ext3: dirindex error pointer issues
[linux-2.6.git] / fs / ext4 / namei.c
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/ext4_fs.h>
32 #include <linux/ext4_jbd2.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.h>
39 #include <linux/smp_lock.h>
40
41 #include "namei.h"
42 #include "xattr.h"
43 #include "acl.h"
44
45 /*
46  * define how far ahead to read directories while searching them.
47  */
48 #define NAMEI_RA_CHUNKS  2
49 #define NAMEI_RA_BLOCKS  4
50 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
52
53 static struct buffer_head *ext4_append(handle_t *handle,
54                                         struct inode *inode,
55                                         u32 *block, int *err)
56 {
57         struct buffer_head *bh;
58
59         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60
61         if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT4_I(inode)->i_disksize = inode->i_size;
64                 ext4_journal_get_write_access(handle,bh);
65         }
66         return bh;
67 }
68
69 #ifndef assert
70 #define assert(test) J_ASSERT(test)
71 #endif
72
73 #ifndef swap
74 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
75 #endif
76
77 #ifdef DX_DEBUG
78 #define dxtrace(command) command
79 #else
80 #define dxtrace(command)
81 #endif
82
83 struct fake_dirent
84 {
85         __le32 inode;
86         __le16 rec_len;
87         u8 name_len;
88         u8 file_type;
89 };
90
91 struct dx_countlimit
92 {
93         __le16 limit;
94         __le16 count;
95 };
96
97 struct dx_entry
98 {
99         __le32 hash;
100         __le32 block;
101 };
102
103 /*
104  * dx_root_info is laid out so that if it should somehow get overlaid by a
105  * dirent the two low bits of the hash version will be zero.  Therefore, the
106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
107  */
108
109 struct dx_root
110 {
111         struct fake_dirent dot;
112         char dot_name[4];
113         struct fake_dirent dotdot;
114         char dotdot_name[4];
115         struct dx_root_info
116         {
117                 __le32 reserved_zero;
118                 u8 hash_version;
119                 u8 info_length; /* 8 */
120                 u8 indirect_levels;
121                 u8 unused_flags;
122         }
123         info;
124         struct dx_entry entries[0];
125 };
126
127 struct dx_node
128 {
129         struct fake_dirent fake;
130         struct dx_entry entries[0];
131 };
132
133
134 struct dx_frame
135 {
136         struct buffer_head *bh;
137         struct dx_entry *entries;
138         struct dx_entry *at;
139 };
140
141 struct dx_map_entry
142 {
143         u32 hash;
144         u32 offs;
145 };
146
147 #ifdef CONFIG_EXT4_INDEX
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct dentry *dentry,
159                                  struct inode *dir,
160                                  struct dx_hash_info *hinfo,
161                                  struct dx_frame *frame,
162                                  int *err);
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
168                 struct dx_map_entry *offsets, int count);
169 static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
172                                  struct dx_frame *frame,
173                                  struct dx_frame *frames,
174                                  __u32 *start_hash);
175 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
176                        struct ext4_dir_entry_2 **res_dir, int *err);
177 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
178                              struct inode *inode);
179
180 /*
181  * Future: use high four bits of block for coalesce-on-delete flags
182  * Mask them off for now.
183  */
184
185 static inline unsigned dx_get_block (struct dx_entry *entry)
186 {
187         return le32_to_cpu(entry->block) & 0x00ffffff;
188 }
189
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191 {
192         entry->block = cpu_to_le32(value);
193 }
194
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
196 {
197         return le32_to_cpu(entry->hash);
198 }
199
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201 {
202         entry->hash = cpu_to_le32(value);
203 }
204
205 static inline unsigned dx_get_count (struct dx_entry *entries)
206 {
207         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208 }
209
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
211 {
212         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213 }
214
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216 {
217         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218 }
219
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221 {
222         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223 }
224
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226 {
227         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
228                 EXT4_DIR_REC_LEN(2) - infosize;
229         return 0? 20: entry_space / sizeof(struct dx_entry);
230 }
231
232 static inline unsigned dx_node_limit (struct inode *dir)
233 {
234         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
235         return 0? 22: entry_space / sizeof(struct dx_entry);
236 }
237
238 /*
239  * Debug
240  */
241 #ifdef DX_DEBUG
242 static void dx_show_index (char * label, struct dx_entry *entries)
243 {
244         int i, n = dx_get_count (entries);
245         printk("%s index ", label);
246         for (i = 0; i < n; i++) {
247                 printk("%x->%u ", i? dx_get_hash(entries + i) :
248                                 0, dx_get_block(entries + i));
249         }
250         printk("\n");
251 }
252
253 struct stats
254 {
255         unsigned names;
256         unsigned space;
257         unsigned bcount;
258 };
259
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
261                                  int size, int show_names)
262 {
263         unsigned names = 0, space = 0;
264         char *base = (char *) de;
265         struct dx_hash_info h = *hinfo;
266
267         printk("names: ");
268         while ((char *) de < base + size)
269         {
270                 if (de->inode)
271                 {
272                         if (show_names)
273                         {
274                                 int len = de->name_len;
275                                 char *name = de->name;
276                                 while (len--) printk("%c", *name++);
277                                 ext4fs_dirhash(de->name, de->name_len, &h);
278                                 printk(":%x.%u ", h.hash,
279                                        ((char *) de - base));
280                         }
281                         space += EXT4_DIR_REC_LEN(de->name_len);
282                         names++;
283                 }
284                 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
285         }
286         printk("(%i)\n", names);
287         return (struct stats) { names, space, 1 };
288 }
289
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291                              struct dx_entry *entries, int levels)
292 {
293         unsigned blocksize = dir->i_sb->s_blocksize;
294         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295         unsigned bcount = 0;
296         struct buffer_head *bh;
297         int err;
298         printk("%i indexed blocks...\n", count);
299         for (i = 0; i < count; i++, entries++)
300         {
301                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303                 struct stats stats;
304                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
305                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
306                 stats = levels?
307                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
309                 names += stats.names;
310                 space += stats.space;
311                 bcount += stats.bcount;
312                 brelse (bh);
313         }
314         if (bcount)
315                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
316                         names, space/bcount,(space/bcount)*100/blocksize);
317         return (struct stats) { names, space, bcount};
318 }
319 #endif /* DX_DEBUG */
320
321 /*
322  * Probe for a directory leaf block to search.
323  *
324  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325  * error in the directory index, and the caller should fall back to
326  * searching the directory normally.  The callers of dx_probe **MUST**
327  * check for this error code, and make sure it never gets reflected
328  * back to userspace.
329  */
330 static struct dx_frame *
331 dx_probe(struct dentry *dentry, struct inode *dir,
332          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333 {
334         unsigned count, indirect;
335         struct dx_entry *at, *entries, *p, *q, *m;
336         struct dx_root *root;
337         struct buffer_head *bh;
338         struct dx_frame *frame = frame_in;
339         u32 hash;
340
341         frame->bh = NULL;
342         if (dentry)
343                 dir = dentry->d_parent->d_inode;
344         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
345                 goto fail;
346         root = (struct dx_root *) bh->b_data;
347         if (root->info.hash_version != DX_HASH_TEA &&
348             root->info.hash_version != DX_HASH_HALF_MD4 &&
349             root->info.hash_version != DX_HASH_LEGACY) {
350                 ext4_warning(dir->i_sb, __FUNCTION__,
351                              "Unrecognised inode hash code %d",
352                              root->info.hash_version);
353                 brelse(bh);
354                 *err = ERR_BAD_DX_DIR;
355                 goto fail;
356         }
357         hinfo->hash_version = root->info.hash_version;
358         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
359         if (dentry)
360                 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
361         hash = hinfo->hash;
362
363         if (root->info.unused_flags & 1) {
364                 ext4_warning(dir->i_sb, __FUNCTION__,
365                              "Unimplemented inode hash flags: %#06x",
366                              root->info.unused_flags);
367                 brelse(bh);
368                 *err = ERR_BAD_DX_DIR;
369                 goto fail;
370         }
371
372         if ((indirect = root->info.indirect_levels) > 1) {
373                 ext4_warning(dir->i_sb, __FUNCTION__,
374                              "Unimplemented inode hash depth: %#06x",
375                              root->info.indirect_levels);
376                 brelse(bh);
377                 *err = ERR_BAD_DX_DIR;
378                 goto fail;
379         }
380
381         entries = (struct dx_entry *) (((char *)&root->info) +
382                                        root->info.info_length);
383         assert(dx_get_limit(entries) == dx_root_limit(dir,
384                                                       root->info.info_length));
385         dxtrace (printk("Look up %x", hash));
386         while (1)
387         {
388                 count = dx_get_count(entries);
389                 assert (count && count <= dx_get_limit(entries));
390                 p = entries + 1;
391                 q = entries + count - 1;
392                 while (p <= q)
393                 {
394                         m = p + (q - p)/2;
395                         dxtrace(printk("."));
396                         if (dx_get_hash(m) > hash)
397                                 q = m - 1;
398                         else
399                                 p = m + 1;
400                 }
401
402                 if (0) // linear search cross check
403                 {
404                         unsigned n = count - 1;
405                         at = entries;
406                         while (n--)
407                         {
408                                 dxtrace(printk(","));
409                                 if (dx_get_hash(++at) > hash)
410                                 {
411                                         at--;
412                                         break;
413                                 }
414                         }
415                         assert (at == p - 1);
416                 }
417
418                 at = p - 1;
419                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
420                 frame->bh = bh;
421                 frame->entries = entries;
422                 frame->at = at;
423                 if (!indirect--) return frame;
424                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
425                         goto fail2;
426                 at = entries = ((struct dx_node *) bh->b_data)->entries;
427                 assert (dx_get_limit(entries) == dx_node_limit (dir));
428                 frame++;
429         }
430 fail2:
431         while (frame >= frame_in) {
432                 brelse(frame->bh);
433                 frame--;
434         }
435 fail:
436         return NULL;
437 }
438
439 static void dx_release (struct dx_frame *frames)
440 {
441         if (frames[0].bh == NULL)
442                 return;
443
444         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445                 brelse(frames[1].bh);
446         brelse(frames[0].bh);
447 }
448
449 /*
450  * This function increments the frame pointer to search the next leaf
451  * block, and reads in the necessary intervening nodes if the search
452  * should be necessary.  Whether or not the search is necessary is
453  * controlled by the hash parameter.  If the hash value is even, then
454  * the search is only continued if the next block starts with that
455  * hash value.  This is used if we are searching for a specific file.
