Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[~shefty/rdma-dev.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
51 static struct buffer_head *ext4_append(handle_t *handle,
52                                         struct inode *inode,
53                                         ext4_lblk_t *block)
54 {
55         struct buffer_head *bh;
56         int err = 0;
57
58         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59                      ((inode->i_size >> 10) >=
60                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61                 return ERR_PTR(-ENOSPC);
62
63         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64
65         bh = ext4_bread(handle, inode, *block, 1, &err);
66         if (!bh)
67                 return ERR_PTR(err);
68         inode->i_size += inode->i_sb->s_blocksize;
69         EXT4_I(inode)->i_disksize = inode->i_size;
70         err = ext4_journal_get_write_access(handle, bh);
71         if (err) {
72                 brelse(bh);
73                 ext4_std_error(inode->i_sb, err);
74                 return ERR_PTR(err);
75         }
76         return bh;
77 }
78
79 static int ext4_dx_csum_verify(struct inode *inode,
80                                struct ext4_dir_entry *dirent);
81
82 typedef enum {
83         EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85
86 #define ext4_read_dirblock(inode, block, type) \
87         __ext4_read_dirblock((inode), (block), (type), __LINE__)
88
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90                                               ext4_lblk_t block,
91                                               dirblock_type_t type,
92                                               unsigned int line)
93 {
94         struct buffer_head *bh;
95         struct ext4_dir_entry *dirent;
96         int err = 0, is_dx_block = 0;
97
98         bh = ext4_bread(NULL, inode, block, 0, &err);
99         if (!bh) {
100                 if (err == 0) {
101                         ext4_error_inode(inode, __func__, line, block,
102                                                "Directory hole found");
103                         return ERR_PTR(-EIO);
104                 }
105                 __ext4_warning(inode->i_sb, __func__, line,
106                                "error reading directory block "
107                                "(ino %lu, block %lu)", inode->i_ino,
108                                (unsigned long) block);
109                 return ERR_PTR(err);
110         }
111         dirent = (struct ext4_dir_entry *) bh->b_data;
112         /* Determine whether or not we have an index block */
113         if (is_dx(inode)) {
114                 if (block == 0)
115                         is_dx_block = 1;
116                 else if (ext4_rec_len_from_disk(dirent->rec_len,
117                                                 inode->i_sb->s_blocksize) ==
118                          inode->i_sb->s_blocksize)
119                         is_dx_block = 1;
120         }
121         if (!is_dx_block && type == INDEX) {
122                 ext4_error_inode(inode, __func__, line, block,
123                        "directory leaf block found instead of index block");
124                 return ERR_PTR(-EIO);
125         }
126         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
127                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
128             buffer_verified(bh))
129                 return bh;
130
131         /*
132          * An empty leaf block can get mistaken for a index block; for
133          * this reason, we can only check the index checksum when the
134          * caller is sure it should be an index block.
135          */
136         if (is_dx_block && type == INDEX) {
137                 if (ext4_dx_csum_verify(inode, dirent))
138                         set_buffer_verified(bh);
139                 else {
140                         ext4_error_inode(inode, __func__, line, block,
141                                 "Directory index failed checksum");
142                         brelse(bh);
143                         return ERR_PTR(-EIO);
144                 }
145         }
146         if (!is_dx_block) {
147                 if (ext4_dirent_csum_verify(inode, dirent))
148                         set_buffer_verified(bh);
149                 else {
150                         ext4_error_inode(inode, __func__, line, block,
151                                 "Directory block failed checksum");
152                         brelse(bh);
153                         return ERR_PTR(-EIO);
154                 }
155         }
156         return bh;
157 }
158
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168
169 struct fake_dirent
170 {
171         __le32 inode;
172         __le16 rec_len;
173         u8 name_len;
174         u8 file_type;
175 };
176
177 struct dx_countlimit
178 {
179         __le16 limit;
180         __le16 count;
181 };
182
183 struct dx_entry
184 {
185         __le32 hash;
186         __le32 block;
187 };
188
189 /*
190  * dx_root_info is laid out so that if it should somehow get overlaid by a
191  * dirent the two low bits of the hash version will be zero.  Therefore, the
192  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
193  */
194
195 struct dx_root
196 {
197         struct fake_dirent dot;
198         char dot_name[4];
199         struct fake_dirent dotdot;
200         char dotdot_name[4];
201         struct dx_root_info
202         {
203                 __le32 reserved_zero;
204                 u8 hash_version;
205                 u8 info_length; /* 8 */
206                 u8 indirect_levels;
207                 u8 unused_flags;
208         }
209         info;
210         struct dx_entry entries[0];
211 };
212
213 struct dx_node
214 {
215         struct fake_dirent fake;
216         struct dx_entry entries[0];
217 };
218
219
220 struct dx_frame
221 {
222         struct buffer_head *bh;
223         struct dx_entry *entries;
224         struct dx_entry *at;
225 };
226
227 struct dx_map_entry
228 {
229         u32 hash;
230         u16 offs;
231         u16 size;
232 };
233
234 /*
235  * This goes at the end of each htree block.
236  */
237 struct dx_tail {
238         u32 dt_reserved;
239         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
240 };
241
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253                                  struct inode *dir,
254                                  struct dx_hash_info *hinfo,
255                                  struct dx_frame *frame,
256                                  int *err);
257 static void dx_release(struct dx_frame *frames);
258 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
259                        struct dx_hash_info *hinfo, struct dx_map_entry map[]);
260 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
261 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
262                 struct dx_map_entry *offsets, int count, unsigned blocksize);
263 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
264 static void dx_insert_block(struct dx_frame *frame,
265                                         u32 hash, ext4_lblk_t block);
266 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
267                                  struct dx_frame *frame,
268                                  struct dx_frame *frames,
269                                  __u32 *start_hash);
270 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
271                 const struct qstr *d_name,
272                 struct ext4_dir_entry_2 **res_dir,
273                 int *err);
274 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
275                              struct inode *inode);
276
277 /* checksumming functions */
278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
279                             unsigned int blocksize)
280 {
281         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
282         t->det_rec_len = ext4_rec_len_to_disk(
283                         sizeof(struct ext4_dir_entry_tail), blocksize);
284         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
285 }
286
287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
289                                                    struct ext4_dir_entry *de)
290 {
291         struct ext4_dir_entry_tail *t;
292
293 #ifdef PARANOID
294         struct ext4_dir_entry *d, *top;
295
296         d = de;
297         top = (struct ext4_dir_entry *)(((void *)de) +
298                 (EXT4_BLOCK_SIZE(inode->i_sb) -
299                 sizeof(struct ext4_dir_entry_tail)));
300         while (d < top && d->rec_len)
301                 d = (struct ext4_dir_entry *)(((void *)d) +
302                     le16_to_cpu(d->rec_len));
303
304         if (d != top)
305                 return NULL;
306
307         t = (struct ext4_dir_entry_tail *)d;
308 #else
309         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
310 #endif
311
312         if (t->det_reserved_zero1 ||
313             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
314             t->det_reserved_zero2 ||
315             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
316                 return NULL;
317
318         return t;
319 }
320
321 static __le32 ext4_dirent_csum(struct inode *inode,
322                                struct ext4_dir_entry *dirent, int size)
323 {
324         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
325         struct ext4_inode_info *ei = EXT4_I(inode);
326         __u32 csum;
327
328         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
329         return cpu_to_le32(csum);
330 }
331
332 static void warn_no_space_for_csum(struct inode *inode)
333 {
334         ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
335                      "checksum.  Please run e2fsck -D.", inode->i_ino);
336 }
337
338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
339 {
340         struct ext4_dir_entry_tail *t;
341
342         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
343                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
344                 return 1;
345
346         t = get_dirent_tail(inode, dirent);
347         if (!t) {
348                 warn_no_space_for_csum(inode);
349                 return 0;
350         }
351
352         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
353                                                 (void *)t - (void *)dirent))
354                 return 0;
355
356         return 1;
357 }
358
359 static void ext4_dirent_csum_set(struct inode *inode,
360                                  struct ext4_dir_entry *dirent)
361 {
362         struct ext4_dir_entry_tail *t;
363
364         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
365                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
366                 return;
367
368         t = get_dirent_tail(inode, dirent);
369         if (!t) {
370                 warn_no_space_for_csum(inode);
371                 return;
372         }
373
374         t->det_checksum = ext4_dirent_csum(inode, dirent,
375                                            (void *)t - (void *)dirent);
376 }
377
378 int ext4_handle_dirty_dirent_node(handle_t *handle,
379                                   struct inode *inode,
380                                   struct buffer_head *bh)
381 {
382         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
383         return ext4_handle_dirty_metadata(handle, inode, bh);
384 }
385
386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
387                                                struct ext4_dir_entry *dirent,
388                                                int *offset)
389 {
390         struct ext4_dir_entry *dp;
391         struct dx_root_info *root;
392         int count_offset;
393
394         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
395                 count_offset = 8;
396         else if (le16_to_cpu(dirent->rec_len) == 12) {
397                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
398                 if (le16_to_cpu(dp->rec_len) !=
399                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
400                         return NULL;
401                 root = (struct dx_root_info *)(((void *)dp + 12));
402                 if (root->reserved_zero ||
403                     root->info_length != sizeof(struct dx_root_info))
404                         return NULL;
405                 count_offset = 32;
406         } else
407                 return NULL;
408
409         if (offset)
410                 *offset = count_offset;
411         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
412 }
413
414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
415                            int count_offset, int count, struct dx_tail *t)
416 {
417         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
418         struct ext4_inode_info *ei = EXT4_I(inode);
419         __u32 csum, old_csum;
420         int size;
421
422         size = count_offset + (count * sizeof(struct dx_entry));
423         old_csum = t->dt_checksum;
424         t->dt_checksum = 0;
425         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
426         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
427         t->dt_checksum = old_csum;
428
429         return cpu_to_le32(csum);
430 }
431
432 static int ext4_dx_csum_verify(struct inode *inode,
433                                struct ext4_dir_entry *dirent)
434 {
435         struct dx_countlimit *c;
436         struct dx_tail *t;
437         int count_offset, limit, count;
438
439         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
440                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
441                 return 1;
442
443         c = get_dx_countlimit(inode, dirent, &count_offset);
444         if (!c) {
445                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
446                 return 1;
447         }
448         limit = le16_to_cpu(c->limit);
449         count = le16_to_cpu(c->count);
450         if (count_offset + (limit * sizeof(struct dx_entry)) >
451             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
452                 warn_no_space_for_csum(inode);
453                 return 1;
454         }
455         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
456
457         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
458                                             count, t))
459                 return 0;
460         return 1;
461 }
462
463 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
464 {
465         struct dx_countlimit *c;
466         struct dx_tail *t;
467         int count_offset, limit, count;
468
469         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
470                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
471                 return;
472
473         c = get_dx_countlimit(inode, dirent, &count_offset);
474         if (!c) {
475                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
476                 return;
477         }
478         limit = le16_to_cpu(c->limit);
479         count = le16_to_cpu(c->count);
480         if (count_offset + (limit * sizeof(struct dx_entry)) >
481             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
482                 warn_no_space_for_csum(inode);
483                 return;
484         }
485         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
486
487         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
488 }
489
490 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
491                                             struct inode *inode,
492                                             struct buffer_head *bh)
493 {
494         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
495         return ext4_handle_dirty_metadata(handle, inode, bh);
496 }
497
498 /*
499  * p is at least 6 bytes before the end of page
500  */
501 static inline struct ext4_dir_entry_2 *
502 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
503 {
504         return (struct ext4_dir_entry_2 *)((char *)p +
505                 ext4_rec_len_from_disk(p->rec_len, blocksize));
506 }
507
508 /*
509  * Future: use high four bits of block for coalesce-on-delete flags
510  * Mask them off for now.
