, &de);
2148         if (!bh)
2149                 goto end_unlink;
2150
2151         inode = dentry->d_inode;
2152
2153         retval = -EIO;
2154         if (le32_to_cpu(de->inode) != inode->i_ino)
2155                 goto end_unlink;
2156
2157         if (!inode->i_nlink) {
2158                 ext4_warning (inode->i_sb, "ext4_unlink",
2159                               "Deleting nonexistent file (%lu), %d",
2160                               inode->i_ino, inode->i_nlink);
2161                 inode->i_nlink = 1;
2162         }
2163         retval = ext4_delete_entry(handle, dir, de, bh);
2164         if (retval)
2165                 goto end_unlink;
2166         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2167         ext4_update_dx_flag(dir);
2168         ext4_mark_inode_dirty(handle, dir);
2169         drop_nlink(inode);
2170         if (!inode->i_nlink)
2171                 ext4_orphan_add(handle, inode);
2172         inode->i_ctime = ext4_current_time(inode);
2173         ext4_mark_inode_dirty(handle, inode);
2174         retval = 0;
2175
2176 end_unlink:
2177         ext4_journal_stop(handle);
2178         brelse (bh);
2179         return retval;
2180 }
2181
2182 static int ext4_symlink (struct inode * dir,
2183                 struct dentry *dentry, const char * symname)
2184 {
2185         handle_t *handle;
2186         struct inode * inode;
2187         int l, err, retries = 0;
2188
2189         l = strlen(symname)+1;
2190         if (l > dir->i_sb->s_blocksize)
2191                 return -ENAMETOOLONG;
2192
2193 retry:
2194         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2195                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2196                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2197         if (IS_ERR(handle))
2198                 return PTR_ERR(handle);
2199
2200         if (IS_DIRSYNC(dir))
2201                 handle->h_sync = 1;
2202
2203         inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2204         err = PTR_ERR(inode);
2205         if (IS_ERR(inode))
2206                 goto out_stop;
2207
2208         if (l > sizeof (EXT4_I(inode)->i_data)) {
2209                 inode->i_op = &ext4_symlink_inode_operations;
2210                 ext4_set_aops(inode);
2211                 /*
2212                  * page_symlink() calls into ext4_prepare/commit_write.
2213                  * We have a transaction open.  All is sweetness.  It also sets
2214                  * i_size in generic_commit_write().
2215                  */
2216                 err = __page_symlink(inode, symname, l,
2217                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2218                 if (err) {
2219                         clear_nlink(inode);
2220                         ext4_mark_inode_dirty(handle, inode);
2221                         iput (inode);
2222                         goto out_stop;
2223                 }
2224         } else {
2225                 /* clear the extent format for fast symlink */
2226                 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2227                 inode->i_op = &ext4_fast_symlink_inode_operations;
2228                 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2229                 inode->i_size = l-1;
2230         }
2231         EXT4_I(inode)->i_disksize = inode->i_size;
2232         err = ext4_add_nondir(handle, dentry, inode);
2233 out_stop:
2234         ext4_journal_stop(handle);
2235         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2236                 goto retry;
2237         return err;
2238 }
2239
2240 static int ext4_link (struct dentry * old_dentry,
2241                 struct inode * dir, struct dentry *dentry)
2242 {
2243         handle_t *handle;
2244         struct inode *inode = old_dentry->d_inode;
2245         int err, retries = 0;
2246
2247         if (EXT4_DIR_LINK_MAX(inode))
2248                 return -EMLINK;
2249
2250         /*
2251          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2252          * otherwise has the potential to corrupt the orphan inode list.
