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         int size;
424         __u32 dummy_csum = 0;
425         int offset = offsetof(struct dx_tail, dt_checksum);
426
427         size = count_offset + (count * sizeof(struct dx_entry));
428         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
431
432         return cpu_to_le32(csum);
433 }
434
435 static int ext4_dx_csum_verify(struct inode *inode,
436                                struct ext4_dir_entry *dirent)
437 {
438         struct dx_countlimit *c;
439         struct dx_tail *t;
440         int count_offset, limit, count;
441
442         if (!ext4_has_metadata_csum(inode->i_sb))
443                 return 1;
444
445         c = get_dx_countlimit(inode, dirent, &count_offset);
446         if (!c) {
447                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
448                 return 0;
449         }
450         limit = le16_to_cpu(c->limit);
451         count = le16_to_cpu(c->count);
452         if (count_offset + (limit * sizeof(struct dx_entry)) >
453             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454                 warn_no_space_for_csum(inode);
455                 return 0;
456         }
457         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458
459         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460                                             count, t))
461                 return 0;
462         return 1;
463 }
464
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466 {
467         struct dx_countlimit *c;
468         struct dx_tail *t;
469         int count_offset, limit, count;
470
471         if (!ext4_has_metadata_csum(inode->i_sb))
472                 return;
473
474         c = get_dx_countlimit(inode, dirent, &count_offset);
475         if (!c) {
476                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
477                 return;
478         }
479         limit = le16_to_cpu(c->limit);
480         count = le16_to_cpu(c->count);
481         if (count_offset + (limit * sizeof(struct dx_entry)) >
482             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483                 warn_no_space_for_csum(inode);
484                 return;
485         }
486         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487
488         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492                                             struct inode *inode,
493                                             struct buffer_head *bh)
494 {
495         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496         return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498
499 /*
500  * p is at least 6 bytes before the end of page
501  */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505         return (struct ext4_dir_entry_2 *)((char *)p +
506                 ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508
509 /*
510  * Future: use high four bits of block for coalesce-on-delete flags
511  * Mask them off for now.
512  */
513
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516         return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521         entry->block = cpu_to_le32(value);
522 }
523
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526         return le32_to_cpu(entry->hash);
527 }
528
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531         entry->hash = cpu_to_le32(value);
532 }
533
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557                 EXT4_DIR_REC_LEN(2) - infosize;
558
559         if (ext4_has_metadata_csum(dir->i_sb))
560                 entry_space -= sizeof(struct dx_tail);
561         return entry_space / sizeof(struct dx_entry);
562 }
563
564 static inline unsigned dx_node_limit(struct inode *dir)
565 {
566         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567
568         if (ext4_has_metadata_csum(dir->i_sb))
569                 entry_space -= sizeof(struct dx_tail);
570         return entry_space / sizeof(struct dx_entry);
571 }
572
573 /*
574  * Debug
575  */
576 #ifdef DX_DEBUG
577 static void dx_show_index(char * label, struct dx_entry *entries)
578 {
579         int i, n = dx_get_count (entries);
580         printk(KERN_DEBUG "%s index", label);
581         for (i = 0; i < n; i++) {
582                 printk(KERN_CONT " %x->%lu",
583                        i ? dx_get_hash(entries + i) : 0,
584                        (unsigned long)dx_get_block(entries + i));
585         }
586         printk(KERN_CONT "\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 fscrypt_str fname_crypto_str =
616                                         FSTR_INIT(NULL, 0);
617                                 int res = 0;
618
619                                 name  = de->name;
620                                 len = de->name_len;
621                                 if (ext4_encrypted_inode(dir))
622                                         res = fscrypt_get_encryption_info(dir);
623                                 if (res) {
624                                         printk(KERN_WARNING "Error setting up"
625                                                " fname crypto: %d\n", res);
626                                 }
627                                 if (!fscrypt_has_encryption_key(dir)) {
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                                         struct fscrypt_str de_name =
637                                                 FSTR_INIT(name, len);
638
639                                         /* Directory is encrypted */
640                                         res = fscrypt_fname_alloc_buffer(
641                                                 dir, len,
642                                                 &fname_crypto_str);
643                                         if (res)
644                                                 printk(KERN_WARNING "Error "
645                                                         "allocating crypto "
646                                                         "buffer--skipping "
647                                                         "crypto\n");
648                                         res = fscrypt_fname_disk_to_usr(dir,
649                                                 0, 0, &de_name,
650                                                 &fname_crypto_str);
651                                         if (res) {
652                                                 printk(KERN_WARNING "Error "