456  *
457  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
458  *
459  * This function returns 1 if the caller should continue to search,
460  * or 0 if it should not.  If there is an error reading one of the
461  * index blocks, it will a negative error code.
462  *
463  * If start_hash is non-null, it will be filled in with the starting
464  * hash of the next page.
465  */
466 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
467                                  struct dx_frame *frame,
468                                  struct dx_frame *frames,
469                                  __u32 *start_hash)
470 {
471         struct dx_frame *p;
472         struct buffer_head *bh;
473         int err, num_frames = 0;
474         __u32 bhash;
475
476         p = frame;
477         /*
478          * Find the next leaf page by incrementing the frame pointer.
479          * If we run out of entries in the interior node, loop around and
480          * increment pointer in the parent node.  When we break out of
481          * this loop, num_frames indicates the number of interior
482          * nodes need to be read.
483          */
484         while (1) {
485                 if (++(p->at) < p->entries + dx_get_count(p->entries))
486                         break;
487                 if (p == frames)
488                         return 0;
489                 num_frames++;
490                 p--;
491         }
492
493         /*
494          * If the hash is 1, then continue only if the next page has a
495          * continuation hash of any value.  This is used for readdir
496          * handling.  Otherwise, check to see if the hash matches the
497          * desired contiuation hash.  If it doesn't, return since
498          * there's no point to read in the successive index pages.
499          */
500         bhash = dx_get_hash(p->at);
501         if (start_hash)
502                 *start_hash = bhash;
503         if ((hash & 1) == 0) {
504                 if ((bhash & ~1) != hash)
505                         return 0;
506         }
507         /*
508          * If the hash is HASH_NB_ALWAYS, we always go to the next
509          * block so no check is necessary
510          */
511         while (num_frames--) {
512                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
513                                       0, &err)))
514                         return err; /* Failure */
515                 p++;
516                 brelse (p->bh);
517                 p->bh = bh;
518                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
519         }
520         return 1;
521 }
522
523
524 /*
525  * p is at least 6 bytes before the end of page
526  */
527 static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
528 {
529         return (struct ext4_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
530 }
531
532 /*
533  * This function fills a red-black tree with information from a
534  * directory block.  It returns the number directory entries loaded
535  * into the tree.  If there is an error it is returned in err.
536  */
537 static int htree_dirblock_to_tree(struct file *dir_file,
538                                   struct inode *dir, int block,
539                                   struct dx_hash_info *hinfo,
540                                   __u32 start_hash, __u32 start_minor_hash)
541 {
542         struct buffer_head *bh;
543         struct ext4_dir_entry_2 *de, *top;
544         int err, count = 0;
545
546         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
548                 return err;
549
550         de = (struct ext4_dir_entry_2 *) bh->b_data;
551         top = (struct ext4_dir_entry_2 *) ((char *) de +
552                                            dir->i_sb->s_blocksize -
553                                            EXT4_DIR_REC_LEN(0));
554         for (; de < top; de = ext4_next_entry(de)) {
555                 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
556                                         (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
557                                                 +((char *)de - bh->b_data))) {
558                         /* On error, skip the f_pos to the next block. */
559                         dir_file->f_pos = (dir_file->f_pos |
560                                         (dir->i_sb->s_blocksize - 1)) + 1;
561                         brelse (bh);
562                         return count;
563                 }
564                 ext4fs_dirhash(de->name, de->name_len, hinfo);
565                 if ((hinfo->hash < start_hash) ||
566                     ((hinfo->hash == start_hash) &&
567                      (hinfo->minor_hash < start_minor_hash)))
568                         continue;
569                 if (de->inode == 0)
570                         continue;
571                 if ((err = ext4_htree_store_dirent(dir_file,
572                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
573                         brelse(bh);
574                         return err;
575                 }
576                 count++;
577         }
578         brelse(bh);
579         return count;
580 }
581
582
583 /*
584  * This function fills a red-black tree with information from a
585  * directory.  We start scanning the directory in hash order, starting
586  * at start_hash and start_minor_hash.
587  *
588  * This function returns the number of entries inserted into the tree,
589  * or a negative error code.
590  */
591 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
592                          __u32 start_minor_hash, __u32 *next_hash)
593 {
594         struct dx_hash_info hinfo;
595         struct ext4_dir_entry_2 *de;
596         struct dx_frame frames[2], *frame;
597         struct inode *dir;
598         int block, err;
599         int count = 0;
600         int ret;
601         __u32 hashval;
602
603         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
604                        start_minor_hash));
605         dir = dir_file->f_path.dentry->d_inode;
606         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
607                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
608                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
609                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
610                                                start_hash, start_minor_hash);
611                 *next_hash = ~0;
612                 return count;
613         }
614         hinfo.hash = start_hash;
615         hinfo.minor_hash = 0;
616         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
617         if (!frame)
618                 return err;
619
620         /* Add '.' and '..' from the htree header */
621         if (!start_hash && !start_minor_hash) {
622                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
623                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
624                         goto errout;
625                 count++;
626         }
627         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
628                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
629                 de = ext4_next_entry(de);
630                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
631                         goto errout;
632                 count++;
633         }
634
635         while (1) {
636                 block = dx_get_block(frame->at);
637                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
638                                              start_hash, start_minor_hash);
639                 if (ret < 0) {
640                         err = ret;
641                         goto errout;
642                 }
643                 count += ret;
644                 hashval = ~0;
645                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
646                                             frame, frames, &hashval);
647                 *next_hash = hashval;
648                 if (ret < 0) {
649                         err = ret;
650                         goto errout;
651                 }
652                 /*
653                  * Stop if:  (a) there are no more entries, or
654                  * (b) we have inserted at least one entry and the
655                  * next hash value is not a continuation
656                  */
657                 if ((ret == 0) ||
658                     (count && ((hashval & 1) == 0)))
659                         break;
660         }
661         dx_release(frames);
662         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
663                        count, *next_hash));
664         return count;
665 errout:
666         dx_release(frames);
667         return (err);
668 }
669
670
671 /*
672  * Directory block splitting, compacting
673  */
674
675 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
676                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
677 {
678         int count = 0;
679         char *base = (char *) de;
680         struct dx_hash_info h = *hinfo;
681
682         while ((char *) de < base + size)
683         {
684                 if (de->name_len && de->inode) {
685                         ext4fs_dirhash(de->name, de->name_len, &h);
686                         map_tail--;
687                         map_tail->hash = h.hash;
688                         map_tail->offs = (u32) ((char *) de - base);
689                         count++;
690                         cond_resched();
691                 }
692                 /* XXX: do we need to check rec_len == 0 case? -Chris */
693                 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
694         }
695         return count;
696 }
697
698 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
699 {
700         struct dx_map_entry *p, *q, *top = map + count - 1;
701         int more;
702         /* Combsort until bubble sort doesn't suck */
703         while (count > 2) {
704                 count = count*10/13;
705                 if (count - 9 < 2) /* 9, 10 -> 11 */
706                         count = 11;
707                 for (p = top, q = p - count; q >= map; p--, q--)
708                         if (p->hash < q->hash)
709                                 swap(*p, *q);
710         }
711         /* Garden variety bubble sort */
712         do {
713                 more = 0;
714                 q = top;
715                 while (q-- > map) {
716                         if (q[1].hash >= q[0].hash)
717                                 continue;
718                         swap(*(q+1), *q);
719                         more = 1;
720                 }
721         } while(more);
722 }
723
724 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
725 {
726         struct dx_entry *entries = frame->entries;
727         struct dx_entry *old = frame->at, *new = old + 1;
728         int count = dx_get_count(entries);
729
730         assert(count < dx_get_limit(entries));
731         assert(old < entries + count);
732         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
733         dx_set_hash(new, hash);
734         dx_set_block(new, block);
735         dx_set_count(entries, count + 1);
736 }
737 #endif
738
739
740 static void ext4_update_dx_flag(struct inode *inode)
741 {
742         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
743                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
744                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
745 }
746
747 /*
748  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
749  *
750  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
751  * `de != NULL' is guaranteed by caller.
