Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[sfrench/cifs-2.6.git] / fs / ext4 / namei.c
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *      Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *      Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *      Theodore Ts'o, 2002
25  */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38
39 #include "xattr.h"
40 #include "acl.h"
41
42 #include <trace/events/ext4.h>
43 /*
44  * define how far ahead to read directories while searching them.
45  */
46 #define NAMEI_RA_CHUNKS  2
47 #define NAMEI_RA_BLOCKS  4
48 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49
50 static struct buffer_head *ext4_append(handle_t *handle,
51                                         struct inode *inode,
52                                         ext4_lblk_t *block)
53 {
54         struct buffer_head *bh;
55         int err;
56
57         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58                      ((inode->i_size >> 10) >=
59                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60                 return ERR_PTR(-ENOSPC);
61
62         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
64         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65         if (IS_ERR(bh))
66                 return bh;
67         inode->i_size += inode->i_sb->s_blocksize;
68         EXT4_I(inode)->i_disksize = inode->i_size;
69         BUFFER_TRACE(bh, "get_write_access");
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), __func__, __LINE__)
88
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90                                                 ext4_lblk_t block,
91                                                 dirblock_type_t type,
92                                                 const char *func,
93                                                 unsigned int line)
94 {
95         struct buffer_head *bh;
96         struct ext4_dir_entry *dirent;
97         int is_dx_block = 0;
98
99         bh = ext4_bread(NULL, inode, block, 0);
100         if (IS_ERR(bh)) {
101                 __ext4_warning(inode->i_sb, func, line,
102                                "inode #%lu: lblock %lu: comm %s: "
103                                "error %ld reading directory block",
104                                inode->i_ino, (unsigned long)block,
105                                current->comm, PTR_ERR(bh));
106
107                 return bh;
108         }
109         if (!bh) {
110                 ext4_error_inode(inode, func, line, block,
111                                  "Directory hole found");
112                 return ERR_PTR(-EFSCORRUPTED);
113         }
114         dirent = (struct ext4_dir_entry *) bh->b_data;
115         /* Determine whether or not we have an index block */
116         if (is_dx(inode)) {
117                 if (block == 0)
118                         is_dx_block = 1;
119                 else if (ext4_rec_len_from_disk(dirent->rec_len,
120                                                 inode->i_sb->s_blocksize) ==
121                          inode->i_sb->s_blocksize)
122                         is_dx_block = 1;
123         }
124         if (!is_dx_block && type == INDEX) {
125                 ext4_error_inode(inode, func, line, block,
126                        "directory leaf block found instead of index block");
127                 return ERR_PTR(-EFSCORRUPTED);
128         }
129         if (!ext4_has_metadata_csum(inode->i_sb) ||
130             buffer_verified(bh))
131                 return bh;
132
133         /*
134          * An empty leaf block can get mistaken for a index block; for
135          * this reason, we can only check the index checksum when the
136          * caller is sure it should be an index block.
137          */
138         if (is_dx_block && type == INDEX) {
139                 if (ext4_dx_csum_verify(inode, dirent))
140                         set_buffer_verified(bh);
141                 else {
142                         ext4_error_inode(inode, func, line, block,
143                                          "Directory index failed checksum");
144                         brelse(bh);
145                         return ERR_PTR(-EFSBADCRC);
146                 }
147         }
148         if (!is_dx_block) {
149                 if (ext4_dirent_csum_verify(inode, dirent))
150                         set_buffer_verified(bh);
151                 else {
152                         ext4_error_inode(inode, func, line, block,
153                                          "Directory block failed checksum");
154                         brelse(bh);
155                         return ERR_PTR(-EFSBADCRC);
156                 }
157         }
158         return bh;
159 }
160
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
164
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
170
171 struct fake_dirent
172 {
173         __le32 inode;
174         __le16 rec_len;
175         u8 name_len;
176         u8 file_type;
177 };
178
179 struct dx_countlimit
180 {
181         __le16 limit;
182         __le16 count;
183 };
184
185 struct dx_entry
186 {
187         __le32 hash;
188         __le32 block;
189 };
190
191 /*
192  * dx_root_info is laid out so that if it should somehow get overlaid by a
193  * dirent the two low bits of the hash version will be zero.  Therefore, the
194  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
195  */
196
197 struct dx_root
198 {
199         struct fake_dirent dot;
200         char dot_name[4];
201         struct fake_dirent dotdot;
202         char dotdot_name[4];
203         struct dx_root_info
204         {
205                 __le32 reserved_zero;
206                 u8 hash_version;
207                 u8 info_length; /* 8 */
208                 u8 indirect_levels;
209                 u8 unused_flags;
210         }
211         info;
212         struct dx_entry entries[0];
213 };
214
215 struct dx_node
216 {
217         struct fake_dirent fake;
218         struct dx_entry entries[0];
219 };
220
221
222 struct dx_frame
223 {
224         struct buffer_head *bh;
225         struct dx_entry *entries;
226         struct dx_entry *at;
227 };
228
229 struct dx_map_entry
230 {
231         u32 hash;
232         u16 offs;
233         u16 size;
234 };
235
236 /*
237  * This goes at the end of each htree block.
238  */
239 struct dx_tail {
240         u32 dt_reserved;
241         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
242 };
243
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255                                  struct inode *dir,
256                                  struct dx_hash_info *hinfo,
257                                  struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260                        unsigned blocksize, struct dx_hash_info *hinfo,
261                        struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264                 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267                                         u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269                                  struct dx_frame *frame,
270                                  struct dx_frame *frames,
271                                  __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273                 struct ext4_filename *fname,
274                 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276                              struct inode *dir, struct inode *inode);
277
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280                             unsigned int blocksize)
281 {
282         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283         t->det_rec_len = ext4_rec_len_to_disk(
284                         sizeof(struct ext4_dir_entry_tail), blocksize);
285         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286 }
287
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290                                                    struct ext4_dir_entry *de)
291 {
292         struct ext4_dir_entry_tail *t;
293
294 #ifdef PARANOID
295         struct ext4_dir_entry *d, *top;
296
297         d = de;
298         top = (struct ext4_dir_entry *)(((void *)de) +
299                 (EXT4_BLOCK_SIZE(inode->i_sb) -
300                 sizeof(struct ext4_dir_entry_tail)));
301         while (d < top && d->rec_len)
302                 d = (struct ext4_dir_entry *)(((void *)d) +
303                     le16_to_cpu(d->rec_len));
304
305         if (d != top)
306                 return NULL;
307
308         t = (struct ext4_dir_entry_tail *)d;
309 #else
310         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
312
313         if (t->det_reserved_zero1 ||
314             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315             t->det_reserved_zero2 ||
316             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317                 return NULL;
318
319         return t;
320 }
321
322 static __le32 ext4_dirent_csum(struct inode *inode,
323                                struct ext4_dir_entry *dirent, int size)
324 {
325         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326         struct ext4_inode_info *ei = EXT4_I(inode);
327         __u32 csum;
328
329         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330         return cpu_to_le32(csum);
331 }
332
333 #define warn_no_space_for_csum(inode)                                   \
334         __warn_no_space_for_csum((inode), __func__, __LINE__)
335
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337                                      unsigned int line)
338 {
339         __ext4_warning_inode(inode, func, line,
340                 "No space for directory leaf checksum. Please run e2fsck -D.");
341 }
342
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344 {
345         struct ext4_dir_entry_tail *t;
346
347         if (!ext4_has_metadata_csum(inode->i_sb))
348                 return 1;
349
350         t = get_dirent_tail(inode, dirent);
351         if (!t) {
352                 warn_no_space_for_csum(inode);
353                 return 0;
354         }
355
356         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357                                                 (void *)t - (void *)dirent))
358                 return 0;
359
360         return 1;
361 }
362
363 static void ext4_dirent_csum_set(struct inode *inode,
364                                  struct ext4_dir_entry *dirent)
365 {
366         struct ext4_dir_entry_tail *t;
367
368         if (!ext4_has_metadata_csum(inode->i_sb))
369                 return;
370
371         t = get_dirent_tail(inode, dirent);
372         if (!t) {
373                 warn_no_space_for_csum(inode);
374                 return;
375         }
376
377         t->det_checksum = ext4_dirent_csum(inode, dirent,
378                                            (void *)t - (void *)dirent);
379 }
380
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382                                   struct inode *inode,
383                                   struct buffer_head *bh)
384 {
385         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386         return ext4_handle_dirty_metadata(handle, inode, bh);
387 }
388
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390                                                struct ext4_dir_entry *dirent,
391                                                int *offset)
392 {
393         struct ext4_dir_entry *dp;
394         struct dx_root_info *root;
395         int count_offset;
396
397         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398                 count_offset = 8;
399         else if (le16_to_cpu(dirent->rec_len) == 12) {
400                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401                 if (le16_to_cpu(dp->rec_len) !=
402                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403                         return NULL;
404                 root = (struct dx_root_info *)(((void *)dp + 12));
405                 if (root->reserved_zero ||
406                     root->info_length != sizeof(struct dx_root_info))
407                         return NULL;
408                 count_offset = 32;
409         } else
410                 return NULL;
411
412         if (offset)
413                 *offset = count_offset;
414         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415 }
416
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418                            int count_offset, int count, struct dx_tail *t)
419 {
420         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421         struct ext4_inode_info *ei = EXT4_I(inode);
422         __u32 csum;
423         __le32 save_csum;
424         int size;
425
426         size = count_offset + (count * sizeof(struct dx_entry));
427         save_csum = t->dt_checksum;
428         t->dt_checksum = 0;
429         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430         csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
431         t->dt_checksum = save_csum;
432
433         return cpu_to_le32(csum);
434 }
435
436 static int ext4_dx_csum_verify(struct inode *inode,
437                                struct ext4_dir_entry *dirent)
438 {
439         struct dx_countlimit *c;
440         struct dx_tail *t;
441         int count_offset, limit, count;
442
443         if (!ext4_has_metadata_csum(inode->i_sb))
444                 return 1;
445
446         c = get_dx_countlimit(inode, dirent, &count_offset);
447         if (!c) {
448                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
449                 return 1;
450         }
451         limit = le16_to_cpu(c->limit);
452         count = le16_to_cpu(c->count);
453         if (count_offset + (limit * sizeof(struct dx_entry)) >
454             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455                 warn_no_space_for_csum(inode);
456                 return 1;
457         }
458         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459
460         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461                                             count, t))
462                 return 0;
463         return 1;
464 }
465
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 {
468         struct dx_countlimit *c;
469         struct dx_tail *t;
470         int count_offset, limit, count;
471
472         if (!ext4_has_metadata_csum(inode->i_sb))
473                 return;
474
475         c = get_dx_countlimit(inode, dirent, &count_offset);
476         if (!c) {
477                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
478                 return;
479         }
480         limit = le16_to_cpu(c->limit);
481         count = le16_to_cpu(c->count);
482         if (count_offset + (limit * sizeof(struct dx_entry)) >
483             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484                 warn_no_space_for_csum(inode);
485                 return;
486         }
487         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488
489         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490 }
491
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493                                             struct inode *inode,
494                                             struct buffer_head *bh)
495 {
496         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497         return ext4_handle_dirty_metadata(handle, inode, bh);
498 }
499
500 /*
501  * p is at least 6 bytes before the end of page
502  */
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 {
506         return (struct ext4_dir_entry_2 *)((char *)p +
507                 ext4_rec_len_from_disk(p->rec_len, blocksize));
508 }
509
510 /*
511  * Future: use high four bits of block for coalesce-on-delete flags
512  * Mask them off for now.