2253          */
2254         if (inode->i_nlink == 0)
2255                 return -ENOENT;
2256
2257 retry:
2258         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2259                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2260         if (IS_ERR(handle))
2261                 return PTR_ERR(handle);
2262
2263         if (IS_DIRSYNC(dir))
2264                 handle->h_sync = 1;
2265
2266         inode->i_ctime = ext4_current_time(inode);
2267         ext4_inc_count(handle, inode);
2268         atomic_inc(&inode->i_count);
2269
2270         err = ext4_add_nondir(handle, dentry, inode);
2271         ext4_journal_stop(handle);
2272         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2273                 goto retry;
2274         return err;
2275 }
2276
2277 #define PARENT_INO(buffer) \
2278         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2279
2280 /*
2281  * Anybody can rename anything with this: the permission checks are left to the
2282  * higher-level routines.
2283  */
2284 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2285                            struct inode * new_dir,struct dentry *new_dentry)
2286 {
2287         handle_t *handle;
2288         struct inode * old_inode, * new_inode;
2289         struct buffer_head * old_bh, * new_bh, * dir_bh;
2290         struct ext4_dir_entry_2 * old_de, * new_de;
2291         int retval;
2292
2293         old_bh = new_bh = dir_bh = NULL;
2294
2295         /* Initialize quotas before so that eventual writes go
2296          * in separate transaction */
2297         if (new_dentry->d_inode)
2298                 DQUOT_INIT(new_dentry->d_inode);
2299         handle = ext4_journal_start(old_dir, 2 *
2300                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2301                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2302         if (IS_ERR(handle))
2303                 return PTR_ERR(handle);
2304
2305         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2306                 handle->h_sync = 1;
2307
2308         old_bh = ext4_find_entry (old_dentry, &old_de);
2309         /*
2310          *  Check for inode number is _not_ due to possible IO errors.
2311          *  We might rmdir the source, keep it as pwd of some process
2312          *  and merrily kill the link to whatever was created under the
2313          *  same name. Goodbye sticky bit ;-<
2314          */
2315         old_inode = old_dentry->d_inode;
2316         retval = -ENOENT;
2317         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2318                 goto end_rename;
2319
2320         new_inode = new_dentry->d_inode;
2321         new_bh = ext4_find_entry (new_dentry, &new_de);
2322         if (new_bh) {
2323                 if (!new_inode) {
2324                         brelse (new_bh);
2325                         new_bh = NULL;
2326                 }
2327         }
2328         if (S_ISDIR(old_inode->i_mode)) {
2329                 if (new_inode) {
2330                         retval = -ENOTEMPTY;
2331                         if (!empty_dir (new_inode))
2332                                 goto end_rename;
2333                 }
2334                 retval = -EIO;
2335                 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2336                 if (!dir_bh)
2337                         goto end_rename;
2338                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2339                         goto end_rename;
2340                 retval = -EMLINK;
2341                 if (!new_inode && new_dir!=old_dir &&
2342                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2343                         goto end_rename;
2344         }
2345         if (!new_bh) {
2346                 retval = ext4_add_entry (handle, new_dentry, old_inode);
2347                 if (retval)
2348                         goto end_rename;
2349         } else {
2350                 BUFFER_TRACE(new_bh, "get write access");
2351                 ext4_journal_get_write_access(handle, new_bh);
2352                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2353                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2354                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2355                         new_de->file_type = old_de->file_type;
2356                 new_dir->i_version++;
2357                 new_dir->i_ctime = new_dir->i_mtime =
2358                                         ext4_current_time(new_dir);
2359                 ext4_mark_inode_dirty(handle, new_dir);
2360                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2361                 ext4_journal_dirty_metadata(handle, new_bh);
2362                 brelse(new_bh);
2363                 new_bh = NULL;
2364         }
2365
2366         /*
2367          * Like most other Unix systems, set the ctime for inodes on a
2368          * rename.