511  */
512
513 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
514 {
515         return le32_to_cpu(entry->block) & 0x00ffffff;
516 }
517
518 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
519 {
520         entry->block = cpu_to_le32(value);
521 }
522
523 static inline unsigned dx_get_hash(struct dx_entry *entry)
524 {
525         return le32_to_cpu(entry->hash);
526 }
527
528 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
529 {
530         entry->hash = cpu_to_le32(value);
531 }
532
533 static inline unsigned dx_get_count(struct dx_entry *entries)
534 {
535         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
536 }
537
538 static inline unsigned dx_get_limit(struct dx_entry *entries)
539 {
540         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
541 }
542
543 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
544 {
545         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
546 }
547
548 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
549 {
550         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
551 }
552
553 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
554 {
555         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
556                 EXT4_DIR_REC_LEN(2) - infosize;
557
558         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
559                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
560                 entry_space -= sizeof(struct dx_tail);
561         return entry_space / sizeof(struct dx_entry);
562 }
563
564 static inline unsigned dx_node_limit(struct inode *dir)
565 {
566         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567
568         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
569                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
570                 entry_space -= sizeof(struct dx_tail);
571         return entry_space / sizeof(struct dx_entry);
572 }
573
574 /*
575  * Debug
576  */
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
579 {
580         int i, n = dx_get_count (entries);
581         printk(KERN_DEBUG "%s index ", label);
582         for (i = 0; i < n; i++) {
583                 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584                                 0, (unsigned long)dx_get_block(entries + i));
585         }
586         printk("\n");
587 }
588
589 struct stats
590 {
591         unsigned names;
592         unsigned space;
593         unsigned bcount;
594 };
595
596 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
597                                  int size, int show_names)
598 {
599         unsigned names = 0, space = 0;
600         char *base = (char *) de;
601         struct dx_hash_info h = *hinfo;
602
603         printk("names: ");
604         while ((char *) de < base + size)
605         {
606                 if (de->inode)
607                 {
608                         if (show_names)
609                         {
610                                 int len = de->name_len;
611                                 char *name = de->name;
612                                 while (len--) printk("%c", *name++);
613                                 ext4fs_dirhash(de->name, de->name_len, &h);
614                                 printk(":%x.%u ", h.hash,
615                                        (unsigned) ((char *) de - base));
616                         }
617                         space += EXT4_DIR_REC_LEN(de->name_len);
618                         names++;
619                 }
620                 de = ext4_next_entry(de, size);
621         }
622         printk("(%i)\n", names);
623         return (struct stats) { names, space, 1 };
624 }
625
626 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
627                              struct dx_entry *entries, int levels)
628 {
629         unsigned blocksize = dir->i_sb->s_blocksize;
630         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
631         unsigned bcount = 0;
632         struct buffer_head *bh;
633         int err;
634         printk("%i indexed blocks...\n", count);
635         for (i = 0; i < count; i++, entries++)
636         {
637                 ext4_lblk_t block = dx_get_block(entries);
638                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
639                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
640                 struct stats stats;
641                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
642                 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
643                 stats = levels?
644                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
645                    dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
646                 names += stats.names;
647                 space += stats.space;
648                 bcount += stats.bcount;
649                 brelse(bh);
650         }
651         if (bcount)
652                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
653                        levels ? "" : "   ", names, space/bcount,
654                        (space/bcount)*100/blocksize);
655         return (struct stats) { names, space, bcount};
656 }
657 #endif /* DX_DEBUG */
658
659 /*
660  * Probe for a directory leaf block to search.
661  *
662  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
663  * error in the directory index, and the caller should fall back to
664  * searching the directory normally.  The callers of dx_probe **MUST**
665  * check for this error code, and make sure it never gets reflected
666  * back to userspace.
667  */
668 static struct dx_frame *
669 dx_probe(const struct qstr *d_name, struct inode *dir,
670          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
671 {
672         unsigned count, indirect;
673         struct dx_entry *at, *entries, *p, *q, *m;
674         struct dx_root *root;
675         struct buffer_head *bh;
676         struct dx_frame *frame = frame_in;
677         u32 hash;
678
679         frame->bh = NULL;
680         bh = ext4_read_dirblock(dir, 0, INDEX);
681         if (IS_ERR(bh)) {
682                 *err = PTR_ERR(bh);
683                 goto fail;
684         }
685         root = (struct dx_root *) bh->b_data;
686         if (root->info.hash_version != DX_HASH_TEA &&
687             root->info.hash_version != DX_HASH_HALF_MD4 &&
688             root->info.hash_version != DX_HASH_LEGACY) {
689                 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
690                              root->info.hash_version);
691                 brelse(bh);
692                 *err = ERR_BAD_DX_DIR;
693                 goto fail;
694         }
695         hinfo->hash_version = root->info.hash_version;
696         if (hinfo->hash_version <= DX_HASH_TEA)
697                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
698         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
699         if (d_name)
700                 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
701         hash = hinfo->hash;
702
703         if (root->info.unused_flags & 1) {
704                 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
705                              root->info.unused_flags);
706                 brelse(bh);
707                 *err = ERR_BAD_DX_DIR;
708                 goto fail;
709         }
710
711         if ((indirect = root->info.indirect_levels) > 1) {
712                 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
713                              root->info.indirect_levels);
714                 brelse(bh);
715                 *err = ERR_BAD_DX_DIR;
716                 goto fail;
717         }
718
719         entries = (struct dx_entry *) (((char *)&root->info) +
720                                        root->info.info_length);
721
722         if (dx_get_limit(entries) != dx_root_limit(dir,
723                                                    root->info.info_length)) {
724                 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
725                 brelse(bh);
726                 *err = ERR_BAD_DX_DIR;
727                 goto fail;
728         }
729
730         dxtrace(printk("Look up %x", hash));
731         while (1)
732         {
733                 count = dx_get_count(entries);
734                 if (!count || count > dx_get_limit(entries)) {
735                         ext4_warning(dir->i_sb,
736                                      "dx entry: no count or count > limit");
737                         brelse(bh);
738                         *err = ERR_BAD_DX_DIR;
739                         goto fail2;
740                 }
741
742                 p = entries + 1;
743                 q = entries + count - 1;
744                 while (p <= q)
745                 {
746                         m = p + (q - p)/2;
747                         dxtrace(printk("."));
748                         if (dx_get_hash(m) > hash)
749                                 q = m - 1;
750                         else
751                                 p = m + 1;
752                 }
753
754                 if (0) // linear search cross check
755                 {
756                         unsigned n = count - 1;
757                         at = entries;
758                         while (n--)
759                         {
760                                 dxtrace(printk(","));
761                                 if (dx_get_hash(++at) > hash)
762                                 {
763                                         at--;
764                                         break;
765                                 }
766                         }
767                         assert (at == p - 1);
768                 }
769
770                 at = p - 1;
771                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
772                 frame->bh = bh;
773                 frame->entries = entries;
774                 frame->at = at;
775                 if (!indirect--) return frame;
776                 bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
777                 if (IS_ERR(bh)) {
778                         *err = PTR_ERR(bh);
779                         goto fail2;
780                 }
781                 entries = ((struct dx_node *) bh->b_data)->entries;
782
783                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
784                         ext4_warning(dir->i_sb,
785                                      "dx entry: limit != node limit");
786                         brelse(bh);
787                         *err = ERR_BAD_DX_DIR;
788                         goto fail2;
789                 }
790                 frame++;
791                 frame->bh = NULL;
792         }
793 fail2:
794         while (frame >= frame_in) {
795                 brelse(frame->bh);
796                 frame--;
797         }
798 fail:
799         if (*err == ERR_BAD_DX_DIR)
800                 ext4_warning(dir->i_sb,
801                              "Corrupt dir inode %lu, running e2fsck is "
802                              "recommended.", dir->i_ino);
803         return NULL;
804 }
805
806 static void dx_release (struct dx_frame *frames)
807 {
808         if (frames[0].bh == NULL)
809                 return;
810
811         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
812                 brelse(frames[1].bh);
813         brelse(frames[0].bh);
814 }
815
816 /*
817  * This function increments the frame pointer to search the next leaf
818  * block, and reads in the necessary intervening nodes if the search
819  * should be necessary.  Whether or not the search is necessary is
820  * controlled by the hash parameter.  If the hash value is even, then
821  * the search is only continued if the next block starts with that
822  * hash value.  This is used if we are searching for a specific file.
823  *
824  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
825  *
826  * This function returns 1 if the caller should continue to search,
827  * or 0 if it should not.  If there is an error reading one of the
828  * index blocks, it will a negative error code.
829  *
830  * If start_hash is non-null, it will be filled in with the starting
831  * hash of the next page.
832  */
833 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
834                                  struct dx_frame *frame,
835                                  struct dx_frame *frames,
836                                  __u32 *start_hash)
837 {
838         struct dx_frame *p;
839         struct buffer_head *bh;
840         int num_frames = 0;
841         __u32 bhash;
842
843         p = frame;
844         /*
845          * Find the next leaf page by incrementing the frame pointer.
846          * If we run out of entries in the interior node, loop around and
847          * increment pointer in the parent node.  When we break out of
848          * this loop, num_frames indicates the number of interior
849          * nodes need to be read.
850          */
851         while (1) {
852                 if (++(p->at) < p->entries + dx_get_count(p->entries))
853                         break;
854                 if (p == frames)
855                         return 0;
856                 num_frames++;
857                 p--;
858         }
859
860         /*
861          * If the hash is 1, then continue only if the next page has a
862          * continuation hash of any value.  This is used for readdir
863          * handling.  Otherwise, check to see if the hash matches the
864          * desired contiuation hash.  If it doesn't, return since
865          * there's no point to read in the successive index pages.
866          */
867         bhash = dx_get_hash(p->at);
868         if (start_hash)
869                 *start_hash = bhash;
870         if ((hash & 1) == 0) {
871                 if ((bhash & ~1) != hash)
872                         return 0;
873         }
874         /*
875          * If the hash is HASH_NB_ALWAYS, we always go to the next
876          * block so no check is necessary
877          */
878         while (num_frames--) {
879                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
880                 if (IS_ERR(bh))
881                         return PTR_ERR(bh);
882                 p++;
883                 brelse(p->bh);
884                 p->bh = bh;
885                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
886         }
887         return 1;
888 }
889
890
891 /*
892  * This function fills a red-black tree with information from a
893  * directory block.  It returns the number directory entries loaded
894  * into the tree.  If there is an error it is returned in err.
895  */
896 static int htree_dirblock_to_tree(struct file *dir_file,
897                                   struct inode *dir, ext4_lblk_t block,
898                                   struct dx_hash_info *hinfo,
899                                   __u32 start_hash, __u32 start_minor_hash)
900 {
901         struct buffer_head *bh;
902         struct ext4_dir_entry_2 *de, *top;
903         int err = 0, count = 0;
904
905         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
906                                                         (unsigned long)block));
907         bh = ext4_read_dirblock(dir, block, DIRENT);
908         if (IS_ERR(bh))
909                 return PTR_ERR(bh);
910
911         de = (struct ext4_dir_entry_2 *) bh->b_data;
912         top = (struct ext4_dir_entry_2 *) ((char *) de +
913                                            dir->i_sb->s_blocksize -
914                                            EXT4_DIR_REC_LEN(0));
915         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
916                 if (ext4_check_dir_entry(dir, NULL, de, bh,
917                                 bh->b_data, bh->b_size,
918                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
919                                          + ((char *)de - bh->b_data))) {
920                         /* On error, skip the f_pos to the next block. */
921                         dir_file->f_pos = (dir_file->f_pos |
922                                         (dir->i_sb->s_blocksize - 1)) + 1;
923                         brelse(bh);
924                         return count;
925                 }
926                 ext4fs_dirhash(de->name, de->name_len, hinfo);
927                 if ((hinfo->hash < start_hash) ||
928                     ((hinfo->hash == start_hash) &&
929                      (hinfo->minor_hash < start_minor_hash)))
930                         continue;
931                 if (de->inode == 0)
932                         continue;
933                 if ((err = ext4_htree_store_dirent(dir_file,
934                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
935                         brelse(bh);
936                         return err;
937                 }
938                 count++;
939         }
940         brelse(bh);
941         return count;
942 }
943
944
945 /*
946  * This function fills a red-black tree with information from a
947  * directory.  We start scanning the directory in hash order, starting
948  * at start_hash and start_minor_hash.
