Merge master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
[linux-2.6] / fs / ext4 / namei.c
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/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                 ext4fs_dirhash(de->name, de->name_len, hinfo);
556                 if ((hinfo->hash < start_hash) ||
557                     ((hinfo->hash == start_hash) &&
558                      (hinfo->minor_hash < start_minor_hash)))
559                         continue;
560                 if (de->inode == 0)
561                         continue;
562                 if ((err = ext4_htree_store_dirent(dir_file,
563                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
564                         brelse(bh);
565                         return err;
566                 }
567                 count++;
568         }
569         brelse(bh);
570         return count;
571 }
572
573
574 /*
575  * This function fills a red-black tree with information from a
576  * directory.  We start scanning the directory in hash order, starting
577  * at start_hash and start_minor_hash.
578  *
579  * This function returns the number of entries inserted into the tree,
580  * or a negative error code.
581  */
582 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
583                          __u32 start_minor_hash, __u32 *next_hash)
584 {
585         struct dx_hash_info hinfo;
586         struct ext4_dir_entry_2 *de;
587         struct dx_frame frames[2], *frame;
588         struct inode *dir;
589         int block, err;
590         int count = 0;
591         int ret;
592         __u32 hashval;
593
594         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
595                        start_minor_hash));
596         dir = dir_file->f_dentry->d_inode;
597         if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
598                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
599                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
600                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
601                                                start_hash, start_minor_hash);
602                 *next_hash = ~0;
603                 return count;
604         }
605         hinfo.hash = start_hash;
606         hinfo.minor_hash = 0;
607         frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
608         if (!frame)
609                 return err;
610
611         /* Add '.' and '..' from the htree header */
612         if (!start_hash && !start_minor_hash) {
613                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
614                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
615                         goto errout;
616                 count++;
617         }
618         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
619                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
620                 de = ext4_next_entry(de);
621                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
622                         goto errout;
623                 count++;
624         }
625
626         while (1) {
627                 block = dx_get_block(frame->at);
628                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
629                                              start_hash, start_minor_hash);
630                 if (ret < 0) {
631                         err = ret;
632                         goto errout;
633                 }
634                 count += ret;
635                 hashval = ~0;
636                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
637                                             frame, frames, &hashval);
638                 *next_hash = hashval;
639                 if (ret < 0) {
640                         err = ret;
641                         goto errout;
642                 }
643                 /*
644                  * Stop if:  (a) there are no more entries, or
645                  * (b) we have inserted at least one entry and the
646                  * next hash value is not a continuation
647                  */
648                 if ((ret == 0) ||
649                     (count && ((hashval & 1) == 0)))
650                         break;
651         }
652         dx_release(frames);
653         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
654                        count, *next_hash));
655         return count;
656 errout:
657         dx_release(frames);
658         return (err);
659 }
660
661
662 /*
663  * Directory block splitting, compacting
664  */
665
666 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
667                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
668 {
669         int count = 0;
670         char *base = (char *) de;
671         struct dx_hash_info h = *hinfo;
672
673         while ((char *) de < base + size)
674         {
675                 if (de->name_len && de->inode) {
676                         ext4fs_dirhash(de->name, de->name_len, &h);
677                         map_tail--;
678                         map_tail->hash = h.hash;
679                         map_tail->offs = (u32) ((char *) de - base);
680                         count++;
681                         cond_resched();
682                 }
683                 /* XXX: do we need to check rec_len == 0 case? -Chris */
684                 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
685         }
686         return count;
687 }
688
689 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
690 {
691         struct dx_map_entry *p, *q, *top = map + count - 1;
692         int more;
693         /* Combsort until bubble sort doesn't suck */
694         while (count > 2) {
695                 count = count*10/13;
696                 if (count - 9 < 2) /* 9, 10 -> 11 */
697                         count = 11;
698                 for (p = top, q = p - count; q >= map; p--, q--)
699                         if (p->hash < q->hash)
700                                 swap(*p, *q);
701         }
702         /* Garden variety bubble sort */
703         do {
704                 more = 0;
705                 q = top;
706                 while (q-- > map) {
707                         if (q[1].hash >= q[0].hash)
708                                 continue;
709                         swap(*(q+1), *q);
710                         more = 1;
711                 }
712         } while(more);
713 }
714
715 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
716 {
717         struct dx_entry *entries = frame->entries;
718         struct dx_entry *old = frame->at, *new = old + 1;
719         int count = dx_get_count(entries);
720
721         assert(count < dx_get_limit(entries));
722         assert(old < entries + count);
723         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
724         dx_set_hash(new, hash);
725         dx_set_block(new, block);
726         dx_set_count(entries, count + 1);
727 }
728 #endif
729
730
731 static void ext4_update_dx_flag(struct inode *inode)
732 {
733         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
734                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
735                 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
736 }
737
738 /*
739  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
740  *
741  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
742  * `de != NULL' is guaranteed by caller.
743  */
744 static inline int ext4_match (int len, const char * const name,
745                               struct ext4_dir_entry_2 * de)
746 {
747         if (len != de->name_len)
748                 return 0;
749         if (!de->inode)
750                 return 0;
751         return !memcmp(name, de->name, len);
752 }
753
754 /*
755  * Returns 0 if not found, -1 on failure, and 1 on success
756  */
757 static inline int search_dirblock(struct buffer_head * bh,
758                                   struct inode *dir,
759                                   struct dentry *dentry,
760                                   unsigned long offset,
761                                   struct ext4_dir_entry_2 ** res_dir)
762 {
763         struct ext4_dir_entry_2 * de;
764         char * dlimit;
765         int de_len;
766         const char *name = dentry->d_name.name;
767         int namelen = dentry->d_name.len;
768
769         de = (struct ext4_dir_entry_2 *) bh->b_data;
770         dlimit = bh->b_data + dir->i_sb->s_blocksize;
771         while ((char *) de < dlimit) {
772                 /* this code is executed quadratically often */
773                 /* do minimal checking `by hand' */
774
775                 if ((char *) de + namelen <= dlimit &&
776                     ext4_match (namelen, name, de)) {
777                         /* found a match - just to be sure, do a full check */
778                         if (!ext4_check_dir_entry("ext4_find_entry",
779                                                   dir, de, bh, offset))
780                                 return -1;
781                         *res_dir = de;
782                         return 1;
783                 }
784                 /* prevent looping on a bad block */
785                 de_len = le16_to_cpu(de->rec_len);
786                 if (de_len <= 0)
787                         return -1;
788                 offset += de_len;
789                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
790         }
791         return 0;
792 }
793
794
795 /*
796  *      ext4_find_entry()
797  *
798  * finds an entry in the specified directory with the wanted name. It
799  * returns the cache buffer in which the entry was found, and the entry
800  * itself (as a parameter - res_dir). It does NOT read the inode of the
801  * entry - you'll have to do that yourself if you want to.
802  *
803  * The returned buffer_head has ->b_count elevated.  The caller is expected
804  * to brelse() it when appropriate.
