Merge branch 'core/stacktrace' of git://git.kernel.org/pub/scm/linux/kernel/git/tip...
[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  * p is at least 6 bytes before the end of page
187  */
188 static inline struct ext4_dir_entry_2 *
189 ext4_next_entry(struct ext4_dir_entry_2 *p)
190 {
191         return (struct ext4_dir_entry_2 *)((char *)p +
192                 ext4_rec_len_from_disk(p->rec_len));
193 }
194
195 /*
196  * Future: use high four bits of block for coalesce-on-delete flags
197  * Mask them off for now.
198  */
199
200 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
201 {
202         return le32_to_cpu(entry->block) & 0x00ffffff;
203 }
204
205 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
206 {
207         entry->block = cpu_to_le32(value);
208 }
209
210 static inline unsigned dx_get_hash (struct dx_entry *entry)
211 {
212         return le32_to_cpu(entry->hash);
213 }
214
215 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
216 {
217         entry->hash = cpu_to_le32(value);
218 }
219
220 static inline unsigned dx_get_count (struct dx_entry *entries)
221 {
222         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
223 }
224
225 static inline unsigned dx_get_limit (struct dx_entry *entries)
226 {
227         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
228 }
229
230 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
231 {
232         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
233 }
234
235 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
236 {
237         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
238 }
239
240 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
241 {
242         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
243                 EXT4_DIR_REC_LEN(2) - infosize;
244         return entry_space / sizeof(struct dx_entry);
245 }
246
247 static inline unsigned dx_node_limit (struct inode *dir)
248 {
249         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
250         return entry_space / sizeof(struct dx_entry);
251 }
252
253 /*
254  * Debug
255  */
256 #ifdef DX_DEBUG
257 static void dx_show_index (char * label, struct dx_entry *entries)
258 {
259         int i, n = dx_get_count (entries);
260         printk("%s index ", label);
261         for (i = 0; i < n; i++) {
262                 printk("%x->%lu ", i? dx_get_hash(entries + i) :
263                                 0, (unsigned long)dx_get_block(entries + i));
264         }
265         printk("\n");
266 }
267
268 struct stats
269 {
270         unsigned names;
271         unsigned space;
272         unsigned bcount;
273 };
274
275 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
276                                  int size, int show_names)
277 {
278         unsigned names = 0, space = 0;
279         char *base = (char *) de;
280         struct dx_hash_info h = *hinfo;
281
282         printk("names: ");
283         while ((char *) de < base + size)
284         {
285                 if (de->inode)
286                 {
287                         if (show_names)
288                         {
289                                 int len = de->name_len;
290                                 char *name = de->name;
291                                 while (len--) printk("%c", *name++);
292                                 ext4fs_dirhash(de->name, de->name_len, &h);
293                                 printk(":%x.%u ", h.hash,
294                                        ((char *) de - base));
295                         }
296                         space += EXT4_DIR_REC_LEN(de->name_len);
297                         names++;
298                 }
299                 de = ext4_next_entry(de);
300         }
301         printk("(%i)\n", names);
302         return (struct stats) { names, space, 1 };
303 }
304
305 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
306                              struct dx_entry *entries, int levels)
307 {
308         unsigned blocksize = dir->i_sb->s_blocksize;
309         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
310         unsigned bcount = 0;
311         struct buffer_head *bh;
312         int err;
313         printk("%i indexed blocks...\n", count);
314         for (i = 0; i < count; i++, entries++)
315         {
316                 ext4_lblk_t block = dx_get_block(entries);
317                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
318                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
319                 struct stats stats;
320                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
321                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
322                 stats = levels?
323                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
324                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
325                 names += stats.names;
326                 space += stats.space;
327                 bcount += stats.bcount;
328                 brelse (bh);
329         }
330         if (bcount)
331                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
332                         names, space/bcount,(space/bcount)*100/blocksize);
333         return (struct stats) { names, space, bcount};
334 }
335 #endif /* DX_DEBUG */
336
337 /*
338  * Probe for a directory leaf block to search.
339  *
340  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
341  * error in the directory index, and the caller should fall back to
342  * searching the directory normally.  The callers of dx_probe **MUST**
343  * check for this error code, and make sure it never gets reflected
344  * back to userspace.
345  */
346 static struct dx_frame *
347 dx_probe(struct dentry *dentry, struct inode *dir,
348          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
349 {
350         unsigned count, indirect;
351         struct dx_entry *at, *entries, *p, *q, *m;
352         struct dx_root *root;
353         struct buffer_head *bh;
354         struct dx_frame *frame = frame_in;
355         u32 hash;
356
357         frame->bh = NULL;
358         if (dentry)
359                 dir = dentry->d_parent->d_inode;
360         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
361                 goto fail;
362         root = (struct dx_root *) bh->b_data;
363         if (root->info.hash_version != DX_HASH_TEA &&
364             root->info.hash_version != DX_HASH_HALF_MD4 &&
365             root->info.hash_version != DX_HASH_LEGACY) {
366                 ext4_warning(dir->i_sb, __func__,
367                              "Unrecognised inode hash code %d",
368                              root->info.hash_version);
369                 brelse(bh);
370                 *err = ERR_BAD_DX_DIR;
371                 goto fail;
372         }
373         hinfo->hash_version = root->info.hash_version;
374         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
375         if (dentry)
376                 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
377         hash = hinfo->hash;
378
379         if (root->info.unused_flags & 1) {
380                 ext4_warning(dir->i_sb, __func__,
381                              "Unimplemented inode hash flags: %#06x",
382                              root->info.unused_flags);
383                 brelse(bh);
384                 *err = ERR_BAD_DX_DIR;
385                 goto fail;
386         }
387
388         if ((indirect = root->info.indirect_levels) > 1) {
389                 ext4_warning(dir->i_sb, __func__,
390                              "Unimplemented inode hash depth: %#06x",
391                              root->info.indirect_levels);
392                 brelse(bh);
393                 *err = ERR_BAD_DX_DIR;
394                 goto fail;
395         }
396
397         entries = (struct dx_entry *) (((char *)&root->info) +
398                                        root->info.info_length);
399
400         if (dx_get_limit(entries) != dx_root_limit(dir,
401                                                    root->info.info_length)) {
402                 ext4_warning(dir->i_sb, __func__,
403                              "dx entry: limit != root limit");
404                 brelse(bh);
405                 *err = ERR_BAD_DX_DIR;
406                 goto fail;
407         }
408
409         dxtrace (printk("Look up %x", hash));
410         while (1)
411         {
412                 count = dx_get_count(entries);
413                 if (!count || count > dx_get_limit(entries)) {
414                         ext4_warning(dir->i_sb, __func__,
415                                      "dx entry: no count or count > limit");
416                         brelse(bh);
417                         *err = ERR_BAD_DX_DIR;
418                         goto fail2;
419                 }
420
421                 p = entries + 1;
422                 q = entries + count - 1;
423                 while (p <= q)
424                 {
425                         m = p + (q - p)/2;
426                         dxtrace(printk("."));
427                         if (dx_get_hash(m) > hash)
428                                 q = m - 1;
429                         else
430                                 p = m + 1;
431                 }
432
433                 if (0) // linear search cross check
434                 {
435                         unsigned n = count - 1;
436                         at = entries;
437                         while (n--)
438                         {
439                                 dxtrace(printk(","));
440                                 if (dx_get_hash(++at) > hash)
441                                 {
442                                         at--;
443                                         break;
444                                 }
445                         }
446                         assert (at == p - 1);
447                 }
448
449                 at = p - 1;
450                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
451                 frame->bh = bh;
452                 frame->entries = entries;
453                 frame->at = at;
454                 if (!indirect--) return frame;
455                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
456                         goto fail2;
457                 at = entries = ((struct dx_node *) bh->b_data)->entries;
458                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
459                         ext4_warning(dir->i_sb, __func__,
460                                      "dx entry: limit != node limit");
461                         brelse(bh);
462                         *err = ERR_BAD_DX_DIR;
463                         goto fail2;
464                 }
465                 frame++;
466                 frame->bh = NULL;
467         }
468 fail2:
469         while (frame >= frame_in) {
470                 brelse(frame->bh);
471                 frame--;
472         }
473 fail:
474         if (*err == ERR_BAD_DX_DIR)
475                 ext4_warning(dir->i_sb, __func__,
476                              "Corrupt dir inode %ld, running e2fsck is "
477                              "recommended.", dir->i_ino);
478         return NULL;
479 }
480
481 static void dx_release (struct dx_frame *frames)
482 {
483         if (frames[0].bh == NULL)
484                 return;
485
486         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
487                 brelse(frames[1].bh);
488         brelse(frames[0].bh);
489 }
490
491 /*
492  * This function increments the frame pointer to search the next leaf
493  * block, and reads in the necessary intervening nodes if the search
494  * should be necessary.  Whether or not the search is necessary is
495  * controlled by the hash parameter.  If the hash value is even, then
496  * the search is only continued if the next block starts with that
497  * hash value.  This is used if we are searching for a specific file.