513  */
514
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 {
517         return le32_to_cpu(entry->block) & 0x00ffffff;
518 }
519
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 {
522         entry->block = cpu_to_le32(value);
523 }
524
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 {
527         return le32_to_cpu(entry->hash);
528 }
529
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 {
532         entry->hash = cpu_to_le32(value);
533 }
534
535 static inline unsigned dx_get_count(struct dx_entry *entries)
536 {
537         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538 }
539
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 {
542         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543 }
544
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 {
547         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548 }
549
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 {
552         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553 }
554
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 {
557         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558                 EXT4_DIR_REC_LEN(2) - infosize;
559
560         if (ext4_has_metadata_csum(dir->i_sb))
561                 entry_space -= sizeof(struct dx_tail);
562         return entry_space / sizeof(struct dx_entry);
563 }
564
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568
569         if (ext4_has_metadata_csum(dir->i_sb))
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 inode *dir,
597                                 struct dx_hash_info *hinfo,
598                                 struct ext4_dir_entry_2 *de,
599                                 int size, int show_names)
600 {
601         unsigned names = 0, space = 0;
602         char *base = (char *) de;
603         struct dx_hash_info h = *hinfo;
604
605         printk("names: ");
606         while ((char *) de < base + size)
607         {
608                 if (de->inode)
609                 {
610                         if (show_names)
611                         {
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613                                 int len;
614                                 char *name;
615                                 struct ext4_str fname_crypto_str
616                                         = {.name = NULL, .len = 0};
617                                 int res = 0;
618
619                                 name  = de->name;
620                                 len = de->name_len;
621                                 if (ext4_encrypted_inode(inode))
622                                         res = ext4_get_encryption_info(dir);
623                                 if (res) {
624                                         printk(KERN_WARNING "Error setting up"
625                                                " fname crypto: %d\n", res);
626                                 }
627                                 if (ctx == NULL) {
628                                         /* Directory is not encrypted */
629                                         ext4fs_dirhash(de->name,
630                                                 de->name_len, &h);
631                                         printk("%*.s:(U)%x.%u ", len,
632                                                name, h.hash,
633                                                (unsigned) ((char *) de
634                                                            - base));
635                                 } else {
636                                         /* Directory is encrypted */
637                                         res = ext4_fname_crypto_alloc_buffer(
638                                                 ctx, de->name_len,
639                                                 &fname_crypto_str);
640                                         if (res < 0) {
641                                                 printk(KERN_WARNING "Error "
642                                                         "allocating crypto "
643                                                         "buffer--skipping "
644                                                         "crypto\n");
645                                                 ctx = NULL;
646                                         }
647                                         res = ext4_fname_disk_to_usr(ctx, NULL, de,
648                                                         &fname_crypto_str);
649                                         if (res < 0) {
650                                                 printk(KERN_WARNING "Error "
651                                                         "converting filename "
652                                                         "from disk to usr"
653                                                         "\n");
654                                                 name = "??";
655                                                 len = 2;
656                                         } else {
657                                                 name = fname_crypto_str.name;
658                                                 len = fname_crypto_str.len;
659                                         }
660                                         ext4fs_dirhash(de->name, de->name_len,
661                                                        &h);
662                                         printk("%*.s:(E)%x.%u ", len, name,
663                                                h.hash, (unsigned) ((char *) de
664                                                                    - base));
665                                         ext4_fname_crypto_free_buffer(
666                                                 &fname_crypto_str);
667                                 }
668 #else
669                                 int len = de->name_len;
670                                 char *name = de->name;
671                                 ext4fs_dirhash(de->name, de->name_len, &h);
672                                 printk("%*.s:%x.%u ", len, name, h.hash,
673                                        (unsigned) ((char *) de - base));
674 #endif
675                         }
676                         space += EXT4_DIR_REC_LEN(de->name_len);
677                         names++;
678                 }
679                 de = ext4_next_entry(de, size);
680         }
681         printk("(%i)\n", names);
682         return (struct stats) { names, space, 1 };
683 }
684
685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686                              struct dx_entry *entries, int levels)
687 {
688         unsigned blocksize = dir->i_sb->s_blocksize;
689         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690         unsigned bcount = 0;
691         struct buffer_head *bh;
692         printk("%i indexed blocks...\n", count);
693         for (i = 0; i < count; i++, entries++)
694         {
695                 ext4_lblk_t block = dx_get_block(entries);
696                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
697                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698                 struct stats stats;
699                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
700                 bh = ext4_bread(NULL,dir, block, 0);
701                 if (!bh || IS_ERR(bh))
702                         continue;
703                 stats = levels?
704                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706                         bh->b_data, blocksize, 0);
707                 names += stats.names;
708                 space += stats.space;
709                 bcount += stats.bcount;
710                 brelse(bh);
711         }
712         if (bcount)
713                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714                        levels ? "" : "   ", names, space/bcount,
715                        (space/bcount)*100/blocksize);
716         return (struct stats) { names, space, bcount};
717 }
718 #endif /* DX_DEBUG */
719
720 /*
721  * Probe for a directory leaf block to search.
722  *
723  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724  * error in the directory index, and the caller should fall back to
725  * searching the directory normally.  The callers of dx_probe **MUST**
726  * check for this error code, and make sure it never gets reflected
727  * back to userspace.
728  */
729 static struct dx_frame *
730 dx_probe(struct ext4_filename *fname, struct inode *dir,
731          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
732 {
733         unsigned count, indirect;
734         struct dx_entry *at, *entries, *p, *q, *m;
735         struct dx_root *root;
736         struct dx_frame *frame = frame_in;
737         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
738         u32 hash;
739
740         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741         if (IS_ERR(frame->bh))
742                 return (struct dx_frame *) frame->bh;
743
744         root = (struct dx_root *) frame->bh->b_data;
745         if (root->info.hash_version != DX_HASH_TEA &&
746             root->info.hash_version != DX_HASH_HALF_MD4 &&
747             root->info.hash_version != DX_HASH_LEGACY) {
748                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749                                    root->info.hash_version);
750                 goto fail;
751         }
752         if (fname)
753                 hinfo = &fname->hinfo;
754         hinfo->hash_version = root->info.hash_version;
755         if (hinfo->hash_version <= DX_HASH_TEA)
756                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758         if (fname && fname_name(fname))
759                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
760         hash = hinfo->hash;
761
762         if (root->info.unused_flags & 1) {
763                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764                                    root->info.unused_flags);
765                 goto fail;
766         }
767
768         indirect = root->info.indirect_levels;
769         if (indirect > 1) {
770                 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771                                    root->info.indirect_levels);
772                 goto fail;
773         }
774
775         entries = (struct dx_entry *)(((char *)&root->info) +
776                                       root->info.info_length);
777
778         if (dx_get_limit(entries) != dx_root_limit(dir,
779                                                    root->info.info_length)) {
780                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781                                    dx_get_limit(entries),
782                                    dx_root_limit(dir, root->info.info_length));
783                 goto fail;
784         }
785
786         dxtrace(printk("Look up %x", hash));
787         while (1) {
788                 count = dx_get_count(entries);
789                 if (!count || count > dx_get_limit(entries)) {
790                         ext4_warning_inode(dir,
791                                            "dx entry: count %u beyond limit %u",
792                                            count, dx_get_limit(entries));
793                         goto fail;
794                 }
795
796                 p = entries + 1;
797                 q = entries + count - 1;
798                 while (p <= q) {
799                         m = p + (q - p) / 2;
800                         dxtrace(printk("."));
801                         if (dx_get_hash(m) > hash)
802                                 q = m - 1;
803                         else
804                                 p = m + 1;
805                 }
806
807                 if (0) { // linear search cross check
808                         unsigned n = count - 1;
809                         at = entries;
810                         while (n--)
811                         {
812                                 dxtrace(printk(","));
813                                 if (dx_get_hash(++at) > hash)
814                                 {
815                                         at--;
816                                         break;
817                                 }
818                         }
819                         assert (at == p - 1);
820                 }
821
822                 at = p - 1;
823                 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824                                dx_get_block(at)));
825                 frame->entries = entries;
826                 frame->at = at;
827                 if (!indirect--)
828                         return frame;
829                 frame++;
830                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831                 if (IS_ERR(frame->bh)) {
832                         ret_err = (struct dx_frame *) frame->bh;
833                         frame->bh = NULL;
834                         goto fail;
835                 }
836                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
837
838                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
839                         ext4_warning_inode(dir,
840                                 "dx entry: limit %u != node limit %u",
841                                 dx_get_limit(entries), dx_node_limit(dir));
842                         goto fail;
843                 }
844         }
845 fail:
846         while (frame >= frame_in) {
847                 brelse(frame->bh);
848                 frame--;
849         }
850
851         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852                 ext4_warning_inode(dir,
853                         "Corrupt directory, running e2fsck is recommended");
854         return ret_err;
855 }
856
857 static void dx_release(struct dx_frame *frames)
858 {
859         if (frames[0].bh == NULL)
860                 return;
861
862         if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863                 brelse(frames[1].bh);
864         brelse(frames[0].bh);
865 }
866
867 /*
868  * This function increments the frame pointer to search the next leaf
869  * block, and reads in the necessary intervening nodes if the search
870  * should be necessary.  Whether or not the search is necessary is
871  * controlled by the hash parameter.  If the hash value is even, then
872  * the search is only continued if the next block starts with that
873  * hash value.  This is used if we are searching for a specific file.
874  *
875  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876  *
877  * This function returns 1 if the caller should continue to search,
878  * or 0 if it should not.  If there is an error reading one of the
879  * index blocks, it will a negative error code.
880  *
881  * If start_hash is non-null, it will be filled in with the starting
882  * hash of the next page.
883  */
884 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885                                  struct dx_frame *frame,
886                                  struct dx_frame *frames,
887                                  __u32 *start_hash)
888 {
889         struct dx_frame *p;
890         struct buffer_head *bh;
891         int num_frames = 0;
892         __u32 bhash;
893
894         p = frame;
895         /*
896          * Find the next leaf page by incrementing the frame pointer.
897          * If we run out of entries in the interior node, loop around and
898          * increment pointer in the parent node.  When we break out of
899          * this loop, num_frames indicates the number of interior
900          * nodes need to be read.
901          */
902         while (1) {
903                 if (++(p->at) < p->entries + dx_get_count(p->entries))
904                         break;
905                 if (p == frames)
906                         return 0;
907                 num_frames++;
908                 p--;
909         }
910
911         /*
912          * If the hash is 1, then continue only if the next page has a
913          * continuation hash of any value.  This is used for readdir
914          * handling.  Otherwise, check to see if the hash matches the
915          * desired contiuation hash.  If it doesn't, return since
916          * there's no point to read in the successive index pages.
917          */
918         bhash = dx_get_hash(p->at);
919         if (start_hash)
920                 *start_hash = bhash;
921         if ((hash & 1) == 0) {
922                 if ((bhash & ~1) != hash)
923                         return 0;
924         }
925         /*
926          * If the hash is HASH_NB_ALWAYS, we always go to the next
927          * block so no check is necessary
928          */
929         while (num_frames--) {
930                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
931                 if (IS_ERR(bh))
932                         return PTR_ERR(bh);
933                 p++;
934                 brelse(p->bh);
935                 p->bh = bh;
936                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
937         }
938         return 1;
939 }
940
941
942 /*
943  * This function fills a red-black tree with information from a
944  * directory block.  It returns the number directory entries loaded
945  * into the tree.  If there is an error it is returned in err.