752  */
753 static inline int ext4_match (int len, const char * const name,
754                               struct ext4_dir_entry_2 * de)
755 {
756         if (len != de->name_len)
757                 return 0;
758         if (!de->inode)
759                 return 0;
760         return !memcmp(name, de->name, len);
761 }
762
763 /*
764  * Returns 0 if not found, -1 on failure, and 1 on success
765  */
766 static inline int search_dirblock(struct buffer_head * bh,
767                                   struct inode *dir,
768                                   struct dentry *dentry,
769                                   unsigned long offset,
770                                   struct ext4_dir_entry_2 ** res_dir)
771 {
772         struct ext4_dir_entry_2 * de;
773         char * dlimit;
774         int de_len;
775         const char *name = dentry->d_name.name;
776         int namelen = dentry->d_name.len;
777
778         de = (struct ext4_dir_entry_2 *) bh->b_data;
779         dlimit = bh->b_data + dir->i_sb->s_blocksize;
780         while ((char *) de < dlimit) {
781                 /* this code is executed quadratically often */
782                 /* do minimal checking `by hand' */
783
784                 if ((char *) de + namelen <= dlimit &&
785                     ext4_match (namelen, name, de)) {
786                         /* found a match - just to be sure, do a full check */
787                         if (!ext4_check_dir_entry("ext4_find_entry",
788                                                   dir, de, bh, offset))
789                                 return -1;
790                         *res_dir = de;
791                         return 1;
792                 }
793                 /* prevent looping on a bad block */
794                 de_len = le16_to_cpu(de->rec_len);
795                 if (de_len <= 0)
796                         return -1;
797                 offset += de_len;
798                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
799         }
800         return 0;
801 }
802
803
804 /*
805  *      ext4_find_entry()
806  *
807  * finds an entry in the specified directory with the wanted name. It
808  * returns the cache buffer in which the entry was found, and the entry
809  * itself (as a parameter - res_dir). It does NOT read the inode of the
810  * entry - you'll have to do that yourself if you want to.
811  *
812  * The returned buffer_head has ->b_count elevated.  The caller is expected
813  * to brelse() it when appropriate.
814  */
815 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
816                                         struct ext4_dir_entry_2 ** res_dir)
817 {
818         struct super_block * sb;
819         struct buffer_head * bh_use[NAMEI_RA_SIZE];
820         struct buffer_head * bh, *ret = NULL;
821         unsigned long start, block, b;
822         int ra_max = 0;         /* Number of bh's in the readahead
823                                    buffer, bh_use[] */
824         int ra_ptr = 0;         /* Current index into readahead
825                                    buffer */
826         int num = 0;
827         int nblocks, i, err;
828         struct inode *dir = dentry->d_parent->d_inode;
829         int namelen;
830         const u8 *name;
831         unsigned blocksize;
832
833         *res_dir = NULL;
834         sb = dir->i_sb;
835         blocksize = sb->s_blocksize;
836         namelen = dentry->d_name.len;
837         name = dentry->d_name.name;
838         if (namelen > EXT4_NAME_LEN)
839                 return NULL;
840 #ifdef CONFIG_EXT4_INDEX
841         if (is_dx(dir)) {
842                 bh = ext4_dx_find_entry(dentry, res_dir, &err);
843                 /*
844                  * On success, or if the error was file not found,
845                  * return.  Otherwise, fall back to doing a search the
846                  * old fashioned way.
847                  */
848                 if (bh || (err != ERR_BAD_DX_DIR))
849                         return bh;
850                 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
851         }
852 #endif
853         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
854         start = EXT4_I(dir)->i_dir_start_lookup;
855         if (start >= nblocks)
856                 start = 0;
857         block = start;
858 restart:
859         do {
860                 /*
861                  * We deal with the read-ahead logic here.
862                  */
863                 if (ra_ptr >= ra_max) {
864                         /* Refill the readahead buffer */
865                         ra_ptr = 0;
866                         b = block;
867                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
868                                 /*
869                                  * Terminate if we reach the end of the
870                                  * directory and must wrap, or if our
871                                  * search has finished at this block.
872                                  */
873                                 if (b >= nblocks || (num && block == start)) {
874                                         bh_use[ra_max] = NULL;
875                                         break;
876                                 }
877                                 num++;
878                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
879                                 bh_use[ra_max] = bh;
880                                 if (bh)
881                                         ll_rw_block(READ_META, 1, &bh);
882                         }
883                 }
884                 if ((bh = bh_use[ra_ptr++]) == NULL)
885                         goto next;
886                 wait_on_buffer(bh);
887                 if (!buffer_uptodate(bh)) {
888                         /* read error, skip block & hope for the best */
889                         ext4_error(sb, __FUNCTION__, "reading directory #%lu "
890                                    "offset %lu", dir->i_ino, block);
891                         brelse(bh);
892                         goto next;
893                 }
894                 i = search_dirblock(bh, dir, dentry,
895                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
896                 if (i == 1) {
897                         EXT4_I(dir)->i_dir_start_lookup = block;
898                         ret = bh;
899                         goto cleanup_and_exit;
900                 } else {
901                         brelse(bh);
902                         if (i < 0)
903                                 goto cleanup_and_exit;
904                 }
905         next:
906                 if (++block >= nblocks)
907                         block = 0;
908         } while (block != start);
909
910         /*
911          * If the directory has grown while we were searching, then
912          * search the last part of the directory before giving up.
913          */
914         block = nblocks;
915         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
916         if (block < nblocks) {
917                 start = 0;
918                 goto restart;
919         }
920
921 cleanup_and_exit:
922         /* Clean up the read-ahead blocks */
923         for (; ra_ptr < ra_max; ra_ptr++)
924                 brelse (bh_use[ra_ptr]);
925         return ret;
926 }
927
928 #ifdef CONFIG_EXT4_INDEX
929 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
930                        struct ext4_dir_entry_2 **res_dir, int *err)
931 {
932         struct super_block * sb;
933         struct dx_hash_info     hinfo;
934         u32 hash;
935         struct dx_frame frames[2], *frame;
936         struct ext4_dir_entry_2 *de, *top;
937         struct buffer_head *bh;
938         unsigned long block;
939         int retval;
940         int namelen = dentry->d_name.len;
941         const u8 *name = dentry->d_name.name;
942         struct inode *dir = dentry->d_parent->d_inode;
943
944         sb = dir->i_sb;
945         /* NFS may look up ".." - look at dx_root directory block */
946         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
947                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
948                         return NULL;
949         } else {
950                 frame = frames;
951                 frame->bh = NULL;                       /* for dx_release() */
952                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
953                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
954         }
955         hash = hinfo.hash;
956         do {
957                 block = dx_get_block(frame->at);
958                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
959                         goto errout;
960                 de = (struct ext4_dir_entry_2 *) bh->b_data;
961                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
962                                        EXT4_DIR_REC_LEN(0));
963                 for (; de < top; de = ext4_next_entry(de))
964                 if (ext4_match (namelen, name, de)) {
965                         if (!ext4_check_dir_entry("ext4_find_entry",
966                                                   dir, de, bh,
967                                   (block<<EXT4_BLOCK_SIZE_BITS(sb))
968                                           +((char *)de - bh->b_data))) {
969                                 brelse (bh);
970                                 *err = ERR_BAD_DX_DIR;
971                                 goto errout;
972                         }
973                         *res_dir = de;
974                         dx_release (frames);
975                         return bh;
976                 }
977                 brelse (bh);
978                 /* Check to see if we should continue to search */
979                 retval = ext4_htree_next_block(dir, hash, frame,
980                                                frames, NULL);
981                 if (retval < 0) {
982                         ext4_warning(sb, __FUNCTION__,
983                              "error reading index page in directory #%lu",
984                              dir->i_ino);
985                         *err = retval;
986                         goto errout;
987                 }
988         } while (retval == 1);
989
990         *err = -ENOENT;
991 errout:
992         dxtrace(printk("%s not found\n", name));
993         dx_release (frames);
994         return NULL;
995 }
996 #endif
997
998 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
999 {
1000         struct inode * inode;
1001         struct ext4_dir_entry_2 * de;
1002         struct buffer_head * bh;
1003
1004         if (dentry->d_name.len > EXT4_NAME_LEN)
1005                 return ERR_PTR(-ENAMETOOLONG);
1006
1007         bh = ext4_find_entry(dentry, &de);
1008         inode = NULL;
1009         if (bh) {
1010                 unsigned long ino = le32_to_cpu(de->inode);
1011                 brelse (bh);
1012                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1013                         ext4_error(dir->i_sb, "ext4_lookup",
1014                                    "bad inode number: %lu", ino);
1015                         inode = NULL;
1016                 } else
1017                         inode = iget(dir->i_sb, ino);
1018
1019                 if (!inode)
1020                         return ERR_PTR(-EACCES);
1021         }
1022         return d_splice_alias(inode, dentry);