498  *
499  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
500  *
501  * This function returns 1 if the caller should continue to search,
502  * or 0 if it should not.  If there is an error reading one of the
503  * index blocks, it will a negative error code.
504  *
505  * If start_hash is non-null, it will be filled in with the starting
506  * hash of the next page.
507  */
508 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
509                                  struct dx_frame *frame,
510                                  struct dx_frame *frames,
511                                  __u32 *start_hash)
512 {
513         struct dx_frame *p;
514         struct buffer_head *bh;
515         int err, num_frames = 0;
516         __u32 bhash;
517
518         p = frame;
519         /*
520          * Find the next leaf page by incrementing the frame pointer.
521          * If we run out of entries in the interior node, loop around and
522          * increment pointer in the parent node.  When we break out of
523          * this loop, num_frames indicates the number of interior
524          * nodes need to be read.
525          */
526         while (1) {
527                 if (++(p->at) < p->entries + dx_get_count(p->entries))
528                         break;
529                 if (p == frames)
530                         return 0;
531                 num_frames++;
532                 p--;
533         }
534
535         /*
536          * If the hash is 1, then continue only if the next page has a
537          * continuation hash of any value.  This is used for readdir
538          * handling.  Otherwise, check to see if the hash matches the
539          * desired contiuation hash.  If it doesn't, return since
540          * there's no point to read in the successive index pages.
541          */
542         bhash = dx_get_hash(p->at);
543         if (start_hash)
544                 *start_hash = bhash;
545         if ((hash & 1) == 0) {
546                 if ((bhash & ~1) != hash)
547                         return 0;
548         }
549         /*
550          * If the hash is HASH_NB_ALWAYS, we always go to the next
551          * block so no check is necessary
552          */
553         while (num_frames--) {
554                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
555                                       0, &err)))
556                         return err; /* Failure */
557                 p++;
558                 brelse (p->bh);
559                 p->bh = bh;
560                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
561         }
562         return 1;
563 }
564
565
566 /*
567  * This function fills a red-black tree with information from a
568  * directory block.  It returns the number directory entries loaded
569  * into the tree.  If there is an error it is returned in err.
570  */
571 static int htree_dirblock_to_tree(struct file *dir_file,
572                                   struct inode *dir, ext4_lblk_t block,
573                                   struct dx_hash_info *hinfo,
574                                   __u32 start_hash, __u32 start_minor_hash)
575 {
576         struct buffer_head *bh;
577         struct ext4_dir_entry_2 *de, *top;
578         int err, count = 0;
579
580         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
581                                                         (unsigned long)block));
582         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
583                 return err;
584
585         de = (struct ext4_dir_entry_2 *) bh->b_data;
586         top = (struct ext4_dir_entry_2 *) ((char *) de +
587                                            dir->i_sb->s_blocksize -
588                                            EXT4_DIR_REC_LEN(0));
589         for (; de < top; de = ext4_next_entry(de)) {
590                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
591                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
592                                                 +((char *)de - bh->b_data))) {
593                         /* On error, skip the f_pos to the next block. */
594                         dir_file->f_pos = (dir_file->f_pos |
595                                         (dir->i_sb->s_blocksize - 1)) + 1;
596                         brelse (bh);
597                         return count;
598                 }
599                 ext4fs_dirhash(de->name, de->name_len, hinfo);
600                 if ((hinfo->hash < start_hash) ||
601                     ((hinfo->hash == start_hash) &&
602                      (hinfo->minor_hash < start_minor_hash)))
603                         continue;
604                 if (de->inode == 0)
605                         continue;
606                 if ((err = ext4_htree_store_dirent(dir_file,
607                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
608                         brelse(bh);
609                         return err;
610                 }
611                 count++;
612         }
613         brelse(bh);
614         return count;
615 }
616
617
618 /*
619  * This function fills a red-black tree with information from a
620  * directory.  We start scanning the directory in hash order, starting
621  * at start_hash and start_minor_hash.
622  *
623  * This function returns the number of entries inserted into the tree,
624  * or a negative error code.
625  */
626 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
627                          __u32 start_minor_hash, __u32 *next_hash)
628 {
629         struct dx_hash_info hinfo;
630         struct ext4_dir_entry_2 *de;
631         struct dx_frame frames[2], *frame;
632         struct inode *dir;
633         ext4_lblk_t block;
634         int count = 0;
635         int ret, err;
636         __u32 hashval;
637
638         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
639                        start_minor_hash));
640         dir = dir_file->f_path.dentry->d_inode;
641         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
642                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
643                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
644                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
645                                                start_hash, start_minor_hash);
646                 *next_hash = ~0;
647                 return count;
648         }
649         hinfo.hash = start_hash;
650         hinfo.minor_hash = 0;
651         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
652         if (!frame)
653                 return err;
654
655         /* Add '.' and '..' from the htree header */
656         if (!start_hash && !start_minor_hash) {
657                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
658                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
659                         goto errout;
660                 count++;
661         }
662         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
663                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
664                 de = ext4_next_entry(de);
665                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
666                         goto errout;
667                 count++;
668         }
669
670         while (1) {
671                 block = dx_get_block(frame->at);
672                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
673                                              start_hash, start_minor_hash);
674                 if (ret < 0) {
675                         err = ret;
676                         goto errout;
677                 }
678                 count += ret;
679                 hashval = ~0;
680                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
681                                             frame, frames, &hashval);
682                 *next_hash = hashval;
683                 if (ret < 0) {
684                         err = ret;
685                         goto errout;
686                 }
687                 /*
688                  * Stop if:  (a) there are no more entries, or
689                  * (b) we have inserted at least one entry and the
690                  * next hash value is not a continuation
691                  */
692                 if ((ret == 0) ||
693                     (count && ((hashval & 1) == 0)))
694                         break;
695         }
696         dx_release(frames);
697         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
698                        count, *next_hash));
699         return count;
700 errout:
701         dx_release(frames);
702         return (err);
703 }
704
705
706 /*
707  * Directory block splitting, compacting
708  */
709
710 /*
711  * Create map of hash values, offsets, and sizes, stored at end of block.
712  * Returns number of entries mapped.
713  */
714 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
715                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
716 {
717         int count = 0;
718         char *base = (char *) de;
719         struct dx_hash_info h = *hinfo;
720
721         while ((char *) de < base + size)
722         {
723                 if (de->name_len && de->inode) {
724                         ext4fs_dirhash(de->name, de->name_len, &h);
725                         map_tail--;
726                         map_tail->hash = h.hash;
727                         map_tail->offs = (u16) ((char *) de - base);
728                         map_tail->size = le16_to_cpu(de->rec_len);
729                         count++;
730                         cond_resched();
731                 }
732                 /* XXX: do we need to check rec_len == 0 case? -Chris */
733                 de = ext4_next_entry(de);
734         }
735         return count;
736 }
737
738 /* Sort map by hash value */
739 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
740 {
741         struct dx_map_entry *p, *q, *top = map + count - 1;
742         int more;
743         /* Combsort until bubble sort doesn't suck */
744         while (count > 2) {
745                 count = count*10/13;
746                 if (count - 9 < 2) /* 9, 10 -> 11 */
747                         count = 11;
748                 for (p = top, q = p - count; q >= map; p--, q--)
749                         if (p->hash < q->hash)
750                                 swap(*p, *q);
751         }
752         /* Garden variety bubble sort */
753         do {
754                 more = 0;
755                 q = top;
756                 while (q-- > map) {
757                         if (q[1].hash >= q[0].hash)
758                                 continue;
759                         swap(*(q+1), *q);
760                         more = 1;
761                 }
762         } while(more);
763 }
764
765 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
766 {
767         struct dx_entry *entries = frame->entries;
768         struct dx_entry *old = frame->at, *new = old + 1;
769         int count = dx_get_count(entries);
770
771         assert(count < dx_get_limit(entries));
772         assert(old < entries + count);
773         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
774         dx_set_hash(new, hash);
775         dx_set_block(new, block);
776         dx_set_count(entries, count + 1);
777 }
778
779 static void ext4_update_dx_flag(struct inode *inode)
780 {
781         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
782                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
783                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
784 }
785
786 /*
787  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
788  *
789  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
790  * `de != NULL' is guaranteed by caller.