805  */
806 static struct buffer_head * ext4_find_entry (struct dentry *dentry,
807                                         struct ext4_dir_entry_2 ** res_dir)
808 {
809         struct super_block * sb;
810         struct buffer_head * bh_use[NAMEI_RA_SIZE];
811         struct buffer_head * bh, *ret = NULL;
812         unsigned long start, block, b;
813         int ra_max = 0;         /* Number of bh's in the readahead
814                                    buffer, bh_use[] */
815         int ra_ptr = 0;         /* Current index into readahead
816                                    buffer */
817         int num = 0;
818         int nblocks, i, err;
819         struct inode *dir = dentry->d_parent->d_inode;
820         int namelen;
821         const u8 *name;
822         unsigned blocksize;
823
824         *res_dir = NULL;
825         sb = dir->i_sb;
826         blocksize = sb->s_blocksize;
827         namelen = dentry->d_name.len;
828         name = dentry->d_name.name;
829         if (namelen > EXT4_NAME_LEN)
830                 return NULL;
831 #ifdef CONFIG_EXT4_INDEX
832         if (is_dx(dir)) {
833                 bh = ext4_dx_find_entry(dentry, res_dir, &err);
834                 /*
835                  * On success, or if the error was file not found,
836                  * return.  Otherwise, fall back to doing a search the
837                  * old fashioned way.
838                  */
839                 if (bh || (err != ERR_BAD_DX_DIR))
840                         return bh;
841                 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
842         }
843 #endif
844         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
845         start = EXT4_I(dir)->i_dir_start_lookup;
846         if (start >= nblocks)
847                 start = 0;
848         block = start;
849 restart:
850         do {
851                 /*
852                  * We deal with the read-ahead logic here.
853                  */
854                 if (ra_ptr >= ra_max) {
855                         /* Refill the readahead buffer */
856                         ra_ptr = 0;
857                         b = block;
858                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
859                                 /*
860                                  * Terminate if we reach the end of the
861                                  * directory and must wrap, or if our
862                                  * search has finished at this block.
863                                  */
864                                 if (b >= nblocks || (num && block == start)) {
865                                         bh_use[ra_max] = NULL;
866                                         break;
867                                 }
868                                 num++;
869                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
870                                 bh_use[ra_max] = bh;
871                                 if (bh)
872                                         ll_rw_block(READ_META, 1, &bh);
873                         }
874                 }
875                 if ((bh = bh_use[ra_ptr++]) == NULL)
876                         goto next;
877                 wait_on_buffer(bh);
878                 if (!buffer_uptodate(bh)) {
879                         /* read error, skip block & hope for the best */
880                         ext4_error(sb, __FUNCTION__, "reading directory #%lu "
881                                    "offset %lu", dir->i_ino, block);
882                         brelse(bh);
883                         goto next;
884                 }
885                 i = search_dirblock(bh, dir, dentry,
886                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
887                 if (i == 1) {
888                         EXT4_I(dir)->i_dir_start_lookup = block;
889                         ret = bh;
890                         goto cleanup_and_exit;
891                 } else {
892                         brelse(bh);
893                         if (i < 0)
894                                 goto cleanup_and_exit;
895                 }
896         next:
897                 if (++block >= nblocks)
898                         block = 0;
899         } while (block != start);
900
901         /*
902          * If the directory has grown while we were searching, then
903          * search the last part of the directory before giving up.
904          */
905         block = nblocks;
906         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
907         if (block < nblocks) {
908                 start = 0;
909                 goto restart;
910         }
911
912 cleanup_and_exit:
913         /* Clean up the read-ahead blocks */
914         for (; ra_ptr < ra_max; ra_ptr++)
915                 brelse (bh_use[ra_ptr]);
916         return ret;
917 }
918
919 #ifdef CONFIG_EXT4_INDEX
920 static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
921                        struct ext4_dir_entry_2 **res_dir, int *err)
922 {
923         struct super_block * sb;
924         struct dx_hash_info     hinfo;
925         u32 hash;
926         struct dx_frame frames[2], *frame;
927         struct ext4_dir_entry_2 *de, *top;
928         struct buffer_head *bh;
929         unsigned long block;
930         int retval;
931         int namelen = dentry->d_name.len;
932         const u8 *name = dentry->d_name.name;
933         struct inode *dir = dentry->d_parent->d_inode;
934
935         sb = dir->i_sb;
936         /* NFS may look up ".." - look at dx_root directory block */
937         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
938                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
939                         return NULL;
940         } else {
941                 frame = frames;
942                 frame->bh = NULL;                       /* for dx_release() */
943                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
944                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
945         }
946         hash = hinfo.hash;
947         do {
948                 block = dx_get_block(frame->at);
949                 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
950                         goto errout;
951                 de = (struct ext4_dir_entry_2 *) bh->b_data;
952                 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
953                                        EXT4_DIR_REC_LEN(0));
954                 for (; de < top; de = ext4_next_entry(de))
955                 if (ext4_match (namelen, name, de)) {
956                         if (!ext4_check_dir_entry("ext4_find_entry",
957                                                   dir, de, bh,
958                                   (block<<EXT4_BLOCK_SIZE_BITS(sb))
959                                           +((char *)de - bh->b_data))) {
960                                 brelse (bh);
961                                 goto errout;
962                         }
963                         *res_dir = de;
964                         dx_release (frames);
965                         return bh;
966                 }
967                 brelse (bh);
968                 /* Check to see if we should continue to search */
969                 retval = ext4_htree_next_block(dir, hash, frame,
970                                                frames, NULL);
971                 if (retval < 0) {
972                         ext4_warning(sb, __FUNCTION__,
973                              "error reading index page in directory #%lu",
974                              dir->i_ino);
975                         *err = retval;
976                         goto errout;
977                 }
978         } while (retval == 1);
979
980         *err = -ENOENT;
981 errout:
982         dxtrace(printk("%s not found\n", name));
983         dx_release (frames);
984         return NULL;
985 }
986 #endif
987
988 static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
989 {
990         struct inode * inode;
991         struct ext4_dir_entry_2 * de;
992         struct buffer_head * bh;
993
994         if (dentry->d_name.len > EXT4_NAME_LEN)
995                 return ERR_PTR(-ENAMETOOLONG);
996
997         bh = ext4_find_entry(dentry, &de);
998         inode = NULL;
999         if (bh) {
1000                 unsigned long ino = le32_to_cpu(de->inode);
1001                 brelse (bh);
1002                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1003                         ext4_error(dir->i_sb, "ext4_lookup",
1004                                    "bad inode number: %lu", ino);
1005                         inode = NULL;
1006                 } else
1007                         inode = iget(dir->i_sb, ino);
1008
1009                 if (!inode)
1010                         return ERR_PTR(-EACCES);
1011         }
1012         return d_splice_alias(inode, dentry);
1013 }
1014
1015
1016 struct dentry *ext4_get_parent(struct dentry *child)
1017 {
1018         unsigned long ino;
1019         struct dentry *parent;
1020         struct inode *inode;
1021         struct dentry dotdot;
1022         struct ext4_dir_entry_2 * de;
1023         struct buffer_head *bh;
1024
1025         dotdot.d_name.name = "..";
1026         dotdot.d_name.len = 2;
1027         dotdot.d_parent = child; /* confusing, isn't it! */
1028
1029         bh = ext4_find_entry(&dotdot, &de);
1030         inode = NULL;
1031         if (!bh)
1032                 return ERR_PTR(-ENOENT);
1033         ino = le32_to_cpu(de->inode);
1034         brelse(bh);
1035
1036         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1037                 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1038                            "bad inode number: %lu", ino);
1039                 inode = NULL;
1040         } else
1041                 inode = iget(child->d_inode->i_sb, ino);
1042
1043         if (!inode)
1044                 return ERR_PTR(-EACCES);
1045
1046         parent = d_alloc_anon(inode);
1047         if (!parent) {
1048                 iput(inode);
1049                 parent = ERR_PTR(-ENOMEM);
1050         }
1051         return parent;
