vfs: make it possible to access the dentry hash/len as one 64-bit entry
[linux-3.10.git] / fs / ext4 / namei.c
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39
40 #include "xattr.h"
41 #include "acl.h"
42
43 #include <trace/events/ext4.h>
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext4_append(handle_t *handle,
53                                         struct inode *inode,
54                                         ext4_lblk_t *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         bh = ext4_bread(handle, inode, *block, 1, err);
61         if (bh) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT4_I(inode)->i_disksize = inode->i_size;
64                 *err = ext4_journal_get_write_access(handle, bh);
65                 if (*err) {
66                         brelse(bh);
67                         bh = NULL;
68                 }
69         }
70         return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #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         u16 offs;
145         u16 size;
146 };
147
148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t 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(const struct qstr *d_name,
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, unsigned blocksize,
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, unsigned blocksize);
169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
170 static void dx_insert_block(struct dx_frame *frame,
171                                         u32 hash, ext4_lblk_t block);
172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
173                                  struct dx_frame *frame,
174                                  struct dx_frame *frames,
175                                  __u32 *start_hash);
176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177                 const struct qstr *d_name,
178                 struct ext4_dir_entry_2 **res_dir,
179                 int *err);
180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
181                              struct inode *inode);
182
183 /*
184  * p is at least 6 bytes before the end of page
185  */
186 static inline struct ext4_dir_entry_2 *
187 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
188 {
189         return (struct ext4_dir_entry_2 *)((char *)p +
190                 ext4_rec_len_from_disk(p->rec_len, blocksize));
191 }
192
193 /*
194  * Future: use high four bits of block for coalesce-on-delete flags
195  * Mask them off for now.
196  */
197
198 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
199 {
200         return le32_to_cpu(entry->block) & 0x00ffffff;
201 }
202
203 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
204 {
205         entry->block = cpu_to_le32(value);
206 }
207
208 static inline unsigned dx_get_hash(struct dx_entry *entry)
209 {
210         return le32_to_cpu(entry->hash);
211 }
212
213 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
214 {
215         entry->hash = cpu_to_le32(value);
216 }
217
218 static inline unsigned dx_get_count(struct dx_entry *entries)
219 {
220         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
221 }
222
223 static inline unsigned dx_get_limit(struct dx_entry *entries)
224 {
225         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
226 }
227
228 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
229 {
230         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
231 }
232
233 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
234 {
235         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
236 }
237
238 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
239 {
240         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
241                 EXT4_DIR_REC_LEN(2) - infosize;
242         return entry_space / sizeof(struct dx_entry);
243 }
244
245 static inline unsigned dx_node_limit(struct inode *dir)
246 {
247         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
248         return entry_space / sizeof(struct dx_entry);
249 }
250
251 /*
252  * Debug
253  */
254 #ifdef DX_DEBUG
255 static void dx_show_index(char * label, struct dx_entry *entries)
256 {
257         int i, n = dx_get_count (entries);
258         printk(KERN_DEBUG "%s index ", label);
259         for (i = 0; i < n; i++) {
260                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
261                                 0, (unsigned long)dx_get_block(entries + i));
262         }
263         printk("\n");
264 }
265
266 struct stats
267 {
268         unsigned names;
269         unsigned space;
270         unsigned bcount;
271 };
272
273 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
274                                  int size, int show_names)
275 {
276         unsigned names = 0, space = 0;
277         char *base = (char *) de;
278         struct dx_hash_info h = *hinfo;
279
280         printk("names: ");
281         while ((char *) de < base + size)
282         {
283                 if (de->inode)
284                 {
285                         if (show_names)
286                         {
287                                 int len = de->name_len;
288                                 char *name = de->name;
289                                 while (len--) printk("%c", *name++);
290                                 ext4fs_dirhash(de->name, de->name_len, &h);
291                                 printk(":%x.%u ", h.hash,
292                                        (unsigned) ((char *) de - base));
293                         }
294                         space += EXT4_DIR_REC_LEN(de->name_len);
295                         names++;
296                 }
297                 de = ext4_next_entry(de, size);
298         }
299         printk("(%i)\n", names);
300         return (struct stats) { names, space, 1 };
301 }
302
303 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
304                              struct dx_entry *entries, int levels)
305 {
306         unsigned blocksize = dir->i_sb->s_blocksize;
307         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
308         unsigned bcount = 0;
309         struct buffer_head *bh;
310         int err;
311         printk("%i indexed blocks...\n", count);
312         for (i = 0; i < count; i++, entries++)
313         {
314                 ext4_lblk_t block = dx_get_block(entries);
315                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
316                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317                 struct stats stats;
318                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
319                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
320                 stats = levels?
321                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
323                 names += stats.names;
324                 space += stats.space;
325                 bcount += stats.bcount;
326                 brelse(bh);
327         }
328         if (bcount)
329                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
330                        levels ? "" : "   ", names, space/bcount,
331                        (space/bcount)*100/blocksize);
332         return (struct stats) { names, space, bcount};
333 }
334 #endif /* DX_DEBUG */
335
336 /*
337  * Probe for a directory leaf block to search.
338  *
339  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340  * error in the directory index, and the caller should fall back to
341  * searching the directory normally.  The callers of dx_probe **MUST**
342  * check for this error code, and make sure it never gets reflected
343  * back to userspace.
344  */
345 static struct dx_frame *
346 dx_probe(const struct qstr *d_name, struct inode *dir,
347          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
348 {
349         unsigned count, indirect;
350         struct dx_entry *at, *entries, *p, *q, *m;
351         struct dx_root *root;
352         struct buffer_head *bh;
353         struct dx_frame *frame = frame_in;
354         u32 hash;
355
356         frame->bh = NULL;
357         if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
358                 goto fail;
359         root = (struct dx_root *) bh->b_data;
360         if (root->info.hash_version != DX_HASH_TEA &&
361             root->info.hash_version != DX_HASH_HALF_MD4 &&
362             root->info.hash_version != DX_HASH_LEGACY) {
363                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
364                              root->info.hash_version);
365                 brelse(bh);
366                 *err = ERR_BAD_DX_DIR;
367                 goto fail;
368         }
369         hinfo->hash_version = root->info.hash_version;
370         if (hinfo->hash_version <= DX_HASH_TEA)
371                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
372         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
373         if (d_name)
374                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
375         hash = hinfo->hash;
376
377         if (root->info.unused_flags & 1) {
378                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
379                              root->info.unused_flags);
380                 brelse(bh);
381                 *err = ERR_BAD_DX_DIR;
382                 goto fail;
383         }
384
385         if ((indirect = root->info.indirect_levels) > 1) {
386                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
387                              root->info.indirect_levels);
388                 brelse(bh);
389                 *err = ERR_BAD_DX_DIR;
390                 goto fail;
391         }
392
393         entries = (struct dx_entry *) (((char *)&root->info) +
394                                        root->info.info_length);
395
396         if (dx_get_limit(entries) != dx_root_limit(dir,
397                                                    root->info.info_length)) {
398                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
399                 brelse(bh);
400                 *err = ERR_BAD_DX_DIR;
401                 goto fail;
402         }
403
404         dxtrace(printk("Look up %x", hash));
405         while (1)
406         {
407                 count = dx_get_count(entries);
408                 if (!count || count > dx_get_limit(entries)) {
409                         ext4_warning(dir->i_sb,
410                                      "dx entry: no count or count > limit");
411                         brelse(bh);
412                         *err = ERR_BAD_DX_DIR;
413                         goto fail2;
414                 }
415
416                 p = entries + 1;
417                 q = entries + count - 1;
418                 while (p <= q)
419                 {
420                         m = p + (q - p)/2;
421                         dxtrace(printk("."));
422                         if (dx_get_hash(m) > hash)
423                                 q = m - 1;
424                         else
425                                 p = m + 1;
426                 }
427
428                 if (0) // linear search cross check
429                 {
430                         unsigned n = count - 1;
431                         at = entries;
432                         while (n--)
433                         {
434                                 dxtrace(printk(","));
435                                 if (dx_get_hash(++at) > hash)
436                                 {
437                                         at--;
438                                         break;
439                                 }
440                         }
441                         assert (at == p - 1);
442                 }
443
444                 at = p - 1;
445                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446                 frame->bh = bh;
447                 frame->entries = entries;
448                 frame->at = at;
449                 if (!indirect--) return frame;
450                 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
451                         goto fail2;
452                 at = entries = ((struct dx_node *) bh->b_data)->entries;
453                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
454                         ext4_warning(dir->i_sb,
455                                      "dx entry: limit != node limit");
456                         brelse(bh);
457                         *err = ERR_BAD_DX_DIR;
458                         goto fail2;
459                 }
460                 frame++;
461                 frame->bh = NULL;
462         }
463 fail2:
464         while (frame >= frame_in) {
465                 brelse(frame->bh);
466                 frame--;
467         }
468 fail:
469         if (*err == ERR_BAD_DX_DIR)
470                 ext4_warning(dir->i_sb,
471                              "Corrupt dir inode %lu, running e2fsck is "
472                              "recommended.", dir->i_ino);
473         return NULL;
474 }
475
476 static void dx_release (struct dx_frame *frames)
477 {
478         if (frames[0].bh == NULL)
479                 return;
480
481         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482                 brelse(frames[1].bh);
483         brelse(frames[0].bh);
484 }
485
486 /*
487  * This function increments the frame pointer to search the next leaf
488  * block, and reads in the necessary intervening nodes if the search
489  * should be necessary.  Whether or not the search is necessary is
490  * controlled by the hash parameter.  If the hash value is even, then
491  * the search is only continued if the next block starts with that
492  * hash value.  This is used if we are searching for a specific file.