791  */
792 static inline int ext4_match (int len, const char * const name,
793                               struct ext4_dir_entry_2 * de)
794 {
795         if (len != de->name_len)
796                 return 0;
797         if (!de->inode)
798                 return 0;
799         return !memcmp(name, de->name, len);
800 }
801
802 /*
803  * Returns 0 if not found, -1 on failure, and 1 on success
804  */
805 static inline int search_dirblock(struct buffer_head * bh,
806                                   struct inode *dir,
807                                   struct dentry *dentry,
808                                   unsigned long offset,
809                                   struct ext4_dir_entry_2 ** res_dir)
810 {
811         struct ext4_dir_entry_2 * de;
812         char * dlimit;
813         int de_len;
814         const char *name = dentry->d_name.name;
815         int namelen = dentry->d_name.len;
816
817         de = (struct ext4_dir_entry_2 *) bh->b_data;
818         dlimit = bh->b_data + dir->i_sb->s_blocksize;
819         while ((char *) de < dlimit) {
820                 /* this code is executed quadratically often */
821                 /* do minimal checking `by hand' */
822
823                 if ((char *) de + namelen <= dlimit &&
824                     ext4_match (namelen, name, de)) {
825                         /* found a match - just to be sure, do a full check */
826                         if (!ext4_check_dir_entry("ext4_find_entry",
827                                                   dir, de, bh, offset))
828                                 return -1;
829                         *res_dir = de;
830                         return 1;
831                 }
832                 /* prevent looping on a bad block */
833                 de_len = ext4_rec_len_from_disk(de->rec_len);
834                 if (de_len <= 0)
835                         return -1;
836                 offset += de_len;
837                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
838         }
839         return 0;
840 }
841
842
843 /*
844  *      ext4_find_entry()
845  *
846  * finds an entry in the specified directory with the wanted name. It
847  * returns the cache buffer in which the entry was found, and the entry
848  * itself (as a parameter - res_dir). It does NOT read the inode of the
849  * entry - you'll have to do that yourself if you want to.
850  *
851  * The returned buffer_head has ->b_count elevated.  The caller is expected
852  * to brelse() it when appropriate.
853  */
854 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
855                                         struct ext4_dir_entry_2 ** res_dir)
856 {
857         struct super_block * sb;
858         struct buffer_head * bh_use[NAMEI_RA_SIZE];
859         struct buffer_head * bh, *ret = NULL;
860         ext4_lblk_t start, block, b;
861         int ra_max = 0;         /* Number of bh's in the readahead
862                                    buffer, bh_use[] */
863         int ra_ptr = 0;         /* Current index into readahead
864                                    buffer */
865         int num = 0;
866         ext4_lblk_t  nblocks;
867         int i, err;
868         struct inode *dir = dentry->d_parent->d_inode;
869         int namelen;
870
871         *res_dir = NULL;
872         sb = dir->i_sb;
873         namelen = dentry->d_name.len;
874         if (namelen > EXT4_NAME_LEN)
875                 return NULL;
876         if (is_dx(dir)) {
877                 bh = ext4_dx_find_entry(dentry, res_dir, &err);
878                 /*
879                  * On success, or if the error was file not found,
880                  * return.  Otherwise, fall back to doing a search the
881                  * old fashioned way.
882                  */
883                 if (bh || (err != ERR_BAD_DX_DIR))
884                         return bh;
885                 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
886         }
887         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
888         start = EXT4_I(dir)->i_dir_start_lookup;
889         if (start >= nblocks)
890                 start = 0;
891         block = start;
892 restart:
893         do {
894                 /*
895                  * We deal with the read-ahead logic here.
896                  */
897                 if (ra_ptr >= ra_max) {
898                         /* Refill the readahead buffer */
899                         ra_ptr = 0;
900                         b = block;
901                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
902                                 /*
903                                  * Terminate if we reach the end of the
904                                  * directory and must wrap, or if our
905                                  * search has finished at this block.
906                                  */
907                                 if (b >= nblocks || (num && block == start)) {
908                                         bh_use[ra_max] = NULL;
909                                         break;
910                                 }
911                                 num++;
912                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
913                                 bh_use[ra_max] = bh;
914                                 if (bh)
915                                         ll_rw_block(READ_META, 1, &bh);
916                         }
917                 }
918                 if ((bh = bh_use[ra_ptr++]) == NULL)
919                         goto next;
920                 wait_on_buffer(bh);
921                 if (!buffer_uptodate(bh)) {
922                         /* read error, skip block & hope for the best */
923                         ext4_error(sb, __func__, "reading directory #%lu "
924                                    "offset %lu", dir->i_ino,
925                                    (unsigned long)block);
926                         brelse(bh);
927                         goto next;
928                 }
929                 i = search_dirblock(bh, dir, dentry,
930                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
931                 if (i == 1) {
932                         EXT4_I(dir)->i_dir_start_lookup = block;
933                         ret = bh;
934                         goto cleanup_and_exit;
935                 } else {
936                         brelse(bh);
937                         if (i < 0)
938                                 goto cleanup_and_exit;
939                 }
940         next:
941                 if (++block >= nblocks)
942                         block = 0;
943         } while (block != start);
944
945         /*
946          * If the directory has grown while we were searching, then
947          * search the last part of the directory before giving up.
948          */
949         block = nblocks;
950         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
951         if (block < nblocks) {
952                 start = 0;
953                 goto restart;
954         }
955
956 cleanup_and_exit:
957         /* Clean up the read-ahead blocks */
958         for (; ra_ptr < ra_max; ra_ptr++)
959                 brelse (bh_use[ra_ptr]);
960         return ret;
961 }
962
963 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
964                        struct ext4_dir_entry_2 **res_dir, int *err)
965 {
966         struct super_block * sb;
967         struct dx_hash_info     hinfo;
968         u32 hash;
969         struct dx_frame frames[2], *frame;
970         struct ext4_dir_entry_2 *de, *top;
971         struct buffer_head *bh;
972         ext4_lblk_t block;
973         int retval;
974         int namelen = dentry->d_name.len;
975         const u8 *name = dentry->d_name.name;
976         struct inode *dir = dentry->d_parent->d_inode;
977
978         sb = dir->i_sb;
979         /* NFS may look up ".." - look at dx_root directory block */
980         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
981                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
982                         return NULL;
983         } else {
984                 frame = frames;
985                 frame->bh = NULL;                       /* for dx_release() */
986                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
987                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
988         }
989         hash = hinfo.hash;
990         do {
991                 block = dx_get_block(frame->at);
992                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
993                         goto errout;
994                 de = (struct ext4_dir_entry_2 *) bh->b_data;
995                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
996                                        EXT4_DIR_REC_LEN(0));
997                 for (; de < top; de = ext4_next_entry(de)) {
998                         int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
999                                   + ((char *) de - bh->b_data);
1000
1001                         if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1002                                 brelse(bh);
1003                                 *err = ERR_BAD_DX_DIR;
1004                                 goto errout;
1005                         }
1006
1007                         if (ext4_match(namelen, name, de)) {
1008                                 *res_dir = de;
1009                                 dx_release(frames);
1010                                 return bh;
1011                         }
1012                 }
1013                 brelse (bh);
1014                 /* Check to see if we should continue to search */
1015                 retval = ext4_htree_next_block(dir, hash, frame,
1016                                                frames, NULL);
1017                 if (retval < 0) {
1018                         ext4_warning(sb, __func__,
1019                              "error reading index page in directory #%lu",
1020                              dir->i_ino);
1021                         *err = retval;
1022                         goto errout;
1023                 }
1024         } while (retval == 1);
1025
1026         *err = -ENOENT;
1027 errout:
1028         dxtrace(printk("%s not found\n", name));
1029         dx_release (frames);
1030         return NULL;
1031 }
1032
1033 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1034 {
1035         struct inode * inode;
1036         struct ext4_dir_entry_2 * de;
1037         struct buffer_head * bh;
1038
1039         if (dentry->d_name.len > EXT4_NAME_LEN)
1040                 return ERR_PTR(-ENAMETOOLONG);
1041
1042         bh = ext4_find_entry(dentry, &de);
1043         inode = NULL;
1044         if (bh) {
1045                 unsigned long ino = le32_to_cpu(de->inode);
1046                 brelse (bh);
1047                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1048                         ext4_error(dir->i_sb, "ext4_lookup",
1049                                    "bad inode number: %lu", ino);
1050                         return ERR_PTR(-EIO);
1051                 }
1052                 inode = ext4_iget(dir->i_sb, ino);
1053                 if (IS_ERR(inode))
1054                         return ERR_CAST(inode);
1055         }
1056         return d_splice_alias(inode, dentry);
1057 }
1058
1059
1060 struct dentry *ext4_get_parent(struct dentry *child)
1061 {
1062         unsigned long ino;
1063         struct dentry *parent;
1064         struct inode *inode;
1065         struct dentry dotdot;
1066         struct ext4_dir_entry_2 * de;
1067         struct buffer_head *bh;
1068
1069         dotdot.d_name.name = "..";
1070         dotdot.d_name.len = 2;
1071         dotdot.d_parent = child; /* confusing, isn't it! */
1072
1073         bh = ext4_find_entry(&dotdot, &de);
1074         inode = NULL;
1075         if (!bh)
1076                 return ERR_PTR(-ENOENT);
1077         ino = le32_to_cpu(de->inode);
1078         brelse(bh);
1079
1080         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1081                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1082                            "bad inode number: %lu", ino);
1083                 return ERR_PTR(-EIO);
1084         }
1085
1086         inode = ext4_iget(child->d_inode->i_sb, ino);
1087         if (IS_ERR(inode))
1088                 return ERR_CAST(inode);
1089
1090         parent = d_alloc_anon(inode);
1091         if (!parent) {
1092                 iput(inode);
1093                 parent = ERR_PTR(-ENOMEM);
1094         }
1095         return parent;
1096 }
1097
1098 #define S_SHIFT 12
1099 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1100         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1101         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1102         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1103         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1104         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1105         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1106         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1107 };
1108
1109 static inline void ext4_set_de_type(struct super_block *sb,
1110                                 struct ext4_dir_entry_2 *de,
1111                                 umode_t mode) {
1112         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1113                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1114 }
1115
1116 /*
1117  * Move count entries from end of map between two memory locations.