946  */
947 static int htree_dirblock_to_tree(struct file *dir_file,
948                                   struct inode *dir, ext4_lblk_t block,
949                                   struct dx_hash_info *hinfo,
950                                   __u32 start_hash, __u32 start_minor_hash)
951 {
952         struct buffer_head *bh;
953         struct ext4_dir_entry_2 *de, *top;
954         int err = 0, count = 0;
955         struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
956
957         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958                                                         (unsigned long)block));
959         bh = ext4_read_dirblock(dir, block, DIRENT);
960         if (IS_ERR(bh))
961                 return PTR_ERR(bh);
962
963         de = (struct ext4_dir_entry_2 *) bh->b_data;
964         top = (struct ext4_dir_entry_2 *) ((char *) de +
965                                            dir->i_sb->s_blocksize -
966                                            EXT4_DIR_REC_LEN(0));
967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
968         /* Check if the directory is encrypted */
969         if (ext4_encrypted_inode(dir)) {
970                 err = ext4_get_encryption_info(dir);
971                 if (err < 0) {
972                         brelse(bh);
973                         return err;
974                 }
975                 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
976                                                      &fname_crypto_str);
977                 if (err < 0) {
978                         brelse(bh);
979                         return err;
980                 }
981         }
982 #endif
983         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984                 if (ext4_check_dir_entry(dir, NULL, de, bh,
985                                 bh->b_data, bh->b_size,
986                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987                                          + ((char *)de - bh->b_data))) {
988                         /* silently ignore the rest of the block */
989                         break;
990                 }
991                 ext4fs_dirhash(de->name, de->name_len, hinfo);
992                 if ((hinfo->hash < start_hash) ||
993                     ((hinfo->hash == start_hash) &&
994                      (hinfo->minor_hash < start_minor_hash)))
995                         continue;
996                 if (de->inode == 0)
997                         continue;
998                 if (!ext4_encrypted_inode(dir)) {
999                         tmp_str.name = de->name;
1000                         tmp_str.len = de->name_len;
1001                         err = ext4_htree_store_dirent(dir_file,
1002                                    hinfo->hash, hinfo->minor_hash, de,
1003                                    &tmp_str);
1004                 } else {
1005                         int save_len = fname_crypto_str.len;
1006
1007                         /* Directory is encrypted */
1008                         err = ext4_fname_disk_to_usr(dir, hinfo, de,
1009                                                      &fname_crypto_str);
1010                         if (err < 0) {
1011                                 count = err;
1012                                 goto errout;
1013                         }
1014                         err = ext4_htree_store_dirent(dir_file,
1015                                    hinfo->hash, hinfo->minor_hash, de,
1016                                         &fname_crypto_str);
1017                         fname_crypto_str.len = save_len;
1018                 }
1019                 if (err != 0) {
1020                         count = err;
1021                         goto errout;
1022                 }
1023                 count++;
1024         }
1025 errout:
1026         brelse(bh);
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028         ext4_fname_crypto_free_buffer(&fname_crypto_str);
1029 #endif
1030         return count;
1031 }
1032
1033
1034 /*
1035  * This function fills a red-black tree with information from a
1036  * directory.  We start scanning the directory in hash order, starting
1037  * at start_hash and start_minor_hash.
1038  *
1039  * This function returns the number of entries inserted into the tree,
1040  * or a negative error code.
1041  */
1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043                          __u32 start_minor_hash, __u32 *next_hash)
1044 {
1045         struct dx_hash_info hinfo;
1046         struct ext4_dir_entry_2 *de;
1047         struct dx_frame frames[2], *frame;
1048         struct inode *dir;
1049         ext4_lblk_t block;
1050         int count = 0;
1051         int ret, err;
1052         __u32 hashval;
1053         struct ext4_str tmp_str;
1054
1055         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056                        start_hash, start_minor_hash));
1057         dir = file_inode(dir_file);
1058         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060                 if (hinfo.hash_version <= DX_HASH_TEA)
1061                         hinfo.hash_version +=
1062                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064                 if (ext4_has_inline_data(dir)) {
1065                         int has_inline_data = 1;
1066                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1067                                                         &hinfo, start_hash,
1068                                                         start_minor_hash,
1069                                                         &has_inline_data);
1070                         if (has_inline_data) {
1071                                 *next_hash = ~0;
1072                                 return count;
1073                         }
1074                 }
1075                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076                                                start_hash, start_minor_hash);
1077                 *next_hash = ~0;
1078                 return count;
1079         }
1080         hinfo.hash = start_hash;
1081         hinfo.minor_hash = 0;
1082         frame = dx_probe(NULL, dir, &hinfo, frames);
1083         if (IS_ERR(frame))
1084                 return PTR_ERR(frame);
1085
1086         /* Add '.' and '..' from the htree header */
1087         if (!start_hash && !start_minor_hash) {
1088                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089                 tmp_str.name = de->name;
1090                 tmp_str.len = de->name_len;
1091                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1092                                               de, &tmp_str);
1093                 if (err != 0)
1094                         goto errout;
1095                 count++;
1096         }
1097         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100                 tmp_str.name = de->name;
1101                 tmp_str.len = de->name_len;
1102                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1103                                               de, &tmp_str);
1104                 if (err != 0)
1105                         goto errout;
1106                 count++;
1107         }
1108
1109         while (1) {
1110                 if (fatal_signal_pending(current)) {
1111                         err = -ERESTARTSYS;
1112                         goto errout;
1113                 }
1114                 cond_resched();
1115                 block = dx_get_block(frame->at);
1116                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1117                                              start_hash, start_minor_hash);
1118                 if (ret < 0) {
1119                         err = ret;
1120                         goto errout;
1121                 }
1122                 count += ret;
1123                 hashval = ~0;
1124                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1125                                             frame, frames, &hashval);
1126                 *next_hash = hashval;
1127                 if (ret < 0) {
1128                         err = ret;
1129                         goto errout;
1130                 }
1131                 /*
1132                  * Stop if:  (a) there are no more entries, or
1133                  * (b) we have inserted at least one entry and the
1134                  * next hash value is not a continuation
1135                  */
1136                 if ((ret == 0) ||
1137                     (count && ((hashval & 1) == 0)))
1138                         break;
1139         }
1140         dx_release(frames);
1141         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1142                        "next hash: %x\n", count, *next_hash));
1143         return count;
1144 errout:
1145         dx_release(frames);
1146         return (err);
1147 }
1148
1149 static inline int search_dirblock(struct buffer_head *bh,
1150                                   struct inode *dir,
1151                                   struct ext4_filename *fname,
1152                                   const struct qstr *d_name,
1153                                   unsigned int offset,
1154                                   struct ext4_dir_entry_2 **res_dir)
1155 {
1156         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1157                                fname, d_name, offset, res_dir);
1158 }
1159
1160 /*
1161  * Directory block splitting, compacting
1162  */
1163
1164 /*
1165  * Create map of hash values, offsets, and sizes, stored at end of block.
1166  * Returns number of entries mapped.
1167  */
1168 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1169                        unsigned blocksize, struct dx_hash_info *hinfo,
1170                        struct dx_map_entry *map_tail)
1171 {
1172         int count = 0;
1173         char *base = (char *) de;
1174         struct dx_hash_info h = *hinfo;
1175
1176         while ((char *) de < base + blocksize) {
1177                 if (de->name_len && de->inode) {
1178                         ext4fs_dirhash(de->name, de->name_len, &h);
1179                         map_tail--;
1180                         map_tail->hash = h.hash;
1181                         map_tail->offs = ((char *) de - base)>>2;
1182                         map_tail->size = le16_to_cpu(de->rec_len);
1183                         count++;
1184                         cond_resched();
1185                 }
1186                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1187                 de = ext4_next_entry(de, blocksize);
1188         }
1189         return count;
1190 }
1191
1192 /* Sort map by hash value */
1193 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1194 {
1195         struct dx_map_entry *p, *q, *top = map + count - 1;
1196         int more;
1197         /* Combsort until bubble sort doesn't suck */
1198         while (count > 2) {
1199                 count = count*10/13;
1200                 if (count - 9 < 2) /* 9, 10 -> 11 */
1201                         count = 11;
1202                 for (p = top, q = p - count; q >= map; p--, q--)
1203                         if (p->hash < q->hash)
1204                                 swap(*p, *q);
1205         }
1206         /* Garden variety bubble sort */
1207         do {
1208                 more = 0;
1209                 q = top;
1210                 while (q-- > map) {
1211                         if (q[1].hash >= q[0].hash)
1212                                 continue;
1213                         swap(*(q+1), *q);
1214                         more = 1;
1215                 }
1216         } while(more);
1217 }
1218
1219 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1220 {
1221         struct dx_entry *entries = frame->entries;
1222         struct dx_entry *old = frame->at, *new = old + 1;
1223         int count = dx_get_count(entries);
1224
1225         assert(count < dx_get_limit(entries));
1226         assert(old < entries + count);
1227         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1228         dx_set_hash(new, hash);
1229         dx_set_block(new, block);
1230         dx_set_count(entries, count + 1);
1231 }
1232
1233 /*
1234  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1235  *
1236  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1237  * `de != NULL' is guaranteed by caller.
1238  */
1239 static inline int ext4_match(struct ext4_filename *fname,
1240                              struct ext4_dir_entry_2 *de)
1241 {
1242         const void *name = fname_name(fname);
1243         u32 len = fname_len(fname);
1244
1245         if (!de->inode)
1246                 return 0;
1247
1248 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1249         if (unlikely(!name)) {
1250                 if (fname->usr_fname->name[0] == '_') {
1251                         int ret;
1252                         if (de->name_len < 16)
1253                                 return 0;
1254                         ret = memcmp(de->name + de->name_len - 16,
1255                                      fname->crypto_buf.name + 8, 16);
1256                         return (ret == 0) ? 1 : 0;
1257                 }
1258                 name = fname->crypto_buf.name;
1259                 len = fname->crypto_buf.len;
1260         }
1261 #endif
1262         if (de->name_len != len)
1263                 return 0;
1264         return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1265 }
1266
1267 /*
1268  * Returns 0 if not found, -1 on failure, and 1 on success
1269  */
1270 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1271                     struct inode *dir, struct ext4_filename *fname,
1272                     const struct qstr *d_name,
1273                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1274 {
1275         struct ext4_dir_entry_2 * de;
1276         char * dlimit;
1277         int de_len;
1278         int res;
1279
1280         de = (struct ext4_dir_entry_2 *)search_buf;
1281         dlimit = search_buf + buf_size;
1282         while ((char *) de < dlimit) {
1283                 /* this code is executed quadratically often */
1284                 /* do minimal checking `by hand' */
1285                 if ((char *) de + de->name_len <= dlimit) {
1286                         res = ext4_match(fname, de);
1287                         if (res < 0) {
1288                                 res = -1;
1289                                 goto return_result;
1290                         }
1291                         if (res > 0) {
1292                                 /* found a match - just to be sure, do
1293                                  * a full check */
1294                                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1295                                                 bh->b_data,
1296                                                  bh->b_size, offset)) {
1297                                         res = -1;
1298                                         goto return_result;
1299                                 }
1300                                 *res_dir = de;
1301                                 res = 1;
1302                                 goto return_result;
1303                         }
1304
1305                 }
1306                 /* prevent looping on a bad block */
1307                 de_len = ext4_rec_len_from_disk(de->rec_len,
1308                                                 dir->i_sb->s_blocksize);
1309                 if (de_len <= 0) {
1310                         res = -1;
1311                         goto return_result;
1312                 }
1313                 offset += de_len;
1314                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1315         }
1316
1317         res = 0;
1318 return_result:
1319         return res;
1320 }
1321
1322 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1323                                struct ext4_dir_entry *de)
1324 {
1325         struct super_block *sb = dir->i_sb;
1326
1327         if (!is_dx(dir))
1328                 return 0;
1329         if (block == 0)
1330                 return 1;
1331         if (de->inode == 0 &&
1332             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1333                         sb->s_blocksize)
1334                 return 1;
1335         return 0;
1336 }
1337
1338 /*
1339  *      ext4_find_entry()
1340  *
1341  * finds an entry in the specified directory with the wanted name. It
1342  * returns the cache buffer in which the entry was found, and the entry
1343  * itself (as a parameter - res_dir). It does NOT read the inode of the
1344  * entry - you'll have to do that yourself if you want to.