493  *
494  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
495  *
496  * This function returns 1 if the caller should continue to search,
497  * or 0 if it should not.  If there is an error reading one of the
498  * index blocks, it will a negative error code.
499  *
500  * If start_hash is non-null, it will be filled in with the starting
501  * hash of the next page.
502  */
503 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
504                                  struct dx_frame *frame,
505                                  struct dx_frame *frames,
506                                  __u32 *start_hash)
507 {
508         struct dx_frame *p;
509         struct buffer_head *bh;
510         int err, num_frames = 0;
511         __u32 bhash;
512
513         p = frame;
514         /*
515          * Find the next leaf page by incrementing the frame pointer.
516          * If we run out of entries in the interior node, loop around and
517          * increment pointer in the parent node.  When we break out of
518          * this loop, num_frames indicates the number of interior
519          * nodes need to be read.
520          */
521         while (1) {
522                 if (++(p->at) < p->entries + dx_get_count(p->entries))
523                         break;
524                 if (p == frames)
525                         return 0;
526                 num_frames++;
527                 p--;
528         }
529
530         /*
531          * If the hash is 1, then continue only if the next page has a
532          * continuation hash of any value.  This is used for readdir
533          * handling.  Otherwise, check to see if the hash matches the
534          * desired contiuation hash.  If it doesn't, return since
535          * there's no point to read in the successive index pages.
536          */
537         bhash = dx_get_hash(p->at);
538         if (start_hash)
539                 *start_hash = bhash;
540         if ((hash & 1) == 0) {
541                 if ((bhash & ~1) != hash)
542                         return 0;
543         }
544         /*
545          * If the hash is HASH_NB_ALWAYS, we always go to the next
546          * block so no check is necessary
547          */
548         while (num_frames--) {
549                 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
550                                       0, &err)))
551                         return err; /* Failure */
552                 p++;
553                 brelse(p->bh);
554                 p->bh = bh;
555                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
556         }
557         return 1;
558 }
559
560
561 /*
562  * This function fills a red-black tree with information from a
563  * directory block.  It returns the number directory entries loaded
564  * into the tree.  If there is an error it is returned in err.
565  */
566 static int htree_dirblock_to_tree(struct file *dir_file,
567                                   struct inode *dir, ext4_lblk_t block,
568                                   struct dx_hash_info *hinfo,
569                                   __u32 start_hash, __u32 start_minor_hash)
570 {
571         struct buffer_head *bh;
572         struct ext4_dir_entry_2 *de, *top;
573         int err, count = 0;
574
575         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
576                                                         (unsigned long)block));
577         if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
578                 return err;
579
580         de = (struct ext4_dir_entry_2 *) bh->b_data;
581         top = (struct ext4_dir_entry_2 *) ((char *) de +
582                                            dir->i_sb->s_blocksize -
583                                            EXT4_DIR_REC_LEN(0));
584         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
585                 if (ext4_check_dir_entry(dir, NULL, de, bh,
586                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
587                                          + ((char *)de - bh->b_data))) {
588                         /* On error, skip the f_pos to the next block. */
589                         dir_file->f_pos = (dir_file->f_pos |
590                                         (dir->i_sb->s_blocksize - 1)) + 1;
591                         brelse(bh);
592                         return count;
593                 }
594                 ext4fs_dirhash(de->name, de->name_len, hinfo);
595                 if ((hinfo->hash < start_hash) ||
596                     ((hinfo->hash == start_hash) &&
597                      (hinfo->minor_hash < start_minor_hash)))
598                         continue;
599                 if (de->inode == 0)
600                         continue;
601                 if ((err = ext4_htree_store_dirent(dir_file,
602                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
603                         brelse(bh);
604                         return err;
605                 }
606                 count++;
607         }
608         brelse(bh);
609         return count;
610 }
611
612
613 /*
614  * This function fills a red-black tree with information from a
615  * directory.  We start scanning the directory in hash order, starting
616  * at start_hash and start_minor_hash.
617  *
618  * This function returns the number of entries inserted into the tree,
619  * or a negative error code.
620  */
621 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
622                          __u32 start_minor_hash, __u32 *next_hash)
623 {
624         struct dx_hash_info hinfo;
625         struct ext4_dir_entry_2 *de;
626         struct dx_frame frames[2], *frame;
627         struct inode *dir;
628         ext4_lblk_t block;
629         int count = 0;
630         int ret, err;
631         __u32 hashval;
632
633         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
634                        start_hash, start_minor_hash));
635         dir = dir_file->f_path.dentry->d_inode;
636         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
637                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
638                 if (hinfo.hash_version <= DX_HASH_TEA)
639                         hinfo.hash_version +=
640                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
641                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
642                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
643                                                start_hash, start_minor_hash);
644                 *next_hash = ~0;
645                 return count;
646         }
647         hinfo.hash = start_hash;
648         hinfo.minor_hash = 0;
649         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
650         if (!frame)
651                 return err;
652
653         /* Add '.' and '..' from the htree header */
654         if (!start_hash && !start_minor_hash) {
655                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
656                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
657                         goto errout;
658                 count++;
659         }
660         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
661                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
662                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
663                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
664                         goto errout;
665                 count++;
666         }
667
668         while (1) {
669                 block = dx_get_block(frame->at);
670                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
671                                              start_hash, start_minor_hash);
672                 if (ret < 0) {
673                         err = ret;
674                         goto errout;
675                 }
676                 count += ret;
677                 hashval = ~0;
678                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
679                                             frame, frames, &hashval);
680                 *next_hash = hashval;
681                 if (ret < 0) {
682                         err = ret;
683                         goto errout;
684                 }
685                 /*
686                  * Stop if:  (a) there are no more entries, or
687                  * (b) we have inserted at least one entry and the
688                  * next hash value is not a continuation
689                  */
690                 if ((ret == 0) ||
691                     (count && ((hashval & 1) == 0)))
692                         break;
693         }
694         dx_release(frames);
695         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
696                        "next hash: %x\n", count, *next_hash));
697         return count;
698 errout:
699         dx_release(frames);
700         return (err);
701 }
702
703
704 /*
705  * Directory block splitting, compacting
706  */
707
708 /*
709  * Create map of hash values, offsets, and sizes, stored at end of block.
710  * Returns number of entries mapped.
711  */
712 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
713                        struct dx_hash_info *hinfo,
714                        struct dx_map_entry *map_tail)
715 {
716         int count = 0;
717         char *base = (char *) de;
718         struct dx_hash_info h = *hinfo;
719
720         while ((char *) de < base + blocksize) {
721                 if (de->name_len && de->inode) {
722                         ext4fs_dirhash(de->name, de->name_len, &h);
723                         map_tail--;
724                         map_tail->hash = h.hash;
725                         map_tail->offs = ((char *) de - base)>>2;
726                         map_tail->size = le16_to_cpu(de->rec_len);
727                         count++;
728                         cond_resched();
729                 }
730                 /* XXX: do we need to check rec_len == 0 case? -Chris */
731                 de = ext4_next_entry(de, blocksize);
732         }
733         return count;
734 }
735
736 /* Sort map by hash value */
737 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
738 {
739         struct dx_map_entry *p, *q, *top = map + count - 1;
740         int more;
741         /* Combsort until bubble sort doesn't suck */
742         while (count > 2) {
743                 count = count*10/13;
744                 if (count - 9 < 2) /* 9, 10 -> 11 */
745                         count = 11;
746                 for (p = top, q = p - count; q >= map; p--, q--)
747                         if (p->hash < q->hash)
748                                 swap(*p, *q);
749         }
750         /* Garden variety bubble sort */
751         do {
752                 more = 0;
753                 q = top;
754                 while (q-- > map) {
755                         if (q[1].hash >= q[0].hash)
756                                 continue;
757                         swap(*(q+1), *q);
758                         more = 1;
759                 }
760         } while(more);
761 }
762
763 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
764 {
765         struct dx_entry *entries = frame->entries;
766         struct dx_entry *old = frame->at, *new = old + 1;
767         int count = dx_get_count(entries);
768
769         assert(count < dx_get_limit(entries));
770         assert(old < entries + count);
771         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
772         dx_set_hash(new, hash);
773         dx_set_block(new, block);
774         dx_set_count(entries, count + 1);
775 }
776
777 static void ext4_update_dx_flag(struct inode *inode)
778 {
779         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
780                                      EXT4_FEATURE_COMPAT_DIR_INDEX))
781                 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
782 }
783
784 /*
785  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
786  *
787  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
788  * `de != NULL' is guaranteed by caller.
789  */
790 static inline int ext4_match (int len, const char * const name,
791                               struct ext4_dir_entry_2 * de)
792 {
793         if (len != de->name_len)
794                 return 0;
795         if (!de->inode)
796                 return 0;
797         return !memcmp(name, de->name, len);
798 }
799
800 /*
801  * Returns 0 if not found, -1 on failure, and 1 on success
802  */
803 static inline int search_dirblock(struct buffer_head *bh,
804                                   struct inode *dir,
805                                   const struct qstr *d_name,
806                                   unsigned int offset,
807                                   struct ext4_dir_entry_2 ** res_dir)
808 {
809         struct ext4_dir_entry_2 * de;
810         char * dlimit;
811         int de_len;
812         const char *name = d_name->name;
813         int namelen = d_name->len;
814
815         de = (struct ext4_dir_entry_2 *) bh->b_data;
816         dlimit = bh->b_data + dir->i_sb->s_blocksize;
817         while ((char *) de < dlimit) {
818                 /* this code is executed quadratically often */
819                 /* do minimal checking `by hand' */
820
821                 if ((char *) de + namelen <= dlimit &&
822                     ext4_match (namelen, name, de)) {
823                         /* found a match - just to be sure, do a full check */
824                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
825                                 return -1;
826                         *res_dir = de;
827                         return 1;
828                 }
829                 /* prevent looping on a bad block */
830                 de_len = ext4_rec_len_from_disk(de->rec_len,
831                                                 dir->i_sb->s_blocksize);
832                 if (de_len <= 0)
833                         return -1;
834                 offset += de_len;
835                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
836         }
837         return 0;
838 }
839
840
841 /*
842  *      ext4_find_entry()
843  *
844  * finds an entry in the specified directory with the wanted name. It
845  * returns the cache buffer in which the entry was found, and the entry
846  * itself (as a parameter - res_dir). It does NOT read the inode of the
847  * entry - you'll have to do that yourself if you want to.