949  *
950  * This function returns the number of entries inserted into the tree,
951  * or a negative error code.
952  */
953 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
954                          __u32 start_minor_hash, __u32 *next_hash)
955 {
956         struct dx_hash_info hinfo;
957         struct ext4_dir_entry_2 *de;
958         struct dx_frame frames[2], *frame;
959         struct inode *dir;
960         ext4_lblk_t block;
961         int count = 0;
962         int ret, err;
963         __u32 hashval;
964
965         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
966                        start_hash, start_minor_hash));
967         dir = dir_file->f_path.dentry->d_inode;
968         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
969                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
970                 if (hinfo.hash_version <= DX_HASH_TEA)
971                         hinfo.hash_version +=
972                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
973                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
974                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
975                                                start_hash, start_minor_hash);
976                 *next_hash = ~0;
977                 return count;
978         }
979         hinfo.hash = start_hash;
980         hinfo.minor_hash = 0;
981         frame = dx_probe(NULL, dir, &hinfo, frames, &err);
982         if (!frame)
983                 return err;
984
985         /* Add '.' and '..' from the htree header */
986         if (!start_hash && !start_minor_hash) {
987                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
988                 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
989                         goto errout;
990                 count++;
991         }
992         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
993                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
994                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
995                 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
996                         goto errout;
997                 count++;
998         }
999
1000         while (1) {
1001                 block = dx_get_block(frame->at);
1002                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1003                                              start_hash, start_minor_hash);
1004                 if (ret < 0) {
1005                         err = ret;
1006                         goto errout;
1007                 }
1008                 count += ret;
1009                 hashval = ~0;
1010                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1011                                             frame, frames, &hashval);
1012                 *next_hash = hashval;
1013                 if (ret < 0) {
1014                         err = ret;
1015                         goto errout;
1016                 }
1017                 /*
1018                  * Stop if:  (a) there are no more entries, or
1019                  * (b) we have inserted at least one entry and the
1020                  * next hash value is not a continuation
1021                  */
1022                 if ((ret == 0) ||
1023                     (count && ((hashval & 1) == 0)))
1024                         break;
1025         }
1026         dx_release(frames);
1027         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1028                        "next hash: %x\n", count, *next_hash));
1029         return count;
1030 errout:
1031         dx_release(frames);
1032         return (err);
1033 }
1034
1035 static inline int search_dirblock(struct buffer_head *bh,
1036                                   struct inode *dir,
1037                                   const struct qstr *d_name,
1038                                   unsigned int offset,
1039                                   struct ext4_dir_entry_2 **res_dir)
1040 {
1041         return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1042                           d_name, offset, res_dir);
1043 }
1044
1045 /*
1046  * Directory block splitting, compacting
1047  */
1048
1049 /*
1050  * Create map of hash values, offsets, and sizes, stored at end of block.
1051  * Returns number of entries mapped.
1052  */
1053 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1054                        struct dx_hash_info *hinfo,
1055                        struct dx_map_entry *map_tail)
1056 {
1057         int count = 0;
1058         char *base = (char *) de;
1059         struct dx_hash_info h = *hinfo;
1060
1061         while ((char *) de < base + blocksize) {
1062                 if (de->name_len && de->inode) {
1063                         ext4fs_dirhash(de->name, de->name_len, &h);
1064                         map_tail--;
1065                         map_tail->hash = h.hash;
1066                         map_tail->offs = ((char *) de - base)>>2;
1067                         map_tail->size = le16_to_cpu(de->rec_len);
1068                         count++;
1069                         cond_resched();
1070                 }
1071                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1072                 de = ext4_next_entry(de, blocksize);
1073         }
1074         return count;
1075 }
1076
1077 /* Sort map by hash value */
1078 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1079 {
1080         struct dx_map_entry *p, *q, *top = map + count - 1;
1081         int more;
1082         /* Combsort until bubble sort doesn't suck */
1083         while (count > 2) {
1084                 count = count*10/13;
1085                 if (count - 9 < 2) /* 9, 10 -> 11 */
1086                         count = 11;
1087                 for (p = top, q = p - count; q >= map; p--, q--)
1088                         if (p->hash < q->hash)
1089                                 swap(*p, *q);
1090         }
1091         /* Garden variety bubble sort */
1092         do {
1093                 more = 0;
1094                 q = top;
1095                 while (q-- > map) {
1096                         if (q[1].hash >= q[0].hash)
1097                                 continue;
1098                         swap(*(q+1), *q);
1099                         more = 1;
1100                 }
1101         } while(more);
1102 }
1103
1104 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1105 {
1106         struct dx_entry *entries = frame->entries;
1107         struct dx_entry *old = frame->at, *new = old + 1;
1108         int count = dx_get_count(entries);
1109
1110         assert(count < dx_get_limit(entries));
1111         assert(old < entries + count);
1112         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1113         dx_set_hash(new, hash);
1114         dx_set_block(new, block);
1115         dx_set_count(entries, count + 1);
1116 }
1117
1118 /*
1119  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1120  *
1121  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1122  * `de != NULL' is guaranteed by caller.
1123  */
1124 static inline int ext4_match (int len, const char * const name,
1125                               struct ext4_dir_entry_2 * de)
1126 {
1127         if (len != de->name_len)
1128                 return 0;
1129         if (!de->inode)
1130                 return 0;
1131         return !memcmp(name, de->name, len);
1132 }
1133
1134 /*
1135  * Returns 0 if not found, -1 on failure, and 1 on success
1136  */
1137 int search_dir(struct buffer_head *bh,
1138                char *search_buf,
1139                int buf_size,
1140                struct inode *dir,
1141                const struct qstr *d_name,
1142                unsigned int offset,
1143                struct ext4_dir_entry_2 **res_dir)
1144 {
1145         struct ext4_dir_entry_2 * de;
1146         char * dlimit;
1147         int de_len;
1148         const char *name = d_name->name;
1149         int namelen = d_name->len;
1150
1151         de = (struct ext4_dir_entry_2 *)search_buf;
1152         dlimit = search_buf + buf_size;
1153         while ((char *) de < dlimit) {
1154                 /* this code is executed quadratically often */
1155                 /* do minimal checking `by hand' */
1156
1157                 if ((char *) de + namelen <= dlimit &&
1158                     ext4_match (namelen, name, de)) {
1159                         /* found a match - just to be sure, do a full check */
1160                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1161                                                  bh->b_size, offset))
1162                                 return -1;
1163                         *res_dir = de;
1164                         return 1;
1165                 }
1166                 /* prevent looping on a bad block */
1167                 de_len = ext4_rec_len_from_disk(de->rec_len,
1168                                                 dir->i_sb->s_blocksize);
1169                 if (de_len <= 0)
1170                         return -1;
1171                 offset += de_len;
1172                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1173         }
1174         return 0;
1175 }
1176
1177 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1178                                struct ext4_dir_entry *de)
1179 {
1180         struct super_block *sb = dir->i_sb;
1181
1182         if (!is_dx(dir))
1183                 return 0;
1184         if (block == 0)
1185                 return 1;
1186         if (de->inode == 0 &&
1187             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1188                         sb->s_blocksize)
1189                 return 1;
1190         return 0;
1191 }
1192
1193 /*
1194  *      ext4_find_entry()
1195  *
1196  * finds an entry in the specified directory with the wanted name. It
1197  * returns the cache buffer in which the entry was found, and the entry
1198  * itself (as a parameter - res_dir). It does NOT read the inode of the
1199  * entry - you'll have to do that yourself if you want to.
1200  *
1201  * The returned buffer_head has ->b_count elevated.  The caller is expected
1202  * to brelse() it when appropriate.
1203  */
1204 static struct buffer_head * ext4_find_entry (struct inode *dir,
1205                                         const struct qstr *d_name,
1206                                         struct ext4_dir_entry_2 **res_dir,
1207                                         int *inlined)
1208 {
1209         struct super_block *sb;
1210         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1211         struct buffer_head *bh, *ret = NULL;
1212         ext4_lblk_t start, block, b;
1213         const u8 *name = d_name->name;
1214         int ra_max = 0;         /* Number of bh's in the readahead
1215                                    buffer, bh_use[] */
1216         int ra_ptr = 0;         /* Current index into readahead
1217                                    buffer */
1218         int num = 0;
1219         ext4_lblk_t  nblocks;
1220         int i, err;
1221         int namelen;
1222
1223         *res_dir = NULL;
1224         sb = dir->i_sb;
1225         namelen = d_name->len;
1226         if (namelen > EXT4_NAME_LEN)
1227                 return NULL;
1228
1229         if (ext4_has_inline_data(dir)) {
1230                 int has_inline_data = 1;
1231                 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1232                                              &has_inline_data);
1233                 if (has_inline_data) {
1234                         if (inlined)
1235                                 *inlined = 1;
1236                         return ret;
1237                 }
1238         }
1239
1240         if ((namelen <= 2) && (name[0] == '.') &&
1241             (name[1] == '.' || name[1] == '\0')) {
1242                 /*
1243                  * "." or ".." will only be in the first block
1244                  * NFS may look up ".."; "." should be handled by the VFS
1245                  */
1246                 block = start = 0;
1247                 nblocks = 1;
1248                 goto restart;
1249         }
1250         if (is_dx(dir)) {
1251                 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1252                 /*
1253                  * On success, or if the error was file not found,
1254                  * return.  Otherwise, fall back to doing a search the
1255                  * old fashioned way.
1256                  */
1257                 if (bh || (err != ERR_BAD_DX_DIR))
1258                         return bh;
1259                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1260                                "falling back\n"));
1261         }
1262         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1263         start = EXT4_I(dir)->i_dir_start_lookup;
1264         if (start >= nblocks)
1265                 start = 0;
1266         block = start;
1267 restart:
1268         do {
1269                 /*
1270                  * We deal with the read-ahead logic here.
1271                  */
1272                 if (ra_ptr >= ra_max) {
1273                         /* Refill the readahead buffer */
1274                         ra_ptr = 0;
1275                         b = block;
1276                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1277                                 /*
1278                                  * Terminate if we reach the end of the
1279                                  * directory and must wrap, or if our
1280                                  * search has finished at this block.
1281                                  */
1282                                 if (b >= nblocks || (num && block == start)) {
1283                                         bh_use[ra_max] = NULL;
1284                                         break;
1285                                 }
1286                                 num++;
1287                                 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1288                                 bh_use[ra_max] = bh;
1289                                 if (bh)
1290                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1291                                                     1, &bh);
1292                         }
1293                 }
1294                 if ((bh = bh_use[ra_ptr++]) == NULL)
1295                         goto next;
1296                 wait_on_buffer(bh);
1297                 if (!buffer_uptodate(bh)) {
1298                         /* read error, skip block & hope for the best */
1299                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1300                                          (unsigned long) block);
1301                         brelse(bh);
1302                         goto next;
1303                 }
1304                 if (!buffer_verified(bh) &&
1305                     !is_dx_internal_node(dir, block,
1306                                          (struct ext4_dir_entry *)bh->b_data) &&
1307                     !ext4_dirent_csum_verify(dir,
1308                                 (struct ext4_dir_entry *)bh->b_data)) {
1309                         EXT4_ERROR_INODE(dir, "checksumming directory "
1310                                          "block %lu", (unsigned long)block);
1311                         brelse(bh);
1312                         goto next;
1313                 }
1314                 set_buffer_verified(bh);
1315                 i = search_dirblock(bh, dir, d_name,
1316                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1317                 if (i == 1) {
1318                         EXT4_I(dir)->i_dir_start_lookup = block;
1319                         ret = bh;
1320                         goto cleanup_and_exit;
1321                 } else {
1322                         brelse(bh);
1323                         if (i < 0)
1324                                 goto cleanup_and_exit;
1325                 }
1326         next:
1327                 if (++block >= nblocks)
1328                         block = 0;
1329         } while (block != start);
1330
1331         /*
1332          * If the directory has grown while we were searching, then
1333          * search the last part of the directory before giving up.