1118  * Returns pointer to last entry moved.
1119  */
1120 static struct ext4_dir_entry_2 *
1121 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1122 {
1123         unsigned rec_len = 0;
1124
1125         while (count--) {
1126                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1127                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1128                 memcpy (to, de, rec_len);
1129                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1130                                 ext4_rec_len_to_disk(rec_len);
1131                 de->inode = 0;
1132                 map++;
1133                 to += rec_len;
1134         }
1135         return (struct ext4_dir_entry_2 *) (to - rec_len);
1136 }
1137
1138 /*
1139  * Compact each dir entry in the range to the minimal rec_len.
1140  * Returns pointer to last entry in range.
1141  */
1142 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1143 {
1144         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1145         unsigned rec_len = 0;
1146
1147         prev = to = de;
1148         while ((char*)de < base + size) {
1149                 next = ext4_next_entry(de);
1150                 if (de->inode && de->name_len) {
1151                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1152                         if (de > to)
1153                                 memmove(to, de, rec_len);
1154                         to->rec_len = ext4_rec_len_to_disk(rec_len);
1155                         prev = to;
1156                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1157                 }
1158                 de = next;
1159         }
1160         return prev;
1161 }
1162
1163 /*
1164  * Split a full leaf block to make room for a new dir entry.
1165  * Allocate a new block, and move entries so that they are approx. equally full.
1166  * Returns pointer to de in block into which the new entry will be inserted.
1167  */
1168 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1169                         struct buffer_head **bh,struct dx_frame *frame,
1170                         struct dx_hash_info *hinfo, int *error)
1171 {
1172         unsigned blocksize = dir->i_sb->s_blocksize;
1173         unsigned count, continued;
1174         struct buffer_head *bh2;
1175         ext4_lblk_t newblock;
1176         u32 hash2;
1177         struct dx_map_entry *map;
1178         char *data1 = (*bh)->b_data, *data2;
1179         unsigned split, move, size, i;
1180         struct ext4_dir_entry_2 *de = NULL, *de2;
1181         int     err = 0;
1182
1183         bh2 = ext4_append (handle, dir, &newblock, &err);
1184         if (!(bh2)) {
1185                 brelse(*bh);
1186                 *bh = NULL;
1187                 goto errout;
1188         }
1189
1190         BUFFER_TRACE(*bh, "get_write_access");
1191         err = ext4_journal_get_write_access(handle, *bh);
1192         if (err)
1193                 goto journal_error;
1194
1195         BUFFER_TRACE(frame->bh, "get_write_access");
1196         err = ext4_journal_get_write_access(handle, frame->bh);
1197         if (err)
1198                 goto journal_error;
1199
1200         data2 = bh2->b_data;
1201
1202         /* create map in the end of data2 block */
1203         map = (struct dx_map_entry *) (data2 + blocksize);
1204         count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1205                              blocksize, hinfo, map);
1206         map -= count;
1207         dx_sort_map (map, count);
1208         /* Split the existing block in the middle, size-wise */
1209         size = 0;
1210         move = 0;
1211         for (i = count-1; i >= 0; i--) {
1212                 /* is more than half of this entry in 2nd half of the block? */
1213                 if (size + map[i].size/2 > blocksize/2)
1214                         break;
1215                 size += map[i].size;
1216                 move++;
1217         }
1218         /* map index at which we will split */
1219         split = count - move;
1220         hash2 = map[split].hash;
1221         continued = hash2 == map[split - 1].hash;
1222         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1223                         (unsigned long)dx_get_block(frame->at),
1224                                         hash2, split, count-split));
1225
1226         /* Fancy dance to stay within two buffers */
1227         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1228         de = dx_pack_dirents(data1,blocksize);
1229         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1230         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
1231         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1232         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1233
1234         /* Which block gets the new entry? */
1235         if (hinfo->hash >= hash2)
1236         {
1237                 swap(*bh, bh2);
1238                 de = de2;
1239         }
1240         dx_insert_block (frame, hash2 + continued, newblock);
1241         err = ext4_journal_dirty_metadata (handle, bh2);
1242         if (err)
1243                 goto journal_error;
1244         err = ext4_journal_dirty_metadata (handle, frame->bh);
1245         if (err)
1246                 goto journal_error;
1247         brelse (bh2);
1248         dxtrace(dx_show_index ("frame", frame->entries));
1249         return de;
1250
1251 journal_error:
1252         brelse(*bh);
1253         brelse(bh2);
1254         *bh = NULL;
1255         ext4_std_error(dir->i_sb, err);
1256 errout:
1257         *error = err;
1258         return NULL;
1259 }
1260
1261 /*
1262  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1263  * it points to a directory entry which is guaranteed to be large
1264  * enough for new directory entry.  If de is NULL, then
1265  * add_dirent_to_buf will attempt search the directory block for
1266  * space.  It will return -ENOSPC if no space is available, and -EIO
1267  * and -EEXIST if directory entry already exists.
1268  *
1269  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1270  * all other cases bh is released.
1271  */
1272 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1273                              struct inode *inode, struct ext4_dir_entry_2 *de,
1274                              struct buffer_head * bh)
1275 {
1276         struct inode    *dir = dentry->d_parent->d_inode;
1277         const char      *name = dentry->d_name.name;
1278         int             namelen = dentry->d_name.len;
1279         unsigned long   offset = 0;
1280         unsigned short  reclen;
1281         int             nlen, rlen, err;
1282         char            *top;
1283
1284         reclen = EXT4_DIR_REC_LEN(namelen);
1285         if (!de) {
1286                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1287                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1288                 while ((char *) de <= top) {
1289                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1290                                                   bh, offset)) {
1291                                 brelse (bh);
1292                                 return -EIO;
1293                         }
1294                         if (ext4_match (namelen, name, de)) {
1295                                 brelse (bh);
1296                                 return -EEXIST;
1297                         }
1298                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1299                         rlen = ext4_rec_len_from_disk(de->rec_len);
1300                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1301                                 break;
1302                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1303                         offset += rlen;
1304                 }
1305                 if ((char *) de > top)
1306                         return -ENOSPC;
1307         }
1308         BUFFER_TRACE(bh, "get_write_access");
1309         err = ext4_journal_get_write_access(handle, bh);
1310         if (err) {
1311                 ext4_std_error(dir->i_sb, err);
1312                 brelse(bh);
1313                 return err;
1314         }
1315
1316         /* By now the buffer is marked for journaling */
1317         nlen = EXT4_DIR_REC_LEN(de->name_len);
1318         rlen = ext4_rec_len_from_disk(de->rec_len);
1319         if (de->inode) {
1320                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1321                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1322                 de->rec_len = ext4_rec_len_to_disk(nlen);
1323                 de = de1;
1324         }
1325         de->file_type = EXT4_FT_UNKNOWN;
1326         if (inode) {
1327                 de->inode = cpu_to_le32(inode->i_ino);
1328                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1329         } else
1330                 de->inode = 0;
1331         de->name_len = namelen;
1332         memcpy (de->name, name, namelen);
1333         /*
1334          * XXX shouldn't update any times until successful
1335          * completion of syscall, but too many callers depend
1336          * on this.
1337          *
1338          * XXX similarly, too many callers depend on
1339          * ext4_new_inode() setting the times, but error
1340          * recovery deletes the inode, so the worst that can
1341          * happen is that the times are slightly out of date
1342          * and/or different from the directory change time.
1343          */
1344         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1345         ext4_update_dx_flag(dir);
1346         dir->i_version++;
1347         ext4_mark_inode_dirty(handle, dir);
1348         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1349         err = ext4_journal_dirty_metadata(handle, bh);
1350         if (err)
1351                 ext4_std_error(dir->i_sb, err);
1352         brelse(bh);
1353         return 0;
1354 }
1355
1356 /*
1357  * This converts a one block unindexed directory to a 3 block indexed
1358  * directory, and adds the dentry to the indexed directory.