848  *
849  * The returned buffer_head has ->b_count elevated.  The caller is expected
850  * to brelse() it when appropriate.
851  */
852 static struct buffer_head * ext4_find_entry (struct inode *dir,
853                                         const struct qstr *d_name,
854                                         struct ext4_dir_entry_2 ** res_dir)
855 {
856         struct super_block *sb;
857         struct buffer_head *bh_use[NAMEI_RA_SIZE];
858         struct buffer_head *bh, *ret = NULL;
859         ext4_lblk_t start, block, b;
860         const u8 *name = d_name->name;
861         int ra_max = 0;         /* Number of bh's in the readahead
862                                    buffer, bh_use[] */
863         int ra_ptr = 0;         /* Current index into readahead
864                                    buffer */
865         int num = 0;
866         ext4_lblk_t  nblocks;
867         int i, err;
868         int namelen;
869
870         *res_dir = NULL;
871         sb = dir->i_sb;
872         namelen = d_name->len;
873         if (namelen > EXT4_NAME_LEN)
874                 return NULL;
875         if ((namelen <= 2) && (name[0] == '.') &&
876             (name[1] == '.' || name[1] == '\0')) {
877                 /*
878                  * "." or ".." will only be in the first block
879                  * NFS may look up ".."; "." should be handled by the VFS
880                  */
881                 block = start = 0;
882                 nblocks = 1;
883                 goto restart;
884         }
885         if (is_dx(dir)) {
886                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
887                 /*
888                  * On success, or if the error was file not found,
889                  * return.  Otherwise, fall back to doing a search the
890                  * old fashioned way.
891                  */
892                 if (bh || (err != ERR_BAD_DX_DIR))
893                         return bh;
894                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
895                                "falling back\n"));
896         }
897         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
898         start = EXT4_I(dir)->i_dir_start_lookup;
899         if (start >= nblocks)
900                 start = 0;
901         block = start;
902 restart:
903         do {
904                 /*
905                  * We deal with the read-ahead logic here.
906                  */
907                 if (ra_ptr >= ra_max) {
908                         /* Refill the readahead buffer */
909                         ra_ptr = 0;
910                         b = block;
911                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
912                                 /*
913                                  * Terminate if we reach the end of the
914                                  * directory and must wrap, or if our
915                                  * search has finished at this block.
916                                  */
917                                 if (b >= nblocks || (num && block == start)) {
918                                         bh_use[ra_max] = NULL;
919                                         break;
920                                 }
921                                 num++;
922                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
923                                 bh_use[ra_max] = bh;
924                                 if (bh)
925                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
926                                                     1, &bh);
927                         }
928                 }
929                 if ((bh = bh_use[ra_ptr++]) == NULL)
930                         goto next;
931                 wait_on_buffer(bh);
932                 if (!buffer_uptodate(bh)) {
933                         /* read error, skip block & hope for the best */
934                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
935                                          (unsigned long) block);
936                         brelse(bh);
937                         goto next;
938                 }
939                 i = search_dirblock(bh, dir, d_name,
940                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
941                 if (i == 1) {
942                         EXT4_I(dir)->i_dir_start_lookup = block;
943                         ret = bh;
944                         goto cleanup_and_exit;
945                 } else {
946                         brelse(bh);
947                         if (i < 0)
948                                 goto cleanup_and_exit;
949                 }
950         next:
951                 if (++block >= nblocks)
952                         block = 0;
953         } while (block != start);
954
955         /*
956          * If the directory has grown while we were searching, then
957          * search the last part of the directory before giving up.
958          */
959         block = nblocks;
960         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
961         if (block < nblocks) {
962                 start = 0;
963                 goto restart;
964         }
965
966 cleanup_and_exit:
967         /* Clean up the read-ahead blocks */
968         for (; ra_ptr < ra_max; ra_ptr++)
969                 brelse(bh_use[ra_ptr]);
970         return ret;
971 }
972
973 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
974                        struct ext4_dir_entry_2 **res_dir, int *err)
975 {
976         struct super_block * sb = dir->i_sb;
977         struct dx_hash_info     hinfo;
978         struct dx_frame frames[2], *frame;
979         struct buffer_head *bh;
980         ext4_lblk_t block;
981         int retval;
982
983         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
984                 return NULL;
985         do {
986                 block = dx_get_block(frame->at);
987                 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
988                         goto errout;
989
990                 retval = search_dirblock(bh, dir, d_name,
991                                          block << EXT4_BLOCK_SIZE_BITS(sb),
992                                          res_dir);
993                 if (retval == 1) {      /* Success! */
994                         dx_release(frames);
995                         return bh;
996                 }
997                 brelse(bh);
998                 if (retval == -1) {
999                         *err = ERR_BAD_DX_DIR;
1000                         goto errout;
1001                 }
1002
1003                 /* Check to see if we should continue to search */
1004                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1005                                                frames, NULL);
1006                 if (retval < 0) {
1007                         ext4_warning(sb,
1008                              "error reading index page in directory #%lu",
1009                              dir->i_ino);
1010                         *err = retval;
1011                         goto errout;
1012                 }
1013         } while (retval == 1);
1014
1015         *err = -ENOENT;
1016 errout:
1017         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1018         dx_release (frames);
1019         return NULL;
1020 }
1021
1022 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1023 {
1024         struct inode *inode;
1025         struct ext4_dir_entry_2 *de;
1026         struct buffer_head *bh;
1027
1028         if (dentry->d_name.len > EXT4_NAME_LEN)
1029                 return ERR_PTR(-ENAMETOOLONG);
1030
1031         bh = ext4_find_entry(dir, &dentry->d_name, &de);
1032         inode = NULL;
1033         if (bh) {
1034                 __u32 ino = le32_to_cpu(de->inode);
1035                 brelse(bh);
1036                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1037                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1038                         return ERR_PTR(-EIO);
1039                 }
1040                 inode = ext4_iget(dir->i_sb, ino);
1041                 if (inode == ERR_PTR(-ESTALE)) {
1042                         EXT4_ERROR_INODE(dir,
1043                                          "deleted inode referenced: %u",
1044                                          ino);
1045                         return ERR_PTR(-EIO);
1046                 }
1047         }
1048         return d_splice_alias(inode, dentry);
1049 }
1050
1051
1052 struct dentry *ext4_get_parent(struct dentry *child)
1053 {
1054         __u32 ino;
1055         static const struct qstr dotdot = QSTR_INIT("..", 2);
1056         struct ext4_dir_entry_2 * de;
1057         struct buffer_head *bh;
1058
1059         bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1060         if (!bh)
1061                 return ERR_PTR(-ENOENT);
1062         ino = le32_to_cpu(de->inode);
1063         brelse(bh);
1064
1065         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1066                 EXT4_ERROR_INODE(child->d_inode,
1067                                  "bad parent inode number: %u", ino);
1068                 return ERR_PTR(-EIO);
1069         }
1070
1071         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1072 }
1073
1074 #define S_SHIFT 12
1075 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1076         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1077         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1078         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1079         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1080         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1081         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1082         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1083 };
1084
1085 static inline void ext4_set_de_type(struct super_block *sb,
1086                                 struct ext4_dir_entry_2 *de,
1087                                 umode_t mode) {
1088         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1089                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1090 }
1091
1092 /*
1093  * Move count entries from end of map between two memory locations.
1094  * Returns pointer to last entry moved.
1095  */
1096 static struct ext4_dir_entry_2 *
1097 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1098                 unsigned blocksize)
1099 {
1100         unsigned rec_len = 0;
1101
1102         while (count--) {
1103                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1104                                                 (from + (map->offs<<2));
1105                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1106                 memcpy (to, de, rec_len);
1107                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1108                                 ext4_rec_len_to_disk(rec_len, blocksize);
1109                 de->inode = 0;
1110                 map++;
1111                 to += rec_len;
1112         }
1113         return (struct ext4_dir_entry_2 *) (to - rec_len);
1114 }
1115
1116 /*
1117  * Compact each dir entry in the range to the minimal rec_len.
1118  * Returns pointer to last entry in range.
1119  */
1120 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1121 {
1122         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1123         unsigned rec_len = 0;
1124
1125         prev = to = de;
1126         while ((char*)de < base + blocksize) {
1127                 next = ext4_next_entry(de, blocksize);
1128                 if (de->inode && de->name_len) {
1129                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1130                         if (de > to)
1131                                 memmove(to, de, rec_len);
1132                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1133                         prev = to;
1134                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1135                 }
1136                 de = next;
1137         }
1138         return prev;
1139 }
1140
1141 /*
1142  * Split a full leaf block to make room for a new dir entry.
1143  * Allocate a new block, and move entries so that they are approx. equally full.
1144  * Returns pointer to de in block into which the new entry will be inserted.