1334          */
1335         block = nblocks;
1336         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1337         if (block < nblocks) {
1338                 start = 0;
1339                 goto restart;
1340         }
1341
1342 cleanup_and_exit:
1343         /* Clean up the read-ahead blocks */
1344         for (; ra_ptr < ra_max; ra_ptr++)
1345                 brelse(bh_use[ra_ptr]);
1346         return ret;
1347 }
1348
1349 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1350                        struct ext4_dir_entry_2 **res_dir, int *err)
1351 {
1352         struct super_block * sb = dir->i_sb;
1353         struct dx_hash_info     hinfo;
1354         struct dx_frame frames[2], *frame;
1355         struct buffer_head *bh;
1356         ext4_lblk_t block;
1357         int retval;
1358
1359         if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1360                 return NULL;
1361         do {
1362                 block = dx_get_block(frame->at);
1363                 bh = ext4_read_dirblock(dir, block, DIRENT);
1364                 if (IS_ERR(bh)) {
1365                         *err = PTR_ERR(bh);
1366                         goto errout;
1367                 }
1368                 retval = search_dirblock(bh, dir, d_name,
1369                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1370                                          res_dir);
1371                 if (retval == 1) {      /* Success! */
1372                         dx_release(frames);
1373                         return bh;
1374                 }
1375                 brelse(bh);
1376                 if (retval == -1) {
1377                         *err = ERR_BAD_DX_DIR;
1378                         goto errout;
1379                 }
1380
1381                 /* Check to see if we should continue to search */
1382                 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1383                                                frames, NULL);
1384                 if (retval < 0) {
1385                         ext4_warning(sb,
1386                              "error reading index page in directory #%lu",
1387                              dir->i_ino);
1388                         *err = retval;
1389                         goto errout;
1390                 }
1391         } while (retval == 1);
1392
1393         *err = -ENOENT;
1394 errout:
1395         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1396         dx_release (frames);
1397         return NULL;
1398 }
1399
1400 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1401 {
1402         struct inode *inode;
1403         struct ext4_dir_entry_2 *de;
1404         struct buffer_head *bh;
1405
1406         if (dentry->d_name.len > EXT4_NAME_LEN)
1407                 return ERR_PTR(-ENAMETOOLONG);
1408
1409         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1410         inode = NULL;
1411         if (bh) {
1412                 __u32 ino = le32_to_cpu(de->inode);
1413                 brelse(bh);
1414                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1415                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1416                         return ERR_PTR(-EIO);
1417                 }
1418                 if (unlikely(ino == dir->i_ino)) {
1419                         EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1420                                          dentry->d_name.len,
1421                                          dentry->d_name.name);
1422                         return ERR_PTR(-EIO);
1423                 }
1424                 inode = ext4_iget(dir->i_sb, ino);
1425                 if (inode == ERR_PTR(-ESTALE)) {
1426                         EXT4_ERROR_INODE(dir,
1427                                          "deleted inode referenced: %u",
1428                                          ino);
1429                         return ERR_PTR(-EIO);
1430                 }
1431         }
1432         return d_splice_alias(inode, dentry);
1433 }
1434
1435
1436 struct dentry *ext4_get_parent(struct dentry *child)
1437 {
1438         __u32 ino;
1439         static const struct qstr dotdot = QSTR_INIT("..", 2);
1440         struct ext4_dir_entry_2 * de;
1441         struct buffer_head *bh;
1442
1443         bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1444         if (!bh)
1445                 return ERR_PTR(-ENOENT);
1446         ino = le32_to_cpu(de->inode);
1447         brelse(bh);
1448
1449         if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1450                 EXT4_ERROR_INODE(child->d_inode,
1451                                  "bad parent inode number: %u", ino);
1452                 return ERR_PTR(-EIO);
1453         }
1454
1455         return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1456 }
1457
1458 #define S_SHIFT 12
1459 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1460         [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1461         [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1462         [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1463         [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1464         [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1465         [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1466         [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1467 };
1468
1469 static inline void ext4_set_de_type(struct super_block *sb,
1470                                 struct ext4_dir_entry_2 *de,
1471                                 umode_t mode) {
1472         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1473                 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1474 }
1475
1476 /*
1477  * Move count entries from end of map between two memory locations.
1478  * Returns pointer to last entry moved.
1479  */
1480 static struct ext4_dir_entry_2 *
1481 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1482                 unsigned blocksize)
1483 {
1484         unsigned rec_len = 0;
1485
1486         while (count--) {
1487                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1488                                                 (from + (map->offs<<2));
1489                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1490                 memcpy (to, de, rec_len);
1491                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1492                                 ext4_rec_len_to_disk(rec_len, blocksize);
1493                 de->inode = 0;
1494                 map++;
1495                 to += rec_len;
1496         }
1497         return (struct ext4_dir_entry_2 *) (to - rec_len);
1498 }
1499
1500 /*
1501  * Compact each dir entry in the range to the minimal rec_len.
1502  * Returns pointer to last entry in range.
1503  */
1504 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1505 {
1506         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1507         unsigned rec_len = 0;
1508
1509         prev = to = de;
1510         while ((char*)de < base + blocksize) {
1511                 next = ext4_next_entry(de, blocksize);
1512                 if (de->inode && de->name_len) {
1513                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1514                         if (de > to)
1515                                 memmove(to, de, rec_len);
1516                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1517                         prev = to;
1518                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1519                 }
1520                 de = next;
1521         }
1522         return prev;
1523 }
1524
1525 /*
1526  * Split a full leaf block to make room for a new dir entry.
1527  * Allocate a new block, and move entries so that they are approx. equally full.
1528  * Returns pointer to de in block into which the new entry will be inserted.
1529  */
1530 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1531                         struct buffer_head **bh,struct dx_frame *frame,
1532                         struct dx_hash_info *hinfo, int *error)
1533 {
1534         unsigned blocksize = dir->i_sb->s_blocksize;
1535         unsigned count, continued;
1536         struct buffer_head *bh2;
1537         ext4_lblk_t newblock;
1538         u32 hash2;
1539         struct dx_map_entry *map;
1540         char *data1 = (*bh)->b_data, *data2;
1541         unsigned split, move, size;
1542         struct ext4_dir_entry_2 *de = NULL, *de2;
1543         struct ext4_dir_entry_tail *t;
1544         int     csum_size = 0;
1545         int     err = 0, i;
1546
1547         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1548                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1549                 csum_size = sizeof(struct ext4_dir_entry_tail);
1550
1551         bh2 = ext4_append(handle, dir, &newblock);
1552         if (IS_ERR(bh2)) {
1553                 brelse(*bh);
1554                 *bh = NULL;
1555                 *error = PTR_ERR(bh2);
1556                 return NULL;
1557         }
1558
1559         BUFFER_TRACE(*bh, "get_write_access");
1560         err = ext4_journal_get_write_access(handle, *bh);
1561         if (err)
1562                 goto journal_error;
1563
1564         BUFFER_TRACE(frame->bh, "get_write_access");
1565         err = ext4_journal_get_write_access(handle, frame->bh);
1566         if (err)
1567                 goto journal_error;
1568
1569         data2 = bh2->b_data;
1570
1571         /* create map in the end of data2 block */
1572         map = (struct dx_map_entry *) (data2 + blocksize);
1573         count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1574                              blocksize, hinfo, map);
1575         map -= count;
1576         dx_sort_map(map, count);
1577         /* Split the existing block in the middle, size-wise */
1578         size = 0;
1579         move = 0;
1580         for (i = count-1; i >= 0; i--) {
1581                 /* is more than half of this entry in 2nd half of the block? */
1582                 if (size + map[i].size/2 > blocksize/2)
1583                         break;
1584                 size += map[i].size;
1585                 move++;
1586         }
1587         /* map index at which we will split */
1588         split = count - move;
1589         hash2 = map[split].hash;
1590         continued = hash2 == map[split - 1].hash;
1591         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1592                         (unsigned long)dx_get_block(frame->at),
1593                                         hash2, split, count-split));
1594
1595         /* Fancy dance to stay within two buffers */
1596         de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1597         de = dx_pack_dirents(data1, blocksize);
1598         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1599                                            (char *) de,
1600                                            blocksize);
1601         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1602                                             (char *) de2,
1603                                             blocksize);
1604         if (csum_size) {
1605                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1606                 initialize_dirent_tail(t, blocksize);
1607
1608                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1609                 initialize_dirent_tail(t, blocksize);
1610         }
1611
1612         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1613         dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1614
1615         /* Which block gets the new entry? */
1616         if (hinfo->hash >= hash2)
1617         {
1618                 swap(*bh, bh2);
1619                 de = de2;
1620         }
1621         dx_insert_block(frame, hash2 + continued, newblock);
1622         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1623         if (err)
1624                 goto journal_error;
1625         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1626         if (err)
1627                 goto journal_error;
1628         brelse(bh2);
1629         dxtrace(dx_show_index("frame", frame->entries));
1630         return de;
1631
1632 journal_error:
1633         brelse(*bh);
1634         brelse(bh2);
1635         *bh = NULL;
1636         ext4_std_error(dir->i_sb, err);
1637         *error = err;
1638         return NULL;
1639 }
1640
1641 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1642                       struct buffer_head *bh,
1643                       void *buf, int buf_size,
1644                       const char *name, int namelen,
1645                       struct ext4_dir_entry_2 **dest_de)
1646 {
1647         struct ext4_dir_entry_2 *de;
1648         unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1649         int nlen, rlen;
1650         unsigned int offset = 0;
1651         char *top;
1652
1653         de = (struct ext4_dir_entry_2 *)buf;
1654         top = buf + buf_size - reclen;
1655         while ((char *) de <= top) {
1656                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1657                                          buf, buf_size, offset))
1658                         return -EIO;
1659                 if (ext4_match(namelen, name, de))
1660                         return -EEXIST;
1661                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1662                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1663                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1664                         break;
1665                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1666                 offset += rlen;
1667         }
1668         if ((char *) de > top)
1669                 return -ENOSPC;
1670
1671         *dest_de = de;
1672         return 0;
1673 }
1674
1675 void ext4_insert_dentry(struct inode *inode,
1676                         struct ext4_dir_entry_2 *de,
1677                         int buf_size,
1678                         const char *name, int namelen)
1679 {
1680
1681         int nlen, rlen;
1682
1683         nlen = EXT4_DIR_REC_LEN(de->name_len);
1684         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1685         if (de->inode) {
1686                 struct ext4_dir_entry_2 *de1 =
1687                                 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1688                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1689                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1690                 de = de1;
1691         }
1692         de->file_type = EXT4_FT_UNKNOWN;
1693         de->inode = cpu_to_le32(inode->i_ino);
1694         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1695         de->name_len = namelen;
1696         memcpy(de->name, name, namelen);
1697 }
1698 /*
1699  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1700  * it points to a directory entry which is guaranteed to be large
1701  * enough for new directory entry.  If de is NULL, then
1702  * add_dirent_to_buf will attempt search the directory block for
1703  * space.  It will return -ENOSPC if no space is available, and -EIO
1704  * and -EEXIST if directory entry already exists.
1705  */
1706 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1707                              struct inode *inode, struct ext4_dir_entry_2 *de,
1708                              struct buffer_head *bh)
1709 {
1710         struct inode    *dir = dentry->d_parent->d_inode;
1711         const char      *name = dentry->d_name.name;
1712         int             namelen = dentry->d_name.len;
1713         unsigned int    blocksize = dir->i_sb->s_blocksize;
1714         int             csum_size = 0;
1715         int             err;
1716
1717         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1718                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1719                 csum_size = sizeof(struct ext4_dir_entry_tail);
1720
1721         if (!de) {
1722                 err = ext4_find_dest_de(dir, inode,
1723                                         bh, bh->b_data, blocksize - csum_size,
1724                                         name, namelen, &de);
1725                 if (err)
1726                         return err;
1727         }
1728         BUFFER_TRACE(bh, "get_write_access");
1729         err = ext4_journal_get_write_access(handle, bh);
1730         if (err) {
1731                 ext4_std_error(dir->i_sb, err);
1732                 return err;
1733         }
1734
1735         /* By now the buffer is marked for journaling */
1736         ext4_insert_dentry(inode, de, blocksize, name, namelen);
1737
1738         /*
1739          * XXX shouldn't update any times until successful
1740          * completion of syscall, but too many callers depend
1741          * on this.