1359  */
1360 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1361                             struct inode *inode, struct buffer_head *bh)
1362 {
1363         struct inode    *dir = dentry->d_parent->d_inode;
1364         const char      *name = dentry->d_name.name;
1365         int             namelen = dentry->d_name.len;
1366         struct buffer_head *bh2;
1367         struct dx_root  *root;
1368         struct dx_frame frames[2], *frame;
1369         struct dx_entry *entries;
1370         struct ext4_dir_entry_2 *de, *de2;
1371         char            *data1, *top;
1372         unsigned        len;
1373         int             retval;
1374         unsigned        blocksize;
1375         struct dx_hash_info hinfo;
1376         ext4_lblk_t  block;
1377         struct fake_dirent *fde;
1378
1379         blocksize =  dir->i_sb->s_blocksize;
1380         dxtrace(printk("Creating index\n"));
1381         retval = ext4_journal_get_write_access(handle, bh);
1382         if (retval) {
1383                 ext4_std_error(dir->i_sb, retval);
1384                 brelse(bh);
1385                 return retval;
1386         }
1387         root = (struct dx_root *) bh->b_data;
1388
1389         bh2 = ext4_append (handle, dir, &block, &retval);
1390         if (!(bh2)) {
1391                 brelse(bh);
1392                 return retval;
1393         }
1394         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1395         data1 = bh2->b_data;
1396
1397         /* The 0th block becomes the root, move the dirents out */
1398         fde = &root->dotdot;
1399         de = (struct ext4_dir_entry_2 *)((char *)fde +
1400                 ext4_rec_len_from_disk(fde->rec_len));
1401         len = ((char *) root) + blocksize - (char *) de;
1402         memcpy (data1, de, len);
1403         de = (struct ext4_dir_entry_2 *) data1;
1404         top = data1 + len;
1405         while ((char *)(de2 = ext4_next_entry(de)) < top)
1406                 de = de2;
1407         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1408         /* Initialize the root; the dot dirents already exist */
1409         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1410         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
1411         memset (&root->info, 0, sizeof(root->info));
1412         root->info.info_length = sizeof(root->info);
1413         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1414         entries = root->entries;
1415         dx_set_block (entries, 1);
1416         dx_set_count (entries, 1);
1417         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1418
1419         /* Initialize as for dx_probe */
1420         hinfo.hash_version = root->info.hash_version;
1421         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1422         ext4fs_dirhash(name, namelen, &hinfo);
1423         frame = frames;
1424         frame->entries = entries;
1425         frame->at = entries;
1426         frame->bh = bh;
1427         bh = bh2;
1428         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1429         dx_release (frames);
1430         if (!(de))
1431                 return retval;
1432
1433         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1434 }
1435
1436 /*
1437  *      ext4_add_entry()
1438  *
1439  * adds a file entry to the specified directory, using the same
1440  * semantics as ext4_find_entry(). It returns NULL if it failed.
1441  *
1442  * NOTE!! The inode part of 'de' is left at 0 - which means you
1443  * may not sleep between calling this and putting something into
1444  * the entry, as someone else might have used it while you slept.
1445  */
1446 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1447         struct inode *inode)
1448 {
1449         struct inode *dir = dentry->d_parent->d_inode;
1450         unsigned long offset;
1451         struct buffer_head * bh;
1452         struct ext4_dir_entry_2 *de;
1453         struct super_block * sb;
1454         int     retval;
1455         int     dx_fallback=0;
1456         unsigned blocksize;
1457         ext4_lblk_t block, blocks;
1458
1459         sb = dir->i_sb;
1460         blocksize = sb->s_blocksize;
1461         if (!dentry->d_name.len)
1462                 return -EINVAL;
1463         if (is_dx(dir)) {
1464                 retval = ext4_dx_add_entry(handle, dentry, inode);
1465                 if (!retval || (retval != ERR_BAD_DX_DIR))
1466                         return retval;
1467                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1468                 dx_fallback++;
1469                 ext4_mark_inode_dirty(handle, dir);
1470         }
1471         blocks = dir->i_size >> sb->s_blocksize_bits;
1472         for (block = 0, offset = 0; block < blocks; block++) {
1473                 bh = ext4_bread(handle, dir, block, 0, &retval);
1474                 if(!bh)
1475                         return retval;
1476                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477                 if (retval != -ENOSPC)
1478                         return retval;
1479
1480                 if (blocks == 1 && !dx_fallback &&
1481                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1482                         return make_indexed_dir(handle, dentry, inode, bh);
1483                 brelse(bh);
1484         }
1485         bh = ext4_append(handle, dir, &block, &retval);
1486         if (!bh)
1487                 return retval;
1488         de = (struct ext4_dir_entry_2 *) bh->b_data;
1489         de->inode = 0;
1490         de->rec_len = ext4_rec_len_to_disk(blocksize);
1491         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1492 }
1493
1494 /*
1495  * Returns 0 for success, or a negative error value
1496  */
1497 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1498                              struct inode *inode)
1499 {
1500         struct dx_frame frames[2], *frame;
1501         struct dx_entry *entries, *at;
1502         struct dx_hash_info hinfo;
1503         struct buffer_head * bh;
1504         struct inode *dir = dentry->d_parent->d_inode;
1505         struct super_block * sb = dir->i_sb;
1506         struct ext4_dir_entry_2 *de;
1507         int err;
1508
1509         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1510         if (!frame)
1511                 return err;
1512         entries = frame->entries;
1513         at = frame->at;
1514
1515         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1516                 goto cleanup;
1517
1518         BUFFER_TRACE(bh, "get_write_access");
1519         err = ext4_journal_get_write_access(handle, bh);
1520         if (err)
1521                 goto journal_error;
1522
1523         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1524         if (err != -ENOSPC) {
1525                 bh = NULL;
1526                 goto cleanup;
1527         }
1528
1529         /* Block full, should compress but for now just split */
1530         dxtrace(printk("using %u of %u node entries\n",
1531                        dx_get_count(entries), dx_get_limit(entries)));
1532         /* Need to split index? */
1533         if (dx_get_count(entries) == dx_get_limit(entries)) {
1534                 ext4_lblk_t newblock;
1535                 unsigned icount = dx_get_count(entries);
1536                 int levels = frame - frames;
1537                 struct dx_entry *entries2;
1538                 struct dx_node *node2;
1539                 struct buffer_head *bh2;
1540
1541                 if (levels && (dx_get_count(frames->entries) ==
1542                                dx_get_limit(frames->entries))) {
1543                         ext4_warning(sb, __func__,
1544                                      "Directory index full!");
1545                         err = -ENOSPC;
1546                         goto cleanup;
1547                 }
1548                 bh2 = ext4_append (handle, dir, &newblock, &err);
1549                 if (!(bh2))
1550                         goto cleanup;
1551                 node2 = (struct dx_node *)(bh2->b_data);
1552                 entries2 = node2->entries;
1553                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
1554                 node2->fake.inode = 0;
1555                 BUFFER_TRACE(frame->bh, "get_write_access");
1556                 err = ext4_journal_get_write_access(handle, frame->bh);
1557                 if (err)
1558                         goto journal_error;
1559                 if (levels) {
1560                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1561                         unsigned hash2 = dx_get_hash(entries + icount1);
1562                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1563
1564                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1565                         err = ext4_journal_get_write_access(handle,
1566                                                              frames[0].bh);
1567                         if (err)
1568                                 goto journal_error;
1569
1570                         memcpy ((char *) entries2, (char *) (entries + icount1),
1571                                 icount2 * sizeof(struct dx_entry));
1572                         dx_set_count (entries, icount1);
1573                         dx_set_count (entries2, icount2);
1574                         dx_set_limit (entries2, dx_node_limit(dir));
1575
1576                         /* Which index block gets the new entry? */
1577                         if (at - entries >= icount1) {
1578                                 frame->at = at = at - entries - icount1 + entries2;
1579                                 frame->entries = entries = entries2;
1580                                 swap(frame->bh, bh2);
1581                         }
1582                         dx_insert_block (frames + 0, hash2, newblock);
1583                         dxtrace(dx_show_index ("node", frames[1].entries));
1584                         dxtrace(dx_show_index ("node",
1585                                ((struct dx_node *) bh2->b_data)->entries));
1586                         err = ext4_journal_dirty_metadata(handle, bh2);
1587                         if (err)
1588                                 goto journal_error;
1589                         brelse (bh2);
1590                 } else {
1591                         dxtrace(printk("Creating second level index...\n"));
1592                         memcpy((char *) entries2, (char *) entries,
1593                                icount * sizeof(struct dx_entry));
1594                         dx_set_limit(entries2, dx_node_limit(dir));
1595
1596                         /* Set up root */
1597                         dx_set_count(entries, 1);
1598                         dx_set_block(entries + 0, newblock);
1599                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1600
1601                         /* Add new access path frame */
1602                         frame = frames + 1;
1603                         frame->at = at = at - entries + entries2;
1604                         frame->entries = entries = entries2;
1605                         frame->bh = bh2;
1606                         err = ext4_journal_get_write_access(handle,
1607                                                              frame->bh);
1608                         if (err)
1609                                 goto journal_error;
1610                 }
1611                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1612         }
1613         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1614         if (!de)
1615                 goto cleanup;
1616         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1617         bh = NULL;
1618         goto cleanup;
1619
1620 journal_error:
1621         ext4_std_error(dir->i_sb, err);
1622 cleanup:
1623         if (bh)
1624                 brelse(bh);
1625         dx_release(frames);
1626         return err;
1627 }
1628
1629 /*
1630  * ext4_delete_entry deletes a directory entry by merging it with the
1631  * previous entry
1632  */
1633 static int ext4_delete_entry (handle_t *handle,
1634                               struct inode * dir,
1635                               struct ext4_dir_entry_2 * de_del,
1636                               struct buffer_head * bh)
1637 {
1638         struct ext4_dir_entry_2 * de, * pde;
1639         int i;
1640
1641         i = 0;
1642         pde = NULL;
1643         de = (struct ext4_dir_entry_2 *) bh->b_data;
1644         while (i < bh->b_size) {
1645                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1646                         return -EIO;
1647                 if (de == de_del)  {
1648                         BUFFER_TRACE(bh, "get_write_access");
1649                         ext4_journal_get_write_access(handle, bh);
1650                         if (pde)
1651                                 pde->rec_len = ext4_rec_len_to_disk(
1652                                         ext4_rec_len_from_disk(pde->rec_len) +
1653                                         ext4_rec_len_from_disk(de->rec_len));
1654                         else
1655                                 de->inode = 0;
1656                         dir->i_version++;
1657                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1658                         ext4_journal_dirty_metadata(handle, bh);
1659                         return 0;
1660                 }
1661                 i += ext4_rec_len_from_disk(de->rec_len);
1662                 pde = de;
1663                 de = ext4_next_entry(de);
1664         }
1665         return -ENOENT;
1666 }
1667
1668 /*
1669  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1670  * since this indicates that nlinks count was previously 1.