1052 }
1053
1054 #define S_SHIFT 12
1055 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1056         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1057         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1058         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1059         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1060         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1061         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1062         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1063 };
1064
1065 static inline void ext4_set_de_type(struct super_block *sb,
1066                                 struct ext4_dir_entry_2 *de,
1067                                 umode_t mode) {
1068         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1069                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1070 }
1071
1072 #ifdef CONFIG_EXT4_INDEX
1073 static struct ext4_dir_entry_2 *
1074 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1075 {
1076         unsigned rec_len = 0;
1077
1078         while (count--) {
1079                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1080                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1081                 memcpy (to, de, rec_len);
1082                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1083                                 cpu_to_le16(rec_len);
1084                 de->inode = 0;
1085                 map++;
1086                 to += rec_len;
1087         }
1088         return (struct ext4_dir_entry_2 *) (to - rec_len);
1089 }
1090
1091 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1092 {
1093         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1094         unsigned rec_len = 0;
1095
1096         prev = to = de;
1097         while ((char*)de < base + size) {
1098                 next = (struct ext4_dir_entry_2 *) ((char *) de +
1099                                                     le16_to_cpu(de->rec_len));
1100                 if (de->inode && de->name_len) {
1101                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1102                         if (de > to)
1103                                 memmove(to, de, rec_len);
1104                         to->rec_len = cpu_to_le16(rec_len);
1105                         prev = to;
1106                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1107                 }
1108                 de = next;
1109         }
1110         return prev;
1111 }
1112
1113 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1114                         struct buffer_head **bh,struct dx_frame *frame,
1115                         struct dx_hash_info *hinfo, int *error)
1116 {
1117         unsigned blocksize = dir->i_sb->s_blocksize;
1118         unsigned count, continued;
1119         struct buffer_head *bh2;
1120         u32 newblock;
1121         u32 hash2;
1122         struct dx_map_entry *map;
1123         char *data1 = (*bh)->b_data, *data2;
1124         unsigned split;
1125         struct ext4_dir_entry_2 *de = NULL, *de2;
1126         int     err;
1127
1128         bh2 = ext4_append (handle, dir, &newblock, error);
1129         if (!(bh2)) {
1130                 brelse(*bh);
1131                 *bh = NULL;
1132                 goto errout;
1133         }
1134
1135         BUFFER_TRACE(*bh, "get_write_access");
1136         err = ext4_journal_get_write_access(handle, *bh);
1137         if (err) {
1138         journal_error:
1139                 brelse(*bh);
1140                 brelse(bh2);
1141                 *bh = NULL;
1142                 ext4_std_error(dir->i_sb, err);
1143                 goto errout;
1144         }
1145         BUFFER_TRACE(frame->bh, "get_write_access");
1146         err = ext4_journal_get_write_access(handle, frame->bh);
1147         if (err)
1148                 goto journal_error;
1149
1150         data2 = bh2->b_data;
1151
1152         /* create map in the end of data2 block */
1153         map = (struct dx_map_entry *) (data2 + blocksize);
1154         count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
1155                              blocksize, hinfo, map);
1156         map -= count;
1157         split = count/2; // need to adjust to actual middle
1158         dx_sort_map (map, count);
1159         hash2 = map[split].hash;
1160         continued = hash2 == map[split - 1].hash;
1161         dxtrace(printk("Split block %i at %x, %i/%i\n",
1162                 dx_get_block(frame->at), hash2, split, count-split));
1163
1164         /* Fancy dance to stay within two buffers */
1165         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1166         de = dx_pack_dirents(data1,blocksize);
1167         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1168         de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1169         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1170         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1171
1172         /* Which block gets the new entry? */
1173         if (hinfo->hash >= hash2)
1174         {
1175                 swap(*bh, bh2);
1176                 de = de2;
1177         }
1178         dx_insert_block (frame, hash2 + continued, newblock);
1179         err = ext4_journal_dirty_metadata (handle, bh2);
1180         if (err)
1181                 goto journal_error;
1182         err = ext4_journal_dirty_metadata (handle, frame->bh);
1183         if (err)
1184                 goto journal_error;
1185         brelse (bh2);
1186         dxtrace(dx_show_index ("frame", frame->entries));
1187 errout:
1188         return de;
1189 }
1190 #endif
1191
1192
1193 /*
1194  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1195  * it points to a directory entry which is guaranteed to be large
1196  * enough for new directory entry.  If de is NULL, then
1197  * add_dirent_to_buf will attempt search the directory block for
1198  * space.  It will return -ENOSPC if no space is available, and -EIO
1199  * and -EEXIST if directory entry already exists.
1200  *
1201  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1202  * all other cases bh is released.
1203  */
1204 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1205                              struct inode *inode, struct ext4_dir_entry_2 *de,
1206                              struct buffer_head * bh)
1207 {
1208         struct inode    *dir = dentry->d_parent->d_inode;
1209         const char      *name = dentry->d_name.name;
1210         int             namelen = dentry->d_name.len;
1211         unsigned long   offset = 0;
1212         unsigned short  reclen;
1213         int             nlen, rlen, err;
1214         char            *top;
1215
1216         reclen = EXT4_DIR_REC_LEN(namelen);
1217         if (!de) {
1218                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1219                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1220                 while ((char *) de <= top) {
1221                         if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1222                                                   bh, offset)) {
1223                                 brelse (bh);
1224                                 return -EIO;
1225                         }
1226                         if (ext4_match (namelen, name, de)) {
1227                                 brelse (bh);
1228                                 return -EEXIST;
1229                         }
1230                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1231                         rlen = le16_to_cpu(de->rec_len);
1232                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1233                                 break;
1234                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1235                         offset += rlen;
1236                 }
1237                 if ((char *) de > top)
1238                         return -ENOSPC;
1239         }
1240         BUFFER_TRACE(bh, "get_write_access");
1241         err = ext4_journal_get_write_access(handle, bh);
1242         if (err) {
1243                 ext4_std_error(dir->i_sb, err);
1244                 brelse(bh);
1245                 return err;
1246         }
1247
1248         /* By now the buffer is marked for journaling */
1249         nlen = EXT4_DIR_REC_LEN(de->name_len);
1250         rlen = le16_to_cpu(de->rec_len);
1251         if (de->inode) {
1252                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1253                 de1->rec_len = cpu_to_le16(rlen - nlen);
1254                 de->rec_len = cpu_to_le16(nlen);
1255                 de = de1;
1256         }
1257         de->file_type = EXT4_FT_UNKNOWN;
1258         if (inode) {
1259                 de->inode = cpu_to_le32(inode->i_ino);
1260                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1261         } else
1262                 de->inode = 0;
1263         de->name_len = namelen;
1264         memcpy (de->name, name, namelen);
1265         /*
1266          * XXX shouldn't update any times until successful
1267          * completion of syscall, but too many callers depend
1268          * on this.
1269          *
1270          * XXX similarly, too many callers depend on
1271          * ext4_new_inode() setting the times, but error
1272          * recovery deletes the inode, so the worst that can
1273          * happen is that the times are slightly out of date
1274          * and/or different from the directory change time.
1275          */
1276         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1277         ext4_update_dx_flag(dir);
1278         dir->i_version++;
1279         ext4_mark_inode_dirty(handle, dir);
1280         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1281         err = ext4_journal_dirty_metadata(handle, bh);
1282         if (err)
1283                 ext4_std_error(dir->i_sb, err);
1284         brelse(bh);
1285         return 0;
1286 }
1287
1288 #ifdef CONFIG_EXT4_INDEX
1289 /*
1290  * This converts a one block unindexed directory to a 3 block indexed
1291  * directory, and adds the dentry to the indexed directory.