653                                                         "converting filename "
654                                                         "from disk to usr"
655                                                         "\n");
656                                                 name = "??";
657                                                 len = 2;
658                                         } else {
659                                                 name = fname_crypto_str.name;
660                                                 len = fname_crypto_str.len;
661                                         }
662                                         ext4fs_dirhash(de->name, de->name_len,
663                                                        &h);
664                                         printk("%*.s:(E)%x.%u ", len, name,
665                                                h.hash, (unsigned) ((char *) de
666                                                                    - base));
667                                         fscrypt_fname_free_buffer(
668                                                         &fname_crypto_str);
669                                 }
670 #else
671                                 int len = de->name_len;
672                                 char *name = de->name;
673                                 ext4fs_dirhash(de->name, de->name_len, &h);
674                                 printk("%*.s:%x.%u ", len, name, h.hash,
675                                        (unsigned) ((char *) de - base));
676 #endif
677                         }
678                         space += EXT4_DIR_REC_LEN(de->name_len);
679                         names++;
680                 }
681                 de = ext4_next_entry(de, size);
682         }
683         printk(KERN_CONT "(%i)\n", names);
684         return (struct stats) { names, space, 1 };
685 }
686
687 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688                              struct dx_entry *entries, int levels)
689 {
690         unsigned blocksize = dir->i_sb->s_blocksize;
691         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
692         unsigned bcount = 0;
693         struct buffer_head *bh;
694         printk("%i indexed blocks...\n", count);
695         for (i = 0; i < count; i++, entries++)
696         {
697                 ext4_lblk_t block = dx_get_block(entries);
698                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
699                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
700                 struct stats stats;
701                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
702                 bh = ext4_bread(NULL,dir, block, 0);
703                 if (!bh || IS_ERR(bh))
704                         continue;
705                 stats = levels?
706                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
707                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708                         bh->b_data, blocksize, 0);
709                 names += stats.names;
710                 space += stats.space;
711                 bcount += stats.bcount;
712                 brelse(bh);
713         }
714         if (bcount)
715                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
716                        levels ? "" : "   ", names, space/bcount,
717                        (space/bcount)*100/blocksize);
718         return (struct stats) { names, space, bcount};
719 }
720 #endif /* DX_DEBUG */
721
722 /*
723  * Probe for a directory leaf block to search.
724  *
725  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726  * error in the directory index, and the caller should fall back to
727  * searching the directory normally.  The callers of dx_probe **MUST**
728  * check for this error code, and make sure it never gets reflected
729  * back to userspace.
730  */
731 static struct dx_frame *
732 dx_probe(struct ext4_filename *fname, struct inode *dir,
733          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
734 {
735         unsigned count, indirect;
736         struct dx_entry *at, *entries, *p, *q, *m;
737         struct dx_root *root;
738         struct dx_frame *frame = frame_in;
739         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
740         u32 hash;
741
742         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
743         if (IS_ERR(frame->bh))
744                 return (struct dx_frame *) frame->bh;
745
746         root = (struct dx_root *) frame->bh->b_data;
747         if (root->info.hash_version != DX_HASH_TEA &&
748             root->info.hash_version != DX_HASH_HALF_MD4 &&
749             root->info.hash_version != DX_HASH_LEGACY) {
750                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
751                                    root->info.hash_version);
752                 goto fail;
753         }
754         if (fname)
755                 hinfo = &fname->hinfo;
756         hinfo->hash_version = root->info.hash_version;
757         if (hinfo->hash_version <= DX_HASH_TEA)
758                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760         if (fname && fname_name(fname))
761                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
762         hash = hinfo->hash;
763
764         if (root->info.unused_flags & 1) {
765                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
766                                    root->info.unused_flags);
767                 goto fail;
768         }
769
770         indirect = root->info.indirect_levels;
771         if (indirect > 1) {
772                 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
773                                    root->info.indirect_levels);
774                 goto fail;
775         }
776
777         entries = (struct dx_entry *)(((char *)&root->info) +
778                                       root->info.info_length);
779
780         if (dx_get_limit(entries) != dx_root_limit(dir,
781                                                    root->info.info_length)) {
782                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
783                                    dx_get_limit(entries),
784                                    dx_root_limit(dir, root->info.info_length));
785                 goto fail;
786         }
787
788         dxtrace(printk("Look up %x", hash));
789         while (1) {
790                 count = dx_get_count(entries);