1671  */
1672 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1673 {
1674         inc_nlink(inode);
1675         if (is_dx(inode) && inode->i_nlink > 1) {
1676                 /* limit is 16-bit i_links_count */
1677                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1678                         inode->i_nlink = 1;
1679                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1680                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1681                 }
1682         }
1683 }
1684
1685 /*
1686  * If a directory had nlink == 1, then we should let it be 1. This indicates
1687  * directory has >EXT4_LINK_MAX subdirs.
1688  */
1689 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1690 {
1691         drop_nlink(inode);
1692         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1693                 inc_nlink(inode);
1694 }
1695
1696
1697 static int ext4_add_nondir(handle_t *handle,
1698                 struct dentry *dentry, struct inode *inode)
1699 {
1700         int err = ext4_add_entry(handle, dentry, inode);
1701         if (!err) {
1702                 ext4_mark_inode_dirty(handle, inode);
1703                 d_instantiate(dentry, inode);
1704                 return 0;
1705         }
1706         drop_nlink(inode);
1707         iput(inode);
1708         return err;
1709 }
1710
1711 /*
1712  * By the time this is called, we already have created
1713  * the directory cache entry for the new file, but it
1714  * is so far negative - it has no inode.
1715  *
1716  * If the create succeeds, we fill in the inode information
1717  * with d_instantiate().
1718  */
1719 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1720                 struct nameidata *nd)
1721 {
1722         handle_t *handle;
1723         struct inode * inode;
1724         int err, retries = 0;
1725
1726 retry:
1727         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1728                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1729                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1730         if (IS_ERR(handle))
1731                 return PTR_ERR(handle);
1732
1733         if (IS_DIRSYNC(dir))
1734                 handle->h_sync = 1;
1735
1736         inode = ext4_new_inode (handle, dir, mode);
1737         err = PTR_ERR(inode);
1738         if (!IS_ERR(inode)) {
1739                 inode->i_op = &ext4_file_inode_operations;
1740                 inode->i_fop = &ext4_file_operations;
1741                 ext4_set_aops(inode);
1742                 err = ext4_add_nondir(handle, dentry, inode);
1743         }
1744         ext4_journal_stop(handle);
1745         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1746                 goto retry;
1747         return err;
1748 }
1749
1750 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1751                         int mode, dev_t rdev)
1752 {
1753         handle_t *handle;
1754         struct inode *inode;
1755         int err, retries = 0;
1756
1757         if (!new_valid_dev(rdev))
1758                 return -EINVAL;
1759
1760 retry:
1761         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1762                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1763                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1764         if (IS_ERR(handle))
1765                 return PTR_ERR(handle);
1766
1767         if (IS_DIRSYNC(dir))
1768                 handle->h_sync = 1;
1769
1770         inode = ext4_new_inode (handle, dir, mode);
1771         err = PTR_ERR(inode);
1772         if (!IS_ERR(inode)) {
1773                 init_special_inode(inode, inode->i_mode, rdev);
1774 #ifdef CONFIG_EXT4DEV_FS_XATTR
1775                 inode->i_op = &ext4_special_inode_operations;
1776 #endif
1777                 err = ext4_add_nondir(handle, dentry, inode);
1778         }
1779         ext4_journal_stop(handle);
1780         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1781                 goto retry;
1782         return err;
1783 }
1784
1785 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1786 {
1787         handle_t *handle;
1788         struct inode * inode;
1789         struct buffer_head * dir_block;
1790         struct ext4_dir_entry_2 * de;
1791         int err, retries = 0;
1792
1793         if (EXT4_DIR_LINK_MAX(dir))
1794                 return -EMLINK;
1795
1796 retry:
1797         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1798                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1799                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1800         if (IS_ERR(handle))
1801                 return PTR_ERR(handle);
1802
1803         if (IS_DIRSYNC(dir))
1804                 handle->h_sync = 1;
1805
1806         inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1807         err = PTR_ERR(inode);
1808         if (IS_ERR(inode))
1809                 goto out_stop;
1810
1811         inode->i_op = &ext4_dir_inode_operations;
1812         inode->i_fop = &ext4_dir_operations;
1813         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1814         dir_block = ext4_bread (handle, inode, 0, 1, &err);
1815         if (!dir_block)
1816                 goto out_clear_inode;
1817         BUFFER_TRACE(dir_block, "get_write_access");
1818         ext4_journal_get_write_access(handle, dir_block);
1819         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1820         de->inode = cpu_to_le32(inode->i_ino);
1821         de->name_len = 1;
1822         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
1823         strcpy (de->name, ".");
1824         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1825         de = ext4_next_entry(de);
1826         de->inode = cpu_to_le32(dir->i_ino);
1827         de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1828                                                 EXT4_DIR_REC_LEN(1));
1829         de->name_len = 2;
1830         strcpy (de->name, "..");
1831         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1832         inode->i_nlink = 2;
1833         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1834         ext4_journal_dirty_metadata(handle, dir_block);
1835         brelse (dir_block);
1836         ext4_mark_inode_dirty(handle, inode);
1837         err = ext4_add_entry (handle, dentry, inode);
1838         if (err) {
1839 out_clear_inode:
1840                 clear_nlink(inode);
1841                 ext4_mark_inode_dirty(handle, inode);
1842                 iput (inode);
1843                 goto out_stop;
1844         }
1845         ext4_inc_count(handle, dir);
1846         ext4_update_dx_flag(dir);
1847         ext4_mark_inode_dirty(handle, dir);
1848         d_instantiate(dentry, inode);
1849 out_stop:
1850         ext4_journal_stop(handle);
1851         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1852                 goto retry;
1853         return err;
1854 }
1855
1856 /*
1857  * routine to check that the specified directory is empty (for rmdir)
1858  */
1859 static int empty_dir (struct inode * inode)
1860 {
1861         unsigned long offset;
1862         struct buffer_head * bh;
1863         struct ext4_dir_entry_2 * de, * de1;
1864         struct super_block * sb;
1865         int err = 0;
1866
1867         sb = inode->i_sb;
1868         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1869             !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1870                 if (err)
1871                         ext4_error(inode->i_sb, __func__,
1872                                    "error %d reading directory #%lu offset 0",
1873                                    err, inode->i_ino);
1874                 else
1875                         ext4_warning(inode->i_sb, __func__,
1876                                      "bad directory (dir #%lu) - no data block",
1877                                      inode->i_ino);
1878                 return 1;
1879         }
1880         de = (struct ext4_dir_entry_2 *) bh->b_data;
1881         de1 = ext4_next_entry(de);
1882         if (le32_to_cpu(de->inode) != inode->i_ino ||
1883                         !le32_to_cpu(de1->inode) ||
1884                         strcmp (".", de->name) ||
1885                         strcmp ("..", de1->name)) {
1886                 ext4_warning (inode->i_sb, "empty_dir",
1887                               "bad directory (dir #%lu) - no `.' or `..'",
1888                               inode->i_ino);
1889                 brelse (bh);
1890                 return 1;
1891         }
1892         offset = ext4_rec_len_from_disk(de->rec_len) +
1893                  ext4_rec_len_from_disk(de1->rec_len);
1894         de = ext4_next_entry(de1);
1895         while (offset < inode->i_size ) {
1896                 if (!bh ||
1897                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1898                         err = 0;
1899                         brelse (bh);
1900                         bh = ext4_bread (NULL, inode,
1901                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1902                         if (!bh) {
1903                                 if (err)
1904                                         ext4_error(sb, __func__,
1905                                                    "error %d reading directory"
1906                                                    " #%lu offset %lu",
1907                                                    err, inode->i_ino, offset);
1908                                 offset += sb->s_blocksize;
1909                                 continue;
1910                         }
1911                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1912                 }
1913                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1914                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1915                                                          sb->s_blocksize);
1916                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1917                         continue;
1918                 }
1919                 if (le32_to_cpu(de->inode)) {
1920                         brelse (bh);
1921                         return 0;
1922                 }
1923                 offset += ext4_rec_len_from_disk(de->rec_len);
1924                 de = ext4_next_entry(de);
1925         }
1926         brelse (bh);
1927         return 1;
1928 }
1929
1930 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1931  * such inodes, starting at the superblock, in case we crash before the
1932  * file is closed/deleted, or in case the inode truncate spans multiple
1933  * transactions and the last transaction is not recovered after a crash.