1292  */
1293 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1294                             struct inode *inode, struct buffer_head *bh)
1295 {
1296         struct inode    *dir = dentry->d_parent->d_inode;
1297         const char      *name = dentry->d_name.name;
1298         int             namelen = dentry->d_name.len;
1299         struct buffer_head *bh2;
1300         struct dx_root  *root;
1301         struct dx_frame frames[2], *frame;
1302         struct dx_entry *entries;
1303         struct ext4_dir_entry_2 *de, *de2;
1304         char            *data1, *top;
1305         unsigned        len;
1306         int             retval;
1307         unsigned        blocksize;
1308         struct dx_hash_info hinfo;
1309         u32             block;
1310         struct fake_dirent *fde;
1311
1312         blocksize =  dir->i_sb->s_blocksize;
1313         dxtrace(printk("Creating index\n"));
1314         retval = ext4_journal_get_write_access(handle, bh);
1315         if (retval) {
1316                 ext4_std_error(dir->i_sb, retval);
1317                 brelse(bh);
1318                 return retval;
1319         }
1320         root = (struct dx_root *) bh->b_data;
1321
1322         bh2 = ext4_append (handle, dir, &block, &retval);
1323         if (!(bh2)) {
1324                 brelse(bh);
1325                 return retval;
1326         }
1327         EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1328         data1 = bh2->b_data;
1329
1330         /* The 0th block becomes the root, move the dirents out */
1331         fde = &root->dotdot;
1332         de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1333         len = ((char *) root) + blocksize - (char *) de;
1334         memcpy (data1, de, len);
1335         de = (struct ext4_dir_entry_2 *) data1;
1336         top = data1 + len;
1337         while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1338                 de = de2;
1339         de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1340         /* Initialize the root; the dot dirents already exist */
1341         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1342         de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
1343         memset (&root->info, 0, sizeof(root->info));
1344         root->info.info_length = sizeof(root->info);
1345         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1346         entries = root->entries;
1347         dx_set_block (entries, 1);
1348         dx_set_count (entries, 1);
1349         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1350
1351         /* Initialize as for dx_probe */
1352         hinfo.hash_version = root->info.hash_version;
1353         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1354         ext4fs_dirhash(name, namelen, &hinfo);
1355         frame = frames;
1356         frame->entries = entries;
1357         frame->at = entries;
1358         frame->bh = bh;
1359         bh = bh2;
1360         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1361         dx_release (frames);
1362         if (!(de))
1363                 return retval;
1364
1365         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1366 }
1367 #endif
1368
1369 /*
1370  *      ext4_add_entry()
1371  *
1372  * adds a file entry to the specified directory, using the same
1373  * semantics as ext4_find_entry(). It returns NULL if it failed.
1374  *
1375  * NOTE!! The inode part of 'de' is left at 0 - which means you
1376  * may not sleep between calling this and putting something into
1377  * the entry, as someone else might have used it while you slept.
1378  */
1379 static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
1380         struct inode *inode)
1381 {
1382         struct inode *dir = dentry->d_parent->d_inode;
1383         unsigned long offset;
1384         struct buffer_head * bh;
1385         struct ext4_dir_entry_2 *de;
1386         struct super_block * sb;
1387         int     retval;
1388 #ifdef CONFIG_EXT4_INDEX
1389         int     dx_fallback=0;
1390 #endif
1391         unsigned blocksize;
1392         u32 block, blocks;
1393
1394         sb = dir->i_sb;
1395         blocksize = sb->s_blocksize;
1396         if (!dentry->d_name.len)
1397                 return -EINVAL;
1398 #ifdef CONFIG_EXT4_INDEX
1399         if (is_dx(dir)) {
1400                 retval = ext4_dx_add_entry(handle, dentry, inode);
1401                 if (!retval || (retval != ERR_BAD_DX_DIR))
1402                         return retval;
1403                 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1404                 dx_fallback++;
1405                 ext4_mark_inode_dirty(handle, dir);
1406         }
1407 #endif
1408         blocks = dir->i_size >> sb->s_blocksize_bits;
1409         for (block = 0, offset = 0; block < blocks; block++) {
1410                 bh = ext4_bread(handle, dir, block, 0, &retval);
1411                 if(!bh)
1412                         return retval;
1413                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1414                 if (retval != -ENOSPC)
1415                         return retval;
1416
1417 #ifdef CONFIG_EXT4_INDEX
1418                 if (blocks == 1 && !dx_fallback &&
1419                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1420                         return make_indexed_dir(handle, dentry, inode, bh);
1421 #endif
1422                 brelse(bh);
1423         }
1424         bh = ext4_append(handle, dir, &block, &retval);
1425         if (!bh)
1426                 return retval;
1427         de = (struct ext4_dir_entry_2 *) bh->b_data;
1428         de->inode = 0;
1429         de->rec_len = cpu_to_le16(blocksize);
1430         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1431 }
1432
1433 #ifdef CONFIG_EXT4_INDEX
1434 /*
1435  * Returns 0 for success, or a negative error value
1436  */
1437 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1438                              struct inode *inode)
1439 {
1440         struct dx_frame frames[2], *frame;
1441         struct dx_entry *entries, *at;
1442         struct dx_hash_info hinfo;
1443         struct buffer_head * bh;
1444         struct inode *dir = dentry->d_parent->d_inode;
1445         struct super_block * sb = dir->i_sb;
1446         struct ext4_dir_entry_2 *de;
1447         int err;
1448
1449         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1450         if (!frame)
1451                 return err;
1452         entries = frame->entries;
1453         at = frame->at;
1454
1455         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1456                 goto cleanup;
1457
1458         BUFFER_TRACE(bh, "get_write_access");
1459         err = ext4_journal_get_write_access(handle, bh);
1460         if (err)
1461                 goto journal_error;
1462
1463         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1464         if (err != -ENOSPC) {
1465                 bh = NULL;
1466                 goto cleanup;
1467         }
1468
1469         /* Block full, should compress but for now just split */
1470         dxtrace(printk("using %u of %u node entries\n",
1471                        dx_get_count(entries), dx_get_limit(entries)));
1472         /* Need to split index? */
1473         if (dx_get_count(entries) == dx_get_limit(entries)) {
1474                 u32 newblock;
1475                 unsigned icount = dx_get_count(entries);
1476                 int levels = frame - frames;
1477                 struct dx_entry *entries2;
1478                 struct dx_node *node2;
1479                 struct buffer_head *bh2;
1480
1481                 if (levels && (dx_get_count(frames->entries) ==
1482                                dx_get_limit(frames->entries))) {
1483                         ext4_warning(sb, __FUNCTION__,
1484                                      "Directory index full!");
1485                         err = -ENOSPC;
1486                         goto cleanup;
1487                 }
1488                 bh2 = ext4_append (handle, dir, &newblock, &err);
1489                 if (!(bh2))
1490                         goto cleanup;
1491                 node2 = (struct dx_node *)(bh2->b_data);
1492                 entries2 = node2->entries;
1493                 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1494                 node2->fake.inode = 0;
1495                 BUFFER_TRACE(frame->bh, "get_write_access");
1496                 err = ext4_journal_get_write_access(handle, frame->bh);
1497                 if (err)
1498                         goto journal_error;
1499                 if (levels) {
1500                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1501                         unsigned hash2 = dx_get_hash(entries + icount1);
1502                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1503
1504                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1505                         err = ext4_journal_get_write_access(handle,
1506                                                              frames[0].bh);
1507                         if (err)
1508                                 goto journal_error;