791                 if (!count || count > dx_get_limit(entries)) {
792                         ext4_warning_inode(dir,
793                                            "dx entry: count %u beyond limit %u",
794                                            count, dx_get_limit(entries));
795                         goto fail;
796                 }
797
798                 p = entries + 1;
799                 q = entries + count - 1;
800                 while (p <= q) {
801                         m = p + (q - p) / 2;
802                         dxtrace(printk(KERN_CONT "."));
803                         if (dx_get_hash(m) > hash)
804                                 q = m - 1;
805                         else
806                                 p = m + 1;
807                 }
808
809                 if (0) { // linear search cross check
810                         unsigned n = count - 1;
811                         at = entries;
812                         while (n--)
813                         {
814                                 dxtrace(printk(KERN_CONT ","));
815                                 if (dx_get_hash(++at) > hash)
816                                 {
817                                         at--;
818                                         break;
819                                 }
820                         }
821                         assert (at == p - 1);
822                 }
823
824                 at = p - 1;
825                 dxtrace(printk(KERN_CONT " %x->%u\n",
826                                at == entries ? 0 : dx_get_hash(at),
827                                dx_get_block(at)));
828                 frame->entries = entries;
829                 frame->at = at;
830                 if (!indirect--)
831                         return frame;
832                 frame++;
833                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
834                 if (IS_ERR(frame->bh)) {
835                         ret_err = (struct dx_frame *) frame->bh;
836                         frame->bh = NULL;
837                         goto fail;
838                 }
839                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
840
841                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
842                         ext4_warning_inode(dir,
843                                 "dx entry: limit %u != node limit %u",
844                                 dx_get_limit(entries), dx_node_limit(dir));
845                         goto fail;
846                 }
847         }
848 fail:
849         while (frame >= frame_in) {
850                 brelse(frame->bh);
851                 frame--;
852         }
853
854         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
855                 ext4_warning_inode(dir,
856                         "Corrupt directory, running e2fsck is recommended");
857         return ret_err;
858 }
859
860 static void dx_release(struct dx_frame *frames)
861 {
862         if (frames[0].bh == NULL)
863                 return;
864
865         if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
866                 brelse(frames[1].bh);
867         brelse(frames[0].bh);
868 }
869
870 /*
871  * This function increments the frame pointer to search the next leaf
872  * block, and reads in the necessary intervening nodes if the search
873  * should be necessary.  Whether or not the search is necessary is
874  * controlled by the hash parameter.  If the hash value is even, then
875  * the search is only continued if the next block starts with that
876  * hash value.  This is used if we are searching for a specific file.
877  *
878  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
879  *
880  * This function returns 1 if the caller should continue to search,
881  * or 0 if it should not.  If there is an error reading one of the
882  * index blocks, it will a negative error code.
883  *
884  * If start_hash is non-null, it will be filled in with the starting
885  * hash of the next page.
886  */
887 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
888                                  struct dx_frame *frame,
889                                  struct dx_frame *frames,
890                                  __u32 *start_hash)
891 {
892         struct dx_frame *p;
893         struct buffer_head *bh;
894         int num_frames = 0;
895         __u32 bhash;
896
897         p = frame;
898         /*
899          * Find the next leaf page by incrementing the frame pointer.
900          * If we run out of entries in the interior node, loop around and
901          * increment pointer in the parent node.  When we break out of
902          * this loop, num_frames indicates the number of interior
903          * nodes need to be read.
904          */
905         while (1) {
906                 if (++(p->at) < p->entries + dx_get_count(p->entries))
907                         break;
908                 if (p == frames)
909                         return 0;
910                 num_frames++;
911                 p--;
912         }
913
914         /*
915          * If the hash is 1, then continue only if the next page has a
916          * continuation hash of any value.  This is used for readdir
917          * handling.  Otherwise, check to see if the hash matches the
918          * desired contiuation hash.  If it doesn't, return since
919          * there's no point to read in the successive index pages.
920          */
921         bhash = dx_get_hash(p->at);
922         if (start_hash)
923                 *start_hash = bhash;
924         if ((hash & 1) == 0) {
925                 if ((bhash & ~1) != hash)
926                         return 0;
927         }
928         /*
929          * If the hash is HASH_NB_ALWAYS, we always go to the next
930          * block so no check is necessary
931          */
932         while (num_frames--) {
933                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
934                 if (IS_ERR(bh))
935                         return PTR_ERR(bh);
936                 p++;
937                 brelse(p->bh);
938                 p->bh = bh;
939                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
940         }
941         return 1;
942 }
943
944
945 /*
946  * This function fills a red-black tree with information from a
947  * directory block.  It returns the number directory entries loaded
948  * into the tree.  If there is an error it is returned in err.