1934  *
1935  * At filesystem recovery time, we walk this list deleting unlinked
1936  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1937  */
1938 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1939 {
1940         struct super_block *sb = inode->i_sb;
1941         struct ext4_iloc iloc;
1942         int err = 0, rc;
1943
1944         lock_super(sb);
1945         if (!list_empty(&EXT4_I(inode)->i_orphan))
1946                 goto out_unlock;
1947
1948         /* Orphan handling is only valid for files with data blocks
1949          * being truncated, or files being unlinked. */
1950
1951         /* @@@ FIXME: Observation from aviro:
1952          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1953          * here (on lock_super()), so race with ext4_link() which might bump
1954          * ->i_nlink. For, say it, character device. Not a regular file,
1955          * not a directory, not a symlink and ->i_nlink > 0.
1956          */
1957         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1958                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1959
1960         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1961         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1962         if (err)
1963                 goto out_unlock;
1964
1965         err = ext4_reserve_inode_write(handle, inode, &iloc);
1966         if (err)
1967                 goto out_unlock;
1968
1969         /* Insert this inode at the head of the on-disk orphan list... */
1970         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1971         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1972         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1973         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1974         if (!err)
1975                 err = rc;
1976
1977         /* Only add to the head of the in-memory list if all the
1978          * previous operations succeeded.  If the orphan_add is going to
1979          * fail (possibly taking the journal offline), we can't risk
1980          * leaving the inode on the orphan list: stray orphan-list
1981          * entries can cause panics at unmount time.
1982          *
1983          * This is safe: on error we're going to ignore the orphan list
1984          * anyway on the next recovery. */
1985         if (!err)
1986                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1987
1988         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1989         jbd_debug(4, "orphan inode %lu will point to %d\n",
1990                         inode->i_ino, NEXT_ORPHAN(inode));
1991 out_unlock:
1992         unlock_super(sb);
1993         ext4_std_error(inode->i_sb, err);
1994         return err;
1995 }
1996
1997 /*
1998  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1999  * of such inodes stored on disk, because it is finally being cleaned up.
2000  */
2001 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2002 {
2003         struct list_head *prev;
2004         struct ext4_inode_info *ei = EXT4_I(inode);
2005         struct ext4_sb_info *sbi;
2006         unsigned long ino_next;
2007         struct ext4_iloc iloc;
2008         int err = 0;
2009
2010         lock_super(inode->i_sb);
2011         if (list_empty(&ei->i_orphan)) {
2012                 unlock_super(inode->i_sb);
2013                 return 0;
2014         }
2015
2016         ino_next = NEXT_ORPHAN(inode);
2017         prev = ei->i_orphan.prev;
2018         sbi = EXT4_SB(inode->i_sb);
2019
2020         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2021
2022         list_del_init(&ei->i_orphan);
2023
2024         /* If we're on an error path, we may not have a valid
2025          * transaction handle with which to update the orphan list on
2026          * disk, but we still need to remove the inode from the linked
2027          * list in memory. */
2028         if (!handle)
2029                 goto out;
2030
2031         err = ext4_reserve_inode_write(handle, inode, &iloc);
2032         if (err)
2033                 goto out_err;
2034
2035         if (prev == &sbi->s_orphan) {
2036                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2037                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2038                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2039                 if (err)
2040                         goto out_brelse;
2041                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2042                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
2043         } else {
2044                 struct ext4_iloc iloc2;
2045                 struct inode *i_prev =
2046                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2047
2048                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2049                           i_prev->i_ino, ino_next);
2050                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2051                 if (err)
2052                         goto out_brelse;
2053                 NEXT_ORPHAN(i_prev) = ino_next;
2054                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2055         }
2056         if (err)
2057                 goto out_brelse;
2058         NEXT_ORPHAN(inode) = 0;
2059         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2060
2061 out_err:
2062         ext4_std_error(inode->i_sb, err);
2063 out:
2064         unlock_super(inode->i_sb);
2065         return err;
2066
2067 out_brelse:
2068         brelse(iloc.bh);
2069         goto out_err;
2070 }
2071
2072 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2073 {
2074         int retval;
2075         struct inode * inode;
2076         struct buffer_head * bh;
2077         struct ext4_dir_entry_2 * de;
2078         handle_t *handle;
2079
2080         /* Initialize quotas before so that eventual writes go in
2081          * separate transaction */
2082         DQUOT_INIT(dentry->d_inode);
2083         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2084         if (IS_ERR(handle))
2085                 return PTR_ERR(handle);
2086
2087         retval = -ENOENT;
2088         bh = ext4_find_entry (dentry, &de);
2089         if (!bh)
2090                 goto end_rmdir;
2091
2092         if (IS_DIRSYNC(dir))
2093                 handle->h_sync = 1;
2094
2095         inode = dentry->d_inode;
2096
2097         retval = -EIO;
2098         if (le32_to_cpu(de->inode) != inode->i_ino)
2099                 goto end_rmdir;
2100
2101         retval = -ENOTEMPTY;
2102         if (!empty_dir (inode))
2103                 goto end_rmdir;
2104
2105         retval = ext4_delete_entry(handle, dir, de, bh);
2106         if (retval)
2107                 goto end_rmdir;
2108         if (!EXT4_DIR_LINK_EMPTY(inode))
2109                 ext4_warning (inode->i_sb, "ext4_rmdir",
2110                               "empty directory has too many links (%d)",
2111                               inode->i_nlink);
2112         inode->i_version++;
2113         clear_nlink(inode);
2114         /* There's no need to set i_disksize: the fact that i_nlink is
2115          * zero will ensure that the right thing happens during any
2116          * recovery. */
2117         inode->i_size = 0;
2118         ext4_orphan_add(handle, inode);
2119         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2120         ext4_mark_inode_dirty(handle, inode);
2121         ext4_dec_count(handle, dir);
2122         ext4_update_dx_flag(dir);
2123         ext4_mark_inode_dirty(handle, dir);
2124
2125 end_rmdir:
2126         ext4_journal_stop(handle);
2127         brelse (bh);
2128         return retval;
2129 }
2130
2131 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2132 {
2133         int retval;
2134         struct inode * inode;
2135         struct buffer_head * bh;
2136         struct ext4_dir_entry_2 * de;
2137         handle_t *handle;
2138
2139         /* Initialize quotas before so that eventual writes go
2140          * in separate transaction */
2141         DQUOT_INIT(dentry->d_inode);
2142         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2143         if (IS_ERR(handle))
2144                 return PTR_ERR(handle);
2145
2146         if (IS_DIRSYNC(dir))
2147                 handle->h_sync = 1;
2148
2149         retval = -ENOENT;
2150         bh = ext4_find_entry (dentry, &de);
2151         if (!bh)
2152                 goto end_unlink;
2153
2154         inode = dentry->d_inode;
2155
2156         retval = -EIO;
2157         if (le32_to_cpu(de->inode) != inode->i_ino)
2158                 goto end_unlink;
2159
2160         if (!inode->i_nlink) {
2161                 ext4_warning (inode->i_sb, "ext4_unlink",
2162                               "Deleting nonexistent file (%lu), %d",
2163                               inode->i_ino, inode->i_nlink);
2164                 inode->i_nlink = 1;
2165         }
2166         retval = ext4_delete_entry(handle, dir, de, bh);
2167         if (retval)
2168                 goto end_unlink;
2169         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2170         ext4_update_dx_flag(dir);
2171         ext4_mark_inode_dirty(handle, dir);
2172         drop_nlink(inode);
2173         if (!inode->i_nlink)
2174                 ext4_orphan_add(handle, inode);
2175         inode->i_ctime = ext4_current_time(inode);
2176         ext4_mark_inode_dirty(handle, inode);
2177         retval = 0;
2178
2179 end_unlink:
2180         ext4_journal_stop(handle);
2181         brelse (bh);
2182         return retval;
2183 }
2184
2185 static int ext4_symlink (struct inode * dir,
2186                 struct dentry *dentry, const char * symname)
2187 {
2188         handle_t *handle;
2189         struct inode * inode;
2190         int l, err, retries = 0;
2191
2192         l = strlen(symname)+1;
2193         if (l > dir->i_sb->s_blocksize)
2194                 return -ENAMETOOLONG;
2195
2196 retry:
2197         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2198                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2199                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2200         if (IS_ERR(handle))
2201                 return PTR_ERR(handle);
2202
2203         if (IS_DIRSYNC(dir))
2204                 handle->h_sync = 1;
2205
2206         inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2207         err = PTR_ERR(inode);
2208         if (IS_ERR(inode))
2209                 goto out_stop;
2210
2211         if (l > sizeof (EXT4_I(inode)->i_data)) {
2212                 inode->i_op = &ext4_symlink_inode_operations;
2213                 ext4_set_aops(inode);
2214                 /*
2215                  * page_symlink() calls into ext4_prepare/commit_write.
2216                  * We have a transaction open.  All is sweetness.  It also sets
2217                  * i_size in generic_commit_write().