1742          *
1743          * XXX similarly, too many callers depend on
1744          * ext4_new_inode() setting the times, but error
1745          * recovery deletes the inode, so the worst that can
1746          * happen is that the times are slightly out of date
1747          * and/or different from the directory change time.
1748          */
1749         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1750         ext4_update_dx_flag(dir);
1751         dir->i_version++;
1752         ext4_mark_inode_dirty(handle, dir);
1753         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1754         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1755         if (err)
1756                 ext4_std_error(dir->i_sb, err);
1757         return 0;
1758 }
1759
1760 /*
1761  * This converts a one block unindexed directory to a 3 block indexed
1762  * directory, and adds the dentry to the indexed directory.
1763  */
1764 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1765                             struct inode *inode, struct buffer_head *bh)
1766 {
1767         struct inode    *dir = dentry->d_parent->d_inode;
1768         const char      *name = dentry->d_name.name;
1769         int             namelen = dentry->d_name.len;
1770         struct buffer_head *bh2;
1771         struct dx_root  *root;
1772         struct dx_frame frames[2], *frame;
1773         struct dx_entry *entries;
1774         struct ext4_dir_entry_2 *de, *de2;
1775         struct ext4_dir_entry_tail *t;
1776         char            *data1, *top;
1777         unsigned        len;
1778         int             retval;
1779         unsigned        blocksize;
1780         struct dx_hash_info hinfo;
1781         ext4_lblk_t  block;
1782         struct fake_dirent *fde;
1783         int             csum_size = 0;
1784
1785         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1786                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1787                 csum_size = sizeof(struct ext4_dir_entry_tail);
1788
1789         blocksize =  dir->i_sb->s_blocksize;
1790         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1791         retval = ext4_journal_get_write_access(handle, bh);
1792         if (retval) {
1793                 ext4_std_error(dir->i_sb, retval);
1794                 brelse(bh);
1795                 return retval;
1796         }
1797         root = (struct dx_root *) bh->b_data;
1798
1799         /* The 0th block becomes the root, move the dirents out */
1800         fde = &root->dotdot;
1801         de = (struct ext4_dir_entry_2 *)((char *)fde +
1802                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1803         if ((char *) de >= (((char *) root) + blocksize)) {
1804                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1805                 brelse(bh);
1806                 return -EIO;
1807         }
1808         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1809
1810         /* Allocate new block for the 0th block's dirents */
1811         bh2 = ext4_append(handle, dir, &block);
1812         if (IS_ERR(bh2)) {
1813                 brelse(bh);
1814                 return PTR_ERR(bh2);
1815         }
1816         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1817         data1 = bh2->b_data;
1818
1819         memcpy (data1, de, len);
1820         de = (struct ext4_dir_entry_2 *) data1;
1821         top = data1 + len;
1822         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1823                 de = de2;
1824         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1825                                            (char *) de,
1826                                            blocksize);
1827
1828         if (csum_size) {
1829                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1830                 initialize_dirent_tail(t, blocksize);
1831         }
1832
1833         /* Initialize the root; the dot dirents already exist */
1834         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1835         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1836                                            blocksize);
1837         memset (&root->info, 0, sizeof(root->info));
1838         root->info.info_length = sizeof(root->info);
1839         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1840         entries = root->entries;
1841         dx_set_block(entries, 1);
1842         dx_set_count(entries, 1);
1843         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1844
1845         /* Initialize as for dx_probe */
1846         hinfo.hash_version = root->info.hash_version;
1847         if (hinfo.hash_version <= DX_HASH_TEA)
1848                 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1849         hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1850         ext4fs_dirhash(name, namelen, &hinfo);
1851         frame = frames;
1852         frame->entries = entries;
1853         frame->at = entries;
1854         frame->bh = bh;
1855         bh = bh2;
1856
1857         ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1858         ext4_handle_dirty_dirent_node(handle, dir, bh);
1859
1860         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1861         if (!de) {
1862                 /*
1863                  * Even if the block split failed, we have to properly write
1864                  * out all the changes we did so far. Otherwise we can end up
1865                  * with corrupted filesystem.
1866                  */
1867                 ext4_mark_inode_dirty(handle, dir);
1868                 dx_release(frames);
1869                 return retval;
1870         }
1871         dx_release(frames);
1872
1873         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1874         brelse(bh);
1875         return retval;
1876 }
1877
1878 /*
1879  *      ext4_add_entry()
1880  *
1881  * adds a file entry to the specified directory, using the same
1882  * semantics as ext4_find_entry(). It returns NULL if it failed.
1883  *
1884  * NOTE!! The inode part of 'de' is left at 0 - which means you
1885  * may not sleep between calling this and putting something into
1886  * the entry, as someone else might have used it while you slept.
1887  */
1888 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1889                           struct inode *inode)
1890 {
1891         struct inode *dir = dentry->d_parent->d_inode;
1892         struct buffer_head *bh;
1893         struct ext4_dir_entry_2 *de;
1894         struct ext4_dir_entry_tail *t;
1895         struct super_block *sb;
1896         int     retval;
1897         int     dx_fallback=0;
1898         unsigned blocksize;
1899         ext4_lblk_t block, blocks;
1900         int     csum_size = 0;
1901
1902         if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1903                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1904                 csum_size = sizeof(struct ext4_dir_entry_tail);
1905
1906         sb = dir->i_sb;
1907         blocksize = sb->s_blocksize;
1908         if (!dentry->d_name.len)
1909                 return -EINVAL;
1910
1911         if (ext4_has_inline_data(dir)) {
1912                 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1913                 if (retval < 0)
1914                         return retval;
1915                 if (retval == 1) {
1916                         retval = 0;
1917                         return retval;
1918                 }
1919         }
1920
1921         if (is_dx(dir)) {
1922                 retval = ext4_dx_add_entry(handle, dentry, inode);
1923                 if (!retval || (retval != ERR_BAD_DX_DIR))
1924                         return retval;
1925                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1926                 dx_fallback++;
1927                 ext4_mark_inode_dirty(handle, dir);
1928         }
1929         blocks = dir->i_size >> sb->s_blocksize_bits;
1930         for (block = 0; block < blocks; block++) {
1931                 bh = ext4_read_dirblock(dir, block, DIRENT);
1932                 if (IS_ERR(bh))
1933                         return PTR_ERR(bh);
1934
1935                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1936                 if (retval != -ENOSPC) {
1937                         brelse(bh);
1938                         return retval;
1939                 }
1940
1941                 if (blocks == 1 && !dx_fallback &&
1942                     EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1943                         return make_indexed_dir(handle, dentry, inode, bh);
1944                 brelse(bh);
1945         }
1946         bh = ext4_append(handle, dir, &block);
1947         if (IS_ERR(bh))
1948                 return PTR_ERR(bh);
1949         de = (struct ext4_dir_entry_2 *) bh->b_data;
1950         de->inode = 0;
1951         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1952
1953         if (csum_size) {
1954                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1955                 initialize_dirent_tail(t, blocksize);
1956         }
1957
1958         retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1959         brelse(bh);
1960         if (retval == 0)
1961                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1962         return retval;
1963 }
1964
1965 /*
1966  * Returns 0 for success, or a negative error value
1967  */
1968 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1969                              struct inode *inode)
1970 {
1971         struct dx_frame frames[2], *frame;
1972         struct dx_entry *entries, *at;
1973         struct dx_hash_info hinfo;
1974         struct buffer_head *bh;
1975         struct inode *dir = dentry->d_parent->d_inode;
1976         struct super_block *sb = dir->i_sb;
1977         struct ext4_dir_entry_2 *de;
1978         int err;
1979
1980         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1981         if (!frame)
1982                 return err;
1983         entries = frame->entries;
1984         at = frame->at;
1985         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1986         if (IS_ERR(bh)) {
1987                 err = PTR_ERR(bh);
1988                 bh = NULL;
1989                 goto cleanup;
1990         }
1991
1992         BUFFER_TRACE(bh, "get_write_access");
1993         err = ext4_journal_get_write_access(handle, bh);
1994         if (err)
1995                 goto journal_error;
1996
1997         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1998         if (err != -ENOSPC)
1999                 goto cleanup;
2000
2001         /* Block full, should compress but for now just split */
2002         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2003                        dx_get_count(entries), dx_get_limit(entries)));
2004         /* Need to split index? */
2005         if (dx_get_count(entries) == dx_get_limit(entries)) {
2006                 ext4_lblk_t newblock;
2007                 unsigned icount = dx_get_count(entries);
2008                 int levels = frame - frames;
2009                 struct dx_entry *entries2;
2010                 struct dx_node *node2;
2011                 struct buffer_head *bh2;
2012
2013                 if (levels && (dx_get_count(frames->entries) ==
2014                                dx_get_limit(frames->entries))) {
2015                         ext4_warning(sb, "Directory index full!");
2016                         err = -ENOSPC;
2017                         goto cleanup;
2018                 }
2019                 bh2 = ext4_append(handle, dir, &newblock);
2020                 if (IS_ERR(bh2)) {
2021                         err = PTR_ERR(bh2);
2022                         goto cleanup;
2023                 }
2024                 node2 = (struct dx_node *)(bh2->b_data);
2025                 entries2 = node2->entries;
2026                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2027                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2028                                                            sb->s_blocksize);
2029                 BUFFER_TRACE(frame->bh, "get_write_access");
2030                 err = ext4_journal_get_write_access(handle, frame->bh);
2031                 if (err)
2032                         goto journal_error;
2033                 if (levels) {
2034                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2035                         unsigned hash2 = dx_get_hash(entries + icount1);
2036                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2037                                        icount1, icount2));
2038
2039                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2040                         err = ext4_journal_get_write_access(handle,
2041                                                              frames[0].bh);
2042                         if (err)
2043                                 goto journal_error;
2044
2045                         memcpy((char *) entries2, (char *) (entries + icount1),
2046                                icount2 * sizeof(struct dx_entry));
2047                         dx_set_count(entries, icount1);
2048                         dx_set_count(entries2, icount2);
2049                         dx_set_limit(entries2, dx_node_limit(dir));
2050
2051                         /* Which index block gets the new entry? */
2052                         if (at - entries >= icount1) {
2053                                 frame->at = at = at - entries - icount1 + entries2;
2054                                 frame->entries = entries = entries2;
2055                                 swap(frame->bh, bh2);
2056                         }
2057                         dx_insert_block(frames + 0, hash2, newblock);
2058                         dxtrace(dx_show_index("node", frames[1].entries));
2059                         dxtrace(dx_show_index("node",
2060                                ((struct dx_node *) bh2->b_data)->entries));
2061                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2062                         if (err)
2063                                 goto journal_error;
2064                         brelse (bh2);
2065                 } else {
2066                         dxtrace(printk(KERN_DEBUG
2067                                        "Creating second level index...\n"));
2068                         memcpy((char *) entries2, (char *) entries,
2069                                icount * sizeof(struct dx_entry));
2070                         dx_set_limit(entries2, dx_node_limit(dir));
2071
2072                         /* Set up root */
2073                         dx_set_count(entries, 1);
2074                         dx_set_block(entries + 0, newblock);
2075                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2076
2077                         /* Add new access path frame */
2078                         frame = frames + 1;
2079                         frame->at = at = at - entries + entries2;
2080                         frame->entries = entries = entries2;
2081                         frame->bh = bh2;
2082                         err = ext4_journal_get_write_access(handle,
2083                                                              frame->bh);
2084                         if (err)
2085                                 goto journal_error;
2086                 }
2087                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2088                 if (err) {
2089                         ext4_std_error(inode->i_sb, err);
2090                         goto cleanup;
2091                 }
2092         }
2093         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2094         if (!de)
2095                 goto cleanup;
2096         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2097         goto cleanup;
2098
2099 journal_error:
2100         ext4_std_error(dir->i_sb, err);
2101 cleanup:
2102         brelse(bh);
2103         dx_release(frames);
2104         return err;
2105 }
2106
2107 /*
2108  * ext4_generic_delete_entry deletes a directory entry by merging it
2109  * with the previous entry
2110  */
2111 int ext4_generic_delete_entry(handle_t *handle,
2112                               struct inode *dir,
2113                               struct ext4_dir_entry_2 *de_del,
2114                               struct buffer_head *bh,
2115                               void *entry_buf,
2116                               int buf_size,
2117                               int csum_size)
2118 {
2119         struct ext4_dir_entry_2 *de, *pde;
2120         unsigned int blocksize = dir->i_sb->s_blocksize;
2121         int i;
2122
2123         i = 0;
2124         pde = NULL;
2125         de = (struct ext4_dir_entry_2 *)entry_buf;
2126         while (i < buf_size - csum_size) {
2127                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2128                                          bh->b_data, bh->b_size, i))
2129                         return -EIO;
2130                 if (de == de_del)  {
2131                         if (pde)
2132                                 pde->rec_len = ext4_rec_len_to_disk(
2133                                         ext4_rec_len_from_disk(pde->rec_len,
2134                                                                blocksize) +
2135                                         ext4_rec_len_from_disk(de->rec_len,
2136                                                                blocksize),
2137                                         blocksize);
2138                         else
2139                                 de->inode = 0;
2140                         dir->i_version++;
2141                         return 0;
2142                 }
2143                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2144                 pde = de;
2145                 de = ext4_next_entry(de, blocksize);
2146         }
2147         return -ENOENT;
2148 }
2149
2150 static int ext4_delete_entry(handle_t *handle,
2151                              struct inode *dir,
2152                              struct ext4_dir_entry_2 *de_del,
2153                              struct buffer_head *bh)
2154 {
2155         int err, csum_size = 0;
2156
2157         if (ext4_has_inline_data(dir)) {
2158                 int has_inline_data = 1;
2159                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2160                                                &has_inline_data);
2161                 if (has_inline_data)
2162                         return err;
2163         }
2164
2165         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2166                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2167                 csum_size = sizeof(struct ext4_dir_entry_tail);
2168
2169         BUFFER_TRACE(bh, "get_write_access");
2170         err = ext4_journal_get_write_access(handle, bh);
2171         if (unlikely(err))
2172                 goto out;
2173
2174         err = ext4_generic_delete_entry(handle, dir, de_del,
2175                                         bh, bh->b_data,
2176                                         dir->i_sb->s_blocksize, csum_size);
2177         if (err)
2178                 goto out;
2179
2180         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2181         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2182         if (unlikely(err))
2183                 goto out;
2184
2185         return 0;
2186 out:
2187         if (err != -ENOENT)
2188                 ext4_std_error(dir->i_sb, err);
2189         return err;
2190 }
2191
2192 /*
2193  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2194  * since this indicates that nlinks count was previously 1.