949  */
950 static int htree_dirblock_to_tree(struct file *dir_file,
951                                   struct inode *dir, ext4_lblk_t block,
952                                   struct dx_hash_info *hinfo,
953                                   __u32 start_hash, __u32 start_minor_hash)
954 {
955         struct buffer_head *bh;
956         struct ext4_dir_entry_2 *de, *top;
957         int err = 0, count = 0;
958         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
959
960         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
961                                                         (unsigned long)block));
962         bh = ext4_read_dirblock(dir, block, DIRENT);
963         if (IS_ERR(bh))
964                 return PTR_ERR(bh);
965
966         de = (struct ext4_dir_entry_2 *) bh->b_data;
967         top = (struct ext4_dir_entry_2 *) ((char *) de +
968                                            dir->i_sb->s_blocksize -
969                                            EXT4_DIR_REC_LEN(0));
970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
971         /* Check if the directory is encrypted */
972         if (ext4_encrypted_inode(dir)) {
973                 err = fscrypt_get_encryption_info(dir);
974                 if (err < 0) {
975                         brelse(bh);
976                         return err;
977                 }
978                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
979                                                      &fname_crypto_str);
980                 if (err < 0) {
981                         brelse(bh);
982                         return err;
983                 }
984         }
985 #endif
986         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
987                 if (ext4_check_dir_entry(dir, NULL, de, bh,
988                                 bh->b_data, bh->b_size,
989                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
990                                          + ((char *)de - bh->b_data))) {
991                         /* silently ignore the rest of the block */
992                         break;
993                 }
994                 ext4fs_dirhash(de->name, de->name_len, hinfo);
995                 if ((hinfo->hash < start_hash) ||
996                     ((hinfo->hash == start_hash) &&
997                      (hinfo->minor_hash < start_minor_hash)))
998                         continue;
999                 if (de->inode == 0)
1000                         continue;
1001                 if (!ext4_encrypted_inode(dir)) {
1002                         tmp_str.name = de->name;
1003                         tmp_str.len = de->name_len;
1004                         err = ext4_htree_store_dirent(dir_file,
1005                                    hinfo->hash, hinfo->minor_hash, de,
1006                                    &tmp_str);
1007                 } else {
1008                         int save_len = fname_crypto_str.len;
1009                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1010                                                                 de->name_len);
1011
1012                         /* Directory is encrypted */
1013                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014                                         hinfo->minor_hash, &de_name,
1015                                         &fname_crypto_str);
1016                         if (err) {
1017                                 count = err;
1018                                 goto errout;
1019                         }
1020                         err = ext4_htree_store_dirent(dir_file,
1021                                    hinfo->hash, hinfo->minor_hash, de,
1022                                         &fname_crypto_str);
1023                         fname_crypto_str.len = save_len;
1024                 }
1025                 if (err != 0) {
1026                         count = err;
1027                         goto errout;
1028                 }
1029                 count++;
1030         }
1031 errout:
1032         brelse(bh);
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034         fscrypt_fname_free_buffer(&fname_crypto_str);
1035 #endif
1036         return count;
1037 }
1038
1039
1040 /*
1041  * This function fills a red-black tree with information from a
1042  * directory.  We start scanning the directory in hash order, starting
1043  * at start_hash and start_minor_hash.
1044  *
1045  * This function returns the number of entries inserted into the tree,
1046  * or a negative error code.
1047  */
1048 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049                          __u32 start_minor_hash, __u32 *next_hash)
1050 {
1051         struct dx_hash_info hinfo;
1052         struct ext4_dir_entry_2 *de;
1053         struct dx_frame frames[2], *frame;
1054         struct inode *dir;
1055         ext4_lblk_t block;
1056         int count = 0;
1057         int ret, err;
1058         __u32 hashval;
1059         struct fscrypt_str tmp_str;
1060
1061         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062                        start_hash, start_minor_hash));
1063         dir = file_inode(dir_file);
1064         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066                 if (hinfo.hash_version <= DX_HASH_TEA)
1067                         hinfo.hash_version +=
1068                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070                 if (ext4_has_inline_data(dir)) {
1071                         int has_inline_data = 1;
1072                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1073                                                         &hinfo, start_hash,
1074                                                         start_minor_hash,
1075                                                         &has_inline_data);
1076                         if (has_inline_data) {
1077                                 *next_hash = ~0;
1078                                 return count;
1079                         }
1080                 }
1081                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082                                                start_hash, start_minor_hash);
1083                 *next_hash = ~0;
1084                 return count;