1509
1510                         memcpy ((char *) entries2, (char *) (entries + icount1),
1511                                 icount2 * sizeof(struct dx_entry));
1512                         dx_set_count (entries, icount1);
1513                         dx_set_count (entries2, icount2);
1514                         dx_set_limit (entries2, dx_node_limit(dir));
1515
1516                         /* Which index block gets the new entry? */
1517                         if (at - entries >= icount1) {
1518                                 frame->at = at = at - entries - icount1 + entries2;
1519                                 frame->entries = entries = entries2;
1520                                 swap(frame->bh, bh2);
1521                         }
1522                         dx_insert_block (frames + 0, hash2, newblock);
1523                         dxtrace(dx_show_index ("node", frames[1].entries));
1524                         dxtrace(dx_show_index ("node",
1525                                ((struct dx_node *) bh2->b_data)->entries));
1526                         err = ext4_journal_dirty_metadata(handle, bh2);
1527                         if (err)
1528                                 goto journal_error;
1529                         brelse (bh2);
1530                 } else {
1531                         dxtrace(printk("Creating second level index...\n"));
1532                         memcpy((char *) entries2, (char *) entries,
1533                                icount * sizeof(struct dx_entry));
1534                         dx_set_limit(entries2, dx_node_limit(dir));
1535
1536                         /* Set up root */
1537                         dx_set_count(entries, 1);
1538                         dx_set_block(entries + 0, newblock);
1539                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1540
1541                         /* Add new access path frame */
1542                         frame = frames + 1;
1543                         frame->at = at = at - entries + entries2;
1544                         frame->entries = entries = entries2;
1545                         frame->bh = bh2;
1546                         err = ext4_journal_get_write_access(handle,
1547                                                              frame->bh);
1548                         if (err)
1549                                 goto journal_error;
1550                 }
1551                 ext4_journal_dirty_metadata(handle, frames[0].bh);
1552         }
1553         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1554         if (!de)
1555                 goto cleanup;
1556         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1557         bh = NULL;
1558         goto cleanup;
1559
1560 journal_error:
1561         ext4_std_error(dir->i_sb, err);
1562 cleanup:
1563         if (bh)
1564                 brelse(bh);
1565         dx_release(frames);
1566         return err;
1567 }
1568 #endif
1569
1570 /*
1571  * ext4_delete_entry deletes a directory entry by merging it with the
1572  * previous entry
1573  */
1574 static int ext4_delete_entry (handle_t *handle,
1575                               struct inode * dir,
1576                               struct ext4_dir_entry_2 * de_del,
1577                               struct buffer_head * bh)
1578 {
1579         struct ext4_dir_entry_2 * de, * pde;
1580         int i;
1581
1582         i = 0;
1583         pde = NULL;
1584         de = (struct ext4_dir_entry_2 *) bh->b_data;
1585         while (i < bh->b_size) {
1586                 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1587                         return -EIO;
1588                 if (de == de_del)  {
1589                         BUFFER_TRACE(bh, "get_write_access");
1590                         ext4_journal_get_write_access(handle, bh);
1591                         if (pde)
1592                                 pde->rec_len =
1593                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1594                                                     le16_to_cpu(de->rec_len));
1595                         else
1596                                 de->inode = 0;
1597                         dir->i_version++;
1598                         BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1599                         ext4_journal_dirty_metadata(handle, bh);
1600                         return 0;
1601                 }
1602                 i += le16_to_cpu(de->rec_len);
1603                 pde = de;
1604                 de = (struct ext4_dir_entry_2 *)
1605                         ((char *) de + le16_to_cpu(de->rec_len));
1606         }
1607         return -ENOENT;
1608 }
1609
1610 /*
1611  * ext4_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1612  * do not perform it in these functions.  We perform it at the call site,
1613  * if it is needed.
1614  */
1615 static inline void ext4_inc_count(handle_t *handle, struct inode *inode)
1616 {
1617         inc_nlink(inode);
1618 }
1619
1620 static inline void ext4_dec_count(handle_t *handle, struct inode *inode)
1621 {
1622         drop_nlink(inode);
1623 }
1624
1625 static int ext4_add_nondir(handle_t *handle,
1626                 struct dentry *dentry, struct inode *inode)
1627 {
1628         int err = ext4_add_entry(handle, dentry, inode);
1629         if (!err) {
1630                 ext4_mark_inode_dirty(handle, inode);
1631                 d_instantiate(dentry, inode);
1632                 return 0;
1633         }
1634         ext4_dec_count(handle, inode);
1635         iput(inode);
1636         return err;
1637 }
1638
1639 /*
1640  * By the time this is called, we already have created
1641  * the directory cache entry for the new file, but it
1642  * is so far negative - it has no inode.
1643  *
1644  * If the create succeeds, we fill in the inode information
1645  * with d_instantiate().
1646  */
1647 static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
1648                 struct nameidata *nd)
1649 {
1650         handle_t *handle;
1651         struct inode * inode;
1652         int err, retries = 0;
1653
1654 retry:
1655         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1656                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1657                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1658         if (IS_ERR(handle))
1659                 return PTR_ERR(handle);
1660
1661         if (IS_DIRSYNC(dir))
1662                 handle->h_sync = 1;
1663
1664         inode = ext4_new_inode (handle, dir, mode);
1665         err = PTR_ERR(inode);
1666         if (!IS_ERR(inode)) {
1667                 inode->i_op = &ext4_file_inode_operations;
1668                 inode->i_fop = &ext4_file_operations;
1669                 ext4_set_aops(inode);
1670                 err = ext4_add_nondir(handle, dentry, inode);
1671         }
1672         ext4_journal_stop(handle);
1673         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1674                 goto retry;
1675         return err;
1676 }
1677
1678 static int ext4_mknod (struct inode * dir, struct dentry *dentry,
1679                         int mode, dev_t rdev)
1680 {
1681         handle_t *handle;
1682         struct inode *inode;
1683         int err, retries = 0;
1684
1685         if (!new_valid_dev(rdev))
1686                 return -EINVAL;
1687
1688 retry:
1689         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1690                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1691                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1692         if (IS_ERR(handle))
1693                 return PTR_ERR(handle);
1694
1695         if (IS_DIRSYNC(dir))
1696                 handle->h_sync = 1;
1697
1698         inode = ext4_new_inode (handle, dir, mode);
1699         err = PTR_ERR(inode);
1700         if (!IS_ERR(inode)) {
1701                 init_special_inode(inode, inode->i_mode, rdev);
1702 #ifdef CONFIG_EXT4DEV_FS_XATTR
1703                 inode->i_op = &ext4_special_inode_operations;
1704 #endif
1705                 err = ext4_add_nondir(handle, dentry, inode);
1706         }
1707         ext4_journal_stop(handle);
1708         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1709                 goto retry;
1710         return err;
1711 }
1712
1713 static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1714 {
1715         handle_t *handle;
1716         struct inode * inode;
1717         struct buffer_head * dir_block;
1718         struct ext4_dir_entry_2 * de;
1719         int err, retries = 0;
1720
1721         if (dir->i_nlink >= EXT4_LINK_MAX)
1722                 return -EMLINK;
1723
1724 retry:
1725         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1726                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1727                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1728         if (IS_ERR(handle))
1729                 return PTR_ERR(handle);
1730
1731         if (IS_DIRSYNC(dir))
1732                 handle->h_sync = 1;
1733
1734         inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
1735         err = PTR_ERR(inode);