1023 }
1024
1025
1026 struct dentry *ext4_get_parent(struct dentry *child)
1027 {
1028         unsigned long ino;
1029         struct dentry *parent;
1030         struct inode *inode;
1031         struct dentry dotdot;
1032         struct ext4_dir_entry_2 * de;
1033         struct buffer_head *bh;
1034
1035         dotdot.d_name.name = "..";
1036         dotdot.d_name.len = 2;
1037         dotdot.d_parent = child; /* confusing, isn't it! */
1038
1039         bh = ext4_find_entry(&dotdot, &de);
1040         inode = NULL;
1041         if (!bh)
1042                 return ERR_PTR(-ENOENT);
1043         ino = le32_to_cpu(de->inode);
1044         brelse(bh);
1045
1046         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1047                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1048                            "bad inode number: %lu", ino);
1049                 inode = NULL;
1050         } else
1051                 inode = iget(child->d_inode->i_sb, ino);
1052
1053         if (!inode)
1054                 return ERR_PTR(-EACCES);
1055
1056         parent = d_alloc_anon(inode);
1057         if (!parent) {
1058                 iput(inode);
1059                 parent = ERR_PTR(-ENOMEM);
1060         }
1061         return parent;
1062 }
1063
1064 #define S_SHIFT 12
1065 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1066         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1067         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1068         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1069         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1070         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1071         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1072         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1073 };
1074
1075 static inline void ext4_set_de_type(struct super_block *sb,
1076                                 struct ext4_dir_entry_2 *de,
1077                                 umode_t mode) {
1078         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1079                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1080 }
1081
1082 #ifdef CONFIG_EXT4_INDEX
1083 static struct ext4_dir_entry_2 *
1084 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1085 {
1086         unsigned rec_len = 0;
1087
1088         while (count--) {
1089                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1090                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1091                 memcpy (to, de, rec_len);
1092                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1093                                 cpu_to_le16(rec_len);
1094                 de->inode = 0;
1095                 map++;
1096                 to += rec_len;
1097         }
1098         return (struct ext4_dir_entry_2 *) (to - rec_len);
1099 }
1100
1101 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1102 {
1103         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1104         unsigned rec_len = 0;
1105
1106         prev = to = de;
1107         while ((char*)de < base + size) {
1108                 next = (struct ext4_dir_entry_2 *) ((char *) de +
1109                                                     le16_to_cpu(de->rec_len));
1110                 if (de->inode && de->name_len) {
1111                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1112                         if (de > to)
1113                                 memmove(to, de, rec_len);
1114                         to->rec_len = cpu_to_le16(rec_len);
1115                         prev = to;
1116                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1117                 }
1118                 de = next;
1119         }
1120         return prev;
1121 }
1122
1123 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1124                         struct buffer_head **bh,struct dx_frame *frame,
1125                         struct dx_hash_info *hinfo, int *error)
1126 {
1127         unsigned blocksize = dir->i_sb->s_blocksize;
1128         unsigned count, continued;
1129         struct buffer_head *bh2;
1130         u32 newblock;
1131         u32 hash2;
1132         struct dx_map_entry *map;
1133         char *data1 = (*bh)->b_data, *data2;
1134         unsigned split;
1135         struct ext4_dir_entry_2 *de = NULL, *de2;
1136         int     err = 0;
1137
1138         bh2 = ext4_append (handle, dir, &newblock, &err);
1139         if (!(bh2)) {
1140                 brelse(*bh);
1141                 *bh = NULL;
1142                 goto errout;
1143         }
1144
1145         BUFFER_TRACE(*bh, "get_write_access");
1146         err = ext4_journal_get_write_access(handle, *bh);
1147         if (err)
1148                 goto journal_error;
1149
1150         BUFFER_TRACE(frame->bh, "get_write_access");
1151         err = ext4_journal_get_write_access(handle, frame->bh);
1152         if (err)
1153                 goto journal_error;
1154
1155         data2 = bh2->b_data;
1156
1157         /* create map in the end of data2 block */
1158         map = (struct dx_map_entry *) (data2 + blocksize);
1159         count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1160                              blocksize, hinfo, map);
1161         map -= count;
1162         split = count/2; // need to adjust to actual middle
1163         dx_sort_map (map, count);
1164         hash2 = map[split].hash;
1165         continued = hash2 == map[split - 1].hash;
1166         dxtrace(printk("Split block %i at %x, %i/%i\n",
1167                 dx_get_block(frame->at), hash2, split, count-split));
1168
1169         /* Fancy dance to stay within two buffers */
1170         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1171         de = dx_pack_dirents(data1,blocksize);
1172         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1173         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1174         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1175         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1176
1177         /* Which block gets the new entry? */
1178         if (hinfo->hash >= hash2)
1179         {
1180                 swap(*bh, bh2);
1181                 de = de2;
1182         }
1183         dx_insert_block (frame, hash2 + continued, newblock);
1184         err = ext4_journal_dirty_metadata (handle, bh2);
1185         if (err)
1186                 goto journal_error;
1187         err = ext4_journal_dirty_metadata (handle, frame->bh);
1188         if (err)
1189                 goto journal_error;
1190         brelse (bh2);
1191         dxtrace(dx_show_index ("frame", frame->entries));
1192         return de;
1193
1194 journal_error:
1195         brelse(*bh);
1196         brelse(bh2);
1197         *bh = NULL;
1198         ext4_std_error(dir->i_sb, err);
1199 errout:
1200         *error = err;
1201         return NULL;
1202 }
1203 #endif
1204
1205
1206 /*
1207  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1208  * it points to a directory entry which is guaranteed to be large
1209  * enough for new directory entry.  If de is NULL, then
1210  * add_dirent_to_buf will attempt search the directory block for
1211  * space.  It will return -ENOSPC if no space is available, and -EIO
1212  * and -EEXIST if directory entry already exists.
1213  *
1214  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1215  * all other cases bh is released.
1216  */
1217 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1218                              struct inode *inode, struct ext4_dir_entry_2 *de,
1219                              struct buffer_head * bh)
1220 {
1221         struct inode    *dir = dentry->d_parent->d_inode;
1222         const char      *name = dentry->d_name.name;
1223         int             namelen = dentry->d_name.len;
1224         unsigned long   offset = 0;
1225         unsigned short  reclen;
1226         int             nlen, rlen, err;
1227         char            *top;
1228
1229         reclen = EXT4_DIR_REC_LEN(namelen);
1230         if (!de) {
1231                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1232                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1233                 while ((char *) de <= top) {
1234                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1235                                                   bh, offset)) {
1236                                 brelse (bh);
1237                                 return -EIO;
1238                         }
1239                         if (ext4_match (namelen, name, de)) {
1240                                 brelse (bh);
1241                                 return -EEXIST;
1242                         }
1243                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1244                         rlen = le16_to_cpu(de->rec_len);
1245                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1246                                 break;
1247                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1248                         offset += rlen;
1249                 }
1250                 if ((char *) de > top)
1251                         return -ENOSPC;
1252         }
1253         BUFFER_TRACE(bh, "get_write_access");
1254         err = ext4_journal_get_write_access(handle, bh);
1255         if (err) {
1256                 ext4_std_error(dir->i_sb, err);
1257                 brelse(bh);
1258                 return err;
1259         }
1260
1261         /* By now the buffer is marked for journaling */
1262         nlen = EXT4_DIR_REC_LEN(de->name_len);
1263         rlen = le16_to_cpu(de->rec_len);
1264         if (de->inode) {
1265                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1266                 de1->rec_len = cpu_to_le16(rlen - nlen);
1267                 de->rec_len = cpu_to_le16(nlen);
1268                 de = de1;
1269         }
1270         de->file_type = EXT4_FT_UNKNOWN;
1271         if (inode) {
1272                 de->inode = cpu_to_le32(inode->i_ino);
1273                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1274         } else
1275                 de->inode = 0;
1276         de->name_len = namelen;
1277         memcpy (de->name, name, namelen);
1278         /*
1279          * XXX shouldn't update any times until successful
1280          * completion of syscall, but too many callers depend
1281          * on this.
1282          *
1283          * XXX similarly, too many callers depend on
1284          * ext4_new_inode() setting the times, but error
1285          * recovery deletes the inode, so the worst that can
1286          * happen is that the times are slightly out of date
1287          * and/or different from the directory change time.
1288          */
1289         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1290         ext4_update_dx_flag(dir);
1291         dir->i_version++;
1292         ext4_mark_inode_dirty(handle, dir);
1293         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1294         err = ext4_journal_dirty_metadata(handle, bh);
1295         if (err)
1296                 ext4_std_error(dir->i_sb, err);
1297         brelse(bh);
1298         return 0;
1299 }
1300
1301 #ifdef CONFIG_EXT4_INDEX
1302 /*
1303  * This converts a one block unindexed directory to a 3 block indexed
1304  * directory, and adds the dentry to the indexed directory.