1085         }
1086         hinfo.hash = start_hash;
1087         hinfo.minor_hash = 0;
1088         frame = dx_probe(NULL, dir, &hinfo, frames);
1089         if (IS_ERR(frame))
1090                 return PTR_ERR(frame);
1091
1092         /* Add '.' and '..' from the htree header */
1093         if (!start_hash && !start_minor_hash) {
1094                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095                 tmp_str.name = de->name;
1096                 tmp_str.len = de->name_len;
1097                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1098                                               de, &tmp_str);
1099                 if (err != 0)
1100                         goto errout;
1101                 count++;
1102         }
1103         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106                 tmp_str.name = de->name;
1107                 tmp_str.len = de->name_len;
1108                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109                                               de, &tmp_str);
1110                 if (err != 0)
1111                         goto errout;
1112                 count++;
1113         }
1114
1115         while (1) {
1116                 if (fatal_signal_pending(current)) {
1117                         err = -ERESTARTSYS;
1118                         goto errout;
1119                 }
1120                 cond_resched();
1121                 block = dx_get_block(frame->at);
1122                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123                                              start_hash, start_minor_hash);
1124                 if (ret < 0) {
1125                         err = ret;
1126                         goto errout;
1127                 }
1128                 count += ret;
1129                 hashval = ~0;
1130                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131                                             frame, frames, &hashval);
1132                 *next_hash = hashval;
1133                 if (ret < 0) {
1134                         err = ret;
1135                         goto errout;
1136                 }
1137                 /*
1138                  * Stop if:  (a) there are no more entries, or
1139                  * (b) we have inserted at least one entry and the
1140                  * next hash value is not a continuation
1141                  */
1142                 if ((ret == 0) ||
1143                     (count && ((hashval & 1) == 0)))
1144                         break;
1145         }
1146         dx_release(frames);
1147         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148                        "next hash: %x\n", count, *next_hash));
1149         return count;
1150 errout:
1151         dx_release(frames);
1152         return (err);
1153 }
1154
1155 static inline int search_dirblock(struct buffer_head *bh,
1156                                   struct inode *dir,
1157                                   struct ext4_filename *fname,
1158                                   const struct qstr *d_name,
1159                                   unsigned int offset,
1160                                   struct ext4_dir_entry_2 **res_dir)
1161 {
1162         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163                                fname, d_name, offset, res_dir);
1164 }
1165
1166 /*
1167  * Directory block splitting, compacting
1168  */
1169
1170 /*
1171  * Create map of hash values, offsets, and sizes, stored at end of block.
1172  * Returns number of entries mapped.
1173  */
1174 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175                        unsigned blocksize, struct dx_hash_info *hinfo,
1176                        struct dx_map_entry *map_tail)
1177 {
1178         int count = 0;
1179         char *base = (char *) de;
1180         struct dx_hash_info h = *hinfo;
1181
1182         while ((char *) de < base + blocksize) {
1183                 if (de->name_len && de->inode) {
1184                         ext4fs_dirhash(de->name, de->name_len, &h);
1185                         map_tail--;
1186                         map_tail->hash = h.hash;
1187                         map_tail->offs = ((char *) de - base)>>2;
1188                         map_tail->size = le16_to_cpu(de->rec_len);
1189                         count++;
1190                         cond_resched();
1191                 }
1192                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193                 de = ext4_next_entry(de, blocksize);
1194         }
1195         return count;
1196 }
1197
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200 {
1201         struct dx_map_entry *p, *q, *top = map + count - 1;
1202         int more;
1203         /* Combsort until bubble sort doesn't suck */
1204         while (count > 2) {
1205                 count = count*10/13;
1206                 if (count - 9 < 2) /* 9, 10 -> 11 */
1207                         count = 11;
1208                 for (p = top, q = p - count; q >= map; p--, q--)
1209                         if (p->hash < q->hash)
1210                                 swap(*p, *q);
1211         }
1212         /* Garden variety bubble sort */
1213         do {
1214                 more = 0;
1215                 q = top;
1216                 while (q-- > map) {
1217                         if (q[1].hash >= q[0].hash)
1218                                 continue;
1219                         swap(*(q+1), *q);
1220                         more = 1;
1221                 }
1222         } while(more);
1223 }
1224
1225 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226 {
1227         struct dx_entry *entries = frame->entries;
1228         struct dx_entry *old = frame->at, *new = old + 1;
1229         int count = dx_get_count(entries);
1230
1231         assert(count < dx_get_limit(entries));
1232         assert(old < entries + count);
1233         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234         dx_set_hash(new, hash);
1235         dx_set_block(new, block);
1236         dx_set_count(entries, count + 1);
1237 }
1238
1239 /*
1240  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1241  *
1242  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243  * `de != NULL' is guaranteed by caller.