2369          */
2370         old_inode->i_ctime = ext4_current_time(old_inode);
2371         ext4_mark_inode_dirty(handle, old_inode);
2372
2373         /*
2374          * ok, that's it
2375          */
2376         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2377             old_de->name_len != old_dentry->d_name.len ||
2378             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2379             (retval = ext4_delete_entry(handle, old_dir,
2380                                         old_de, old_bh)) == -ENOENT) {
2381                 /* old_de could have moved from under us during htree split, so
2382                  * make sure that we are deleting the right entry.  We might
2383                  * also be pointing to a stale entry in the unused part of
2384                  * old_bh so just checking inum and the name isn't enough. */
2385                 struct buffer_head *old_bh2;
2386                 struct ext4_dir_entry_2 *old_de2;
2387
2388                 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2389                 if (old_bh2) {
2390                         retval = ext4_delete_entry(handle, old_dir,
2391                                                    old_de2, old_bh2);
2392                         brelse(old_bh2);
2393                 }
2394         }
2395         if (retval) {
2396                 ext4_warning(old_dir->i_sb, "ext4_rename",
2397                                 "Deleting old file (%lu), %d, error=%d",
2398                                 old_dir->i_ino, old_dir->i_nlink, retval);
2399         }
2400
2401         if (new_inode) {
2402                 ext4_dec_count(handle, new_inode);
2403                 new_inode->i_ctime = ext4_current_time(new_inode);
2404         }
2405         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2406         ext4_update_dx_flag(old_dir);
2407         if (dir_bh) {
2408                 BUFFER_TRACE(dir_bh, "get_write_access");
2409                 ext4_journal_get_write_access(handle, dir_bh);
2410                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2411                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2412                 ext4_journal_dirty_metadata(handle, dir_bh);
2413                 ext4_dec_count(handle, old_dir);
2414                 if (new_inode) {
2415                         /* checked empty_dir above, can't have another parent,
2416                          * ext4_dec_count() won't work for many-linked dirs */
2417                         new_inode->i_nlink = 0;
2418                 } else {
2419                         ext4_inc_count(handle, new_dir);
2420                         ext4_update_dx_flag(new_dir);
2421                         ext4_mark_inode_dirty(handle, new_dir);
2422                 }
2423         }
2424         ext4_mark_inode_dirty(handle, old_dir);
2425         if (new_inode) {
2426                 ext4_mark_inode_dirty(handle, new_inode);
2427                 if (!new_inode->i_nlink)
2428                         ext4_orphan_add(handle, new_inode);
2429         }
2430         retval = 0;
2431
2432 end_rename:
2433         brelse (dir_bh);
2434         brelse (old_bh);
2435         brelse (new_bh);
2436         ext4_journal_stop(handle);
2437         return retval;
2438 }
2439
2440 /*
2441  * directories can handle most operations...
2442  */
2443 const struct inode_operations ext4_dir_inode_operations = {
2444         .create         = ext4_create,
2445         .lookup         = ext4_lookup,
2446         .link           = ext4_link,
2447         .unlink         = ext4_unlink,
2448         .symlink        = ext4_symlink,
2449         .mkdir          = ext4_mkdir,
2450         .rmdir          = ext4_rmdir,
2451         .mknod          = ext4_mknod,
2452         .rename         = ext4_rename,
2453         .setattr        = ext4_setattr,
2454 #ifdef CONFIG_EXT4DEV_FS_XATTR
2455         .setxattr       = generic_setxattr,
2456         .getxattr       = generic_getxattr,
2457         .listxattr      = ext4_listxattr,
2458         .removexattr    = generic_removexattr,
2459 #endif
2460         .permission     = ext4_permission,
2461 };
2462
2463 const struct inode_operations ext4_special_inode_operations = {
2464         .setattr        = ext4_setattr,
2465 #ifdef CONFIG_EXT4DEV_FS_XATTR
2466         .setxattr       = generic_setxattr,
2467         .getxattr       = generic_getxattr,
2468         .listxattr      = ext4_listxattr,
2469         .removexattr    = generic_removexattr,
2470 #endif
2471         .permission     = ext4_permission,
2472 };