1145  */
1146 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1147                         struct buffer_head **bh,struct dx_frame *frame,
1148                         struct dx_hash_info *hinfo, int *error)
1149 {
1150         unsigned blocksize = dir->i_sb->s_blocksize;
1151         unsigned count, continued;
1152         struct buffer_head *bh2;
1153         ext4_lblk_t newblock;
1154         u32 hash2;
1155         struct dx_map_entry *map;
1156         char *data1 = (*bh)->b_data, *data2;
1157         unsigned split, move, size;
1158         struct ext4_dir_entry_2 *de = NULL, *de2;
1159         int     err = 0, i;
1160
1161         bh2 = ext4_append (handle, dir, &newblock, &err);
1162         if (!(bh2)) {
1163                 brelse(*bh);
1164                 *bh = NULL;
1165                 goto errout;
1166         }
1167
1168         BUFFER_TRACE(*bh, "get_write_access");
1169         err = ext4_journal_get_write_access(handle, *bh);
1170         if (err)
1171                 goto journal_error;
1172
1173         BUFFER_TRACE(frame->bh, "get_write_access");
1174         err = ext4_journal_get_write_access(handle, frame->bh);
1175         if (err)
1176                 goto journal_error;
1177
1178         data2 = bh2->b_data;
1179
1180         /* create map in the end of data2 block */
1181         map = (struct dx_map_entry *) (data2 + blocksize);
1182         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1183                              blocksize, hinfo, map);
1184         map -= count;
1185         dx_sort_map(map, count);
1186         /* Split the existing block in the middle, size-wise */
1187         size = 0;
1188         move = 0;
1189         for (i = count-1; i >= 0; i--) {
1190                 /* is more than half of this entry in 2nd half of the block? */
1191                 if (size + map[i].size/2 > blocksize/2)
1192                         break;
1193                 size += map[i].size;
1194                 move++;
1195         }
1196         /* map index at which we will split */
1197         split = count - move;
1198         hash2 = map[split].hash;
1199         continued = hash2 == map[split - 1].hash;
1200         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1201                         (unsigned long)dx_get_block(frame->at),
1202                                         hash2, split, count-split));
1203
1204         /* Fancy dance to stay within two buffers */
1205         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1206         de = dx_pack_dirents(data1, blocksize);
1207         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1208                                            blocksize);
1209         de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1210                                             blocksize);
1211         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1212         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1213
1214         /* Which block gets the new entry? */
1215         if (hinfo->hash >= hash2)
1216         {
1217                 swap(*bh, bh2);
1218                 de = de2;
1219         }
1220         dx_insert_block(frame, hash2 + continued, newblock);
1221         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1222         if (err)
1223                 goto journal_error;
1224         err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1225         if (err)
1226                 goto journal_error;
1227         brelse(bh2);
1228         dxtrace(dx_show_index("frame", frame->entries));
1229         return de;
1230
1231 journal_error:
1232         brelse(*bh);
1233         brelse(bh2);
1234         *bh = NULL;
1235         ext4_std_error(dir->i_sb, err);
1236 errout:
1237         *error = err;
1238         return NULL;
1239 }
1240
1241 /*
1242  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1243  * it points to a directory entry which is guaranteed to be large
1244  * enough for new directory entry.  If de is NULL, then
1245  * add_dirent_to_buf will attempt search the directory block for
1246  * space.  It will return -ENOSPC if no space is available, and -EIO
1247  * and -EEXIST if directory entry already exists.
1248  */
1249 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1250                              struct inode *inode, struct ext4_dir_entry_2 *de,
1251                              struct buffer_head *bh)
1252 {
1253         struct inode    *dir = dentry->d_parent->d_inode;
1254         const char      *name = dentry->d_name.name;
1255         int             namelen = dentry->d_name.len;
1256         unsigned int    offset = 0;
1257         unsigned int    blocksize = dir->i_sb->s_blocksize;
1258         unsigned short  reclen;
1259         int             nlen, rlen, err;
1260         char            *top;
1261
1262         reclen = EXT4_DIR_REC_LEN(namelen);
1263         if (!de) {
1264                 de = (struct ext4_dir_entry_2 *)bh->b_data;
1265                 top = bh->b_data + blocksize - reclen;
1266                 while ((char *) de <= top) {
1267                         if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1268                                 return -EIO;
1269                         if (ext4_match(namelen, name, de))
1270                                 return -EEXIST;
1271                         nlen = EXT4_DIR_REC_LEN(de->name_len);
1272                         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1273                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1274                                 break;
1275                         de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1276                         offset += rlen;
1277                 }
1278                 if ((char *) de > top)
1279                         return -ENOSPC;
1280         }
1281         BUFFER_TRACE(bh, "get_write_access");
1282         err = ext4_journal_get_write_access(handle, bh);
1283         if (err) {
1284                 ext4_std_error(dir->i_sb, err);
1285                 return err;
1286         }
1287
1288         /* By now the buffer is marked for journaling */
1289         nlen = EXT4_DIR_REC_LEN(de->name_len);
1290         rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1291         if (de->inode) {
1292                 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1293                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1294                 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1295                 de = de1;
1296         }
1297         de->file_type = EXT4_FT_UNKNOWN;
1298         if (inode) {
1299                 de->inode = cpu_to_le32(inode->i_ino);
1300                 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1301         } else
1302                 de->inode = 0;
1303         de->name_len = namelen;
1304         memcpy(de->name, name, namelen);
1305         /*
1306          * XXX shouldn't update any times until successful
1307          * completion of syscall, but too many callers depend
1308          * on this.
1309          *
1310          * XXX similarly, too many callers depend on
1311          * ext4_new_inode() setting the times, but error
1312          * recovery deletes the inode, so the worst that can
1313          * happen is that the times are slightly out of date
1314          * and/or different from the directory change time.
1315          */
1316         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1317         ext4_update_dx_flag(dir);
1318         dir->i_version++;
1319         ext4_mark_inode_dirty(handle, dir);
1320         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1321         err = ext4_handle_dirty_metadata(handle, dir, bh);
1322         if (err)
1323                 ext4_std_error(dir->i_sb, err);
1324         return 0;
1325 }
1326
1327 /*
1328  * This converts a one block unindexed directory to a 3 block indexed
1329  * directory, and adds the dentry to the indexed directory.
1330  */
1331 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1332                             struct inode *inode, struct buffer_head *bh)
1333 {
1334         struct inode    *dir = dentry->d_parent->d_inode;
1335         const char      *name = dentry->d_name.name;
1336         int             namelen = dentry->d_name.len;
1337         struct buffer_head *bh2;
1338         struct dx_root  *root;
1339         struct dx_frame frames[2], *frame;
1340         struct dx_entry *entries;
1341         struct ext4_dir_entry_2 *de, *de2;
1342         char            *data1, *top;
1343         unsigned        len;
1344         int             retval;
1345         unsigned        blocksize;
1346         struct dx_hash_info hinfo;
1347         ext4_lblk_t  block;
1348         struct fake_dirent *fde;
1349
1350         blocksize =  dir->i_sb->s_blocksize;
1351         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1352         retval = ext4_journal_get_write_access(handle, bh);
1353         if (retval) {
1354                 ext4_std_error(dir->i_sb, retval);
1355                 brelse(bh);
1356                 return retval;
1357         }
1358         root = (struct dx_root *) bh->b_data;
1359
1360         /* The 0th block becomes the root, move the dirents out */
1361         fde = &root->dotdot;
1362         de = (struct ext4_dir_entry_2 *)((char *)fde +
1363                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1364         if ((char *) de >= (((char *) root) + blocksize)) {
1365                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1366                 brelse(bh);
1367                 return -EIO;
1368         }
1369         len = ((char *) root) + blocksize - (char *) de;
1370
1371         /* Allocate new block for the 0th block's dirents */
1372         bh2 = ext4_append(handle, dir, &block, &retval);
1373         if (!(bh2)) {
1374                 brelse(bh);
1375                 return retval;
1376         }
1377         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1378         data1 = bh2->b_data;
1379
1380         memcpy (data1, de, len);
1381         de = (struct ext4_dir_entry_2 *) data1;
1382         top = data1 + len;
1383         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1384                 de = de2;
1385         de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1386                                            blocksize);
1387         /* Initialize the root; the dot dirents already exist */
1388         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1389         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1390                                            blocksize);
1391         memset (&root->info, 0, sizeof(root->info));
1392         root->info.info_length = sizeof(root->info);
1393         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1394         entries = root->entries;
1395         dx_set_block(entries, 1);
1396         dx_set_count(entries, 1);
1397         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1398
1399         /* Initialize as for dx_probe */
1400         hinfo.hash_version = root->info.hash_version;
1401         if (hinfo.hash_version <= DX_HASH_TEA)
1402                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1403         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1404         ext4fs_dirhash(name, namelen, &hinfo);
1405         frame = frames;
1406         frame->entries = entries;
1407         frame->at = entries;
1408         frame->bh = bh;
1409         bh = bh2;
1410
1411         ext4_handle_dirty_metadata(handle, dir, frame->bh);
1412         ext4_handle_dirty_metadata(handle, dir, bh);
1413
1414         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1415         if (!de) {
1416                 /*
1417                  * Even if the block split failed, we have to properly write
1418                  * out all the changes we did so far. Otherwise we can end up
1419                  * with corrupted filesystem.
1420                  */
1421                 ext4_mark_inode_dirty(handle, dir);
1422                 dx_release(frames);
1423                 return retval;
1424         }
1425         dx_release(frames);
1426
1427         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1428         brelse(bh);
1429         return retval;
1430 }
1431
1432 /*
1433  *      ext4_add_entry()
1434  *
1435  * adds a file entry to the specified directory, using the same
1436  * semantics as ext4_find_entry(). It returns NULL if it failed.