2195  */
2196 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2197 {
2198         inc_nlink(inode);
2199         if (is_dx(inode) && inode->i_nlink > 1) {
2200                 /* limit is 16-bit i_links_count */
2201                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2202                         set_nlink(inode, 1);
2203                         EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2204                                               EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2205                 }
2206         }
2207 }
2208
2209 /*
2210  * If a directory had nlink == 1, then we should let it be 1. This indicates
2211  * directory has >EXT4_LINK_MAX subdirs.
2212  */
2213 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2214 {
2215         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2216                 drop_nlink(inode);
2217 }
2218
2219
2220 static int ext4_add_nondir(handle_t *handle,
2221                 struct dentry *dentry, struct inode *inode)
2222 {
2223         int err = ext4_add_entry(handle, dentry, inode);
2224         if (!err) {
2225                 ext4_mark_inode_dirty(handle, inode);
2226                 unlock_new_inode(inode);
2227                 d_instantiate(dentry, inode);
2228                 return 0;
2229         }
2230         drop_nlink(inode);
2231         unlock_new_inode(inode);
2232         iput(inode);
2233         return err;
2234 }
2235
2236 /*
2237  * By the time this is called, we already have created
2238  * the directory cache entry for the new file, but it
2239  * is so far negative - it has no inode.
2240  *
2241  * If the create succeeds, we fill in the inode information
2242  * with d_instantiate().
2243  */
2244 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2245                        bool excl)
2246 {
2247         handle_t *handle;
2248         struct inode *inode;
2249         int err, credits, retries = 0;
2250
2251         dquot_initialize(dir);
2252
2253         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2254                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2255                    EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2256 retry:
2257         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2258                                             NULL, EXT4_HT_DIR, credits);
2259         handle = ext4_journal_current_handle();
2260         err = PTR_ERR(inode);
2261         if (!IS_ERR(inode)) {
2262                 inode->i_op = &ext4_file_inode_operations;
2263                 inode->i_fop = &ext4_file_operations;
2264                 ext4_set_aops(inode);
2265                 err = ext4_add_nondir(handle, dentry, inode);
2266                 if (!err && IS_DIRSYNC(dir))
2267                         ext4_handle_sync(handle);
2268         }
2269         if (handle)
2270                 ext4_journal_stop(handle);
2271         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2272                 goto retry;
2273         return err;
2274 }
2275
2276 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2277                       umode_t mode, dev_t rdev)
2278 {
2279         handle_t *handle;
2280         struct inode *inode;
2281         int err, credits, retries = 0;
2282
2283         if (!new_valid_dev(rdev))
2284                 return -EINVAL;
2285
2286         dquot_initialize(dir);
2287
2288         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2289                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2290                    EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2291 retry:
2292         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2293                                             NULL, EXT4_HT_DIR, credits);
2294         handle = ext4_journal_current_handle();
2295         err = PTR_ERR(inode);
2296         if (!IS_ERR(inode)) {
2297                 init_special_inode(inode, inode->i_mode, rdev);
2298                 inode->i_op = &ext4_special_inode_operations;
2299                 err = ext4_add_nondir(handle, dentry, inode);
2300                 if (!err && IS_DIRSYNC(dir))
2301                         ext4_handle_sync(handle);
2302         }
2303         if (handle)
2304                 ext4_journal_stop(handle);
2305         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2306                 goto retry;
2307         return err;
2308 }
2309
2310 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2311                           struct ext4_dir_entry_2 *de,
2312                           int blocksize, int csum_size,
2313                           unsigned int parent_ino, int dotdot_real_len)
2314 {
2315         de->inode = cpu_to_le32(inode->i_ino);
2316         de->name_len = 1;
2317         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2318                                            blocksize);
2319         strcpy(de->name, ".");
2320         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2321
2322         de = ext4_next_entry(de, blocksize);
2323         de->inode = cpu_to_le32(parent_ino);
2324         de->name_len = 2;
2325         if (!dotdot_real_len)
2326                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2327                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2328                                         blocksize);
2329         else
2330                 de->rec_len = ext4_rec_len_to_disk(
2331                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2332         strcpy(de->name, "..");
2333         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2334
2335         return ext4_next_entry(de, blocksize);
2336 }
2337
2338 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2339                              struct inode *inode)
2340 {
2341         struct buffer_head *dir_block = NULL;
2342         struct ext4_dir_entry_2 *de;
2343         struct ext4_dir_entry_tail *t;
2344         ext4_lblk_t block = 0;
2345         unsigned int blocksize = dir->i_sb->s_blocksize;
2346         int csum_size = 0;
2347         int err;
2348
2349         if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2350                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2351                 csum_size = sizeof(struct ext4_dir_entry_tail);
2352
2353         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2354                 err = ext4_try_create_inline_dir(handle, dir, inode);
2355                 if (err < 0 && err != -ENOSPC)
2356                         goto out;
2357                 if (!err)
2358                         goto out;
2359         }
2360
2361         inode->i_size = 0;
2362         dir_block = ext4_append(handle, inode, &block);
2363         if (IS_ERR(dir_block))
2364                 return PTR_ERR(dir_block);
2365         BUFFER_TRACE(dir_block, "get_write_access");
2366         err = ext4_journal_get_write_access(handle, dir_block);
2367         if (err)
2368                 goto out;
2369         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2370         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2371         set_nlink(inode, 2);
2372         if (csum_size) {
2373                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2374                 initialize_dirent_tail(t, blocksize);
2375         }
2376
2377         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2378         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2379         if (err)
2380                 goto out;
2381         set_buffer_verified(dir_block);
2382 out:
2383         brelse(dir_block);
2384         return err;
2385 }
2386
2387 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2388 {
2389         handle_t *handle;
2390         struct inode *inode;
2391         int err, credits, retries = 0;
2392
2393         if (EXT4_DIR_LINK_MAX(dir))
2394                 return -EMLINK;
2395
2396         dquot_initialize(dir);
2397
2398         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2399                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2400                    EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2401 retry:
2402         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2403                                             &dentry->d_name,
2404                                             0, NULL, EXT4_HT_DIR, credits);
2405         handle = ext4_journal_current_handle();
2406         err = PTR_ERR(inode);
2407         if (IS_ERR(inode))
2408                 goto out_stop;
2409
2410         inode->i_op = &ext4_dir_inode_operations;
2411         inode->i_fop = &ext4_dir_operations;
2412         err = ext4_init_new_dir(handle, dir, inode);
2413         if (err)
2414                 goto out_clear_inode;
2415         err = ext4_mark_inode_dirty(handle, inode);
2416         if (!err)
2417                 err = ext4_add_entry(handle, dentry, inode);
2418         if (err) {
2419 out_clear_inode:
2420                 clear_nlink(inode);
2421                 unlock_new_inode(inode);
2422                 ext4_mark_inode_dirty(handle, inode);
2423                 iput(inode);
2424                 goto out_stop;
2425         }
2426         ext4_inc_count(handle, dir);
2427         ext4_update_dx_flag(dir);
2428         err = ext4_mark_inode_dirty(handle, dir);
2429         if (err)
2430                 goto out_clear_inode;
2431         unlock_new_inode(inode);
2432         d_instantiate(dentry, inode);
2433         if (IS_DIRSYNC(dir))
2434                 ext4_handle_sync(handle);
2435
2436 out_stop:
2437         if (handle)
2438                 ext4_journal_stop(handle);
2439         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2440                 goto retry;
2441         return err;
2442 }
2443
2444 /*
2445  * routine to check that the specified directory is empty (for rmdir)
2446  */
2447 static int empty_dir(struct inode *inode)
2448 {
2449         unsigned int offset;
2450         struct buffer_head *bh;
2451         struct ext4_dir_entry_2 *de, *de1;
2452         struct super_block *sb;
2453         int err = 0;
2454
2455         if (ext4_has_inline_data(inode)) {
2456                 int has_inline_data = 1;
2457
2458                 err = empty_inline_dir(inode, &has_inline_data);
2459                 if (has_inline_data)
2460                         return err;
2461         }
2462
2463         sb = inode->i_sb;
2464         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2465                 EXT4_ERROR_INODE(inode, "invalid size");
2466                 return 1;
2467         }
2468         bh = ext4_read_dirblock(inode, 0, EITHER);
2469         if (IS_ERR(bh))
2470                 return 1;
2471
2472         de = (struct ext4_dir_entry_2 *) bh->b_data;
2473         de1 = ext4_next_entry(de, sb->s_blocksize);
2474         if (le32_to_cpu(de->inode) != inode->i_ino ||
2475                         !le32_to_cpu(de1->inode) ||
2476                         strcmp(".", de->name) ||
2477                         strcmp("..", de1->name)) {
2478                 ext4_warning(inode->i_sb,
2479                              "bad directory (dir #%lu) - no `.' or `..'",
2480                              inode->i_ino);
2481                 brelse(bh);
2482                 return 1;
2483         }
2484         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2485                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2486         de = ext4_next_entry(de1, sb->s_blocksize);
2487         while (offset < inode->i_size) {
2488                 if (!bh ||
2489                     (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2490                         unsigned int lblock;
2491                         err = 0;
2492                         brelse(bh);
2493                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2494                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2495                         if (IS_ERR(bh))
2496                                 return 1;
2497                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2498                 }
2499                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2500                                          bh->b_data, bh->b_size, offset)) {
2501                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2502                                                          sb->s_blocksize);
2503                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2504                         continue;
2505                 }
2506                 if (le32_to_cpu(de->inode)) {
2507                         brelse(bh);
2508                         return 0;
2509                 }
2510                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2511                 de = ext4_next_entry(de, sb->s_blocksize);
2512         }
2513         brelse(bh);
2514         return 1;
2515 }
2516
2517 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2518  * such inodes, starting at the superblock, in case we crash before the
2519  * file is closed/deleted, or in case the inode truncate spans multiple
2520  * transactions and the last transaction is not recovered after a crash.