1305  */
1306 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1307                             struct inode *inode, struct buffer_head *bh)
1308 {
1309         struct inode    *dir = dentry->d_parent->d_inode;
1310         const char      *name = dentry->d_name.name;
1311         int             namelen = dentry->d_name.len;
1312         struct buffer_head *bh2;
1313         struct dx_root  *root;
1314         struct dx_frame frames[2], *frame;
1315         struct dx_entry *entries;
1316         struct ext4_dir_entry_2 *de, *de2;
1317         char            *data1, *top;
1318         unsigned        len;
1319         int             retval;
1320         unsigned        blocksize;
1321         struct dx_hash_info hinfo;
1322         u32             block;
1323         struct fake_dirent *fde;
1324
1325         blocksize =  dir->i_sb->s_blocksize;
1326         dxtrace(printk("Creating index\n"));
1327         retval = ext4_journal_get_write_access(handle, bh);
1328         if (retval) {
1329                 ext4_std_error(dir->i_sb, retval);
1330                 brelse(bh);
1331                 return retval;
1332         }
1333         root = (struct dx_root *) bh->b_data;
1334
1335         bh2 = ext4_append (handle, dir, &block, &retval);
1336         if (!(bh2)) {
1337                 brelse(bh);
1338                 return retval;
1339         }
1340         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1341         data1 = bh2->b_data;
1342
1343         /* The 0th block becomes the root, move the dirents out */
1344         fde = &root->dotdot;
1345         de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1346         len = ((char *) root) + blocksize - (char *) de;
1347         memcpy (data1, de, len);
1348         de = (struct ext4_dir_entry_2 *) data1;
1349         top = data1 + len;
1350         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1351                 de = de2;
1352         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1353         /* Initialize the root; the dot dirents already exist */
1354         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1355         de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1356         memset (&root->info, 0, sizeof(root->info));
1357         root->info.info_length = sizeof(root->info);
1358         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1359         entries = root->entries;
1360         dx_set_block (entries, 1);
1361         dx_set_count (entries, 1);
1362         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1363
1364         /* Initialize as for dx_probe */
1365         hinfo.hash_version = root->info.hash_version;
1366         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1367         ext4fs_dirhash(name, namelen, &hinfo);
1368         frame = frames;
1369         frame->entries = entries;
1370         frame->at = entries;
1371         frame->bh = bh;
1372         bh = bh2;
1373         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1374         dx_release (frames);
1375         if (!(de))
1376                 return retval;
1377
1378         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1379 }
1380 #endif
1381
1382 /*
1383  *      ext4_add_entry()
1384  *
1385  * adds a file entry to the specified directory, using the same
1386  * semantics as ext4_find_entry(). It returns NULL if it failed.
1387  *
1388  * NOTE!! The inode part of 'de' is left at 0 - which means you
1389  * may not sleep between calling this and putting something into
1390  * the entry, as someone else might have used it while you slept.
1391  */
1392 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1393         struct inode *inode)
1394 {
1395         struct inode *dir = dentry->d_parent->d_inode;
1396         unsigned long offset;
1397         struct buffer_head * bh;
1398         struct ext4_dir_entry_2 *de;
1399         struct super_block * sb;
1400         int     retval;
1401 #ifdef CONFIG_EXT4_INDEX
1402         int     dx_fallback=0;
1403 #endif
1404         unsigned blocksize;
1405         u32 block, blocks;
1406
1407         sb = dir->i_sb;
1408         blocksize = sb->s_blocksize;
1409         if (!dentry->d_name.len)
1410                 return -EINVAL;
1411 #ifdef CONFIG_EXT4_INDEX
1412         if (is_dx(dir)) {
1413                 retval = ext4_dx_add_entry(handle, dentry, inode);
1414                 if (!retval || (retval != ERR_BAD_DX_DIR))
1415                         return retval;
1416                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1417                 dx_fallback++;
1418                 ext4_mark_inode_dirty(handle, dir);
1419         }
1420 #endif
1421         blocks = dir->i_size >> sb->s_blocksize_bits;
1422         for (block = 0, offset = 0; block < blocks; block++) {
1423                 bh = ext4_bread(handle, dir, block, 0, &retval);
1424                 if(!bh)
1425                         return retval;
1426                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1427                 if (retval != -ENOSPC)
1428                         return retval;
1429
1430 #ifdef CONFIG_EXT4_INDEX
1431                 if (blocks == 1 && !dx_fallback &&
1432                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1433                         return make_indexed_dir(handle, dentry, inode, bh);
1434 #endif
1435                 brelse(bh);
1436         }
1437         bh = ext4_append(handle, dir, &block, &retval);
1438         if (!bh)
1439                 return retval;
1440         de = (struct ext4_dir_entry_2 *) bh->b_data;
1441         de->inode = 0;
1442         de->rec_len = cpu_to_le16(blocksize);
1443         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1444 }
1445
1446 #ifdef CONFIG_EXT4_INDEX
1447 /*
1448  * Returns 0 for success, or a negative error value
1449  */
1450 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1451                              struct inode *inode)
1452 {
1453         struct dx_frame frames[2], *frame;
1454         struct dx_entry *entries, *at;
1455         struct dx_hash_info hinfo;
1456         struct buffer_head * bh;
1457         struct inode *dir = dentry->d_parent->d_inode;
1458         struct super_block * sb = dir->i_sb;
1459         struct ext4_dir_entry_2 *de;
1460         int err;
1461
1462         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1463         if (!frame)
1464                 return err;
1465         entries = frame->entries;
1466         at = frame->at;
1467
1468         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1469                 goto cleanup;
1470
1471         BUFFER_TRACE(bh, "get_write_access");
1472         err = ext4_journal_get_write_access(handle, bh);
1473         if (err)
1474                 goto journal_error;
1475
1476         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477         if (err != -ENOSPC) {
1478                 bh = NULL;
1479                 goto cleanup;
1480         }
1481
1482         /* Block full, should compress but for now just split */
1483         dxtrace(printk("using %u of %u node entries\n",
1484                        dx_get_count(entries), dx_get_limit(entries)));
1485         /* Need to split index? */
1486         if (dx_get_count(entries) == dx_get_limit(entries)) {
1487                 u32 newblock;
1488                 unsigned icount = dx_get_count(entries);
1489                 int levels = frame - frames;
1490                 struct dx_entry *entries2;
1491                 struct dx_node *node2;
1492                 struct buffer_head *bh2;
1493
1494                 if (levels && (dx_get_count(frames->entries) ==
1495                                dx_get_limit(frames->entries))) {
1496                         ext4_warning(sb, __FUNCTION__,
1497                                      "Directory index full!");
1498                         err = -ENOSPC;
1499                         goto cleanup;
1500                 }
1501                 bh2 = ext4_append (handle, dir, &newblock, &err);
1502                 if (!(bh2))
1503                         goto cleanup;
1504                 node2 = (struct dx_node *)(bh2->b_data);
1505                 entries2 = node2->entries;
1506                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1507                 node2->fake.inode = 0;
1508                 BUFFER_TRACE(frame->bh, "get_write_access");
1509                 err = ext4_journal_get_write_access(handle, frame->bh);
1510                 if (err)
1511                         goto journal_error;
1512                 if (levels) {
1513                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1514                         unsigned hash2 = dx_get_hash(entries + icount1);
1515                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1516
1517                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1518                         err = ext4_journal_get_write_access(handle,
1519                                                              frames[0].bh);
1520                         if (err)
1521                                 goto journal_error;
1522
1523                         memcpy ((char *) entries2, (char *) (entries + icount1),
1524                                 icount2 * sizeof(struct dx_entry));
1525                         dx_set_count (entries, icount1);
1526                         dx_set_count (entries2, icount2);
1527                         dx_set_limit (entries2, dx_node_limit(dir));
1528
1529                         /* Which index block gets the new entry? */
1530                         if (at - entries >= icount1) {
1531                                 frame->at = at = at - entries - icount1 + entries2;
1532                                 frame->entries = entries = entries2;
1533                                 swap(frame->bh, bh2);
1534                         }
1535                         dx_insert_block (frames + 0, hash2, newblock);
1536                         dxtrace(dx_show_index ("node", frames[1].entries));
1537                         dxtrace(dx_show_index ("node",
1538                                ((struct dx_node *) bh2->b_data)->entries));
1539                         err = ext4_journal_dirty_metadata(handle, bh2);
1540                         if (err)
1541                                 goto journal_error;
1542                         brelse (bh2);
1543                 } else {
1544                         dxtrace(printk("Creating second level index...\n"));
1545                         memcpy((char *) entries2, (char *) entries,
1546                                icount * sizeof(struct dx_entry));
1547                         dx_set_limit(entries2, dx_node_limit(dir));
1548
1549                         /* Set up root */
1550                         dx_set_count(entries, 1);
1551                         dx_set_block(entries + 0, newblock);
1552                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1553
1554                         /* Add new access path frame */
1555                         frame = frames + 1;
1556                         frame->at = at = at - entries + entries2;
1557                         frame->entries = entries = entries2;
1558                         frame->bh = bh2;
1559                         err = ext4_journal_get_write_access(handle,
1560                                                              frame->bh);
1561                         if (err)
1562                                 goto journal_error;
1563                 }
1564                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1565         }
1566         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1567         if (!de)
1568                 goto cleanup;
1569         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1570         bh = NULL;
1571         goto cleanup;
1572
1573 journal_error:
1574         ext4_std_error(dir->i_sb, err);
1575 cleanup:
1576         if (bh)
1577                 brelse(bh);
1578         dx_release(frames);
1579         return err;
1580 }
1581 #endif
1582
1583 /*
1584  * ext4_delete_entry deletes a directory entry by merging it with the
1585  * previous entry
1586  */
1587 static int ext4_delete_entry (handle_t *handle,
1588                               struct inode * dir,
1589                               struct ext4_dir_entry_2 * de_del,
1590                               struct buffer_head * bh)
1591 {
1592         struct ext4_dir_entry_2 * de, * pde;
1593         int i;
1594
1595         i = 0;
1596         pde = NULL;
1597         de = (struct ext4_dir_entry_2 *) bh->b_data;
1598         while (i < bh->b_size) {
1599                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1600                         return -EIO;
1601                 if (de == de_del)  {
1602                         BUFFER_TRACE(bh, "get_write_access");
1603                         ext4_journal_get_write_access(handle, bh);
1604                         if (pde)
1605                                 pde->rec_len =
1606                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1607                                                     le16_to_cpu(de->rec_len));
1608                         else
1609                                 de->inode = 0;
1610                         dir->i_version++;
1611                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1612                         ext4_journal_dirty_metadata(handle, bh);
1613                         return 0;
1614                 }
1615                 i += le16_to_cpu(de->rec_len);
1616                 pde = de;
1617                 de = (struct ext4_dir_entry_2 *)
1618                         ((char *) de + le16_to_cpu(de->rec_len));
1619         }
1620         return -ENOENT;
1621 }
1622
1623 static int ext4_add_nondir(handle_t *handle,
1624                 struct dentry *dentry, struct inode *inode)
1625 {
1626         int err = ext4_add_entry(handle, dentry, inode);
1627         if (!err) {
1628                 ext4_mark_inode_dirty(handle, inode);
1629                 d_instantiate(dentry, inode);
1630                 return 0;
1631         }
1632         drop_nlink(inode);
1633         iput(inode);
1634         return err;
1635 }
1636
1637 /*
1638  * By the time this is called, we already have created
1639  * the directory cache entry for the new file, but it
1640  * is so far negative - it has no inode.