1345  *
1346  * The returned buffer_head has ->b_count elevated.  The caller is expected
1347  * to brelse() it when appropriate.
1348  */
1349 static struct buffer_head * ext4_find_entry (struct inode *dir,
1350                                         const struct qstr *d_name,
1351                                         struct ext4_dir_entry_2 **res_dir,
1352                                         int *inlined)
1353 {
1354         struct super_block *sb;
1355         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1356         struct buffer_head *bh, *ret = NULL;
1357         ext4_lblk_t start, block, b;
1358         const u8 *name = d_name->name;
1359         int ra_max = 0;         /* Number of bh's in the readahead
1360                                    buffer, bh_use[] */
1361         int ra_ptr = 0;         /* Current index into readahead
1362                                    buffer */
1363         int num = 0;
1364         ext4_lblk_t  nblocks;
1365         int i, namelen, retval;
1366         struct ext4_filename fname;
1367
1368         *res_dir = NULL;
1369         sb = dir->i_sb;
1370         namelen = d_name->len;
1371         if (namelen > EXT4_NAME_LEN)
1372                 return NULL;
1373
1374         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1375         if (retval)
1376                 return ERR_PTR(retval);
1377
1378         if (ext4_has_inline_data(dir)) {
1379                 int has_inline_data = 1;
1380                 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1381                                              &has_inline_data);
1382                 if (has_inline_data) {
1383                         if (inlined)
1384                                 *inlined = 1;
1385                         goto cleanup_and_exit;
1386                 }
1387         }
1388
1389         if ((namelen <= 2) && (name[0] == '.') &&
1390             (name[1] == '.' || name[1] == '\0')) {
1391                 /*
1392                  * "." or ".." will only be in the first block
1393                  * NFS may look up ".."; "." should be handled by the VFS
1394                  */
1395                 block = start = 0;
1396                 nblocks = 1;
1397                 goto restart;
1398         }
1399         if (is_dx(dir)) {
1400                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1401                 /*
1402                  * On success, or if the error was file not found,
1403                  * return.  Otherwise, fall back to doing a search the
1404                  * old fashioned way.
1405                  */
1406                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1407                         goto cleanup_and_exit;
1408                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1409                                "falling back\n"));
1410         }
1411         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1412         start = EXT4_I(dir)->i_dir_start_lookup;
1413         if (start >= nblocks)
1414                 start = 0;
1415         block = start;
1416 restart:
1417         do {
1418                 /*
1419                  * We deal with the read-ahead logic here.
1420                  */
1421                 if (ra_ptr >= ra_max) {
1422                         /* Refill the readahead buffer */
1423                         ra_ptr = 0;
1424                         b = block;
1425                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1426                                 /*
1427                                  * Terminate if we reach the end of the
1428                                  * directory and must wrap, or if our
1429                                  * search has finished at this block.
1430                                  */
1431                                 if (b >= nblocks || (num && block == start)) {
1432                                         bh_use[ra_max] = NULL;
1433                                         break;
1434                                 }
1435                                 num++;
1436                                 bh = ext4_getblk(NULL, dir, b++, 0);
1437                                 if (IS_ERR(bh)) {
1438                                         if (ra_max == 0) {
1439                                                 ret = bh;
1440                                                 goto cleanup_and_exit;
1441                                         }
1442                                         break;
1443                                 }
1444                                 bh_use[ra_max] = bh;
1445                                 if (bh)
1446                                         ll_rw_block(READ | REQ_META | REQ_PRIO,
1447                                                     1, &bh);
1448                         }
1449                 }
1450                 if ((bh = bh_use[ra_ptr++]) == NULL)
1451                         goto next;
1452                 wait_on_buffer(bh);
1453                 if (!buffer_uptodate(bh)) {
1454                         /* read error, skip block & hope for the best */
1455                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1456                                          (unsigned long) block);
1457                         brelse(bh);
1458                         goto next;
1459                 }
1460                 if (!buffer_verified(bh) &&
1461                     !is_dx_internal_node(dir, block,
1462                                          (struct ext4_dir_entry *)bh->b_data) &&
1463                     !ext4_dirent_csum_verify(dir,
1464                                 (struct ext4_dir_entry *)bh->b_data)) {
1465                         EXT4_ERROR_INODE(dir, "checksumming directory "
1466                                          "block %lu", (unsigned long)block);
1467                         brelse(bh);
1468                         goto next;
1469                 }
1470                 set_buffer_verified(bh);
1471                 i = search_dirblock(bh, dir, &fname, d_name,
1472                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1473                 if (i == 1) {
1474                         EXT4_I(dir)->i_dir_start_lookup = block;
1475                         ret = bh;
1476                         goto cleanup_and_exit;
1477                 } else {
1478                         brelse(bh);
1479                         if (i < 0)
1480                                 goto cleanup_and_exit;
1481                 }
1482         next:
1483                 if (++block >= nblocks)
1484                         block = 0;
1485         } while (block != start);
1486
1487         /*
1488          * If the directory has grown while we were searching, then
1489          * search the last part of the directory before giving up.
1490          */
1491         block = nblocks;
1492         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1493         if (block < nblocks) {
1494                 start = 0;
1495                 goto restart;
1496         }
1497
1498 cleanup_and_exit:
1499         /* Clean up the read-ahead blocks */
1500         for (; ra_ptr < ra_max; ra_ptr++)
1501                 brelse(bh_use[ra_ptr]);
1502         ext4_fname_free_filename(&fname);
1503         return ret;
1504 }
1505
1506 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1507                         struct ext4_filename *fname,
1508                         struct ext4_dir_entry_2 **res_dir)
1509 {
1510         struct super_block * sb = dir->i_sb;
1511         struct dx_frame frames[2], *frame;
1512         const struct qstr *d_name = fname->usr_fname;
1513         struct buffer_head *bh;
1514         ext4_lblk_t block;
1515         int retval;
1516
1517 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1518         *res_dir = NULL;
1519 #endif
1520         frame = dx_probe(fname, dir, NULL, frames);
1521         if (IS_ERR(frame))
1522                 return (struct buffer_head *) frame;
1523         do {
1524                 block = dx_get_block(frame->at);
1525                 bh = ext4_read_dirblock(dir, block, DIRENT);
1526                 if (IS_ERR(bh))
1527                         goto errout;
1528
1529                 retval = search_dirblock(bh, dir, fname, d_name,
1530                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1531                                          res_dir);
1532                 if (retval == 1)
1533                         goto success;
1534                 brelse(bh);
1535                 if (retval == -1) {
1536                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1537                         goto errout;
1538                 }
1539
1540                 /* Check to see if we should continue to search */
1541                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1542                                                frames, NULL);
1543                 if (retval < 0) {
1544                         ext4_warning_inode(dir,
1545                                 "error %d reading directory index block",
1546                                 retval);
1547                         bh = ERR_PTR(retval);
1548                         goto errout;
1549                 }
1550         } while (retval == 1);
1551
1552         bh = NULL;
1553 errout:
1554         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1555 success:
1556         dx_release(frames);
1557         return bh;
1558 }
1559
1560 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1561 {
1562         struct inode *inode;
1563         struct ext4_dir_entry_2 *de;
1564         struct buffer_head *bh;
1565
1566        if (ext4_encrypted_inode(dir)) {
1567                int res = ext4_get_encryption_info(dir);
1568
1569                 /*
1570                  * This should be a properly defined flag for
1571                  * dentry->d_flags when we uplift this to the VFS.
1572                  * d_fsdata is set to (void *) 1 if if the dentry is
1573                  * created while the directory was encrypted and we
1574                  * don't have access to the key.
1575                  */
1576                dentry->d_fsdata = NULL;
1577                if (ext4_encryption_info(dir))
1578                        dentry->d_fsdata = (void *) 1;
1579                d_set_d_op(dentry, &ext4_encrypted_d_ops);
1580                if (res && res != -ENOKEY)
1581                        return ERR_PTR(res);
1582        }
1583
1584         if (dentry->d_name.len > EXT4_NAME_LEN)
1585                 return ERR_PTR(-ENAMETOOLONG);
1586
1587         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1588         if (IS_ERR(bh))
1589                 return (struct dentry *) bh;
1590         inode = NULL;
1591         if (bh) {
1592                 __u32 ino = le32_to_cpu(de->inode);
1593                 brelse(bh);
1594                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1595                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1596                         return ERR_PTR(-EFSCORRUPTED);
1597                 }
1598                 if (unlikely(ino == dir->i_ino)) {
1599                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1600                                          dentry);
1601                         return ERR_PTR(-EFSCORRUPTED);
1602                 }
1603                 inode = ext4_iget_normal(dir->i_sb, ino);
1604                 if (inode == ERR_PTR(-ESTALE)) {
1605                         EXT4_ERROR_INODE(dir,
1606                                          "deleted inode referenced: %u",
1607                                          ino);
1608                         return ERR_PTR(-EFSCORRUPTED);
1609                 }
1610                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1611                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1612                     !ext4_is_child_context_consistent_with_parent(dir,
1613                                                                   inode)) {
1614                         int nokey = ext4_encrypted_inode(inode) &&
1615                                 !ext4_encryption_info(inode);
1616
1617                         iput(inode);
1618                         if (nokey)
1619                                 return ERR_PTR(-ENOKEY);
1620                         ext4_warning(inode->i_sb,
1621                                      "Inconsistent encryption contexts: %lu/%lu",
1622                                      (unsigned long) dir->i_ino,
1623                                      (unsigned long) inode->i_ino);
1624                         return ERR_PTR(-EPERM);
1625                 }
1626         }
1627         return d_splice_alias(inode, dentry);
1628 }
1629
1630
1631 struct dentry *ext4_get_parent(struct dentry *child)
1632 {
1633         __u32 ino;
1634         static const struct qstr dotdot = QSTR_INIT("..", 2);
1635         struct ext4_dir_entry_2 * de;
1636         struct buffer_head *bh;
1637
1638         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1639         if (IS_ERR(bh))
1640                 return (struct dentry *) bh;
1641         if (!bh)
1642                 return ERR_PTR(-ENOENT);
1643         ino = le32_to_cpu(de->inode);
1644         brelse(bh);
1645
1646         if (!ext4_valid_inum(child->d_sb, ino)) {
1647                 EXT4_ERROR_INODE(d_inode(child),
1648                                  "bad parent inode number: %u", ino);
1649                 return ERR_PTR(-EFSCORRUPTED);
1650         }
1651
1652         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1653 }
1654
1655 /*
1656  * Move count entries from end of map between two memory locations.
1657  * Returns pointer to last entry moved.
1658  */
1659 static struct ext4_dir_entry_2 *
1660 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1661                 unsigned blocksize)
1662 {
1663         unsigned rec_len = 0;
1664
1665         while (count--) {
1666                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1667                                                 (from + (map->offs<<2));
1668                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1669                 memcpy (to, de, rec_len);
1670                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1671                                 ext4_rec_len_to_disk(rec_len, blocksize);
1672                 de->inode = 0;
1673                 map++;
1674                 to += rec_len;
1675         }
1676         return (struct ext4_dir_entry_2 *) (to - rec_len);
1677 }
1678
1679 /*
1680  * Compact each dir entry in the range to the minimal rec_len.