1437  *
1438  * NOTE!! The inode part of 'de' is left at 0 - which means you
1439  * may not sleep between calling this and putting something into
1440  * the entry, as someone else might have used it while you slept.
1441  */
1442 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1443                           struct inode *inode)
1444 {
1445         struct inode *dir = dentry->d_parent->d_inode;
1446         struct buffer_head *bh;
1447         struct ext4_dir_entry_2 *de;
1448         struct super_block *sb;
1449         int     retval;
1450         int     dx_fallback=0;
1451         unsigned blocksize;
1452         ext4_lblk_t block, blocks;
1453
1454         sb = dir->i_sb;
1455         blocksize = sb->s_blocksize;
1456         if (!dentry->d_name.len)
1457                 return -EINVAL;
1458         if (is_dx(dir)) {
1459                 retval = ext4_dx_add_entry(handle, dentry, inode);
1460                 if (!retval || (retval != ERR_BAD_DX_DIR))
1461                         return retval;
1462                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1463                 dx_fallback++;
1464                 ext4_mark_inode_dirty(handle, dir);
1465         }
1466         blocks = dir->i_size >> sb->s_blocksize_bits;
1467         for (block = 0; block < blocks; block++) {
1468                 bh = ext4_bread(handle, dir, block, 0, &retval);
1469                 if(!bh)
1470                         return retval;
1471                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1472                 if (retval != -ENOSPC) {
1473                         brelse(bh);
1474                         return retval;
1475                 }
1476
1477                 if (blocks == 1 && !dx_fallback &&
1478                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1479                         return make_indexed_dir(handle, dentry, inode, bh);
1480                 brelse(bh);
1481         }
1482         bh = ext4_append(handle, dir, &block, &retval);
1483         if (!bh)
1484                 return retval;
1485         de = (struct ext4_dir_entry_2 *) bh->b_data;
1486         de->inode = 0;
1487         de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1488         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1489         brelse(bh);
1490         if (retval == 0)
1491                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1492         return retval;
1493 }
1494
1495 /*
1496  * Returns 0 for success, or a negative error value
1497  */
1498 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1499                              struct inode *inode)
1500 {
1501         struct dx_frame frames[2], *frame;
1502         struct dx_entry *entries, *at;
1503         struct dx_hash_info hinfo;
1504         struct buffer_head *bh;
1505         struct inode *dir = dentry->d_parent->d_inode;
1506         struct super_block *sb = dir->i_sb;
1507         struct ext4_dir_entry_2 *de;
1508         int err;
1509
1510         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1511         if (!frame)
1512                 return err;
1513         entries = frame->entries;
1514         at = frame->at;
1515
1516         if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1517                 goto cleanup;
1518
1519         BUFFER_TRACE(bh, "get_write_access");
1520         err = ext4_journal_get_write_access(handle, bh);
1521         if (err)
1522                 goto journal_error;
1523
1524         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1525         if (err != -ENOSPC)
1526                 goto cleanup;
1527
1528         /* Block full, should compress but for now just split */
1529         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1530                        dx_get_count(entries), dx_get_limit(entries)));
1531         /* Need to split index? */
1532         if (dx_get_count(entries) == dx_get_limit(entries)) {
1533                 ext4_lblk_t newblock;
1534                 unsigned icount = dx_get_count(entries);
1535                 int levels = frame - frames;
1536                 struct dx_entry *entries2;
1537                 struct dx_node *node2;
1538                 struct buffer_head *bh2;
1539
1540                 if (levels && (dx_get_count(frames->entries) ==
1541                                dx_get_limit(frames->entries))) {
1542                         ext4_warning(sb, "Directory index full!");
1543                         err = -ENOSPC;
1544                         goto cleanup;
1545                 }
1546                 bh2 = ext4_append (handle, dir, &newblock, &err);
1547                 if (!(bh2))
1548                         goto cleanup;
1549                 node2 = (struct dx_node *)(bh2->b_data);
1550                 entries2 = node2->entries;
1551                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1552                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1553                                                            sb->s_blocksize);
1554                 BUFFER_TRACE(frame->bh, "get_write_access");
1555                 err = ext4_journal_get_write_access(handle, frame->bh);
1556                 if (err)
1557                         goto journal_error;
1558                 if (levels) {
1559                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1560                         unsigned hash2 = dx_get_hash(entries + icount1);
1561                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1562                                        icount1, icount2));
1563
1564                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1565                         err = ext4_journal_get_write_access(handle,
1566                                                              frames[0].bh);
1567                         if (err)
1568                                 goto journal_error;
1569
1570                         memcpy((char *) entries2, (char *) (entries + icount1),
1571                                icount2 * sizeof(struct dx_entry));
1572                         dx_set_count(entries, icount1);
1573                         dx_set_count(entries2, icount2);
1574                         dx_set_limit(entries2, dx_node_limit(dir));
1575
1576                         /* Which index block gets the new entry? */
1577                         if (at - entries >= icount1) {
1578                                 frame->at = at = at - entries - icount1 + entries2;
1579                                 frame->entries = entries = entries2;
1580                                 swap(frame->bh, bh2);
1581                         }
1582                         dx_insert_block(frames + 0, hash2, newblock);
1583                         dxtrace(dx_show_index("node", frames[1].entries));
1584                         dxtrace(dx_show_index("node",
1585                                ((struct dx_node *) bh2->b_data)->entries));
1586                         err = ext4_handle_dirty_metadata(handle, dir, bh2);
1587                         if (err)
1588                                 goto journal_error;
1589                         brelse (bh2);
1590                 } else {
1591                         dxtrace(printk(KERN_DEBUG
1592                                        "Creating second level index...\n"));
1593                         memcpy((char *) entries2, (char *) entries,
1594                                icount * sizeof(struct dx_entry));
1595                         dx_set_limit(entries2, dx_node_limit(dir));
1596
1597                         /* Set up root */
1598                         dx_set_count(entries, 1);
1599                         dx_set_block(entries + 0, newblock);
1600                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1601
1602                         /* Add new access path frame */
1603                         frame = frames + 1;
1604                         frame->at = at = at - entries + entries2;
1605                         frame->entries = entries = entries2;
1606                         frame->bh = bh2;
1607                         err = ext4_journal_get_write_access(handle,
1608                                                              frame->bh);
1609                         if (err)
1610                                 goto journal_error;
1611                 }
1612                 err = ext4_handle_dirty_metadata(handle, dir, frames[0].bh);
1613                 if (err) {
1614                         ext4_std_error(inode->i_sb, err);
1615                         goto cleanup;
1616                 }
1617         }
1618         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1619         if (!de)
1620                 goto cleanup;
1621         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1622         goto cleanup;
1623
1624 journal_error:
1625         ext4_std_error(dir->i_sb, err);
1626 cleanup:
1627         if (bh)
1628                 brelse(bh);
1629         dx_release(frames);
1630         return err;
1631 }
1632
1633 /*
1634  * ext4_delete_entry deletes a directory entry by merging it with the
1635  * previous entry
1636  */
1637 static int ext4_delete_entry(handle_t *handle,
1638                              struct inode *dir,
1639                              struct ext4_dir_entry_2 *de_del,
1640                              struct buffer_head *bh)
1641 {
1642         struct ext4_dir_entry_2 *de, *pde;
1643         unsigned int blocksize = dir->i_sb->s_blocksize;
1644         int i, err;
1645
1646         i = 0;
1647         pde = NULL;
1648         de = (struct ext4_dir_entry_2 *) bh->b_data;
1649         while (i < bh->b_size) {
1650                 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1651                         return -EIO;
1652                 if (de == de_del)  {
1653                         BUFFER_TRACE(bh, "get_write_access");
1654                         err = ext4_journal_get_write_access(handle, bh);
1655                         if (unlikely(err)) {
1656                                 ext4_std_error(dir->i_sb, err);
1657                                 return err;
1658                         }
1659                         if (pde)
1660                                 pde->rec_len = ext4_rec_len_to_disk(
1661                                         ext4_rec_len_from_disk(pde->rec_len,
1662                                                                blocksize) +
1663                                         ext4_rec_len_from_disk(de->rec_len,
1664                                                                blocksize),
1665                                         blocksize);
1666                         else
1667                                 de->inode = 0;
1668                         dir->i_version++;
1669                         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1670                         err = ext4_handle_dirty_metadata(handle, dir, bh);
1671                         if (unlikely(err)) {
1672                                 ext4_std_error(dir->i_sb, err);
1673                                 return err;
1674                         }
1675                         return 0;
1676                 }
1677                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1678                 pde = de;
1679                 de = ext4_next_entry(de, blocksize);
1680         }
1681         return -ENOENT;
1682 }
1683
1684 /*
1685  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1686  * since this indicates that nlinks count was previously 1.
1687  */
1688 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1689 {
1690         inc_nlink(inode);
1691         if (is_dx(inode) && inode->i_nlink > 1) {
1692                 /* limit is 16-bit i_links_count */
1693                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1694                         set_nlink(inode, 1);
1695                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1696                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1697                 }
1698         }
1699 }
1700
1701 /*
1702  * If a directory had nlink == 1, then we should let it be 1. This indicates
1703  * directory has >EXT4_LINK_MAX subdirs.