1641  *
1642  * If the create succeeds, we fill in the inode information
1643  * with d_instantiate().
1644  */
1645 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1646                 struct nameidata *nd)
1647 {
1648         handle_t *handle;
1649         struct inode * inode;
1650         int err, retries = 0;
1651
1652 retry:
1653         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1654                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1655                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1656         if (IS_ERR(handle))
1657                 return PTR_ERR(handle);
1658
1659         if (IS_DIRSYNC(dir))
1660                 handle->h_sync = 1;
1661
1662         inode = ext4_new_inode (handle, dir, mode);
1663         err = PTR_ERR(inode);
1664         if (!IS_ERR(inode)) {
1665                 inode->i_op = &ext4_file_inode_operations;
1666                 inode->i_fop = &ext4_file_operations;
1667                 ext4_set_aops(inode);
1668                 err = ext4_add_nondir(handle, dentry, inode);
1669         }
1670         ext4_journal_stop(handle);
1671         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1672                 goto retry;
1673         return err;
1674 }
1675
1676 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1677                         int mode, dev_t rdev)
1678 {
1679         handle_t *handle;
1680         struct inode *inode;
1681         int err, retries = 0;
1682
1683         if (!new_valid_dev(rdev))
1684                 return -EINVAL;
1685
1686 retry:
1687         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1688                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1689                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1690         if (IS_ERR(handle))
1691                 return PTR_ERR(handle);
1692
1693         if (IS_DIRSYNC(dir))
1694                 handle->h_sync = 1;
1695
1696         inode = ext4_new_inode (handle, dir, mode);
1697         err = PTR_ERR(inode);
1698         if (!IS_ERR(inode)) {
1699                 init_special_inode(inode, inode->i_mode, rdev);
1700 #ifdef CONFIG_EXT4DEV_FS_XATTR
1701                 inode->i_op = &ext4_special_inode_operations;
1702 #endif
1703                 err = ext4_add_nondir(handle, dentry, inode);
1704         }
1705         ext4_journal_stop(handle);
1706         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1707                 goto retry;
1708         return err;
1709 }
1710
1711 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1712 {
1713         handle_t *handle;
1714         struct inode * inode;
1715         struct buffer_head * dir_block;
1716         struct ext4_dir_entry_2 * de;
1717         int err, retries = 0;
1718
1719         if (dir->i_nlink >= EXT4_LINK_MAX)
1720                 return -EMLINK;
1721
1722 retry:
1723         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1724                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1725                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1726         if (IS_ERR(handle))
1727                 return PTR_ERR(handle);
1728
1729         if (IS_DIRSYNC(dir))
1730                 handle->h_sync = 1;
1731
1732         inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1733         err = PTR_ERR(inode);
1734         if (IS_ERR(inode))
1735                 goto out_stop;
1736
1737         inode->i_op = &ext4_dir_inode_operations;
1738         inode->i_fop = &ext4_dir_operations;
1739         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1740         dir_block = ext4_bread (handle, inode, 0, 1, &err);
1741         if (!dir_block) {
1742                 drop_nlink(inode); /* is this nlink == 0? */
1743                 ext4_mark_inode_dirty(handle, inode);
1744                 iput (inode);
1745                 goto out_stop;
1746         }
1747         BUFFER_TRACE(dir_block, "get_write_access");
1748         ext4_journal_get_write_access(handle, dir_block);
1749         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1750         de->inode = cpu_to_le32(inode->i_ino);
1751         de->name_len = 1;
1752         de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1753         strcpy (de->name, ".");
1754         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1755         de = (struct ext4_dir_entry_2 *)
1756                         ((char *) de + le16_to_cpu(de->rec_len));
1757         de->inode = cpu_to_le32(dir->i_ino);
1758         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1759         de->name_len = 2;
1760         strcpy (de->name, "..");
1761         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1762         inode->i_nlink = 2;
1763         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1764         ext4_journal_dirty_metadata(handle, dir_block);
1765         brelse (dir_block);
1766         ext4_mark_inode_dirty(handle, inode);
1767         err = ext4_add_entry (handle, dentry, inode);
1768         if (err) {
1769                 inode->i_nlink = 0;
1770                 ext4_mark_inode_dirty(handle, inode);
1771                 iput (inode);
1772                 goto out_stop;
1773         }
1774         inc_nlink(dir);
1775         ext4_update_dx_flag(dir);
1776         ext4_mark_inode_dirty(handle, dir);
1777         d_instantiate(dentry, inode);
1778 out_stop:
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 /*
1786  * routine to check that the specified directory is empty (for rmdir)
1787  */
1788 static int empty_dir (struct inode * inode)
1789 {
1790         unsigned long offset;
1791         struct buffer_head * bh;
1792         struct ext4_dir_entry_2 * de, * de1;
1793         struct super_block * sb;
1794         int err = 0;
1795
1796         sb = inode->i_sb;
1797         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1798             !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1799                 if (err)
1800                         ext4_error(inode->i_sb, __FUNCTION__,
1801                                    "error %d reading directory #%lu offset 0",
1802                                    err, inode->i_ino);
1803                 else
1804                         ext4_warning(inode->i_sb, __FUNCTION__,
1805                                      "bad directory (dir #%lu) - no data block",
1806                                      inode->i_ino);
1807                 return 1;
1808         }
1809         de = (struct ext4_dir_entry_2 *) bh->b_data;
1810         de1 = (struct ext4_dir_entry_2 *)
1811                         ((char *) de + le16_to_cpu(de->rec_len));
1812         if (le32_to_cpu(de->inode) != inode->i_ino ||
1813                         !le32_to_cpu(de1->inode) ||
1814                         strcmp (".", de->name) ||
1815                         strcmp ("..", de1->name)) {
1816                 ext4_warning (inode->i_sb, "empty_dir",
1817                               "bad directory (dir #%lu) - no `.' or `..'",
1818                               inode->i_ino);
1819                 brelse (bh);
1820                 return 1;
1821         }
1822         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1823         de = (struct ext4_dir_entry_2 *)
1824                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1825         while (offset < inode->i_size ) {
1826                 if (!bh ||
1827                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1828                         err = 0;
1829                         brelse (bh);
1830                         bh = ext4_bread (NULL, inode,
1831                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1832                         if (!bh) {
1833                                 if (err)
1834                                         ext4_error(sb, __FUNCTION__,
1835                                                    "error %d reading directory"
1836                                                    " #%lu offset %lu",
1837                                                    err, inode->i_ino, offset);
1838                                 offset += sb->s_blocksize;
1839                                 continue;
1840                         }
1841                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1842                 }
1843                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1844                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1845                                                          sb->s_blocksize);
1846                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1847                         continue;
1848                 }
1849                 if (le32_to_cpu(de->inode)) {
1850                         brelse (bh);
1851                         return 0;
1852                 }
1853                 offset += le16_to_cpu(de->rec_len);
1854                 de = (struct ext4_dir_entry_2 *)
1855                                 ((char *) de + le16_to_cpu(de->rec_len));
1856         }
1857         brelse (bh);
1858         return 1;
1859 }
1860
1861 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1862  * such inodes, starting at the superblock, in case we crash before the
1863  * file is closed/deleted, or in case the inode truncate spans multiple
1864  * transactions and the last transaction is not recovered after a crash.