1244  */
1245 static inline int ext4_match(struct ext4_filename *fname,
1246                              struct ext4_dir_entry_2 *de)
1247 {
1248         const void *name = fname_name(fname);
1249         u32 len = fname_len(fname);
1250
1251         if (!de->inode)
1252                 return 0;
1253
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255         if (unlikely(!name)) {
1256                 if (fname->usr_fname->name[0] == '_') {
1257                         int ret;
1258                         if (de->name_len < 16)
1259                                 return 0;
1260                         ret = memcmp(de->name + de->name_len - 16,
1261                                      fname->crypto_buf.name + 8, 16);
1262                         return (ret == 0) ? 1 : 0;
1263                 }
1264                 name = fname->crypto_buf.name;
1265                 len = fname->crypto_buf.len;
1266         }
1267 #endif
1268         if (de->name_len != len)
1269                 return 0;
1270         return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1271 }
1272
1273 /*
1274  * Returns 0 if not found, -1 on failure, and 1 on success
1275  */
1276 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277                     struct inode *dir, struct ext4_filename *fname,
1278                     const struct qstr *d_name,
1279                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281         struct ext4_dir_entry_2 * de;
1282         char * dlimit;
1283         int de_len;
1284         int res;
1285
1286         de = (struct ext4_dir_entry_2 *)search_buf;
1287         dlimit = search_buf + buf_size;
1288         while ((char *) de < dlimit) {
1289                 /* this code is executed quadratically often */
1290                 /* do minimal checking `by hand' */
1291                 if ((char *) de + de->name_len <= dlimit) {
1292                         res = ext4_match(fname, de);
1293                         if (res < 0) {
1294                                 res = -1;
1295                                 goto return_result;
1296                         }
1297                         if (res > 0) {
1298                                 /* found a match - just to be sure, do
1299                                  * a full check */
1300                                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1301                                                 bh->b_data,
1302                                                  bh->b_size, offset)) {
1303                                         res = -1;
1304                                         goto return_result;
1305                                 }
1306                                 *res_dir = de;
1307                                 res = 1;
1308                                 goto return_result;
1309                         }
1310
1311                 }
1312                 /* prevent looping on a bad block */
1313                 de_len = ext4_rec_len_from_disk(de->rec_len,
1314                                                 dir->i_sb->s_blocksize);
1315                 if (de_len <= 0) {
1316                         res = -1;
1317                         goto return_result;
1318                 }
1319                 offset += de_len;
1320                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1321         }
1322
1323         res = 0;
1324 return_result:
1325         return res;
1326 }
1327
1328 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329                                struct ext4_dir_entry *de)
1330 {
1331         struct super_block *sb = dir->i_sb;
1332
1333         if (!is_dx(dir))
1334                 return 0;
1335         if (block == 0)
1336                 return 1;
1337         if (de->inode == 0 &&
1338             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339                         sb->s_blocksize)
1340                 return 1;
1341         return 0;
1342 }
1343
1344 /*
1345  *      ext4_find_entry()
1346  *
1347  * finds an entry in the specified directory with the wanted name. It
1348  * returns the cache buffer in which the entry was found, and the entry
1349  * itself (as a parameter - res_dir). It does NOT read the inode of the
1350  * entry - you'll have to do that yourself if you want to.
1351  *
1352  * The returned buffer_head has ->b_count elevated.  The caller is expected
1353  * to brelse() it when appropriate.
1354  */
1355 static struct buffer_head * ext4_find_entry (struct inode *dir,
1356                                         const struct qstr *d_name,
1357                                         struct ext4_dir_entry_2 **res_dir,
1358                                         int *inlined)
1359 {
1360         struct super_block *sb;
1361         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362         struct buffer_head *bh, *ret = NULL;
1363         ext4_lblk_t start, block, b;
1364         const u8 *name = d_name->name;
1365         int ra_max = 0;         /* Number of bh's in the readahead
1366                                    buffer, bh_use[] */
1367         int ra_ptr = 0;         /* Current index into readahead
1368                                    buffer */
1369         int num = 0;
1370         ext4_lblk_t  nblocks;
1371         int i, namelen, retval;
1372         struct ext4_filename fname;
1373
1374         *res_dir = NULL;
1375         sb = dir->i_sb;
1376         namelen = d_name->len;
1377         if (namelen > EXT4_NAME_LEN)
1378                 return NULL;
1379
1380         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1381         if (retval == -ENOENT)
1382                 return NULL;
1383         if (retval)
1384                 return ERR_PTR(retval);
1385
1386         if (ext4_has_inline_data(dir)) {
1387                 int has_inline_data = 1;
1388                 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1389                                              &has_inline_data);
1390                 if (has_inline_data) {
1391                         if (inlined)
1392                                 *inlined = 1;
1393                         goto cleanup_and_exit;
1394                 }
1395         }
1396
1397         if ((namelen <= 2) && (name[0] == '.') &&
1398             (name[1] == '.' || name[1] == '\0')) {
1399                 /*
1400                  * "." or ".." will only be in the first block
1401                  * NFS may look up ".."; "." should be handled by the VFS
1402                  */
1403                 block = start = 0;
1404                 nblocks = 1;
1405                 goto restart;
1406         }
1407         if (is_dx(dir)) {
1408                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1409                 /*
1410                  * On success, or if the error was file not found,
1411                  * return.  Otherwise, fall back to doing a search the
1412                  * old fashioned way.
1413                  */
1414                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1415                         goto cleanup_and_exit;
1416                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1417                                "falling back\n"));
1418         }
1419         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1420         start = EXT4_I(dir)->i_dir_start_lookup;
1421         if (start >= nblocks)
1422                 start = 0;
1423         block = start;
1424 restart:
1425         do {
1426                 /*
1427                  * We deal with the read-ahead logic here.