1736         if (IS_ERR(inode))
1737                 goto out_stop;
1738
1739         inode->i_op = &ext4_dir_inode_operations;
1740         inode->i_fop = &ext4_dir_operations;
1741         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1742         dir_block = ext4_bread (handle, inode, 0, 1, &err);
1743         if (!dir_block) {
1744                 drop_nlink(inode); /* is this nlink == 0? */
1745                 ext4_mark_inode_dirty(handle, inode);
1746                 iput (inode);
1747                 goto out_stop;
1748         }
1749         BUFFER_TRACE(dir_block, "get_write_access");
1750         ext4_journal_get_write_access(handle, dir_block);
1751         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1752         de->inode = cpu_to_le32(inode->i_ino);
1753         de->name_len = 1;
1754         de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
1755         strcpy (de->name, ".");
1756         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1757         de = (struct ext4_dir_entry_2 *)
1758                         ((char *) de + le16_to_cpu(de->rec_len));
1759         de->inode = cpu_to_le32(dir->i_ino);
1760         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
1761         de->name_len = 2;
1762         strcpy (de->name, "..");
1763         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1764         inode->i_nlink = 2;
1765         BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1766         ext4_journal_dirty_metadata(handle, dir_block);
1767         brelse (dir_block);
1768         ext4_mark_inode_dirty(handle, inode);
1769         err = ext4_add_entry (handle, dentry, inode);
1770         if (err) {
1771                 inode->i_nlink = 0;
1772                 ext4_mark_inode_dirty(handle, inode);
1773                 iput (inode);
1774                 goto out_stop;
1775         }
1776         inc_nlink(dir);
1777         ext4_update_dx_flag(dir);
1778         ext4_mark_inode_dirty(handle, dir);
1779         d_instantiate(dentry, inode);
1780 out_stop:
1781         ext4_journal_stop(handle);
1782         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1783                 goto retry;
1784         return err;
1785 }
1786
1787 /*
1788  * routine to check that the specified directory is empty (for rmdir)
1789  */
1790 static int empty_dir (struct inode * inode)
1791 {
1792         unsigned long offset;
1793         struct buffer_head * bh;
1794         struct ext4_dir_entry_2 * de, * de1;
1795         struct super_block * sb;
1796         int err = 0;
1797
1798         sb = inode->i_sb;
1799         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1800             !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
1801                 if (err)
1802                         ext4_error(inode->i_sb, __FUNCTION__,
1803                                    "error %d reading directory #%lu offset 0",
1804                                    err, inode->i_ino);
1805                 else
1806                         ext4_warning(inode->i_sb, __FUNCTION__,
1807                                      "bad directory (dir #%lu) - no data block",
1808                                      inode->i_ino);
1809                 return 1;
1810         }
1811         de = (struct ext4_dir_entry_2 *) bh->b_data;
1812         de1 = (struct ext4_dir_entry_2 *)
1813                         ((char *) de + le16_to_cpu(de->rec_len));
1814         if (le32_to_cpu(de->inode) != inode->i_ino ||
1815                         !le32_to_cpu(de1->inode) ||
1816                         strcmp (".", de->name) ||
1817                         strcmp ("..", de1->name)) {
1818                 ext4_warning (inode->i_sb, "empty_dir",
1819                               "bad directory (dir #%lu) - no `.' or `..'",
1820                               inode->i_ino);
1821                 brelse (bh);
1822                 return 1;
1823         }
1824         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1825         de = (struct ext4_dir_entry_2 *)
1826                         ((char *) de1 + le16_to_cpu(de1->rec_len));
1827         while (offset < inode->i_size ) {
1828                 if (!bh ||
1829                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1830                         err = 0;
1831                         brelse (bh);
1832                         bh = ext4_bread (NULL, inode,
1833                                 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1834                         if (!bh) {
1835                                 if (err)
1836                                         ext4_error(sb, __FUNCTION__,
1837                                                    "error %d reading directory"
1838                                                    " #%lu offset %lu",
1839                                                    err, inode->i_ino, offset);
1840                                 offset += sb->s_blocksize;
1841                                 continue;
1842                         }
1843                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1844                 }
1845                 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1846                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1847                                                          sb->s_blocksize);
1848                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1849                         continue;
1850                 }
1851                 if (le32_to_cpu(de->inode)) {
1852                         brelse (bh);
1853                         return 0;
1854                 }
1855                 offset += le16_to_cpu(de->rec_len);
1856                 de = (struct ext4_dir_entry_2 *)
1857                                 ((char *) de + le16_to_cpu(de->rec_len));
1858         }
1859         brelse (bh);
1860         return 1;
1861 }
1862
1863 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1864  * such inodes, starting at the superblock, in case we crash before the
1865  * file is closed/deleted, or in case the inode truncate spans multiple
1866  * transactions and the last transaction is not recovered after a crash.
1867  *
1868  * At filesystem recovery time, we walk this list deleting unlinked
1869  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1870  */
1871 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1872 {
1873         struct super_block *sb = inode->i_sb;
1874         struct ext4_iloc iloc;
1875         int err = 0, rc;
1876
1877         lock_super(sb);
1878         if (!list_empty(&EXT4_I(inode)->i_orphan))
1879                 goto out_unlock;
1880
1881         /* Orphan handling is only valid for files with data blocks
1882          * being truncated, or files being unlinked. */
1883
1884         /* @@@ FIXME: Observation from aviro:
1885          * I think I can trigger J_ASSERT in ext4_orphan_add().  We block
1886          * here (on lock_super()), so race with ext4_link() which might bump
1887          * ->i_nlink. For, say it, character device. Not a regular file,
1888          * not a directory, not a symlink and ->i_nlink > 0.
1889          */
1890         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1891                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1892
1893         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1894         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1895         if (err)
1896                 goto out_unlock;
1897
1898         err = ext4_reserve_inode_write(handle, inode, &iloc);
1899         if (err)
1900                 goto out_unlock;
1901
1902         /* Insert this inode at the head of the on-disk orphan list... */
1903         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1904         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1905         err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1906         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1907         if (!err)
1908                 err = rc;
1909
1910         /* Only add to the head of the in-memory list if all the
1911          * previous operations succeeded.  If the orphan_add is going to
1912          * fail (possibly taking the journal offline), we can't risk
1913          * leaving the inode on the orphan list: stray orphan-list
1914          * entries can cause panics at unmount time.
1915          *
1916          * This is safe: on error we're going to ignore the orphan list
1917          * anyway on the next recovery. */
1918         if (!err)
1919                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1920
1921         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1922         jbd_debug(4, "orphan inode %lu will point to %d\n",
1923                         inode->i_ino, NEXT_ORPHAN(inode));
1924 out_unlock:
1925         unlock_super(sb);
1926         ext4_std_error(inode->i_sb, err);
1927         return err;
1928 }
1929
1930 /*
1931  * ext4_orphan_del() removes an unlinked or truncated inode from the list
1932  * of such inodes stored on disk, because it is finally being cleaned up.