1865  *
1866  * At filesystem recovery time, we walk this list deleting unlinked
1867  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1868  */
1869 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1870 {
1871         struct super_block *sb = inode->i_sb;
1872         struct ext4_iloc iloc;
1873         int err = 0, rc;
1874
1875         lock_super(sb);
1876         if (!list_empty(&EXT4_I(inode)->i_orphan))
1877                 goto out_unlock;
1878
1879         /* Orphan handling is only valid for files with data blocks
1880          * being truncated, or files being unlinked. */
1881
1882         /* @@@ FIXME: Observation from aviro:
1883          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1884          * here (on lock_super()), so race with ext4_link() which might bump
1885          * ->i_nlink. For, say it, character device. Not a regular file,
1886          * not a directory, not a symlink and ->i_nlink > 0.
1887          */
1888         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1889                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1890
1891         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1892         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1893         if (err)
1894                 goto out_unlock;
1895
1896         err = ext4_reserve_inode_write(handle, inode, &iloc);
1897         if (err)
1898                 goto out_unlock;
1899
1900         /* Insert this inode at the head of the on-disk orphan list... */
1901         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1902         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1903         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1904         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1905         if (!err)
1906                 err = rc;
1907
1908         /* Only add to the head of the in-memory list if all the
1909          * previous operations succeeded.  If the orphan_add is going to
1910          * fail (possibly taking the journal offline), we can't risk
1911          * leaving the inode on the orphan list: stray orphan-list
1912          * entries can cause panics at unmount time.
1913          *
1914          * This is safe: on error we're going to ignore the orphan list
1915          * anyway on the next recovery. */
1916         if (!err)
1917                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1918
1919         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1920         jbd_debug(4, "orphan inode %lu will point to %d\n",
1921                         inode->i_ino, NEXT_ORPHAN(inode));
1922 out_unlock:
1923         unlock_super(sb);
1924         ext4_std_error(inode->i_sb, err);
1925         return err;
1926 }
1927
1928 /*
1929  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1930  * of such inodes stored on disk, because it is finally being cleaned up.
1931  */
1932 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1933 {
1934         struct list_head *prev;
1935         struct ext4_inode_info *ei = EXT4_I(inode);
1936         struct ext4_sb_info *sbi;
1937         unsigned long ino_next;
1938         struct ext4_iloc iloc;
1939         int err = 0;
1940
1941         lock_super(inode->i_sb);
1942         if (list_empty(&ei->i_orphan)) {
1943                 unlock_super(inode->i_sb);
1944                 return 0;
1945         }
1946
1947         ino_next = NEXT_ORPHAN(inode);
1948         prev = ei->i_orphan.prev;
1949         sbi = EXT4_SB(inode->i_sb);
1950
1951         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1952
1953         list_del_init(&ei->i_orphan);
1954
1955         /* If we're on an error path, we may not have a valid
1956          * transaction handle with which to update the orphan list on
1957          * disk, but we still need to remove the inode from the linked
1958          * list in memory. */
1959         if (!handle)
1960                 goto out;
1961
1962         err = ext4_reserve_inode_write(handle, inode, &iloc);
1963         if (err)
1964                 goto out_err;
1965
1966         if (prev == &sbi->s_orphan) {
1967                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1968                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1969                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1970                 if (err)
1971                         goto out_brelse;
1972                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1973                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
1974         } else {
1975                 struct ext4_iloc iloc2;
1976                 struct inode *i_prev =
1977                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
1978
1979                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1980                           i_prev->i_ino, ino_next);
1981                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
1982                 if (err)
1983                         goto out_brelse;
1984                 NEXT_ORPHAN(i_prev) = ino_next;
1985                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
1986         }
1987         if (err)
1988                 goto out_brelse;
1989         NEXT_ORPHAN(inode) = 0;
1990         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
1991
1992 out_err:
1993         ext4_std_error(inode->i_sb, err);
1994 out:
1995         unlock_super(inode->i_sb);
1996         return err;
1997
1998 out_brelse:
1999         brelse(iloc.bh);
2000         goto out_err;
2001 }
2002
2003 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2004 {
2005         int retval;
2006         struct inode * inode;
2007         struct buffer_head * bh;
2008         struct ext4_dir_entry_2 * de;
2009         handle_t *handle;
2010
2011         /* Initialize quotas before so that eventual writes go in
2012          * separate transaction */
2013         DQUOT_INIT(dentry->d_inode);
2014         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2015         if (IS_ERR(handle))
2016                 return PTR_ERR(handle);
2017
2018         retval = -ENOENT;
2019         bh = ext4_find_entry (dentry, &de);
2020         if (!bh)
2021                 goto end_rmdir;
2022
2023         if (IS_DIRSYNC(dir))
2024                 handle->h_sync = 1;
2025
2026         inode = dentry->d_inode;
2027
2028         retval = -EIO;
2029         if (le32_to_cpu(de->inode) != inode->i_ino)
2030                 goto end_rmdir;
2031
2032         retval = -ENOTEMPTY;
2033         if (!empty_dir (inode))
2034                 goto end_rmdir;
2035
2036         retval = ext4_delete_entry(handle, dir, de, bh);
2037         if (retval)
2038                 goto end_rmdir;
2039         if (inode->i_nlink != 2)
2040                 ext4_warning (inode->i_sb, "ext4_rmdir",
2041                               "empty directory has nlink!=2 (%d)",
2042                               inode->i_nlink);
2043         inode->i_version++;
2044         clear_nlink(inode);
2045         /* There's no need to set i_disksize: the fact that i_nlink is
2046          * zero will ensure that the right thing happens during any
2047          * recovery. */
2048         inode->i_size = 0;
2049         ext4_orphan_add(handle, inode);
2050         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2051         ext4_mark_inode_dirty(handle, inode);
2052         drop_nlink(dir);
2053         ext4_update_dx_flag(dir);
2054         ext4_mark_inode_dirty(handle, dir);
2055
2056 end_rmdir:
2057         ext4_journal_stop(handle);
2058         brelse (bh);
2059         return retval;
2060 }
2061
2062 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2063 {
2064         int retval;
2065         struct inode * inode;
2066         struct buffer_head * bh;
2067         struct ext4_dir_entry_2 * de;
2068         handle_t *handle;
2069
2070         /* Initialize quotas before so that eventual writes go
2071          * in separate transaction */
2072         DQUOT_INIT(dentry->d_inode);
2073         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2074         if (IS_ERR(handle))
2075                 return PTR_ERR(handle);
2076
2077         if (IS_DIRSYNC(dir))
2078                 handle->h_sync = 1;
2079
2080         retval = -ENOENT;
2081         bh = ext4_find_entry (dentry, &de);
2082         if (!bh)
2083                 goto end_unlink;
2084
2085         inode = dentry->d_inode;
2086
2087         retval = -EIO;
2088         if (le32_to_cpu(de->inode) != inode->i_ino)
2089                 goto end_unlink;
2090
2091         if (!inode->i_nlink) {
2092                 ext4_warning (inode->i_sb, "ext4_unlink",
2093                               "Deleting nonexistent file (%lu), %d",
2094                               inode->i_ino, inode->i_nlink);
2095                 inode->i_nlink = 1;
2096         }
2097         retval = ext4_delete_entry(handle, dir, de, bh);
2098         if (retval)
2099                 goto end_unlink;
2100         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2101         ext4_update_dx_flag(dir);
2102         ext4_mark_inode_dirty(handle, dir);
2103         drop_nlink(inode);
2104         if (!inode->i_nlink)
2105                 ext4_orphan_add(handle, inode);
2106         inode->i_ctime = dir->i_ctime;
2107         ext4_mark_inode_dirty(handle, inode);
2108         retval = 0;
2109
2110 end_unlink:
2111         ext4_journal_stop(handle);
2112         brelse (bh);
2113         return retval;
2114 }
2115
2116 static int ext4_symlink (struct inode * dir,
2117                 struct dentry *dentry, const char * symname)
2118 {
2119         handle_t *handle;
2120         struct inode * inode;
2121         int l, err, retries = 0;
2122
2123         l = strlen(symname)+1;
2124         if (l > dir->i_sb->s_blocksize)
2125                 return -ENAMETOOLONG;
2126
2127 retry:
2128         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2129                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2130                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2131         if (IS_ERR(handle))
2132                 return PTR_ERR(handle);
2133
2134         if (IS_DIRSYNC(dir))
2135                 handle->h_sync = 1;
2136
2137         inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2138         err = PTR_ERR(inode);
2139         if (IS_ERR(inode))
2140                 goto out_stop;
2141
2142         if (l > sizeof (EXT4_I(inode)->i_data)) {
2143                 inode->i_op = &ext4_symlink_inode_operations;
2144                 ext4_set_aops(inode);
2145                 /*
2146                  * page_symlink() calls into ext4_prepare/commit_write.