1704  */
1705 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1706 {
1707         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
1708                 drop_nlink(inode);
1709 }
1710
1711
1712 static int ext4_add_nondir(handle_t *handle,
1713                 struct dentry *dentry, struct inode *inode)
1714 {
1715         int err = ext4_add_entry(handle, dentry, inode);
1716         if (!err) {
1717                 ext4_mark_inode_dirty(handle, inode);
1718                 d_instantiate(dentry, inode);
1719                 unlock_new_inode(inode);
1720                 return 0;
1721         }
1722         drop_nlink(inode);
1723         unlock_new_inode(inode);
1724         iput(inode);
1725         return err;
1726 }
1727
1728 /*
1729  * By the time this is called, we already have created
1730  * the directory cache entry for the new file, but it
1731  * is so far negative - it has no inode.
1732  *
1733  * If the create succeeds, we fill in the inode information
1734  * with d_instantiate().
1735  */
1736 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1737                        struct nameidata *nd)
1738 {
1739         handle_t *handle;
1740         struct inode *inode;
1741         int err, retries = 0;
1742
1743         dquot_initialize(dir);
1744
1745 retry:
1746         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1747                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1748                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1749         if (IS_ERR(handle))
1750                 return PTR_ERR(handle);
1751
1752         if (IS_DIRSYNC(dir))
1753                 ext4_handle_sync(handle);
1754
1755         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1756         err = PTR_ERR(inode);
1757         if (!IS_ERR(inode)) {
1758                 inode->i_op = &ext4_file_inode_operations;
1759                 inode->i_fop = &ext4_file_operations;
1760                 ext4_set_aops(inode);
1761                 err = ext4_add_nondir(handle, dentry, inode);
1762         }
1763         ext4_journal_stop(handle);
1764         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1765                 goto retry;
1766         return err;
1767 }
1768
1769 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1770                       umode_t mode, dev_t rdev)
1771 {
1772         handle_t *handle;
1773         struct inode *inode;
1774         int err, retries = 0;
1775
1776         if (!new_valid_dev(rdev))
1777                 return -EINVAL;
1778
1779         dquot_initialize(dir);
1780
1781 retry:
1782         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1783                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1784                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1785         if (IS_ERR(handle))
1786                 return PTR_ERR(handle);
1787
1788         if (IS_DIRSYNC(dir))
1789                 ext4_handle_sync(handle);
1790
1791         inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
1792         err = PTR_ERR(inode);
1793         if (!IS_ERR(inode)) {
1794                 init_special_inode(inode, inode->i_mode, rdev);
1795 #ifdef CONFIG_EXT4_FS_XATTR
1796                 inode->i_op = &ext4_special_inode_operations;
1797 #endif
1798                 err = ext4_add_nondir(handle, dentry, inode);
1799         }
1800         ext4_journal_stop(handle);
1801         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1802                 goto retry;
1803         return err;
1804 }
1805
1806 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1807 {
1808         handle_t *handle;
1809         struct inode *inode;
1810         struct buffer_head *dir_block = NULL;
1811         struct ext4_dir_entry_2 *de;
1812         unsigned int blocksize = dir->i_sb->s_blocksize;
1813         int err, retries = 0;
1814
1815         if (EXT4_DIR_LINK_MAX(dir))
1816                 return -EMLINK;
1817
1818         dquot_initialize(dir);
1819
1820 retry:
1821         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1822                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1823                                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1824         if (IS_ERR(handle))
1825                 return PTR_ERR(handle);
1826
1827         if (IS_DIRSYNC(dir))
1828                 ext4_handle_sync(handle);
1829
1830         inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1831                                &dentry->d_name, 0, NULL);
1832         err = PTR_ERR(inode);
1833         if (IS_ERR(inode))
1834                 goto out_stop;
1835
1836         inode->i_op = &ext4_dir_inode_operations;
1837         inode->i_fop = &ext4_dir_operations;
1838         inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1839         dir_block = ext4_bread(handle, inode, 0, 1, &err);
1840         if (!dir_block)
1841                 goto out_clear_inode;
1842         BUFFER_TRACE(dir_block, "get_write_access");
1843         err = ext4_journal_get_write_access(handle, dir_block);
1844         if (err)
1845                 goto out_clear_inode;
1846         de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1847         de->inode = cpu_to_le32(inode->i_ino);
1848         de->name_len = 1;
1849         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1850                                            blocksize);
1851         strcpy(de->name, ".");
1852         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1853         de = ext4_next_entry(de, blocksize);
1854         de->inode = cpu_to_le32(dir->i_ino);
1855         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1856                                            blocksize);
1857         de->name_len = 2;
1858         strcpy(de->name, "..");
1859         ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1860         set_nlink(inode, 2);
1861         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1862         err = ext4_handle_dirty_metadata(handle, inode, dir_block);
1863         if (err)
1864                 goto out_clear_inode;
1865         err = ext4_mark_inode_dirty(handle, inode);
1866         if (!err)
1867                 err = ext4_add_entry(handle, dentry, inode);
1868         if (err) {
1869 out_clear_inode:
1870                 clear_nlink(inode);
1871                 unlock_new_inode(inode);
1872                 ext4_mark_inode_dirty(handle, inode);
1873                 iput(inode);
1874                 goto out_stop;
1875         }
1876         ext4_inc_count(handle, dir);
1877         ext4_update_dx_flag(dir);
1878         err = ext4_mark_inode_dirty(handle, dir);
1879         if (err)
1880                 goto out_clear_inode;
1881         d_instantiate(dentry, inode);
1882         unlock_new_inode(inode);
1883 out_stop:
1884         brelse(dir_block);
1885         ext4_journal_stop(handle);
1886         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1887                 goto retry;
1888         return err;
1889 }
1890
1891 /*
1892  * routine to check that the specified directory is empty (for rmdir)
1893  */
1894 static int empty_dir(struct inode *inode)
1895 {
1896         unsigned int offset;
1897         struct buffer_head *bh;
1898         struct ext4_dir_entry_2 *de, *de1;
1899         struct super_block *sb;
1900         int err = 0;
1901
1902         sb = inode->i_sb;
1903         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1904             !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1905                 if (err)
1906                         EXT4_ERROR_INODE(inode,
1907                                 "error %d reading directory lblock 0", err);
1908                 else
1909                         ext4_warning(inode->i_sb,
1910                                      "bad directory (dir #%lu) - no data block",
1911                                      inode->i_ino);
1912                 return 1;
1913         }
1914         de = (struct ext4_dir_entry_2 *) bh->b_data;
1915         de1 = ext4_next_entry(de, sb->s_blocksize);
1916         if (le32_to_cpu(de->inode) != inode->i_ino ||
1917                         !le32_to_cpu(de1->inode) ||
1918                         strcmp(".", de->name) ||
1919                         strcmp("..", de1->name)) {
1920                 ext4_warning(inode->i_sb,
1921                              "bad directory (dir #%lu) - no `.' or `..'",
1922                              inode->i_ino);
1923                 brelse(bh);
1924                 return 1;
1925         }
1926         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1927                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1928         de = ext4_next_entry(de1, sb->s_blocksize);
1929         while (offset < inode->i_size) {
1930                 if (!bh ||
1931                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1932                         unsigned int lblock;
1933                         err = 0;
1934                         brelse(bh);
1935                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1936                         bh = ext4_bread(NULL, inode, lblock, 0, &err);
1937                         if (!bh) {
1938                                 if (err)
1939                                         EXT4_ERROR_INODE(inode,
1940                                                 "error %d reading directory "
1941                                                 "lblock %u", err, lblock);
1942                                 offset += sb->s_blocksize;
1943                                 continue;
1944                         }
1945                         de = (struct ext4_dir_entry_2 *) bh->b_data;
1946                 }
1947                 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1948                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
1949                                                          sb->s_blocksize);
1950                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1951                         continue;
1952                 }
1953                 if (le32_to_cpu(de->inode)) {
1954                         brelse(bh);
1955                         return 0;
1956                 }
1957                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1958                 de = ext4_next_entry(de, sb->s_blocksize);
1959         }
1960         brelse(bh);
1961         return 1;
1962 }
1963
1964 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1965  * such inodes, starting at the superblock, in case we crash before the
1966  * file is closed/deleted, or in case the inode truncate spans multiple
1967  * transactions and the last transaction is not recovered after a crash.
1968  *
1969  * At filesystem recovery time, we walk this list deleting unlinked
1970  * inodes and truncating linked inodes in ext4_orphan_cleanup().
1971  */
1972 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1973 {
1974         struct super_block *sb = inode->i_sb;
1975         struct ext4_iloc iloc;
1976         int err = 0, rc;
1977
1978         if (!ext4_handle_valid(handle))
1979                 return 0;
1980
1981         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1982         if (!list_empty(&EXT4_I(inode)->i_orphan))
1983                 goto out_unlock;
1984
1985         /*
1986          * Orphan handling is only valid for files with data blocks
1987          * being truncated, or files being unlinked. Note that we either
1988          * hold i_mutex, or the inode can not be referenced from outside,
1989          * so i_nlink should not be bumped due to race
1990          */
1991         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1992                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1993
1994         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1995         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1996         if (err)
1997                 goto out_unlock;
1998
1999         err = ext4_reserve_inode_write(handle, inode, &iloc);
2000         if (err)
2001                 goto out_unlock;
2002         /*
2003          * Due to previous errors inode may be already a part of on-disk
2004          * orphan list. If so skip on-disk list modification.
2005          */
2006         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2007                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2008                         goto mem_insert;
2009
2010         /* Insert this inode at the head of the on-disk orphan list... */
2011         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2012         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2013         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2014         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2015         if (!err)
2016                 err = rc;
2017
2018         /* Only add to the head of the in-memory list if all the
2019          * previous operations succeeded.  If the orphan_add is going to
2020          * fail (possibly taking the journal offline), we can't risk
2021          * leaving the inode on the orphan list: stray orphan-list
2022          * entries can cause panics at unmount time.