1933  */
1934 int ext4_orphan_del(handle_t *handle, struct inode *inode)
1935 {
1936         struct list_head *prev;
1937         struct ext4_inode_info *ei = EXT4_I(inode);
1938         struct ext4_sb_info *sbi;
1939         unsigned long ino_next;
1940         struct ext4_iloc iloc;
1941         int err = 0;
1942
1943         lock_super(inode->i_sb);
1944         if (list_empty(&ei->i_orphan)) {
1945                 unlock_super(inode->i_sb);
1946                 return 0;
1947         }
1948
1949         ino_next = NEXT_ORPHAN(inode);
1950         prev = ei->i_orphan.prev;
1951         sbi = EXT4_SB(inode->i_sb);
1952
1953         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1954
1955         list_del_init(&ei->i_orphan);
1956
1957         /* If we're on an error path, we may not have a valid
1958          * transaction handle with which to update the orphan list on
1959          * disk, but we still need to remove the inode from the linked
1960          * list in memory. */
1961         if (!handle)
1962                 goto out;
1963
1964         err = ext4_reserve_inode_write(handle, inode, &iloc);
1965         if (err)
1966                 goto out_err;
1967
1968         if (prev == &sbi->s_orphan) {
1969                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1970                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1971                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1972                 if (err)
1973                         goto out_brelse;
1974                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1975                 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
1976         } else {
1977                 struct ext4_iloc iloc2;
1978                 struct inode *i_prev =
1979                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
1980
1981                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1982                           i_prev->i_ino, ino_next);
1983                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
1984                 if (err)
1985                         goto out_brelse;
1986                 NEXT_ORPHAN(i_prev) = ino_next;
1987                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
1988         }
1989         if (err)
1990                 goto out_brelse;
1991         NEXT_ORPHAN(inode) = 0;
1992         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
1993
1994 out_err:
1995         ext4_std_error(inode->i_sb, err);
1996 out:
1997         unlock_super(inode->i_sb);
1998         return err;
1999
2000 out_brelse:
2001         brelse(iloc.bh);
2002         goto out_err;
2003 }
2004
2005 static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
2006 {
2007         int retval;
2008         struct inode * inode;
2009         struct buffer_head * bh;
2010         struct ext4_dir_entry_2 * de;
2011         handle_t *handle;
2012
2013         /* Initialize quotas before so that eventual writes go in
2014          * separate transaction */
2015         DQUOT_INIT(dentry->d_inode);
2016         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2017         if (IS_ERR(handle))
2018                 return PTR_ERR(handle);
2019
2020         retval = -ENOENT;
2021         bh = ext4_find_entry (dentry, &de);
2022         if (!bh)
2023                 goto end_rmdir;
2024
2025         if (IS_DIRSYNC(dir))
2026                 handle->h_sync = 1;
2027
2028         inode = dentry->d_inode;
2029
2030         retval = -EIO;
2031         if (le32_to_cpu(de->inode) != inode->i_ino)
2032                 goto end_rmdir;
2033
2034         retval = -ENOTEMPTY;
2035         if (!empty_dir (inode))
2036                 goto end_rmdir;
2037
2038         retval = ext4_delete_entry(handle, dir, de, bh);
2039         if (retval)
2040                 goto end_rmdir;
2041         if (inode->i_nlink != 2)
2042                 ext4_warning (inode->i_sb, "ext4_rmdir",
2043                               "empty directory has nlink!=2 (%d)",
2044                               inode->i_nlink);
2045         inode->i_version++;
2046         clear_nlink(inode);
2047         /* There's no need to set i_disksize: the fact that i_nlink is
2048          * zero will ensure that the right thing happens during any
2049          * recovery. */
2050         inode->i_size = 0;
2051         ext4_orphan_add(handle, inode);
2052         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2053         ext4_mark_inode_dirty(handle, inode);
2054         drop_nlink(dir);
2055         ext4_update_dx_flag(dir);
2056         ext4_mark_inode_dirty(handle, dir);
2057
2058 end_rmdir:
2059         ext4_journal_stop(handle);
2060         brelse (bh);
2061         return retval;
2062 }
2063
2064 static int ext4_unlink(struct inode * dir, struct dentry *dentry)
2065 {
2066         int retval;
2067         struct inode * inode;
2068         struct buffer_head * bh;
2069         struct ext4_dir_entry_2 * de;
2070         handle_t *handle;
2071
2072         /* Initialize quotas before so that eventual writes go
2073          * in separate transaction */
2074         DQUOT_INIT(dentry->d_inode);
2075         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2076         if (IS_ERR(handle))
2077                 return PTR_ERR(handle);
2078
2079         if (IS_DIRSYNC(dir))
2080                 handle->h_sync = 1;
2081
2082         retval = -ENOENT;
2083         bh = ext4_find_entry (dentry, &de);
2084         if (!bh)
2085                 goto end_unlink;
2086
2087         inode = dentry->d_inode;
2088
2089         retval = -EIO;
2090         if (le32_to_cpu(de->inode) != inode->i_ino)
2091                 goto end_unlink;
2092
2093         if (!inode->i_nlink) {
2094                 ext4_warning (inode->i_sb, "ext4_unlink",
2095                               "Deleting nonexistent file (%lu), %d",
2096                               inode->i_ino, inode->i_nlink);
2097                 inode->i_nlink = 1;
2098         }
2099         retval = ext4_delete_entry(handle, dir, de, bh);
2100         if (retval)
2101                 goto end_unlink;
2102         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2103         ext4_update_dx_flag(dir);
2104         ext4_mark_inode_dirty(handle, dir);
2105         drop_nlink(inode);
2106         if (!inode->i_nlink)
2107                 ext4_orphan_add(handle, inode);
2108         inode->i_ctime = dir->i_ctime;
2109         ext4_mark_inode_dirty(handle, inode);
2110         retval = 0;
2111
2112 end_unlink:
2113         ext4_journal_stop(handle);
2114         brelse (bh);
2115         return retval;
2116 }
2117
2118 static int ext4_symlink (struct inode * dir,
2119                 struct dentry *dentry, const char * symname)
2120 {
2121         handle_t *handle;
2122         struct inode * inode;
2123         int l, err, retries = 0;
2124
2125         l = strlen(symname)+1;
2126         if (l > dir->i_sb->s_blocksize)
2127                 return -ENAMETOOLONG;
2128
2129 retry:
2130         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2131                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2132                                         2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2133         if (IS_ERR(handle))
2134                 return PTR_ERR(handle);
2135
2136         if (IS_DIRSYNC(dir))
2137                 handle->h_sync = 1;
2138
2139         inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2140         err = PTR_ERR(inode);
2141         if (IS_ERR(inode))
2142                 goto out_stop;
2143
2144         if (l > sizeof (EXT4_I(inode)->i_data)) {
2145                 inode->i_op = &ext4_symlink_inode_operations;
2146                 ext4_set_aops(inode);
2147                 /*
2148                  * page_symlink() calls into ext4_prepare/commit_write.
2149                  * We have a transaction open.  All is sweetness.  It also sets
2150                  * i_size in generic_commit_write().
2151                  */
2152                 err = __page_symlink(inode, symname, l,
2153                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2154                 if (err) {
2155                         ext4_dec_count(handle, inode);
2156                         ext4_mark_inode_dirty(handle, inode);
2157                         iput (inode);
2158                         goto out_stop;
2159                 }
2160         } else {
2161                 inode->i_op = &ext4_fast_symlink_inode_operations;
2162                 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
2163                 inode->i_size = l-1;
2164         }
2165         EXT4_I(inode)->i_disksize = inode->i_size;
2166         err = ext4_add_nondir(handle, dentry, inode);
2167 out_stop:
2168         ext4_journal_stop(handle);
2169         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2170                 goto retry;
2171         return err;
2172 }
2173
2174 static int ext4_link (struct dentry * old_dentry,
2175                 struct inode * dir, struct dentry *dentry)
2176 {
2177         handle_t *handle;
2178         struct inode *inode = old_dentry->d_inode;
2179         int err, retries = 0;
2180
2181         if (inode->i_nlink >= EXT4_LINK_MAX)
2182                 return -EMLINK;
2183
2184 retry:
2185         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2186                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2187         if (IS_ERR(handle))
2188                 return PTR_ERR(handle);
2189
2190         if (IS_DIRSYNC(dir))
2191                 handle->h_sync = 1;
2192
2193         inode->i_ctime = CURRENT_TIME_SEC;
2194         ext4_inc_count(handle, inode);
2195         atomic_inc(&inode->i_count);
2196
2197         err = ext4_add_nondir(handle, dentry, inode);
2198         ext4_journal_stop(handle);
2199         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2200                 goto retry;
2201         return err;
2202 }
2203
2204 #define PARENT_INO(buffer) \
2205         ((struct ext4_dir_entry_2 *) ((char *) buffer + \
2206         le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
2207
2208 /*
2209  * Anybody can rename anything with this: the permission checks are left to the
2210  * higher-level routines.