1681  * Returns pointer to last entry in range.
1682  */
1683 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1684 {
1685         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1686         unsigned rec_len = 0;
1687
1688         prev = to = de;
1689         while ((char*)de < base + blocksize) {
1690                 next = ext4_next_entry(de, blocksize);
1691                 if (de->inode && de->name_len) {
1692                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1693                         if (de > to)
1694                                 memmove(to, de, rec_len);
1695                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1696                         prev = to;
1697                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1698                 }
1699                 de = next;
1700         }
1701         return prev;
1702 }
1703
1704 /*
1705  * Split a full leaf block to make room for a new dir entry.
1706  * Allocate a new block, and move entries so that they are approx. equally full.
1707  * Returns pointer to de in block into which the new entry will be inserted.
1708  */
1709 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1710                         struct buffer_head **bh,struct dx_frame *frame,
1711                         struct dx_hash_info *hinfo)
1712 {
1713         unsigned blocksize = dir->i_sb->s_blocksize;
1714         unsigned count, continued;
1715         struct buffer_head *bh2;
1716         ext4_lblk_t newblock;
1717         u32 hash2;
1718         struct dx_map_entry *map;
1719         char *data1 = (*bh)->b_data, *data2;
1720         unsigned split, move, size;
1721         struct ext4_dir_entry_2 *de = NULL, *de2;
1722         struct ext4_dir_entry_tail *t;
1723         int     csum_size = 0;
1724         int     err = 0, i;
1725
1726         if (ext4_has_metadata_csum(dir->i_sb))
1727                 csum_size = sizeof(struct ext4_dir_entry_tail);
1728
1729         bh2 = ext4_append(handle, dir, &newblock);
1730         if (IS_ERR(bh2)) {
1731                 brelse(*bh);
1732                 *bh = NULL;
1733                 return (struct ext4_dir_entry_2 *) bh2;
1734         }
1735
1736         BUFFER_TRACE(*bh, "get_write_access");
1737         err = ext4_journal_get_write_access(handle, *bh);
1738         if (err)
1739                 goto journal_error;
1740
1741         BUFFER_TRACE(frame->bh, "get_write_access");
1742         err = ext4_journal_get_write_access(handle, frame->bh);
1743         if (err)
1744                 goto journal_error;
1745
1746         data2 = bh2->b_data;
1747
1748         /* create map in the end of data2 block */
1749         map = (struct dx_map_entry *) (data2 + blocksize);
1750         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1751                              blocksize, hinfo, map);
1752         map -= count;
1753         dx_sort_map(map, count);
1754         /* Split the existing block in the middle, size-wise */
1755         size = 0;
1756         move = 0;
1757         for (i = count-1; i >= 0; i--) {
1758                 /* is more than half of this entry in 2nd half of the block? */
1759                 if (size + map[i].size/2 > blocksize/2)
1760                         break;
1761                 size += map[i].size;
1762                 move++;
1763         }
1764         /* map index at which we will split */
1765         split = count - move;
1766         hash2 = map[split].hash;
1767         continued = hash2 == map[split - 1].hash;
1768         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1769                         (unsigned long)dx_get_block(frame->at),
1770                                         hash2, split, count-split));
1771
1772         /* Fancy dance to stay within two buffers */
1773         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1774                               blocksize);
1775         de = dx_pack_dirents(data1, blocksize);
1776         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1777                                            (char *) de,
1778                                            blocksize);
1779         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1780                                             (char *) de2,
1781                                             blocksize);
1782         if (csum_size) {
1783                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1784                 initialize_dirent_tail(t, blocksize);
1785
1786                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1787                 initialize_dirent_tail(t, blocksize);
1788         }
1789
1790         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1791                         blocksize, 1));
1792         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1793                         blocksize, 1));
1794
1795         /* Which block gets the new entry? */
1796         if (hinfo->hash >= hash2) {
1797                 swap(*bh, bh2);
1798                 de = de2;
1799         }
1800         dx_insert_block(frame, hash2 + continued, newblock);
1801         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1802         if (err)
1803                 goto journal_error;
1804         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1805         if (err)
1806                 goto journal_error;
1807         brelse(bh2);
1808         dxtrace(dx_show_index("frame", frame->entries));
1809         return de;
1810
1811 journal_error:
1812         brelse(*bh);
1813         brelse(bh2);
1814         *bh = NULL;
1815         ext4_std_error(dir->i_sb, err);
1816         return ERR_PTR(err);
1817 }
1818
1819 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1820                       struct buffer_head *bh,
1821                       void *buf, int buf_size,
1822                       struct ext4_filename *fname,
1823                       struct ext4_dir_entry_2 **dest_de)
1824 {
1825         struct ext4_dir_entry_2 *de;
1826         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1827         int nlen, rlen;
1828         unsigned int offset = 0;
1829         char *top;
1830         int res;
1831
1832         de = (struct ext4_dir_entry_2 *)buf;
1833         top = buf + buf_size - reclen;
1834         while ((char *) de <= top) {
1835                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1836                                          buf, buf_size, offset)) {
1837                         res = -EFSCORRUPTED;
1838                         goto return_result;
1839                 }
1840                 /* Provide crypto context and crypto buffer to ext4 match */
1841                 res = ext4_match(fname, de);
1842                 if (res < 0)
1843                         goto return_result;
1844                 if (res > 0) {
1845                         res = -EEXIST;
1846                         goto return_result;
1847                 }
1848                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1849                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1850                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1851                         break;
1852                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1853                 offset += rlen;
1854         }
1855
1856         if ((char *) de > top)
1857                 res = -ENOSPC;
1858         else {
1859                 *dest_de = de;
1860                 res = 0;
1861         }
1862 return_result:
1863         return res;
1864 }
1865
1866 int ext4_insert_dentry(struct inode *dir,
1867                        struct inode *inode,
1868                        struct ext4_dir_entry_2 *de,
1869                        int buf_size,
1870                        struct ext4_filename *fname)
1871 {
1872
1873         int nlen, rlen;
1874
1875         nlen = EXT4_DIR_REC_LEN(de->name_len);
1876         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1877         if (de->inode) {
1878                 struct ext4_dir_entry_2 *de1 =
1879                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1880                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1881                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1882                 de = de1;
1883         }
1884         de->file_type = EXT4_FT_UNKNOWN;
1885         de->inode = cpu_to_le32(inode->i_ino);
1886         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1887         de->name_len = fname_len(fname);
1888         memcpy(de->name, fname_name(fname), fname_len(fname));
1889         return 0;
1890 }
1891
1892 /*
1893  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1894  * it points to a directory entry which is guaranteed to be large
1895  * enough for new directory entry.  If de is NULL, then
1896  * add_dirent_to_buf will attempt search the directory block for
1897  * space.  It will return -ENOSPC if no space is available, and -EIO
1898  * and -EEXIST if directory entry already exists.
1899  */
1900 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1901                              struct inode *dir,
1902                              struct inode *inode, struct ext4_dir_entry_2 *de,
1903                              struct buffer_head *bh)
1904 {
1905         unsigned int    blocksize = dir->i_sb->s_blocksize;
1906         int             csum_size = 0;
1907         int             err;
1908
1909         if (ext4_has_metadata_csum(inode->i_sb))
1910                 csum_size = sizeof(struct ext4_dir_entry_tail);
1911
1912         if (!de) {
1913                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1914                                         blocksize - csum_size, fname, &de);
1915                 if (err)
1916                         return err;
1917         }
1918         BUFFER_TRACE(bh, "get_write_access");
1919         err = ext4_journal_get_write_access(handle, bh);
1920         if (err) {
1921                 ext4_std_error(dir->i_sb, err);
1922                 return err;
1923         }
1924
1925         /* By now the buffer is marked for journaling. Due to crypto operations,
1926          * the following function call may fail */
1927         err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1928         if (err < 0)
1929                 return err;
1930
1931         /*
1932          * XXX shouldn't update any times until successful
1933          * completion of syscall, but too many callers depend
1934          * on this.
1935          *
1936          * XXX similarly, too many callers depend on
1937          * ext4_new_inode() setting the times, but error
1938          * recovery deletes the inode, so the worst that can
1939          * happen is that the times are slightly out of date
1940          * and/or different from the directory change time.
1941          */
1942         dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1943         ext4_update_dx_flag(dir);
1944         dir->i_version++;
1945         ext4_mark_inode_dirty(handle, dir);
1946         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1947         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1948         if (err)
1949                 ext4_std_error(dir->i_sb, err);
1950         return 0;
1951 }
1952
1953 /*
1954  * This converts a one block unindexed directory to a 3 block indexed
1955  * directory, and adds the dentry to the indexed directory.
1956  */
1957 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1958                             struct inode *dir,
1959                             struct inode *inode, struct buffer_head *bh)
1960 {
1961         struct buffer_head *bh2;
1962         struct dx_root  *root;
1963         struct dx_frame frames[2], *frame;
1964         struct dx_entry *entries;
1965         struct ext4_dir_entry_2 *de, *de2;
1966         struct ext4_dir_entry_tail *t;
1967         char            *data1, *top;
1968         unsigned        len;
1969         int             retval;
1970         unsigned        blocksize;
1971         ext4_lblk_t  block;
1972         struct fake_dirent *fde;
1973         int csum_size = 0;
1974
1975         if (ext4_has_metadata_csum(inode->i_sb))
1976                 csum_size = sizeof(struct ext4_dir_entry_tail);
1977
1978         blocksize =  dir->i_sb->s_blocksize;
1979         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1980         BUFFER_TRACE(bh, "get_write_access");
1981         retval = ext4_journal_get_write_access(handle, bh);
1982         if (retval) {
1983                 ext4_std_error(dir->i_sb, retval);
1984                 brelse(bh);
1985                 return retval;
1986         }
1987         root = (struct dx_root *) bh->b_data;
1988
1989         /* The 0th block becomes the root, move the dirents out */
1990         fde = &root->dotdot;
1991         de = (struct ext4_dir_entry_2 *)((char *)fde +
1992                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1993         if ((char *) de >= (((char *) root) + blocksize)) {
1994                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1995                 brelse(bh);
1996                 return -EFSCORRUPTED;
1997         }
1998         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1999
2000         /* Allocate new block for the 0th block's dirents */
2001         bh2 = ext4_append(handle, dir, &block);
2002         if (IS_ERR(bh2)) {
2003                 brelse(bh);
2004                 return PTR_ERR(bh2);
2005         }
2006         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2007         data1 = bh2->b_data;
2008
2009         memcpy (data1, de, len);
2010         de = (struct ext4_dir_entry_2 *) data1;
2011         top = data1 + len;
2012         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2013                 de = de2;
2014         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2015                                            (char *) de,
2016                                            blocksize);
2017
2018         if (csum_size) {
2019                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2020                 initialize_dirent_tail(t, blocksize);
2021         }
2022
2023         /* Initialize the root; the dot dirents already exist */
2024         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2025         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2026                                            blocksize);
2027         memset (&root->info, 0, sizeof(root->info));
2028         root->info.info_length = sizeof(root->info);
2029         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2030         entries = root->entries;
2031         dx_set_block(entries, 1);
2032         dx_set_count(entries, 1);
2033         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2034
2035         /* Initialize as for dx_probe */
2036         fname->hinfo.hash_version = root->info.hash_version;
2037         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2038                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2039         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2040         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2041
2042         memset(frames, 0, sizeof(frames));
2043         frame = frames;
2044         frame->entries = entries;
2045         frame->at = entries;
2046         frame->bh = bh;
2047         bh = bh2;
2048
2049         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2050         if (retval)
2051                 goto out_frames;        
2052         retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2053         if (retval)
2054                 goto out_frames;        
2055
2056         de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2057         if (IS_ERR(de)) {
2058                 retval = PTR_ERR(de);
2059                 goto out_frames;
2060         }
2061         dx_release(frames);
2062
2063         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2064         brelse(bh);
2065         return retval;
2066 out_frames:
2067         /*
2068          * Even if the block split failed, we have to properly write
2069          * out all the changes we did so far. Otherwise we can end up
2070          * with corrupted filesystem.