2023          *
2024          * This is safe: on error we're going to ignore the orphan list
2025          * anyway on the next recovery. */
2026 mem_insert:
2027         if (!err)
2028                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2029
2030         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2031         jbd_debug(4, "orphan inode %lu will point to %d\n",
2032                         inode->i_ino, NEXT_ORPHAN(inode));
2033 out_unlock:
2034         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2035         ext4_std_error(inode->i_sb, err);
2036         return err;
2037 }
2038
2039 /*
2040  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2041  * of such inodes stored on disk, because it is finally being cleaned up.
2042  */
2043 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2044 {
2045         struct list_head *prev;
2046         struct ext4_inode_info *ei = EXT4_I(inode);
2047         struct ext4_sb_info *sbi;
2048         __u32 ino_next;
2049         struct ext4_iloc iloc;
2050         int err = 0;
2051
2052         /* ext4_handle_valid() assumes a valid handle_t pointer */
2053         if (handle && !ext4_handle_valid(handle))
2054                 return 0;
2055
2056         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2057         if (list_empty(&ei->i_orphan))
2058                 goto out;
2059
2060         ino_next = NEXT_ORPHAN(inode);
2061         prev = ei->i_orphan.prev;
2062         sbi = EXT4_SB(inode->i_sb);
2063
2064         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2065
2066         list_del_init(&ei->i_orphan);
2067
2068         /* If we're on an error path, we may not have a valid
2069          * transaction handle with which to update the orphan list on
2070          * disk, but we still need to remove the inode from the linked
2071          * list in memory. */
2072         if (sbi->s_journal && !handle)
2073                 goto out;
2074
2075         err = ext4_reserve_inode_write(handle, inode, &iloc);
2076         if (err)
2077                 goto out_err;
2078
2079         if (prev == &sbi->s_orphan) {
2080                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2081                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2082                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2083                 if (err)
2084                         goto out_brelse;
2085                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2086                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2087         } else {
2088                 struct ext4_iloc iloc2;
2089                 struct inode *i_prev =
2090                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2091
2092                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2093                           i_prev->i_ino, ino_next);
2094                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2095                 if (err)
2096                         goto out_brelse;
2097                 NEXT_ORPHAN(i_prev) = ino_next;
2098                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2099         }
2100         if (err)
2101                 goto out_brelse;
2102         NEXT_ORPHAN(inode) = 0;
2103         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2104
2105 out_err:
2106         ext4_std_error(inode->i_sb, err);
2107 out:
2108         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2109         return err;
2110
2111 out_brelse:
2112         brelse(iloc.bh);
2113         goto out_err;
2114 }
2115
2116 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2117 {
2118         int retval;
2119         struct inode *inode;
2120         struct buffer_head *bh;
2121         struct ext4_dir_entry_2 *de;
2122         handle_t *handle;
2123
2124         /* Initialize quotas before so that eventual writes go in
2125          * separate transaction */
2126         dquot_initialize(dir);
2127         dquot_initialize(dentry->d_inode);
2128
2129         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2130         if (IS_ERR(handle))
2131                 return PTR_ERR(handle);
2132
2133         retval = -ENOENT;
2134         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2135         if (!bh)
2136                 goto end_rmdir;
2137
2138         if (IS_DIRSYNC(dir))
2139                 ext4_handle_sync(handle);
2140
2141         inode = dentry->d_inode;
2142
2143         retval = -EIO;
2144         if (le32_to_cpu(de->inode) != inode->i_ino)
2145                 goto end_rmdir;
2146
2147         retval = -ENOTEMPTY;
2148         if (!empty_dir(inode))
2149                 goto end_rmdir;
2150
2151         retval = ext4_delete_entry(handle, dir, de, bh);
2152         if (retval)
2153                 goto end_rmdir;
2154         if (!EXT4_DIR_LINK_EMPTY(inode))
2155                 ext4_warning(inode->i_sb,
2156                              "empty directory has too many links (%d)",
2157                              inode->i_nlink);
2158         inode->i_version++;
2159         clear_nlink(inode);
2160         /* There's no need to set i_disksize: the fact that i_nlink is
2161          * zero will ensure that the right thing happens during any
2162          * recovery. */
2163         inode->i_size = 0;
2164         ext4_orphan_add(handle, inode);
2165         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2166         ext4_mark_inode_dirty(handle, inode);
2167         ext4_dec_count(handle, dir);
2168         ext4_update_dx_flag(dir);
2169         ext4_mark_inode_dirty(handle, dir);
2170
2171 end_rmdir:
2172         ext4_journal_stop(handle);
2173         brelse(bh);
2174         return retval;
2175 }
2176
2177 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2178 {
2179         int retval;
2180         struct inode *inode;
2181         struct buffer_head *bh;
2182         struct ext4_dir_entry_2 *de;
2183         handle_t *handle;
2184
2185         trace_ext4_unlink_enter(dir, dentry);
2186         /* Initialize quotas before so that eventual writes go
2187          * in separate transaction */
2188         dquot_initialize(dir);
2189         dquot_initialize(dentry->d_inode);
2190
2191         handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2192         if (IS_ERR(handle))
2193                 return PTR_ERR(handle);
2194
2195         if (IS_DIRSYNC(dir))
2196                 ext4_handle_sync(handle);
2197
2198         retval = -ENOENT;
2199         bh = ext4_find_entry(dir, &dentry->d_name, &de);
2200         if (!bh)
2201                 goto end_unlink;
2202
2203         inode = dentry->d_inode;
2204
2205         retval = -EIO;
2206         if (le32_to_cpu(de->inode) != inode->i_ino)
2207                 goto end_unlink;
2208
2209         if (!inode->i_nlink) {
2210                 ext4_warning(inode->i_sb,
2211                              "Deleting nonexistent file (%lu), %d",
2212                              inode->i_ino, inode->i_nlink);
2213                 set_nlink(inode, 1);
2214         }
2215         retval = ext4_delete_entry(handle, dir, de, bh);
2216         if (retval)
2217                 goto end_unlink;
2218         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2219         ext4_update_dx_flag(dir);
2220         ext4_mark_inode_dirty(handle, dir);
2221         drop_nlink(inode);
2222         if (!inode->i_nlink)
2223                 ext4_orphan_add(handle, inode);
2224         inode->i_ctime = ext4_current_time(inode);
2225         ext4_mark_inode_dirty(handle, inode);
2226         retval = 0;
2227
2228 end_unlink:
2229         ext4_journal_stop(handle);
2230         brelse(bh);
2231         trace_ext4_unlink_exit(dentry, retval);
2232         return retval;
2233 }
2234
2235 static int ext4_symlink(struct inode *dir,
2236                         struct dentry *dentry, const char *symname)
2237 {
2238         handle_t *handle;
2239         struct inode *inode;
2240         int l, err, retries = 0;
2241         int credits;
2242
2243         l = strlen(symname)+1;
2244         if (l > dir->i_sb->s_blocksize)
2245                 return -ENAMETOOLONG;
2246
2247         dquot_initialize(dir);
2248
2249         if (l > EXT4_N_BLOCKS * 4) {
2250                 /*
2251                  * For non-fast symlinks, we just allocate inode and put it on
2252                  * orphan list in the first transaction => we need bitmap,
2253                  * group descriptor, sb, inode block, quota blocks, and
2254                  * possibly selinux xattr blocks.
2255                  */
2256                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2257                           EXT4_XATTR_TRANS_BLOCKS;
2258         } else {
2259                 /*
2260                  * Fast symlink. We have to add entry to directory
2261                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2262                  * allocate new inode (bitmap, group descriptor, inode block,
2263                  * quota blocks, sb is already counted in previous macros).
2264                  */
2265                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2266                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2267                           EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2268         }
2269 retry:
2270         handle = ext4_journal_start(dir, credits);
2271         if (IS_ERR(handle))
2272                 return PTR_ERR(handle);
2273
2274         if (IS_DIRSYNC(dir))
2275                 ext4_handle_sync(handle);
2276
2277         inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2278                                &dentry->d_name, 0, NULL);
2279         err = PTR_ERR(inode);
2280         if (IS_ERR(inode))
2281                 goto out_stop;
2282
2283         if (l > EXT4_N_BLOCKS * 4) {
2284                 inode->i_op = &ext4_symlink_inode_operations;
2285                 ext4_set_aops(inode);
2286                 /*
2287                  * We cannot call page_symlink() with transaction started
2288                  * because it calls into ext4_write_begin() which can wait
2289                  * for transaction commit if we are running out of space
2290                  * and thus we deadlock. So we have to stop transaction now
2291                  * and restart it when symlink contents is written.
2292                  * 
2293                  * To keep fs consistent in case of crash, we have to put inode
2294                  * to orphan list in the mean time.