2521  *
2522  * At filesystem recovery time, we walk this list deleting unlinked
2523  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2524  */
2525 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2526 {
2527         struct super_block *sb = inode->i_sb;
2528         struct ext4_iloc iloc;
2529         int err = 0, rc;
2530
2531         if (!EXT4_SB(sb)->s_journal)
2532                 return 0;
2533
2534         mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2535         if (!list_empty(&EXT4_I(inode)->i_orphan))
2536                 goto out_unlock;
2537
2538         /*
2539          * Orphan handling is only valid for files with data blocks
2540          * being truncated, or files being unlinked. Note that we either
2541          * hold i_mutex, or the inode can not be referenced from outside,
2542          * so i_nlink should not be bumped due to race
2543          */
2544         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2545                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2546
2547         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2548         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2549         if (err)
2550                 goto out_unlock;
2551
2552         err = ext4_reserve_inode_write(handle, inode, &iloc);
2553         if (err)
2554                 goto out_unlock;
2555         /*
2556          * Due to previous errors inode may be already a part of on-disk
2557          * orphan list. If so skip on-disk list modification.
2558          */
2559         if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2560                 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2561                         goto mem_insert;
2562
2563         /* Insert this inode at the head of the on-disk orphan list... */
2564         NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2565         EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2566         err = ext4_handle_dirty_super(handle, sb);
2567         rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2568         if (!err)
2569                 err = rc;
2570
2571         /* Only add to the head of the in-memory list if all the
2572          * previous operations succeeded.  If the orphan_add is going to
2573          * fail (possibly taking the journal offline), we can't risk
2574          * leaving the inode on the orphan list: stray orphan-list
2575          * entries can cause panics at unmount time.
2576          *
2577          * This is safe: on error we're going to ignore the orphan list
2578          * anyway on the next recovery. */
2579 mem_insert:
2580         if (!err)
2581                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2582
2583         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2584         jbd_debug(4, "orphan inode %lu will point to %d\n",
2585                         inode->i_ino, NEXT_ORPHAN(inode));
2586 out_unlock:
2587         mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2588         ext4_std_error(inode->i_sb, err);
2589         return err;
2590 }
2591
2592 /*
2593  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2594  * of such inodes stored on disk, because it is finally being cleaned up.
2595  */
2596 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2597 {
2598         struct list_head *prev;
2599         struct ext4_inode_info *ei = EXT4_I(inode);
2600         struct ext4_sb_info *sbi;
2601         __u32 ino_next;
2602         struct ext4_iloc iloc;
2603         int err = 0;
2604
2605         if ((!EXT4_SB(inode->i_sb)->s_journal) &&
2606             !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2607                 return 0;
2608
2609         mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2610         if (list_empty(&ei->i_orphan))
2611                 goto out;
2612
2613         ino_next = NEXT_ORPHAN(inode);
2614         prev = ei->i_orphan.prev;
2615         sbi = EXT4_SB(inode->i_sb);
2616
2617         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2618
2619         list_del_init(&ei->i_orphan);
2620
2621         /* If we're on an error path, we may not have a valid
2622          * transaction handle with which to update the orphan list on
2623          * disk, but we still need to remove the inode from the linked
2624          * list in memory. */
2625         if (!handle)
2626                 goto out;
2627
2628         err = ext4_reserve_inode_write(handle, inode, &iloc);
2629         if (err)
2630                 goto out_err;
2631
2632         if (prev == &sbi->s_orphan) {
2633                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2634                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2635                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2636                 if (err)
2637                         goto out_brelse;
2638                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2639                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2640         } else {
2641                 struct ext4_iloc iloc2;
2642                 struct inode *i_prev =
2643                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2644
2645                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2646                           i_prev->i_ino, ino_next);
2647                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2648                 if (err)
2649                         goto out_brelse;
2650                 NEXT_ORPHAN(i_prev) = ino_next;
2651                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2652         }
2653         if (err)
2654                 goto out_brelse;
2655         NEXT_ORPHAN(inode) = 0;
2656         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2657
2658 out_err:
2659         ext4_std_error(inode->i_sb, err);
2660 out:
2661         mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2662         return err;
2663
2664 out_brelse:
2665         brelse(iloc.bh);
2666         goto out_err;
2667 }
2668
2669 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2670 {
2671         int retval;
2672         struct inode *inode;
2673         struct buffer_head *bh;
2674         struct ext4_dir_entry_2 *de;
2675         handle_t *handle = NULL;
2676
2677         /* Initialize quotas before so that eventual writes go in
2678          * separate transaction */
2679         dquot_initialize(dir);
2680         dquot_initialize(dentry->d_inode);
2681
2682         retval = -ENOENT;
2683         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2684         if (!bh)
2685                 goto end_rmdir;
2686
2687         inode = dentry->d_inode;
2688
2689         retval = -EIO;
2690         if (le32_to_cpu(de->inode) != inode->i_ino)
2691                 goto end_rmdir;
2692
2693         retval = -ENOTEMPTY;
2694         if (!empty_dir(inode))
2695                 goto end_rmdir;
2696
2697         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2698                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2699         if (IS_ERR(handle)) {
2700                 retval = PTR_ERR(handle);
2701                 handle = NULL;
2702                 goto end_rmdir;
2703         }
2704
2705         if (IS_DIRSYNC(dir))
2706                 ext4_handle_sync(handle);
2707
2708         retval = ext4_delete_entry(handle, dir, de, bh);
2709         if (retval)
2710                 goto end_rmdir;
2711         if (!EXT4_DIR_LINK_EMPTY(inode))
2712                 ext4_warning(inode->i_sb,
2713                              "empty directory has too many links (%d)",
2714                              inode->i_nlink);
2715         inode->i_version++;
2716         clear_nlink(inode);
2717         /* There's no need to set i_disksize: the fact that i_nlink is
2718          * zero will ensure that the right thing happens during any
2719          * recovery. */
2720         inode->i_size = 0;
2721         ext4_orphan_add(handle, inode);
2722         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2723         ext4_mark_inode_dirty(handle, inode);
2724         ext4_dec_count(handle, dir);
2725         ext4_update_dx_flag(dir);
2726         ext4_mark_inode_dirty(handle, dir);
2727
2728 end_rmdir:
2729         brelse(bh);
2730         if (handle)
2731                 ext4_journal_stop(handle);
2732         return retval;
2733 }
2734
2735 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2736 {
2737         int retval;
2738         struct inode *inode;
2739         struct buffer_head *bh;
2740         struct ext4_dir_entry_2 *de;
2741         handle_t *handle = NULL;
2742
2743         trace_ext4_unlink_enter(dir, dentry);
2744         /* Initialize quotas before so that eventual writes go
2745          * in separate transaction */
2746         dquot_initialize(dir);
2747         dquot_initialize(dentry->d_inode);
2748
2749         retval = -ENOENT;
2750         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2751         if (!bh)
2752                 goto end_unlink;
2753
2754         inode = dentry->d_inode;
2755
2756         retval = -EIO;
2757         if (le32_to_cpu(de->inode) != inode->i_ino)
2758                 goto end_unlink;
2759
2760         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2761                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2762         if (IS_ERR(handle)) {
2763                 retval = PTR_ERR(handle);
2764                 handle = NULL;
2765                 goto end_unlink;
2766         }
2767
2768         if (IS_DIRSYNC(dir))
2769                 ext4_handle_sync(handle);
2770
2771         if (!inode->i_nlink) {
2772                 ext4_warning(inode->i_sb,
2773                              "Deleting nonexistent file (%lu), %d",
2774                              inode->i_ino, inode->i_nlink);
2775                 set_nlink(inode, 1);
2776         }
2777         retval = ext4_delete_entry(handle, dir, de, bh);
2778         if (retval)
2779                 goto end_unlink;
2780         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2781         ext4_update_dx_flag(dir);
2782         ext4_mark_inode_dirty(handle, dir);
2783         drop_nlink(inode);
2784         if (!inode->i_nlink)
2785                 ext4_orphan_add(handle, inode);
2786         inode->i_ctime = ext4_current_time(inode);
2787         ext4_mark_inode_dirty(handle, inode);
2788         retval = 0;
2789
2790 end_unlink:
2791         brelse(bh);
2792         if (handle)
2793                 ext4_journal_stop(handle);
2794         trace_ext4_unlink_exit(dentry, retval);
2795         return retval;
2796 }
2797
2798 static int ext4_symlink(struct inode *dir,
2799                         struct dentry *dentry, const char *symname)
2800 {
2801         handle_t *handle;
2802         struct inode *inode;
2803         int l, err, retries = 0;
2804         int credits;
2805
2806         l = strlen(symname)+1;
2807         if (l > dir->i_sb->s_blocksize)
2808                 return -ENAMETOOLONG;
2809
2810         dquot_initialize(dir);
2811
2812         if (l > EXT4_N_BLOCKS * 4) {
2813                 /*
2814                  * For non-fast symlinks, we just allocate inode and put it on
2815                  * orphan list in the first transaction => we need bitmap,
2816                  * group descriptor, sb, inode block, quota blocks, and
2817                  * possibly selinux xattr blocks.
2818                  */
2819                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2820                           EXT4_XATTR_TRANS_BLOCKS;
2821         } else {
2822                 /*
2823                  * Fast symlink. We have to add entry to directory
2824                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2825                  * allocate new inode (bitmap, group descriptor, inode block,
2826                  * quota blocks, sb is already counted in previous macros).
2827                  */
2828                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2829                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2830                           EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2831         }
2832 retry:
2833         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2834                                             &dentry->d_name, 0, NULL,
2835                                             EXT4_HT_DIR, credits);
2836         handle = ext4_journal_current_handle();
2837         err = PTR_ERR(inode);
2838         if (IS_ERR(inode))
2839                 goto out_stop;
2840
2841         if (l > EXT4_N_BLOCKS * 4) {
2842                 inode->i_op = &ext4_symlink_inode_operations;
2843                 ext4_set_aops(inode);
2844                 /*
2845                  * We cannot call page_symlink() with transaction started
2846                  * because it calls into ext4_write_begin() which can wait
2847                  * for transaction commit if we are running out of space
2848                  * and thus we deadlock. So we have to stop transaction now
2849                  * and restart it when symlink contents is written.
2850                  * 
2851                  * To keep fs consistent in case of crash, we have to put inode
2852                  * to orphan list in the mean time.