2147                  * We have a transaction open.  All is sweetness.  It also sets
2148                  * i_size in generic_commit_write().
2149                  */
2150                 err = __page_symlink(inode, symname, l,
2151                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2152                 if (err) {
2153                         drop_nlink(inode);
2154                         ext4_mark_inode_dirty(handle, inode);
2155                         iput (inode);
2156                         goto out_stop;
2157                 }
2158         } else {
2159                 inode->i_op = &ext4_fast_symlink_inode_operations;
2160                 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2161                 inode->i_size = l-1;
2162         }
2163         EXT4_I(inode)->i_disksize = inode->i_size;
2164         err = ext4_add_nondir(handle, dentry, inode);
2165 out_stop:
2166         ext4_journal_stop(handle);
2167         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2168                 goto retry;
2169         return err;
2170 }
2171
2172 static int ext4_link (struct dentry * old_dentry,
2173                 struct inode * dir, struct dentry *dentry)
2174 {
2175         handle_t *handle;
2176         struct inode *inode = old_dentry->d_inode;
2177         int err, retries = 0;
2178
2179         if (inode->i_nlink >= EXT4_LINK_MAX)
2180                 return -EMLINK;
2181         /*
2182          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2183          * otherwise has the potential to corrupt the orphan inode list.
2184          */
2185         if (inode->i_nlink == 0)
2186                 return -ENOENT;
2187
2188 retry:
2189         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2190                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2191         if (IS_ERR(handle))
2192                 return PTR_ERR(handle);
2193
2194         if (IS_DIRSYNC(dir))
2195                 handle->h_sync = 1;
2196
2197         inode->i_ctime = CURRENT_TIME_SEC;
2198         inc_nlink(inode);
2199         atomic_inc(&inode->i_count);
2200
2201         err = ext4_add_nondir(handle, dentry, inode);
2202         ext4_journal_stop(handle);
2203         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2204                 goto retry;
2205         return err;
2206 }
2207
2208 #define PARENT_INO(buffer) \
2209         ((struct ext4_dir_entry_2 *) ((char *) buffer + \
2210         le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2211
2212 /*
2213  * Anybody can rename anything with this: the permission checks are left to the
2214  * higher-level routines.
2215  */
2216 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2217                            struct inode * new_dir,struct dentry *new_dentry)
2218 {
2219         handle_t *handle;
2220         struct inode * old_inode, * new_inode;
2221         struct buffer_head * old_bh, * new_bh, * dir_bh;
2222         struct ext4_dir_entry_2 * old_de, * new_de;
2223         int retval;
2224
2225         old_bh = new_bh = dir_bh = NULL;
2226
2227         /* Initialize quotas before so that eventual writes go
2228          * in separate transaction */
2229         if (new_dentry->d_inode)
2230                 DQUOT_INIT(new_dentry->d_inode);
2231         handle = ext4_journal_start(old_dir, 2 *
2232                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2233                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2234         if (IS_ERR(handle))
2235                 return PTR_ERR(handle);
2236
2237         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2238                 handle->h_sync = 1;
2239
2240         old_bh = ext4_find_entry (old_dentry, &old_de);
2241         /*
2242          *  Check for inode number is _not_ due to possible IO errors.
2243          *  We might rmdir the source, keep it as pwd of some process
2244          *  and merrily kill the link to whatever was created under the
2245          *  same name. Goodbye sticky bit ;-<
2246          */
2247         old_inode = old_dentry->d_inode;
2248         retval = -ENOENT;
2249         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2250                 goto end_rename;
2251
2252         new_inode = new_dentry->d_inode;
2253         new_bh = ext4_find_entry (new_dentry, &new_de);
2254         if (new_bh) {
2255                 if (!new_inode) {
2256                         brelse (new_bh);
2257                         new_bh = NULL;
2258                 }
2259         }
2260         if (S_ISDIR(old_inode->i_mode)) {
2261                 if (new_inode) {
2262                         retval = -ENOTEMPTY;
2263                         if (!empty_dir (new_inode))
2264                                 goto end_rename;
2265                 }
2266                 retval = -EIO;
2267                 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2268                 if (!dir_bh)
2269                         goto end_rename;
2270                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2271                         goto end_rename;
2272                 retval = -EMLINK;
2273                 if (!new_inode && new_dir!=old_dir &&
2274                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2275                         goto end_rename;
2276         }
2277         if (!new_bh) {
2278                 retval = ext4_add_entry (handle, new_dentry, old_inode);
2279                 if (retval)
2280                         goto end_rename;
2281         } else {
2282                 BUFFER_TRACE(new_bh, "get write access");
2283                 ext4_journal_get_write_access(handle, new_bh);
2284                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2285                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2286                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2287                         new_de->file_type = old_de->file_type;
2288                 new_dir->i_version++;
2289                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2290                 ext4_journal_dirty_metadata(handle, new_bh);
2291                 brelse(new_bh);
2292                 new_bh = NULL;
2293         }
2294
2295         /*
2296          * Like most other Unix systems, set the ctime for inodes on a
2297          * rename.
2298          */
2299         old_inode->i_ctime = CURRENT_TIME_SEC;
2300         ext4_mark_inode_dirty(handle, old_inode);
2301
2302         /*
2303          * ok, that's it
2304          */
2305         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2306             old_de->name_len != old_dentry->d_name.len ||
2307             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2308             (retval = ext4_delete_entry(handle, old_dir,
2309                                         old_de, old_bh)) == -ENOENT) {
2310                 /* old_de could have moved from under us during htree split, so
2311                  * make sure that we are deleting the right entry.  We might
2312                  * also be pointing to a stale entry in the unused part of
2313                  * old_bh so just checking inum and the name isn't enough. */
2314                 struct buffer_head *old_bh2;
2315                 struct ext4_dir_entry_2 *old_de2;
2316
2317                 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2318                 if (old_bh2) {
2319                         retval = ext4_delete_entry(handle, old_dir,
2320                                                    old_de2, old_bh2);
2321                         brelse(old_bh2);
2322                 }
2323         }
2324         if (retval) {
2325                 ext4_warning(old_dir->i_sb, "ext4_rename",
2326                                 "Deleting old file (%lu), %d, error=%d",
2327                                 old_dir->i_ino, old_dir->i_nlink, retval);
2328         }
2329
2330         if (new_inode) {
2331                 drop_nlink(new_inode);
2332                 new_inode->i_ctime = CURRENT_TIME_SEC;
2333         }
2334         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2335         ext4_update_dx_flag(old_dir);
2336         if (dir_bh) {
2337                 BUFFER_TRACE(dir_bh, "get_write_access");
2338                 ext4_journal_get_write_access(handle, dir_bh);
2339                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2340                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2341                 ext4_journal_dirty_metadata(handle, dir_bh);
2342                 drop_nlink(old_dir);
2343                 if (new_inode) {
2344                         drop_nlink(new_inode);
2345                 } else {
2346                         inc_nlink(new_dir);
2347                         ext4_update_dx_flag(new_dir);
2348                         ext4_mark_inode_dirty(handle, new_dir);
2349                 }
2350         }
2351         ext4_mark_inode_dirty(handle, old_dir);
2352         if (new_inode) {
2353                 ext4_mark_inode_dirty(handle, new_inode);
2354                 if (!new_inode->i_nlink)
2355                         ext4_orphan_add(handle, new_inode);
2356         }
2357         retval = 0;
2358
2359 end_rename:
2360         brelse (dir_bh);
2361         brelse (old_bh);
2362         brelse (new_bh);
2363         ext4_journal_stop(handle);
2364         return retval;
2365 }
2366
2367 /*
2368  * directories can handle most operations...
2369  */
2370 const struct inode_operations ext4_dir_inode_operations = {
2371         .create         = ext4_create,
2372         .lookup         = ext4_lookup,
2373         .link           = ext4_link,
2374         .unlink         = ext4_unlink,
2375         .symlink        = ext4_symlink,
2376         .mkdir          = ext4_mkdir,
2377         .rmdir          = ext4_rmdir,
2378         .mknod          = ext4_mknod,
2379         .rename         = ext4_rename,
2380         .setattr        = ext4_setattr,
2381 #ifdef CONFIG_EXT4DEV_FS_XATTR
2382         .setxattr       = generic_setxattr,
2383         .getxattr       = generic_getxattr,
2384         .listxattr      = ext4_listxattr,
2385         .removexattr    = generic_removexattr,
2386 #endif
2387         .permission     = ext4_permission,
2388 };
2389
2390 const struct inode_operations ext4_special_inode_operations = {
2391         .setattr        = ext4_setattr,
2392 #ifdef CONFIG_EXT4DEV_FS_XATTR
2393         .setxattr       = generic_setxattr,
2394         .getxattr       = generic_getxattr,
2395         .listxattr      = ext4_listxattr,
2396         .removexattr    = generic_removexattr,
2397 #endif
2398         .permission     = ext4_permission,
2399 };