2295                  */
2296                 drop_nlink(inode);
2297                 err = ext4_orphan_add(handle, inode);
2298                 ext4_journal_stop(handle);
2299                 if (err)
2300                         goto err_drop_inode;
2301                 err = __page_symlink(inode, symname, l, 1);
2302                 if (err)
2303                         goto err_drop_inode;
2304                 /*
2305                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2306                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2307                  */
2308                 handle = ext4_journal_start(dir,
2309                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2310                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2311                 if (IS_ERR(handle)) {
2312                         err = PTR_ERR(handle);
2313                         goto err_drop_inode;
2314                 }
2315                 set_nlink(inode, 1);
2316                 err = ext4_orphan_del(handle, inode);
2317                 if (err) {
2318                         ext4_journal_stop(handle);
2319                         clear_nlink(inode);
2320                         goto err_drop_inode;
2321                 }
2322         } else {
2323                 /* clear the extent format for fast symlink */
2324                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2325                 inode->i_op = &ext4_fast_symlink_inode_operations;
2326                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2327                 inode->i_size = l-1;
2328         }
2329         EXT4_I(inode)->i_disksize = inode->i_size;
2330         err = ext4_add_nondir(handle, dentry, inode);
2331 out_stop:
2332         ext4_journal_stop(handle);
2333         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2334                 goto retry;
2335         return err;
2336 err_drop_inode:
2337         unlock_new_inode(inode);
2338         iput(inode);
2339         return err;
2340 }
2341
2342 static int ext4_link(struct dentry *old_dentry,
2343                      struct inode *dir, struct dentry *dentry)
2344 {
2345         handle_t *handle;
2346         struct inode *inode = old_dentry->d_inode;
2347         int err, retries = 0;
2348
2349         if (inode->i_nlink >= EXT4_LINK_MAX)
2350                 return -EMLINK;
2351
2352         dquot_initialize(dir);
2353
2354 retry:
2355         handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2356                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2357         if (IS_ERR(handle))
2358                 return PTR_ERR(handle);
2359
2360         if (IS_DIRSYNC(dir))
2361                 ext4_handle_sync(handle);
2362
2363         inode->i_ctime = ext4_current_time(inode);
2364         ext4_inc_count(handle, inode);
2365         ihold(inode);
2366
2367         err = ext4_add_entry(handle, dentry, inode);
2368         if (!err) {
2369                 ext4_mark_inode_dirty(handle, inode);
2370                 d_instantiate(dentry, inode);
2371         } else {
2372                 drop_nlink(inode);
2373                 iput(inode);
2374         }
2375         ext4_journal_stop(handle);
2376         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2377                 goto retry;
2378         return err;
2379 }
2380
2381 #define PARENT_INO(buffer, size) \
2382         (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2383
2384 /*
2385  * Anybody can rename anything with this: the permission checks are left to the
2386  * higher-level routines.
2387  */
2388 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2389                        struct inode *new_dir, struct dentry *new_dentry)
2390 {
2391         handle_t *handle;
2392         struct inode *old_inode, *new_inode;
2393         struct buffer_head *old_bh, *new_bh, *dir_bh;
2394         struct ext4_dir_entry_2 *old_de, *new_de;
2395         int retval, force_da_alloc = 0;
2396
2397         dquot_initialize(old_dir);
2398         dquot_initialize(new_dir);
2399
2400         old_bh = new_bh = dir_bh = NULL;
2401
2402         /* Initialize quotas before so that eventual writes go
2403          * in separate transaction */
2404         if (new_dentry->d_inode)
2405                 dquot_initialize(new_dentry->d_inode);
2406         handle = ext4_journal_start(old_dir, 2 *
2407                                         EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2408                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2409         if (IS_ERR(handle))
2410                 return PTR_ERR(handle);
2411
2412         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2413                 ext4_handle_sync(handle);
2414
2415         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2416         /*
2417          *  Check for inode number is _not_ due to possible IO errors.
2418          *  We might rmdir the source, keep it as pwd of some process
2419          *  and merrily kill the link to whatever was created under the
2420          *  same name. Goodbye sticky bit ;-<
2421          */
2422         old_inode = old_dentry->d_inode;
2423         retval = -ENOENT;
2424         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2425                 goto end_rename;
2426
2427         new_inode = new_dentry->d_inode;
2428         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2429         if (new_bh) {
2430                 if (!new_inode) {
2431                         brelse(new_bh);
2432                         new_bh = NULL;
2433                 }
2434         }
2435         if (S_ISDIR(old_inode->i_mode)) {
2436                 if (new_inode) {
2437                         retval = -ENOTEMPTY;
2438                         if (!empty_dir(new_inode))
2439                                 goto end_rename;
2440                 }
2441                 retval = -EIO;
2442                 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2443                 if (!dir_bh)
2444                         goto end_rename;
2445                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2446                                 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2447                         goto end_rename;
2448                 retval = -EMLINK;
2449                 if (!new_inode && new_dir != old_dir &&
2450                     EXT4_DIR_LINK_MAX(new_dir))
2451                         goto end_rename;
2452                 BUFFER_TRACE(dir_bh, "get_write_access");
2453                 retval = ext4_journal_get_write_access(handle, dir_bh);
2454                 if (retval)
2455                         goto end_rename;
2456         }
2457         if (!new_bh) {
2458                 retval = ext4_add_entry(handle, new_dentry, old_inode);
2459                 if (retval)
2460                         goto end_rename;
2461         } else {
2462                 BUFFER_TRACE(new_bh, "get write access");
2463                 retval = ext4_journal_get_write_access(handle, new_bh);
2464                 if (retval)
2465                         goto end_rename;
2466                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2467                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2468                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
2469                         new_de->file_type = old_de->file_type;
2470                 new_dir->i_version++;
2471                 new_dir->i_ctime = new_dir->i_mtime =
2472                                         ext4_current_time(new_dir);
2473                 ext4_mark_inode_dirty(handle, new_dir);
2474                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2475                 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2476                 if (unlikely(retval)) {
2477                         ext4_std_error(new_dir->i_sb, retval);
2478                         goto end_rename;
2479                 }
2480                 brelse(new_bh);
2481                 new_bh = NULL;
2482         }
2483
2484         /*
2485          * Like most other Unix systems, set the ctime for inodes on a
2486          * rename.
2487          */
2488         old_inode->i_ctime = ext4_current_time(old_inode);
2489         ext4_mark_inode_dirty(handle, old_inode);
2490
2491         /*
2492          * ok, that's it
2493          */
2494         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2495             old_de->name_len != old_dentry->d_name.len ||
2496             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2497             (retval = ext4_delete_entry(handle, old_dir,
2498                                         old_de, old_bh)) == -ENOENT) {
2499                 /* old_de could have moved from under us during htree split, so
2500                  * make sure that we are deleting the right entry.  We might
2501                  * also be pointing to a stale entry in the unused part of
2502                  * old_bh so just checking inum and the name isn't enough. */
2503                 struct buffer_head *old_bh2;
2504                 struct ext4_dir_entry_2 *old_de2;
2505
2506                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2507                 if (old_bh2) {
2508                         retval = ext4_delete_entry(handle, old_dir,
2509                                                    old_de2, old_bh2);
2510                         brelse(old_bh2);
2511                 }
2512         }
2513         if (retval) {
2514                 ext4_warning(old_dir->i_sb,
2515                                 "Deleting old file (%lu), %d, error=%d",
2516                                 old_dir->i_ino, old_dir->i_nlink, retval);
2517         }
2518
2519         if (new_inode) {
2520                 ext4_dec_count(handle, new_inode);
2521                 new_inode->i_ctime = ext4_current_time(new_inode);
2522         }
2523         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2524         ext4_update_dx_flag(old_dir);
2525         if (dir_bh) {
2526                 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2527                                                 cpu_to_le32(new_dir->i_ino);
2528                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2529                 retval = ext4_handle_dirty_metadata(handle, old_inode, dir_bh);
2530                 if (retval) {
2531                         ext4_std_error(old_dir->i_sb, retval);
2532                         goto end_rename;
2533                 }
2534                 ext4_dec_count(handle, old_dir);
2535                 if (new_inode) {
2536                         /* checked empty_dir above, can't have another parent,
2537                          * ext4_dec_count() won't work for many-linked dirs */
2538                         clear_nlink(new_inode);
2539                 } else {
2540                         ext4_inc_count(handle, new_dir);
2541                         ext4_update_dx_flag(new_dir);
2542                         ext4_mark_inode_dirty(handle, new_dir);
2543                 }
2544         }
2545         ext4_mark_inode_dirty(handle, old_dir);
2546         if (new_inode) {
2547                 ext4_mark_inode_dirty(handle, new_inode);
2548                 if (!new_inode->i_nlink)
2549                         ext4_orphan_add(handle, new_inode);
2550                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2551                         force_da_alloc = 1;
2552         }
2553         retval = 0;
2554
2555 end_rename:
2556         brelse(dir_bh);
2557         brelse(old_bh);
2558         brelse(new_bh);
2559         ext4_journal_stop(handle);
2560         if (retval == 0 && force_da_alloc)
2561                 ext4_alloc_da_blocks(old_inode);
2562         return retval;
2563 }
2564
2565 /*
2566  * directories can handle most operations...
2567  */
2568 const struct inode_operations ext4_dir_inode_operations = {
2569         .create         = ext4_create,
2570         .lookup         = ext4_lookup,
2571         .link           = ext4_link,
2572         .unlink         = ext4_unlink,
2573         .symlink        = ext4_symlink,
2574         .mkdir          = ext4_mkdir,
2575         .rmdir          = ext4_rmdir,
2576         .mknod          = ext4_mknod,
2577         .rename         = ext4_rename,
2578         .setattr        = ext4_setattr,
2579 #ifdef CONFIG_EXT4_FS_XATTR
2580         .setxattr       = generic_setxattr,
2581         .getxattr       = generic_getxattr,
2582         .listxattr      = ext4_listxattr,
2583         .removexattr    = generic_removexattr,
2584 #endif
2585         .get_acl        = ext4_get_acl,
2586         .fiemap         = ext4_fiemap,
2587 };
2588
2589 const struct inode_operations ext4_special_inode_operations = {
2590         .setattr        = ext4_setattr,
2591 #ifdef CONFIG_EXT4_FS_XATTR
2592         .setxattr       = generic_setxattr,
2593         .getxattr       = generic_getxattr,
2594         .listxattr      = ext4_listxattr,
2595         .removexattr    = generic_removexattr,
2596 #endif
2597         .get_acl        = ext4_get_acl,
2598 };