2853                  */
2854                 drop_nlink(inode);
2855                 err = ext4_orphan_add(handle, inode);
2856                 ext4_journal_stop(handle);
2857                 if (err)
2858                         goto err_drop_inode;
2859                 err = __page_symlink(inode, symname, l, 1);
2860                 if (err)
2861                         goto err_drop_inode;
2862                 /*
2863                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2864                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2865                  */
2866                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2867                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2868                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2869                 if (IS_ERR(handle)) {
2870                         err = PTR_ERR(handle);
2871                         goto err_drop_inode;
2872                 }
2873                 set_nlink(inode, 1);
2874                 err = ext4_orphan_del(handle, inode);
2875                 if (err) {
2876                         ext4_journal_stop(handle);
2877                         clear_nlink(inode);
2878                         goto err_drop_inode;
2879                 }
2880         } else {
2881                 /* clear the extent format for fast symlink */
2882                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2883                 inode->i_op = &ext4_fast_symlink_inode_operations;
2884                 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2885                 inode->i_size = l-1;
2886         }
2887         EXT4_I(inode)->i_disksize = inode->i_size;
2888         err = ext4_add_nondir(handle, dentry, inode);
2889         if (!err && IS_DIRSYNC(dir))
2890                 ext4_handle_sync(handle);
2891
2892 out_stop:
2893         if (handle)
2894                 ext4_journal_stop(handle);
2895         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2896                 goto retry;
2897         return err;
2898 err_drop_inode:
2899         unlock_new_inode(inode);
2900         iput(inode);
2901         return err;
2902 }
2903
2904 static int ext4_link(struct dentry *old_dentry,
2905                      struct inode *dir, struct dentry *dentry)
2906 {
2907         handle_t *handle;
2908         struct inode *inode = old_dentry->d_inode;
2909         int err, retries = 0;
2910
2911         if (inode->i_nlink >= EXT4_LINK_MAX)
2912                 return -EMLINK;
2913
2914         dquot_initialize(dir);
2915
2916 retry:
2917         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2918                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2919                  EXT4_INDEX_EXTRA_TRANS_BLOCKS));
2920         if (IS_ERR(handle))
2921                 return PTR_ERR(handle);
2922
2923         if (IS_DIRSYNC(dir))
2924                 ext4_handle_sync(handle);
2925
2926         inode->i_ctime = ext4_current_time(inode);
2927         ext4_inc_count(handle, inode);
2928         ihold(inode);
2929
2930         err = ext4_add_entry(handle, dentry, inode);
2931         if (!err) {
2932                 ext4_mark_inode_dirty(handle, inode);
2933                 d_instantiate(dentry, inode);
2934         } else {
2935                 drop_nlink(inode);
2936                 iput(inode);
2937         }
2938         ext4_journal_stop(handle);
2939         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2940                 goto retry;
2941         return err;
2942 }
2943
2944
2945 /*
2946  * Try to find buffer head where contains the parent block.
2947  * It should be the inode block if it is inlined or the 1st block
2948  * if it is a normal dir.
2949  */
2950 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2951                                         struct inode *inode,
2952                                         int *retval,
2953                                         struct ext4_dir_entry_2 **parent_de,
2954                                         int *inlined)
2955 {
2956         struct buffer_head *bh;
2957
2958         if (!ext4_has_inline_data(inode)) {
2959                 bh = ext4_read_dirblock(inode, 0, EITHER);
2960                 if (IS_ERR(bh)) {
2961                         *retval = PTR_ERR(bh);
2962                         return NULL;
2963                 }
2964                 *parent_de = ext4_next_entry(
2965                                         (struct ext4_dir_entry_2 *)bh->b_data,
2966                                         inode->i_sb->s_blocksize);
2967                 return bh;
2968         }
2969
2970         *inlined = 1;
2971         return ext4_get_first_inline_block(inode, parent_de, retval);
2972 }
2973
2974 /*
2975  * Anybody can rename anything with this: the permission checks are left to the
2976  * higher-level routines.
2977  */
2978 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2979                        struct inode *new_dir, struct dentry *new_dentry)
2980 {
2981         handle_t *handle;
2982         struct inode *old_inode, *new_inode;
2983         struct buffer_head *old_bh, *new_bh, *dir_bh;
2984         struct ext4_dir_entry_2 *old_de, *new_de;
2985         int retval, force_da_alloc = 0;
2986         int inlined = 0, new_inlined = 0;
2987         struct ext4_dir_entry_2 *parent_de;
2988
2989         dquot_initialize(old_dir);
2990         dquot_initialize(new_dir);
2991
2992         old_bh = new_bh = dir_bh = NULL;
2993
2994         /* Initialize quotas before so that eventual writes go
2995          * in separate transaction */
2996         if (new_dentry->d_inode)
2997                 dquot_initialize(new_dentry->d_inode);
2998         handle = ext4_journal_start(old_dir, EXT4_HT_DIR,
2999                 (2 * EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
3000                  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3001         if (IS_ERR(handle))
3002                 return PTR_ERR(handle);
3003
3004         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
3005                 ext4_handle_sync(handle);
3006
3007         old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
3008         /*
3009          *  Check for inode number is _not_ due to possible IO errors.
3010          *  We might rmdir the source, keep it as pwd of some process
3011          *  and merrily kill the link to whatever was created under the
3012          *  same name. Goodbye sticky bit ;-<
3013          */
3014         old_inode = old_dentry->d_inode;
3015         retval = -ENOENT;
3016         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
3017                 goto end_rename;
3018
3019         new_inode = new_dentry->d_inode;
3020         new_bh = ext4_find_entry(new_dir, &new_dentry->d_name,
3021                                  &new_de, &new_inlined);
3022         if (new_bh) {
3023                 if (!new_inode) {
3024                         brelse(new_bh);
3025                         new_bh = NULL;
3026                 }
3027         }
3028         if (S_ISDIR(old_inode->i_mode)) {
3029                 if (new_inode) {
3030                         retval = -ENOTEMPTY;
3031                         if (!empty_dir(new_inode))
3032                                 goto end_rename;
3033                 }
3034                 retval = -EIO;
3035                 dir_bh = ext4_get_first_dir_block(handle, old_inode,
3036                                                   &retval, &parent_de,
3037                                                   &inlined);
3038                 if (!dir_bh)
3039                         goto end_rename;
3040                 if (le32_to_cpu(parent_de->inode) != old_dir->i_ino)
3041                         goto end_rename;
3042                 retval = -EMLINK;
3043                 if (!new_inode && new_dir != old_dir &&
3044                     EXT4_DIR_LINK_MAX(new_dir))
3045                         goto end_rename;
3046                 BUFFER_TRACE(dir_bh, "get_write_access");
3047                 retval = ext4_journal_get_write_access(handle, dir_bh);
3048                 if (retval)
3049                         goto end_rename;
3050         }
3051         if (!new_bh) {
3052                 retval = ext4_add_entry(handle, new_dentry, old_inode);
3053                 if (retval)
3054                         goto end_rename;
3055         } else {
3056                 BUFFER_TRACE(new_bh, "get write access");
3057                 retval = ext4_journal_get_write_access(handle, new_bh);
3058                 if (retval)
3059                         goto end_rename;
3060                 new_de->inode = cpu_to_le32(old_inode->i_ino);
3061                 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3062                                               EXT4_FEATURE_INCOMPAT_FILETYPE))
3063                         new_de->file_type = old_de->file_type;
3064                 new_dir->i_version++;
3065                 new_dir->i_ctime = new_dir->i_mtime =
3066                                         ext4_current_time(new_dir);
3067                 ext4_mark_inode_dirty(handle, new_dir);
3068                 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3069                 if (!new_inlined) {
3070                         retval = ext4_handle_dirty_dirent_node(handle,
3071                                                                new_dir, new_bh);
3072                         if (unlikely(retval)) {
3073                                 ext4_std_error(new_dir->i_sb, retval);
3074                                 goto end_rename;
3075                         }
3076                 }
3077                 brelse(new_bh);
3078                 new_bh = NULL;
3079         }
3080
3081         /*
3082          * Like most other Unix systems, set the ctime for inodes on a
3083          * rename.
3084          */
3085         old_inode->i_ctime = ext4_current_time(old_inode);
3086         ext4_mark_inode_dirty(handle, old_inode);
3087
3088         /*
3089          * ok, that's it
3090          */
3091         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3092             old_de->name_len != old_dentry->d_name.len ||
3093             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3094             (retval = ext4_delete_entry(handle, old_dir,
3095                                         old_de, old_bh)) == -ENOENT) {
3096                 /* old_de could have moved from under us during htree split, so
3097                  * make sure that we are deleting the right entry.  We might
3098                  * also be pointing to a stale entry in the unused part of
3099                  * old_bh so just checking inum and the name isn't enough. */
3100                 struct buffer_head *old_bh2;
3101                 struct ext4_dir_entry_2 *old_de2;
3102
3103                 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
3104                                           &old_de2, NULL);
3105                 if (old_bh2) {
3106                         retval = ext4_delete_entry(handle, old_dir,
3107                                                    old_de2, old_bh2);
3108                         brelse(old_bh2);
3109                 }
3110         }
3111         if (retval) {
3112                 ext4_warning(old_dir->i_sb,
3113                                 "Deleting old file (%lu), %d, error=%d",
3114                                 old_dir->i_ino, old_dir->i_nlink, retval);
3115         }
3116
3117         if (new_inode) {
3118                 ext4_dec_count(handle, new_inode);
3119                 new_inode->i_ctime = ext4_current_time(new_inode);
3120         }
3121         old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3122         ext4_update_dx_flag(old_dir);
3123         if (dir_bh) {
3124                 parent_de->inode = cpu_to_le32(new_dir->i_ino);
3125                 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3126                 if (!inlined) {
3127                         if (is_dx(old_inode)) {
3128                                 retval = ext4_handle_dirty_dx_node(handle,
3129                                                                    old_inode,
3130                                                                    dir_bh);
3131                         } else {
3132                                 retval = ext4_handle_dirty_dirent_node(handle,
3133                                                         old_inode, dir_bh);
3134                         }
3135                 } else {
3136                         retval = ext4_mark_inode_dirty(handle, old_inode);
3137                 }
3138                 if (retval) {
3139                         ext4_std_error(old_dir->i_sb, retval);
3140                         goto end_rename;
3141                 }
3142                 ext4_dec_count(handle, old_dir);
3143                 if (new_inode) {
3144                         /* checked empty_dir above, can't have another parent,
3145                          * ext4_dec_count() won't work for many-linked dirs */
3146                         clear_nlink(new_inode);
3147                 } else {
3148                         ext4_inc_count(handle, new_dir);
3149                         ext4_update_dx_flag(new_dir);
3150                         ext4_mark_inode_dirty(handle, new_dir);
3151                 }
3152         }
3153         ext4_mark_inode_dirty(handle, old_dir);
3154         if (new_inode) {
3155                 ext4_mark_inode_dirty(handle, new_inode);
3156                 if (!new_inode->i_nlink)
3157                         ext4_orphan_add(handle, new_inode);
3158                 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3159                         force_da_alloc = 1;
3160         }
3161         retval = 0;
3162
3163 end_rename:
3164         brelse(dir_bh);
3165         brelse(old_bh);
3166         brelse(new_bh);
3167         ext4_journal_stop(handle);
3168         if (retval == 0 && force_da_alloc)
3169                 ext4_alloc_da_blocks(old_inode);
3170         return retval;
3171 }
3172
3173 /*
3174  * directories can handle most operations...
3175  */
3176 const struct inode_operations ext4_dir_inode_operations = {
3177         .create         = ext4_create,
3178         .lookup         = ext4_lookup,
3179         .link           = ext4_link,
3180         .unlink         = ext4_unlink,
3181         .symlink        = ext4_symlink,
3182         .mkdir          = ext4_mkdir,
3183         .rmdir          = ext4_rmdir,
3184         .mknod          = ext4_mknod,
3185         .rename         = ext4_rename,
3186         .setattr        = ext4_setattr,
3187         .setxattr       = generic_setxattr,
3188         .getxattr       = generic_getxattr,
3189         .listxattr      = ext4_listxattr,
3190         .removexattr    = generic_removexattr,
3191         .get_acl        = ext4_get_acl,
3192         .fiemap         = ext4_fiemap,
3193 };
3194
3195 const struct inode_operations ext4_special_inode_operations = {
3196         .setattr        = ext4_setattr,
3197         .setxattr       = generic_setxattr,
3198         .getxattr       = generic_getxattr,
3199         .listxattr      = ext4_listxattr,
3200         .removexattr    = generic_removexattr,
3201         .get_acl        = ext4_get_acl,
3202 };