2218                  */
2219                 err = __page_symlink(inode, symname, l,
2220                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2221                 if (err) {
2222                         clear_nlink(inode);
2223                         ext4_mark_inode_dirty(handle, inode);
2224                         iput (inode);
2225                         goto out_stop;
2226                 }
2227         } else {
2228                 /* clear the extent format for fast symlink */
2229                 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2230                 inode->i_op = &ext4_fast_symlink_inode_operations;
2231                 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2232                 inode->i_size = l-1;
2233         }
2234         EXT4_I(inode)->i_disksize = inode->i_size;
2235         err = ext4_add_nondir(handle, dentry, inode);
2236 out_stop:
2237         ext4_journal_stop(handle);
2238         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2239                 goto retry;
2240         return err;
2241 }
2242
2243 static int ext4_link (struct dentry * old_dentry,
2244                 struct inode * dir, struct dentry *dentry)
2245 {
2246         handle_t *handle;
2247         struct inode *inode = old_dentry->d_inode;
2248         int err, retries = 0;
2249
2250         if (EXT4_DIR_LINK_MAX(inode))
2251                 return -EMLINK;
2252
2253         /*
2254          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2255          * otherwise has the potential to corrupt the orphan inode list.
2256          */
2257         if (inode->i_nlink == 0)
2258                 return -ENOENT;
2259
2260 retry:
2261         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2262                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2263         if (IS_ERR(handle))
2264                 return PTR_ERR(handle);
2265
2266         if (IS_DIRSYNC(dir))
2267                 handle->h_sync = 1;
2268
2269         inode->i_ctime = ext4_current_time(inode);
2270         ext4_inc_count(handle, inode);
2271         atomic_inc(&inode->i_count);
2272
2273         err = ext4_add_nondir(handle, dentry, inode);
2274         ext4_journal_stop(handle);
2275         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2276                 goto retry;
2277         return err;
2278 }
2279
2280 #define PARENT_INO(buffer) \
2281         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2282
2283 /*
2284  * Anybody can rename anything with this: the permission checks are left to the
2285  * higher-level routines.
2286  */
2287 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2288                            struct inode * new_dir,struct dentry *new_dentry)
2289 {
2290         handle_t *handle;
2291         struct inode * old_inode, * new_inode;
2292         struct buffer_head * old_bh, * new_bh, * dir_bh;
2293         struct ext4_dir_entry_2 * old_de, * new_de;
2294         int retval;
2295
2296         old_bh = new_bh = dir_bh = NULL;
2297
2298         /* Initialize quotas before so that eventual writes go
2299          * in separate transaction */
2300         if (new_dentry->d_inode)
2301                 DQUOT_INIT(new_dentry->d_inode);
2302         handle = ext4_journal_start(old_dir, 2 *
2303                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2304                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2305         if (IS_ERR(handle))
2306                 return PTR_ERR(handle);
2307
2308         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2309                 handle->h_sync = 1;
2310
2311         old_bh = ext4_find_entry (old_dentry, &old_de);
2312         /*
2313          *  Check for inode number is _not_ due to possible IO errors.
2314          *  We might rmdir the source, keep it as pwd of some process
2315          *  and merrily kill the link to whatever was created under the
2316          *  same name. Goodbye sticky bit ;-<
2317          */
2318         old_inode = old_dentry->d_inode;
2319         retval = -ENOENT;
2320         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2321                 goto end_rename;
2322
2323         new_inode = new_dentry->d_inode;
2324         new_bh = ext4_find_entry (new_dentry, &new_de);
2325         if (new_bh) {
2326                 if (!new_inode) {
2327                         brelse (new_bh);
2328                         new_bh = NULL;
2329                 }
2330         }
2331         if (S_ISDIR(old_inode->i_mode)) {
2332                 if (new_inode) {
2333                         retval = -ENOTEMPTY;
2334                         if (!empty_dir (new_inode))
2335                                 goto end_rename;
2336                 }
2337                 retval = -EIO;
2338                 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2339                 if (!dir_bh)
2340                         goto end_rename;
2341                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2342                         goto end_rename;
2343                 retval = -EMLINK;
2344                 if (!new_inode && new_dir!=old_dir &&
2345                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2346                         goto end_rename;
2347         }
2348         if (!new_bh) {
2349                 retval = ext4_add_entry (handle, new_dentry, old_inode);
2350                 if (retval)
2351                         goto end_rename;
2352         } else {
2353                 BUFFER_TRACE(new_bh, "get write access");
2354                 ext4_journal_get_write_access(handle, new_bh);
2355                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2356                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2357                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2358                         new_de->file_type = old_de->file_type;
2359                 new_dir->i_version++;
2360                 new_dir->i_ctime = new_dir->i_mtime =
2361                                         ext4_current_time(new_dir);
2362                 ext4_mark_inode_dirty(handle, new_dir);
2363                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2364                 ext4_journal_dirty_metadata(handle, new_bh);
2365                 brelse(new_bh);
2366                 new_bh = NULL;
2367         }
2368
2369         /*
2370          * Like most other Unix systems, set the ctime for inodes on a
2371          * rename.
2372          */
2373         old_inode->i_ctime = ext4_current_time(old_inode);
2374         ext4_mark_inode_dirty(handle, old_inode);
2375
2376         /*
2377          * ok, that's it
2378          */
2379         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2380             old_de->name_len != old_dentry->d_name.len ||
2381             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2382             (retval = ext4_delete_entry(handle, old_dir,
2383                                         old_de, old_bh)) == -ENOENT) {
2384                 /* old_de could have moved from under us during htree split, so
2385                  * make sure that we are deleting the right entry.  We might
2386                  * also be pointing to a stale entry in the unused part of
2387                  * old_bh so just checking inum and the name isn't enough. */
2388                 struct buffer_head *old_bh2;
2389                 struct ext4_dir_entry_2 *old_de2;
2390
2391                 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2392                 if (old_bh2) {
2393                         retval = ext4_delete_entry(handle, old_dir,
2394                                                    old_de2, old_bh2);
2395                         brelse(old_bh2);
2396                 }
2397         }
2398         if (retval) {
2399                 ext4_warning(old_dir->i_sb, "ext4_rename",
2400                                 "Deleting old file (%lu), %d, error=%d",
2401                                 old_dir->i_ino, old_dir->i_nlink, retval);
2402         }
2403
2404         if (new_inode) {
2405                 ext4_dec_count(handle, new_inode);
2406                 new_inode->i_ctime = ext4_current_time(new_inode);
2407         }
2408         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2409         ext4_update_dx_flag(old_dir);
2410         if (dir_bh) {
2411                 BUFFER_TRACE(dir_bh, "get_write_access");
2412                 ext4_journal_get_write_access(handle, dir_bh);
2413                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2414                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2415                 ext4_journal_dirty_metadata(handle, dir_bh);
2416                 ext4_dec_count(handle, old_dir);
2417                 if (new_inode) {
2418                         /* checked empty_dir above, can't have another parent,
2419                          * ext4_dec_count() won't work for many-linked dirs */
2420                         new_inode->i_nlink = 0;
2421                 } else {
2422                         ext4_inc_count(handle, new_dir);
2423                         ext4_update_dx_flag(new_dir);
2424                         ext4_mark_inode_dirty(handle, new_dir);
2425                 }
2426         }
2427         ext4_mark_inode_dirty(handle, old_dir);
2428         if (new_inode) {
2429                 ext4_mark_inode_dirty(handle, new_inode);
2430                 if (!new_inode->i_nlink)
2431                         ext4_orphan_add(handle, new_inode);
2432         }
2433         retval = 0;
2434
2435 end_rename:
2436         brelse (dir_bh);
2437         brelse (old_bh);
2438         brelse (new_bh);
2439         ext4_journal_stop(handle);
2440         return retval;
2441 }
2442
2443 /*
2444  * directories can handle most operations...
2445  */
2446 const struct inode_operations ext4_dir_inode_operations = {
2447         .create         = ext4_create,
2448         .lookup         = ext4_lookup,
2449         .link           = ext4_link,
2450         .unlink         = ext4_unlink,
2451         .symlink        = ext4_symlink,
2452         .mkdir          = ext4_mkdir,
2453         .rmdir          = ext4_rmdir,
2454         .mknod          = ext4_mknod,
2455         .rename         = ext4_rename,
2456         .setattr        = ext4_setattr,
2457 #ifdef CONFIG_EXT4DEV_FS_XATTR
2458         .setxattr       = generic_setxattr,
2459         .getxattr       = generic_getxattr,
2460         .listxattr      = ext4_listxattr,
2461         .removexattr    = generic_removexattr,
2462 #endif
2463         .permission     = ext4_permission,
2464 };
2465
2466 const struct inode_operations ext4_special_inode_operations = {
2467         .setattr        = ext4_setattr,
2468 #ifdef CONFIG_EXT4DEV_FS_XATTR
2469         .setxattr       = generic_setxattr,
2470         .getxattr       = generic_getxattr,
2471         .listxattr      = ext4_listxattr,
2472         .removexattr    = generic_removexattr,
2473 #endif
2474         .permission     = ext4_permission,
2475 };