2211  */
2212 static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
2213                            struct inode * new_dir,struct dentry *new_dentry)
2214 {
2215         handle_t *handle;
2216         struct inode * old_inode, * new_inode;
2217         struct buffer_head * old_bh, * new_bh, * dir_bh;
2218         struct ext4_dir_entry_2 * old_de, * new_de;
2219         int retval;
2220
2221         old_bh = new_bh = dir_bh = NULL;
2222
2223         /* Initialize quotas before so that eventual writes go
2224          * in separate transaction */
2225         if (new_dentry->d_inode)
2226                 DQUOT_INIT(new_dentry->d_inode);
2227         handle = ext4_journal_start(old_dir, 2 *
2228                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2229                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2230         if (IS_ERR(handle))
2231                 return PTR_ERR(handle);
2232
2233         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2234                 handle->h_sync = 1;
2235
2236         old_bh = ext4_find_entry (old_dentry, &old_de);
2237         /*
2238          *  Check for inode number is _not_ due to possible IO errors.
2239          *  We might rmdir the source, keep it as pwd of some process
2240          *  and merrily kill the link to whatever was created under the
2241          *  same name. Goodbye sticky bit ;-<
2242          */
2243         old_inode = old_dentry->d_inode;
2244         retval = -ENOENT;
2245         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2246                 goto end_rename;
2247
2248         new_inode = new_dentry->d_inode;
2249         new_bh = ext4_find_entry (new_dentry, &new_de);
2250         if (new_bh) {
2251                 if (!new_inode) {
2252                         brelse (new_bh);
2253                         new_bh = NULL;
2254                 }
2255         }
2256         if (S_ISDIR(old_inode->i_mode)) {
2257                 if (new_inode) {
2258                         retval = -ENOTEMPTY;
2259                         if (!empty_dir (new_inode))
2260                                 goto end_rename;
2261                 }
2262                 retval = -EIO;
2263                 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
2264                 if (!dir_bh)
2265                         goto end_rename;
2266                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2267                         goto end_rename;
2268                 retval = -EMLINK;
2269                 if (!new_inode && new_dir!=old_dir &&
2270                                 new_dir->i_nlink >= EXT4_LINK_MAX)
2271                         goto end_rename;
2272         }
2273         if (!new_bh) {
2274                 retval = ext4_add_entry (handle, new_dentry, old_inode);
2275                 if (retval)
2276                         goto end_rename;
2277         } else {
2278                 BUFFER_TRACE(new_bh, "get write access");
2279                 ext4_journal_get_write_access(handle, new_bh);
2280                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2281                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2282                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2283                         new_de->file_type = old_de->file_type;
2284                 new_dir->i_version++;
2285                 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2286                 ext4_journal_dirty_metadata(handle, new_bh);
2287                 brelse(new_bh);
2288                 new_bh = NULL;
2289         }
2290
2291         /*
2292          * Like most other Unix systems, set the ctime for inodes on a
2293          * rename.
2294          */
2295         old_inode->i_ctime = CURRENT_TIME_SEC;
2296         ext4_mark_inode_dirty(handle, old_inode);
2297
2298         /*
2299          * ok, that's it
2300          */
2301         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2302             old_de->name_len != old_dentry->d_name.len ||
2303             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2304             (retval = ext4_delete_entry(handle, old_dir,
2305                                         old_de, old_bh)) == -ENOENT) {
2306                 /* old_de could have moved from under us during htree split, so
2307                  * make sure that we are deleting the right entry.  We might
2308                  * also be pointing to a stale entry in the unused part of
2309                  * old_bh so just checking inum and the name isn't enough. */
2310                 struct buffer_head *old_bh2;
2311                 struct ext4_dir_entry_2 *old_de2;
2312
2313                 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
2314                 if (old_bh2) {
2315                         retval = ext4_delete_entry(handle, old_dir,
2316                                                    old_de2, old_bh2);
2317                         brelse(old_bh2);
2318                 }
2319         }
2320         if (retval) {
2321                 ext4_warning(old_dir->i_sb, "ext4_rename",
2322                                 "Deleting old file (%lu), %d, error=%d",
2323                                 old_dir->i_ino, old_dir->i_nlink, retval);
2324         }
2325
2326         if (new_inode) {
2327                 drop_nlink(new_inode);
2328                 new_inode->i_ctime = CURRENT_TIME_SEC;
2329         }
2330         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2331         ext4_update_dx_flag(old_dir);
2332         if (dir_bh) {
2333                 BUFFER_TRACE(dir_bh, "get_write_access");
2334                 ext4_journal_get_write_access(handle, dir_bh);
2335                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2336                 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2337                 ext4_journal_dirty_metadata(handle, dir_bh);
2338                 drop_nlink(old_dir);
2339                 if (new_inode) {
2340                         drop_nlink(new_inode);
2341                 } else {
2342                         inc_nlink(new_dir);
2343                         ext4_update_dx_flag(new_dir);
2344                         ext4_mark_inode_dirty(handle, new_dir);
2345                 }
2346         }
2347         ext4_mark_inode_dirty(handle, old_dir);
2348         if (new_inode) {
2349                 ext4_mark_inode_dirty(handle, new_inode);
2350                 if (!new_inode->i_nlink)
2351                         ext4_orphan_add(handle, new_inode);
2352         }
2353         retval = 0;
2354
2355 end_rename:
2356         brelse (dir_bh);
2357         brelse (old_bh);
2358         brelse (new_bh);
2359         ext4_journal_stop(handle);
2360         return retval;
2361 }
2362
2363 /*
2364  * directories can handle most operations...
2365  */
2366 struct inode_operations ext4_dir_inode_operations = {
2367         .create         = ext4_create,
2368         .lookup         = ext4_lookup,
2369         .link           = ext4_link,
2370         .unlink         = ext4_unlink,
2371         .symlink        = ext4_symlink,
2372         .mkdir          = ext4_mkdir,
2373         .rmdir          = ext4_rmdir,
2374         .mknod          = ext4_mknod,
2375         .rename         = ext4_rename,
2376         .setattr        = ext4_setattr,
2377 #ifdef CONFIG_EXT4DEV_FS_XATTR
2378         .setxattr       = generic_setxattr,
2379         .getxattr       = generic_getxattr,
2380         .listxattr      = ext4_listxattr,
2381         .removexattr    = generic_removexattr,
2382 #endif
2383         .permission     = ext4_permission,
2384 };
2385
2386 struct inode_operations ext4_special_inode_operations = {
2387         .setattr        = ext4_setattr,
2388 #ifdef CONFIG_EXT4DEV_FS_XATTR
2389         .setxattr       = generic_setxattr,
2390         .getxattr       = generic_getxattr,
2391         .listxattr      = ext4_listxattr,
2392         .removexattr    = generic_removexattr,
2393 #endif
2394         .permission     = ext4_permission,
2395 };