2071          */
2072         ext4_mark_inode_dirty(handle, dir);
2073         dx_release(frames);
2074         return retval;
2075 }
2076
2077 /*
2078  *      ext4_add_entry()
2079  *
2080  * adds a file entry to the specified directory, using the same
2081  * semantics as ext4_find_entry(). It returns NULL if it failed.
2082  *
2083  * NOTE!! The inode part of 'de' is left at 0 - which means you
2084  * may not sleep between calling this and putting something into
2085  * the entry, as someone else might have used it while you slept.
2086  */
2087 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2088                           struct inode *inode)
2089 {
2090         struct inode *dir = d_inode(dentry->d_parent);
2091         struct buffer_head *bh = NULL;
2092         struct ext4_dir_entry_2 *de;
2093         struct ext4_dir_entry_tail *t;
2094         struct super_block *sb;
2095         struct ext4_filename fname;
2096         int     retval;
2097         int     dx_fallback=0;
2098         unsigned blocksize;
2099         ext4_lblk_t block, blocks;
2100         int     csum_size = 0;
2101
2102         if (ext4_has_metadata_csum(inode->i_sb))
2103                 csum_size = sizeof(struct ext4_dir_entry_tail);
2104
2105         sb = dir->i_sb;
2106         blocksize = sb->s_blocksize;
2107         if (!dentry->d_name.len)
2108                 return -EINVAL;
2109
2110         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2111         if (retval)
2112                 return retval;
2113
2114         if (ext4_has_inline_data(dir)) {
2115                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2116                 if (retval < 0)
2117                         goto out;
2118                 if (retval == 1) {
2119                         retval = 0;
2120                         goto out;
2121                 }
2122         }
2123
2124         if (is_dx(dir)) {
2125                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2126                 if (!retval || (retval != ERR_BAD_DX_DIR))
2127                         goto out;
2128                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2129                 dx_fallback++;
2130                 ext4_mark_inode_dirty(handle, dir);
2131         }
2132         blocks = dir->i_size >> sb->s_blocksize_bits;
2133         for (block = 0; block < blocks; block++) {
2134                 bh = ext4_read_dirblock(dir, block, DIRENT);
2135                 if (IS_ERR(bh)) {
2136                         retval = PTR_ERR(bh);
2137                         bh = NULL;
2138                         goto out;
2139                 }
2140                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2141                                            NULL, bh);
2142                 if (retval != -ENOSPC)
2143                         goto out;
2144
2145                 if (blocks == 1 && !dx_fallback &&
2146                     ext4_has_feature_dir_index(sb)) {
2147                         retval = make_indexed_dir(handle, &fname, dir,
2148                                                   inode, bh);
2149                         bh = NULL; /* make_indexed_dir releases bh */
2150                         goto out;
2151                 }
2152                 brelse(bh);
2153         }
2154         bh = ext4_append(handle, dir, &block);
2155         if (IS_ERR(bh)) {
2156                 retval = PTR_ERR(bh);
2157                 bh = NULL;
2158                 goto out;
2159         }
2160         de = (struct ext4_dir_entry_2 *) bh->b_data;
2161         de->inode = 0;
2162         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2163
2164         if (csum_size) {
2165                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2166                 initialize_dirent_tail(t, blocksize);
2167         }
2168
2169         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2170 out:
2171         ext4_fname_free_filename(&fname);
2172         brelse(bh);
2173         if (retval == 0)
2174                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2175         return retval;
2176 }
2177
2178 /*
2179  * Returns 0 for success, or a negative error value
2180  */
2181 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2182                              struct inode *dir, struct inode *inode)
2183 {
2184         struct dx_frame frames[2], *frame;
2185         struct dx_entry *entries, *at;
2186         struct buffer_head *bh;
2187         struct super_block *sb = dir->i_sb;
2188         struct ext4_dir_entry_2 *de;
2189         int err;
2190
2191         frame = dx_probe(fname, dir, NULL, frames);
2192         if (IS_ERR(frame))
2193                 return PTR_ERR(frame);
2194         entries = frame->entries;
2195         at = frame->at;
2196         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2197         if (IS_ERR(bh)) {
2198                 err = PTR_ERR(bh);
2199                 bh = NULL;
2200                 goto cleanup;
2201         }
2202
2203         BUFFER_TRACE(bh, "get_write_access");
2204         err = ext4_journal_get_write_access(handle, bh);
2205         if (err)
2206                 goto journal_error;
2207
2208         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2209         if (err != -ENOSPC)
2210                 goto cleanup;
2211
2212         /* Block full, should compress but for now just split */
2213         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2214                        dx_get_count(entries), dx_get_limit(entries)));
2215         /* Need to split index? */
2216         if (dx_get_count(entries) == dx_get_limit(entries)) {
2217                 ext4_lblk_t newblock;
2218                 unsigned icount = dx_get_count(entries);
2219                 int levels = frame - frames;
2220                 struct dx_entry *entries2;
2221                 struct dx_node *node2;
2222                 struct buffer_head *bh2;
2223
2224                 if (levels && (dx_get_count(frames->entries) ==
2225                                dx_get_limit(frames->entries))) {
2226                         ext4_warning_inode(dir, "Directory index full!");
2227                         err = -ENOSPC;
2228                         goto cleanup;
2229                 }
2230                 bh2 = ext4_append(handle, dir, &newblock);
2231                 if (IS_ERR(bh2)) {
2232                         err = PTR_ERR(bh2);
2233                         goto cleanup;
2234                 }
2235                 node2 = (struct dx_node *)(bh2->b_data);
2236                 entries2 = node2->entries;
2237                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2238                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2239                                                            sb->s_blocksize);
2240                 BUFFER_TRACE(frame->bh, "get_write_access");
2241                 err = ext4_journal_get_write_access(handle, frame->bh);
2242                 if (err)
2243                         goto journal_error;
2244                 if (levels) {
2245                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2246                         unsigned hash2 = dx_get_hash(entries + icount1);
2247                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2248                                        icount1, icount2));
2249
2250                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2251                         err = ext4_journal_get_write_access(handle,
2252                                                              frames[0].bh);
2253                         if (err)
2254                                 goto journal_error;
2255
2256                         memcpy((char *) entries2, (char *) (entries + icount1),
2257                                icount2 * sizeof(struct dx_entry));
2258                         dx_set_count(entries, icount1);
2259                         dx_set_count(entries2, icount2);
2260                         dx_set_limit(entries2, dx_node_limit(dir));
2261
2262                         /* Which index block gets the new entry? */
2263                         if (at - entries >= icount1) {
2264                                 frame->at = at = at - entries - icount1 + entries2;
2265                                 frame->entries = entries = entries2;
2266                                 swap(frame->bh, bh2);
2267                         }
2268                         dx_insert_block(frames + 0, hash2, newblock);
2269                         dxtrace(dx_show_index("node", frames[1].entries));
2270                         dxtrace(dx_show_index("node",
2271                                ((struct dx_node *) bh2->b_data)->entries));
2272                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2273                         if (err)
2274                                 goto journal_error;
2275                         brelse (bh2);
2276                 } else {
2277                         dxtrace(printk(KERN_DEBUG
2278                                        "Creating second level index...\n"));
2279                         memcpy((char *) entries2, (char *) entries,
2280                                icount * sizeof(struct dx_entry));
2281                         dx_set_limit(entries2, dx_node_limit(dir));
2282
2283                         /* Set up root */
2284                         dx_set_count(entries, 1);
2285                         dx_set_block(entries + 0, newblock);
2286                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2287
2288                         /* Add new access path frame */
2289                         frame = frames + 1;
2290                         frame->at = at = at - entries + entries2;
2291                         frame->entries = entries = entries2;
2292                         frame->bh = bh2;
2293                         err = ext4_journal_get_write_access(handle,
2294                                                              frame->bh);
2295                         if (err)
2296                                 goto journal_error;
2297                 }
2298                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2299                 if (err) {
2300                         ext4_std_error(inode->i_sb, err);
2301                         goto cleanup;
2302                 }
2303         }
2304         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2305         if (IS_ERR(de)) {
2306                 err = PTR_ERR(de);
2307                 goto cleanup;
2308         }
2309         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2310         goto cleanup;
2311
2312 journal_error:
2313         ext4_std_error(dir->i_sb, err);
2314 cleanup:
2315         brelse(bh);
2316         dx_release(frames);
2317         return err;
2318 }
2319
2320 /*
2321  * ext4_generic_delete_entry deletes a directory entry by merging it
2322  * with the previous entry
2323  */
2324 int ext4_generic_delete_entry(handle_t *handle,
2325                               struct inode *dir,
2326                               struct ext4_dir_entry_2 *de_del,
2327                               struct buffer_head *bh,
2328                               void *entry_buf,
2329                               int buf_size,
2330                               int csum_size)
2331 {
2332         struct ext4_dir_entry_2 *de, *pde;
2333         unsigned int blocksize = dir->i_sb->s_blocksize;
2334         int i;
2335
2336         i = 0;
2337         pde = NULL;
2338         de = (struct ext4_dir_entry_2 *)entry_buf;
2339         while (i < buf_size - csum_size) {
2340                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2341                                          bh->b_data, bh->b_size, i))
2342                         return -EFSCORRUPTED;
2343                 if (de == de_del)  {
2344                         if (pde)
2345                                 pde->rec_len = ext4_rec_len_to_disk(
2346                                         ext4_rec_len_from_disk(pde->rec_len,
2347                                                                blocksize) +
2348                                         ext4_rec_len_from_disk(de->rec_len,
2349                                                                blocksize),
2350                                         blocksize);
2351                         else
2352                                 de->inode = 0;
2353                         dir->i_version++;
2354                         return 0;
2355                 }
2356                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2357                 pde = de;
2358                 de = ext4_next_entry(de, blocksize);
2359         }
2360         return -ENOENT;
2361 }
2362
2363 static int ext4_delete_entry(handle_t *handle,
2364                              struct inode *dir,
2365                              struct ext4_dir_entry_2 *de_del,
2366                              struct buffer_head *bh)
2367 {
2368         int err, csum_size = 0;
2369
2370         if (ext4_has_inline_data(dir)) {
2371                 int has_inline_data = 1;
2372                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2373                                                &has_inline_data);
2374                 if (has_inline_data)
2375                         return err;
2376         }
2377
2378         if (ext4_has_metadata_csum(dir->i_sb))
2379                 csum_size = sizeof(struct ext4_dir_entry_tail);
2380
2381         BUFFER_TRACE(bh, "get_write_access");
2382         err = ext4_journal_get_write_access(handle, bh);
2383         if (unlikely(err))
2384                 goto out;
2385
2386         err = ext4_generic_delete_entry(handle, dir, de_del,
2387                                         bh, bh->b_data,
2388                                         dir->i_sb->s_blocksize, csum_size);
2389         if (err)
2390                 goto out;
2391
2392         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2393         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2394         if (unlikely(err))
2395                 goto out;
2396
2397         return 0;
2398 out:
2399         if (err != -ENOENT)
2400                 ext4_std_error(dir->i_sb, err);
2401         return err;
2402 }
2403
2404 /*
2405  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2406  * since this indicates that nlinks count was previously 1.