1428                  */
1429                 if (ra_ptr >= ra_max) {
1430                         /* Refill the readahead buffer */
1431                         ra_ptr = 0;
1432                         b = block;
1433                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1434                                 /*
1435                                  * Terminate if we reach the end of the
1436                                  * directory and must wrap, or if our
1437                                  * search has finished at this block.
1438                                  */
1439                                 if (b >= nblocks || (num && block == start)) {
1440                                         bh_use[ra_max] = NULL;
1441                                         break;
1442                                 }
1443                                 num++;
1444                                 bh = ext4_getblk(NULL, dir, b++, 0);
1445                                 if (IS_ERR(bh)) {
1446                                         if (ra_max == 0) {
1447                                                 ret = bh;
1448                                                 goto cleanup_and_exit;
1449                                         }
1450                                         break;
1451                                 }
1452                                 bh_use[ra_max] = bh;
1453                                 if (bh)
1454                                         ll_rw_block(REQ_OP_READ,
1455                                                     REQ_META | REQ_PRIO,
1456                                                     1, &bh);
1457                         }
1458                 }
1459                 if ((bh = bh_use[ra_ptr++]) == NULL)
1460                         goto next;
1461                 wait_on_buffer(bh);
1462                 if (!buffer_uptodate(bh)) {
1463                         /* read error, skip block & hope for the best */
1464                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1465                                          (unsigned long) block);
1466                         brelse(bh);
1467                         goto next;
1468                 }
1469                 if (!buffer_verified(bh) &&
1470                     !is_dx_internal_node(dir, block,
1471                                          (struct ext4_dir_entry *)bh->b_data) &&
1472                     !ext4_dirent_csum_verify(dir,
1473                                 (struct ext4_dir_entry *)bh->b_data)) {
1474                         EXT4_ERROR_INODE(dir, "checksumming directory "
1475                                          "block %lu", (unsigned long)block);
1476                         brelse(bh);
1477                         goto next;
1478                 }
1479                 set_buffer_verified(bh);
1480                 i = search_dirblock(bh, dir, &fname, d_name,
1481                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1482                 if (i == 1) {
1483                         EXT4_I(dir)->i_dir_start_lookup = block;
1484                         ret = bh;
1485                         goto cleanup_and_exit;
1486                 } else {
1487                         brelse(bh);
1488                         if (i < 0)
1489                                 goto cleanup_and_exit;
1490                 }
1491         next:
1492                 if (++block >= nblocks)
1493                         block = 0;
1494         } while (block != start);
1495
1496         /*
1497          * If the directory has grown while we were searching, then
1498          * search the last part of the directory before giving up.
1499          */
1500         block = nblocks;
1501         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1502         if (block < nblocks) {
1503                 start = 0;
1504                 goto restart;
1505         }
1506
1507 cleanup_and_exit:
1508         /* Clean up the read-ahead blocks */
1509         for (; ra_ptr < ra_max; ra_ptr++)
1510                 brelse(bh_use[ra_ptr]);
1511         ext4_fname_free_filename(&fname);
1512         return ret;
1513 }
1514
1515 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1516                         struct ext4_filename *fname,
1517                         struct ext4_dir_entry_2 **res_dir)
1518 {
1519         struct super_block * sb = dir->i_sb;
1520         struct dx_frame frames[2], *frame;
1521         const struct qstr *d_name = fname->usr_fname;
1522         struct buffer_head *bh;
1523         ext4_lblk_t block;
1524         int retval;
1525
1526 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1527         *res_dir = NULL;
1528 #endif
1529         frame = dx_probe(fname, dir, NULL, frames);
1530         if (IS_ERR(frame))
1531                 return (struct buffer_head *) frame;
1532         do {
1533                 block = dx_get_block(frame->at);
1534                 bh = ext4_read_dirblock(dir, block, DIRENT);
1535                 if (IS_ERR(bh))
1536                         goto errout;
1537
1538                 retval = search_dirblock(bh, dir, fname, d_name,
1539                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1540                                          res_dir);
1541                 if (retval == 1)
1542                         goto success;
1543                 brelse(bh);
1544                 if (retval == -1) {
1545                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1546                         goto errout;
1547                 }
1548
1549                 /* Check to see if we should continue to search */
1550                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1551                                                frames, NULL);
1552                 if (retval < 0) {
1553                         ext4_warning_inode(dir,
1554                                 "error %d reading directory index block",
1555                                 retval);
1556                         bh = ERR_PTR(retval);
1557                         goto errout;
1558                 }
1559         } while (retval == 1);
1560
1561         bh = NULL;
1562 errout:
1563         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1564 success:
1565         dx_release(frames);
1566         return bh;
1567 }
1568
1569 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1570 {
1571         struct inode *inode;
1572         struct ext4_dir_entry_2 *de;
1573         struct buffer_head *bh;
1574
1575         if (ext4_encrypted_inode(dir)) {
1576                 int res = fscrypt_get_encryption_info(dir);
1577
1578                 /*
1579                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1580                  * created while the directory was encrypted and we
1581                  * have access to the key.