2407  */
2408 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2409 {
2410         inc_nlink(inode);
2411         if (is_dx(inode) && inode->i_nlink > 1) {
2412                 /* limit is 16-bit i_links_count */
2413                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2414                         set_nlink(inode, 1);
2415                         ext4_set_feature_dir_nlink(inode->i_sb);
2416                 }
2417         }
2418 }
2419
2420 /*
2421  * If a directory had nlink == 1, then we should let it be 1. This indicates
2422  * directory has >EXT4_LINK_MAX subdirs.
2423  */
2424 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2425 {
2426         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2427                 drop_nlink(inode);
2428 }
2429
2430
2431 static int ext4_add_nondir(handle_t *handle,
2432                 struct dentry *dentry, struct inode *inode)
2433 {
2434         int err = ext4_add_entry(handle, dentry, inode);
2435         if (!err) {
2436                 ext4_mark_inode_dirty(handle, inode);
2437                 unlock_new_inode(inode);
2438                 d_instantiate(dentry, inode);
2439                 return 0;
2440         }
2441         drop_nlink(inode);
2442         unlock_new_inode(inode);
2443         iput(inode);
2444         return err;
2445 }
2446
2447 /*
2448  * By the time this is called, we already have created
2449  * the directory cache entry for the new file, but it
2450  * is so far negative - it has no inode.
2451  *
2452  * If the create succeeds, we fill in the inode information
2453  * with d_instantiate().
2454  */
2455 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2456                        bool excl)
2457 {
2458         handle_t *handle;
2459         struct inode *inode;
2460         int err, credits, retries = 0;
2461
2462         err = dquot_initialize(dir);
2463         if (err)
2464                 return err;
2465
2466         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2467                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2468 retry:
2469         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2470                                             NULL, EXT4_HT_DIR, credits);
2471         handle = ext4_journal_current_handle();
2472         err = PTR_ERR(inode);
2473         if (!IS_ERR(inode)) {
2474                 inode->i_op = &ext4_file_inode_operations;
2475                 inode->i_fop = &ext4_file_operations;
2476                 ext4_set_aops(inode);
2477                 err = ext4_add_nondir(handle, dentry, inode);
2478                 if (!err && IS_DIRSYNC(dir))
2479                         ext4_handle_sync(handle);
2480         }
2481         if (handle)
2482                 ext4_journal_stop(handle);
2483         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2484                 goto retry;
2485         return err;
2486 }
2487
2488 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2489                       umode_t mode, dev_t rdev)
2490 {
2491         handle_t *handle;
2492         struct inode *inode;
2493         int err, credits, retries = 0;
2494
2495         err = dquot_initialize(dir);
2496         if (err)
2497                 return err;
2498
2499         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2500                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2501 retry:
2502         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2503                                             NULL, EXT4_HT_DIR, credits);
2504         handle = ext4_journal_current_handle();
2505         err = PTR_ERR(inode);
2506         if (!IS_ERR(inode)) {
2507                 init_special_inode(inode, inode->i_mode, rdev);
2508                 inode->i_op = &ext4_special_inode_operations;
2509                 err = ext4_add_nondir(handle, dentry, inode);
2510                 if (!err && IS_DIRSYNC(dir))
2511                         ext4_handle_sync(handle);
2512         }
2513         if (handle)
2514                 ext4_journal_stop(handle);
2515         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2516                 goto retry;
2517         return err;
2518 }
2519
2520 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2521 {
2522         handle_t *handle;
2523         struct inode *inode;
2524         int err, retries = 0;
2525
2526         err = dquot_initialize(dir);
2527         if (err)
2528                 return err;
2529
2530 retry:
2531         inode = ext4_new_inode_start_handle(dir, mode,
2532                                             NULL, 0, NULL,
2533                                             EXT4_HT_DIR,
2534                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2535                           4 + EXT4_XATTR_TRANS_BLOCKS);
2536         handle = ext4_journal_current_handle();
2537         err = PTR_ERR(inode);
2538         if (!IS_ERR(inode)) {
2539                 inode->i_op = &ext4_file_inode_operations;
2540                 inode->i_fop = &ext4_file_operations;
2541                 ext4_set_aops(inode);
2542                 d_tmpfile(dentry, inode);
2543                 err = ext4_orphan_add(handle, inode);
2544                 if (err)
2545                         goto err_unlock_inode;
2546                 mark_inode_dirty(inode);
2547                 unlock_new_inode(inode);
2548         }
2549         if (handle)
2550                 ext4_journal_stop(handle);
2551         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2552                 goto retry;
2553         return err;
2554 err_unlock_inode:
2555         ext4_journal_stop(handle);
2556         unlock_new_inode(inode);
2557         return err;
2558 }
2559
2560 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2561                           struct ext4_dir_entry_2 *de,
2562                           int blocksize, int csum_size,
2563                           unsigned int parent_ino, int dotdot_real_len)
2564 {
2565         de->inode = cpu_to_le32(inode->i_ino);
2566         de->name_len = 1;
2567         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2568                                            blocksize);
2569         strcpy(de->name, ".");
2570         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2571
2572         de = ext4_next_entry(de, blocksize);
2573         de->inode = cpu_to_le32(parent_ino);
2574         de->name_len = 2;
2575         if (!dotdot_real_len)
2576                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2577                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2578                                         blocksize);
2579         else
2580                 de->rec_len = ext4_rec_len_to_disk(
2581                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2582         strcpy(de->name, "..");
2583         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2584
2585         return ext4_next_entry(de, blocksize);
2586 }
2587
2588 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2589                              struct inode *inode)
2590 {
2591         struct buffer_head *dir_block = NULL;
2592         struct ext4_dir_entry_2 *de;
2593         struct ext4_dir_entry_tail *t;
2594         ext4_lblk_t block = 0;
2595         unsigned int blocksize = dir->i_sb->s_blocksize;
2596         int csum_size = 0;
2597         int err;
2598
2599         if (ext4_has_metadata_csum(dir->i_sb))
2600                 csum_size = sizeof(struct ext4_dir_entry_tail);
2601
2602         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2603                 err = ext4_try_create_inline_dir(handle, dir, inode);
2604                 if (err < 0 && err != -ENOSPC)
2605                         goto out;
2606                 if (!err)
2607                         goto out;
2608         }
2609
2610         inode->i_size = 0;
2611         dir_block = ext4_append(handle, inode, &block);
2612         if (IS_ERR(dir_block))
2613                 return PTR_ERR(dir_block);
2614         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2615         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2616         set_nlink(inode, 2);
2617         if (csum_size) {
2618                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2619                 initialize_dirent_tail(t, blocksize);
2620         }
2621
2622         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2623         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2624         if (err)
2625                 goto out;
2626         set_buffer_verified(dir_block);
2627 out:
2628         brelse(dir_block);
2629         return err;
2630 }
2631
2632 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2633 {
2634         handle_t *handle;
2635         struct inode *inode;
2636         int err, credits, retries = 0;
2637
2638         if (EXT4_DIR_LINK_MAX(dir))
2639                 return -EMLINK;
2640
2641         err = dquot_initialize(dir);
2642         if (err)
2643                 return err;
2644
2645         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2646                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2647 retry:
2648         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2649                                             &dentry->d_name,
2650                                             0, NULL, EXT4_HT_DIR, credits);
2651         handle = ext4_journal_current_handle();
2652         err = PTR_ERR(inode);
2653         if (IS_ERR(inode))
2654                 goto out_stop;
2655
2656         inode->i_op = &ext4_dir_inode_operations;
2657         inode->i_fop = &ext4_dir_operations;
2658         err = ext4_init_new_dir(handle, dir, inode);
2659         if (err)
2660                 goto out_clear_inode;
2661         err = ext4_mark_inode_dirty(handle, inode);
2662         if (!err)
2663                 err = ext4_add_entry(handle, dentry, inode);
2664         if (err) {
2665 out_clear_inode:
2666                 clear_nlink(inode);
2667                 unlock_new_inode(inode);
2668                 ext4_mark_inode_dirty(handle, inode);
2669                 iput(inode);
2670                 goto out_stop;
2671         }
2672         ext4_inc_count(handle, dir);
2673         ext4_update_dx_flag(dir);
2674         err = ext4_mark_inode_dirty(handle, dir);
2675         if (err)
2676                 goto out_clear_inode;
2677         unlock_new_inode(inode);
2678         d_instantiate(dentry, inode);
2679         if (IS_DIRSYNC(dir))
2680                 ext4_handle_sync(handle);
2681
2682 out_stop:
2683         if (handle)
2684                 ext4_journal_stop(handle);
2685         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2686                 goto retry;
2687         return err;
2688 }
2689
2690 /*
2691  * routine to check that the specified directory is empty (for rmdir)
2692  */
2693 int ext4_empty_dir(struct inode *inode)
2694 {
2695         unsigned int offset;
2696         struct buffer_head *bh;
2697         struct ext4_dir_entry_2 *de, *de1;
2698         struct super_block *sb;
2699         int err = 0;
2700
2701         if (ext4_has_inline_data(inode)) {
2702                 int has_inline_data = 1;
2703
2704                 err = empty_inline_dir(inode, &has_inline_data);
2705                 if (has_inline_data)
2706                         return err;
2707         }
2708
2709         sb = inode->i_sb;
2710         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2711                 EXT4_ERROR_INODE(inode, "invalid size");
2712                 return 1;
2713         }
2714         bh = ext4_read_dirblock(inode, 0, EITHER);
2715         if (IS_ERR(bh))
2716                 return 1;
2717
2718         de = (struct ext4_dir_entry_2 *) bh->b_data;
2719         de1 = ext4_next_entry(de, sb->s_blocksize);
2720         if (le32_to_cpu(de->inode) != inode->i_ino ||
2721                         le32_to_cpu(de1->inode) == 0 ||
2722                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2723                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2724                 brelse(bh);
2725                 return 1;
2726         }
2727         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2728                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2729         de = ext4_next_entry(de1, sb->s_blocksize);
2730         while (offset < inode->i_size) {
2731                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2732                         unsigned int lblock;
2733                         err = 0;
2734                         brelse(bh);
2735                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2736                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2737                         if (IS_ERR(bh))
2738                                 return 1;
2739                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2740                 }
2741                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2742                                          bh->b_data, bh->b_size, offset)) {
2743                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2744                                                          sb->s_blocksize);
2745                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2746                         continue;
2747                 }
2748                 if (le32_to_cpu(de->inode)) {
2749                         brelse(bh);
2750                         return 0;
2751                 }
2752                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2753                 de = ext4_next_entry(de, sb->s_blocksize);
2754         }
2755         brelse(bh);
2756         return 1;
2757 }
2758
2759 /*
2760  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2761  * such inodes, starting at the superblock, in case we crash before the
2762  * file is closed/deleted, or in case the inode truncate spans multiple
2763  * transactions and the last transaction is not recovered after a crash.
2764  *
2765  * At filesystem recovery time, we walk this list deleting unlinked
2766  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2767  *
2768  * Orphan list manipulation functions must be called under i_mutex unless
2769  * we are just creating the inode or deleting it.
2770  */
2771 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2772 {
2773         struct super_block *sb = inode->i_sb;
2774         struct ext4_sb_info *sbi = EXT4_SB(sb);
2775         struct ext4_iloc iloc;
2776         int err = 0, rc;
2777         bool dirty = false;
2778
2779         if (!sbi->s_journal || is_bad_inode(inode))
2780                 return 0;
2781
2782         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2783                      !inode_is_locked(inode));
2784         /*
2785          * Exit early if inode already is on orphan list. This is a big speedup
2786          * since we don't have to contend on the global s_orphan_lock.
2787          */
2788         if (!list_empty(&EXT4_I(inode)->i_orphan))
2789                 return 0;
2790
2791         /*
2792          * Orphan handling is only valid for files with data blocks
2793          * being truncated, or files being unlinked. Note that we either
2794          * hold i_mutex, or the inode can not be referenced from outside,
2795          * so i_nlink should not be bumped due to race
2796          */
2797         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2798                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2799
2800         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2801         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2802         if (err)
2803                 goto out;
2804
2805         err = ext4_reserve_inode_write(handle, inode, &iloc);
2806         if (err)
2807                 goto out;
2808
2809         mutex_lock(&sbi->s_orphan_lock);
2810         /*
2811          * Due to previous errors inode may be already a part of on-disk
2812          * orphan list. If so skip on-disk list modification.
2813          */
2814         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2815             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2816                 /* Insert this inode at the head of the on-disk orphan list */
2817                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2818                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2819                 dirty = true;
2820         }
2821         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2822         mutex_unlock(&sbi->s_orphan_lock);
2823
2824         if (dirty) {
2825                 err = ext4_handle_dirty_super(handle, sb);
2826                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2827                 if (!err)
2828                         err = rc;
2829                 if (err) {
2830                         /*
2831                          * We have to remove inode from in-memory list if
2832                          * addition to on disk orphan list failed. Stray orphan
2833                          * list entries can cause panics at unmount time.
2834                          */
2835                         mutex_lock(&sbi->s_orphan_lock);
2836                         list_del_init(&EXT4_I(inode)->i_orphan);
2837                         mutex_unlock(&sbi->s_orphan_lock);
2838                 }
2839         }
2840         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2841         jbd_debug(4, "orphan inode %lu will point to %d\n",
2842                         inode->i_ino, NEXT_ORPHAN(inode));
2843 out:
2844         ext4_std_error(sb, err);
2845         return err;
2846 }
2847
2848 /*
2849  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2850  * of such inodes stored on disk, because it is finally being cleaned up.
2851  */
2852 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2853 {
2854         struct list_head *prev;
2855         struct ext4_inode_info *ei = EXT4_I(inode);
2856         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2857         __u32 ino_next;
2858         struct ext4_iloc iloc;
2859         int err = 0;
2860
2861         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2862                 return 0;
2863
2864         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2865                      !inode_is_locked(inode));
2866         /* Do this quick check before taking global s_orphan_lock. */
2867         if (list_empty(&ei->i_orphan))
2868                 return 0;
2869
2870         if (handle) {
2871                 /* Grab inode buffer early before taking global s_orphan_lock */
2872                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2873         }
2874
2875         mutex_lock(&sbi->s_orphan_lock);
2876         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2877
2878         prev = ei->i_orphan.prev;
2879         list_del_init(&ei->i_orphan);
2880
2881         /* If we're on an error path, we may not have a valid
2882          * transaction handle with which to update the orphan list on
2883          * disk, but we still need to remove the inode from the linked
2884          * list in memory. */
2885         if (!handle || err) {
2886                 mutex_unlock(&sbi->s_orphan_lock);
2887                 goto out_err;
2888         }
2889
2890         ino_next = NEXT_ORPHAN(inode);
2891         if (prev == &sbi->s_orphan) {
2892                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2893                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2894                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2895                 if (err) {
2896                         mutex_unlock(&sbi->s_orphan_lock);
2897                         goto out_brelse;
2898                 }
2899                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2900                 mutex_unlock(&sbi->s_orphan_lock);
2901                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2902         } else {
2903                 struct ext4_iloc iloc2;
2904                 struct inode *i_prev =
2905                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2906
2907                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2908                           i_prev->i_ino, ino_next);
2909                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2910                 if (err) {
2911                         mutex_unlock(&sbi->s_orphan_lock);
2912                         goto out_brelse;
2913                 }
2914                 NEXT_ORPHAN(i_prev) = ino_next;
2915                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2916                 mutex_unlock(&sbi->s_orphan_lock);
2917         }
2918         if (err)
2919                 goto out_brelse;
2920         NEXT_ORPHAN(inode) = 0;
2921         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2922 out_err:
2923         ext4_std_error(inode->i_sb, err);
2924         return err;
2925
2926 out_brelse:
2927         brelse(iloc.bh);
2928         goto out_err;
2929 }
2930
2931 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2932 {
2933         int retval;
2934         struct inode *inode;
2935         struct buffer_head *bh;
2936         struct ext4_dir_entry_2 *de;
2937         handle_t *handle = NULL;
2938
2939         /* Initialize quotas before so that eventual writes go in
2940          * separate transaction */
2941         retval = dquot_initialize(dir);
2942         if (retval)
2943                 return retval;
2944         retval = dquot_initialize(d_inode(dentry));
2945         if (retval)
2946                 return retval;
2947
2948         retval = -ENOENT;
2949         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2950         if (IS_ERR(bh))
2951                 return PTR_ERR(bh);
2952         if (!bh)
2953                 goto end_rmdir;
2954
2955         inode = d_inode(dentry);
2956
2957         retval = -EFSCORRUPTED;
2958         if (le32_to_cpu(de->inode) != inode->i_ino)
2959                 goto end_rmdir;
2960
2961         retval = -ENOTEMPTY;
2962         if (!ext4_empty_dir(inode))
2963                 goto end_rmdir;
2964
2965         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2966                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2967         if (IS_ERR(handle)) {
2968                 retval = PTR_ERR(handle);
2969                 handle = NULL;
2970                 goto end_rmdir;
2971         }
2972
2973         if (IS_DIRSYNC(dir))
2974                 ext4_handle_sync(handle);
2975
2976         retval = ext4_delete_entry(handle, dir, de, bh);
2977         if (retval)
2978                 goto end_rmdir;
2979         if (!EXT4_DIR_LINK_EMPTY(inode))
2980                 ext4_warning_inode(inode,
2981                              "empty directory '%.*s' has too many links (%u)",
2982                              dentry->d_name.len, dentry->d_name.name,
2983                              inode->i_nlink);
2984         inode->i_version++;
2985         clear_nlink(inode);
2986         /* There's no need to set i_disksize: the fact that i_nlink is
2987          * zero will ensure that the right thing happens during any
2988          * recovery. */
2989         inode->i_size = 0;
2990         ext4_orphan_add(handle, inode);
2991         inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2992         ext4_mark_inode_dirty(handle, inode);
2993         ext4_dec_count(handle, dir);
2994         ext4_update_dx_flag(dir);
2995         ext4_mark_inode_dirty(handle, dir);
2996
2997 end_rmdir:
2998         brelse(bh);
2999         if (handle)
3000                 ext4_journal_stop(handle);
3001         return retval;
3002 }
3003
3004 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3005 {
3006         int retval;
3007         struct inode *inode;
3008         struct buffer_head *bh;
3009         struct ext4_dir_entry_2 *de;
3010         handle_t *handle = NULL;
3011
3012         trace_ext4_unlink_enter(dir, dentry);
3013         /* Initialize quotas before so that eventual writes go
3014          * in separate transaction */
3015         retval = dquot_initialize(dir);
3016         if (retval)
3017                 return retval;
3018         retval = dquot_initialize(d_inode(dentry));
3019         if (retval)
3020                 return retval;
3021
3022         retval = -ENOENT;
3023         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3024         if (IS_ERR(bh))
3025                 return PTR_ERR(bh);
3026         if (!bh)
3027                 goto end_unlink;
3028
3029         inode = d_inode(dentry);
3030
3031         retval = -EFSCORRUPTED;
3032         if (le32_to_cpu(de->inode) != inode->i_ino)
3033                 goto end_unlink;
3034
3035         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3036                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3037         if (IS_ERR(handle)) {
3038                 retval = PTR_ERR(handle);
3039                 handle = NULL;
3040                 goto end_unlink;
3041         }
3042
3043         if (IS_DIRSYNC(dir))
3044                 ext4_handle_sync(handle);
3045
3046         if (inode->i_nlink == 0) {
3047                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3048                                    dentry->d_name.len, dentry->d_name.name);
3049                 set_nlink(inode, 1);
3050         }
3051         retval = ext4_delete_entry(handle, dir, de, bh);
3052         if (retval)
3053                 goto end_unlink;
3054         dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3055         ext4_update_dx_flag(dir);
3056         ext4_mark_inode_dirty(handle, dir);
3057         drop_nlink(inode);
3058         if (!inode->i_nlink)
3059                 ext4_orphan_add(handle, inode);
3060         inode->i_ctime = ext4_current_time(inode);
3061         ext4_mark_inode_dirty(handle, inode);
3062
3063 end_unlink:
3064         brelse(bh);
3065         if (handle)
3066                 ext4_journal_stop(handle);
3067         trace_ext4_unlink_exit(dentry, retval);
3068         return retval;
3069 }
3070
3071 static int ext4_symlink(struct inode *dir,
3072                         struct dentry *dentry, const char *symname)
3073 {
3074         handle_t *handle;
3075         struct inode *inode;
3076         int err, len = strlen(symname);
3077         int credits;
3078         bool encryption_required;
3079         struct ext4_str disk_link;
3080         struct ext4_encrypted_symlink_data *sd = NULL;
3081
3082         disk_link.len = len + 1;
3083         disk_link.name = (char *) symname;
3084
3085         encryption_required = (ext4_encrypted_inode(dir) ||
3086                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3087         if (encryption_required) {
3088                 err = ext4_get_encryption_info(dir);
3089                 if (err)
3090                         return err;
3091                 if (ext4_encryption_info(dir) == NULL)
3092                         return -EPERM;
3093                 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3094                                  sizeof(struct ext4_encrypted_symlink_data));
3095                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3096                 if (!sd)
3097                         return -ENOMEM;
3098         }
3099
3100         if (disk_link.len > dir->i_sb->s_blocksize) {
3101                 err = -ENAMETOOLONG;
3102                 goto err_free_sd;
3103         }
3104
3105         err = dquot_initialize(dir);
3106         if (err)
3107                 goto err_free_sd;
3108
3109         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3110                 /*
3111                  * For non-fast symlinks, we just allocate inode and put it on
3112                  * orphan list in the first transaction => we need bitmap,
3113                  * group descriptor, sb, inode block, quota blocks, and
3114                  * possibly selinux xattr blocks.
3115                  */
3116                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3117                           EXT4_XATTR_TRANS_BLOCKS;
3118         } else {
3119                 /*
3120                  * Fast symlink. We have to add entry to directory
3121                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3122                  * allocate new inode (bitmap, group descriptor, inode block,
3123                  * quota blocks, sb is already counted in previous macros).
3124                  */
3125                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3126                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3127         }
3128
3129         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3130                                             &dentry->d_name, 0, NULL,
3131                                             EXT4_HT_DIR, credits);
3132         handle = ext4_journal_current_handle();
3133         if (IS_ERR(inode)) {
3134                 if (handle)
3135                         ext4_journal_stop(handle);
3136                 err = PTR_ERR(inode);
3137                 goto err_free_sd;
3138         }
3139
3140         if (encryption_required) {
3141                 struct qstr istr;
3142                 struct ext4_str ostr;
3143
3144                 istr.name = (const unsigned char *) symname;
3145                 istr.len = len;
3146                 ostr.name = sd->encrypted_path;