1582                  */
1583                 if (fscrypt_has_encryption_key(dir))
1584                         fscrypt_set_encrypted_dentry(dentry);
1585                 fscrypt_set_d_op(dentry);
1586                 if (res && res != -ENOKEY)
1587                         return ERR_PTR(res);
1588         }
1589
1590        if (dentry->d_name.len > EXT4_NAME_LEN)
1591                return ERR_PTR(-ENAMETOOLONG);
1592
1593         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1594         if (IS_ERR(bh))
1595                 return (struct dentry *) bh;
1596         inode = NULL;
1597         if (bh) {
1598                 __u32 ino = le32_to_cpu(de->inode);
1599                 brelse(bh);
1600                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1601                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1602                         return ERR_PTR(-EFSCORRUPTED);
1603                 }
1604                 if (unlikely(ino == dir->i_ino)) {
1605                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1606                                          dentry);
1607                         return ERR_PTR(-EFSCORRUPTED);
1608                 }
1609                 inode = ext4_iget_normal(dir->i_sb, ino);
1610                 if (inode == ERR_PTR(-ESTALE)) {
1611                         EXT4_ERROR_INODE(dir,
1612                                          "deleted inode referenced: %u",
1613                                          ino);
1614                         return ERR_PTR(-EFSCORRUPTED);
1615                 }
1616                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1617                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1618                     !fscrypt_has_permitted_context(dir, inode)) {
1619                         int nokey = ext4_encrypted_inode(inode) &&
1620                                 !fscrypt_has_encryption_key(inode);
1621                         if (nokey) {
1622                                 iput(inode);
1623                                 return ERR_PTR(-ENOKEY);
1624                         }
1625                         ext4_warning(inode->i_sb,
1626                                      "Inconsistent encryption contexts: %lu/%lu",
1627                                      (unsigned long) dir->i_ino,
1628                                      (unsigned long) inode->i_ino);
1629                         iput(inode);
1630                         return ERR_PTR(-EPERM);
1631                 }
1632         }
1633         return d_splice_alias(inode, dentry);
1634 }
1635
1636
1637 struct dentry *ext4_get_parent(struct dentry *child)
1638 {
1639         __u32 ino;
1640         static const struct qstr dotdot = QSTR_INIT("..", 2);
1641         struct ext4_dir_entry_2 * de;
1642         struct buffer_head *bh;
1643
1644         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1645         if (IS_ERR(bh))
1646                 return (struct dentry *) bh;
1647         if (!bh)
1648                 return ERR_PTR(-ENOENT);
1649         ino = le32_to_cpu(de->inode);
1650         brelse(bh);
1651
1652         if (!ext4_valid_inum(child->d_sb, ino)) {
1653                 EXT4_ERROR_INODE(d_inode(child),
1654                                  "bad parent inode number: %u", ino);
1655                 return ERR_PTR(-EFSCORRUPTED);
1656         }
1657
1658         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1659 }
1660
1661 /*
1662  * Move count entries from end of map between two memory locations.
1663  * Returns pointer to last entry moved.
1664  */
1665 static struct ext4_dir_entry_2 *
1666 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1667                 unsigned blocksize)
1668 {
1669         unsigned rec_len = 0;
1670
1671         while (count--) {
1672                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1673                                                 (from + (map->offs<<2));
1674                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1675                 memcpy (to, de, rec_len);
1676                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1677                                 ext4_rec_len_to_disk(rec_len, blocksize);
1678                 de->inode = 0;
1679                 map++;
1680                 to += rec_len;
1681         }
1682         return (struct ext4_dir_entry_2 *) (to - rec_len);
1683 }
1684
1685 /*
1686  * Compact each dir entry in the range to the minimal rec_len.
1687  * Returns pointer to last entry in range.
1688  */
1689 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1690 {
1691         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1692         unsigned rec_len = 0;
1693
1694         prev = to = de;
1695         while ((char*)de < base + blocksize) {
1696                 next = ext4_next_entry(de, blocksize);
1697                 if (de->inode && de->name_len) {
1698                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1699                         if (de > to)
1700                                 memmove(to, de, rec_len);
1701                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1702                         prev = to;
1703                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1704                 }
1705                 de = next;
1706         }
1707         return prev;
1708 }
1709
1710 /*
1711  * Split a full leaf block to make room for a new dir entry.
1712  * Allocate a new block, and move entries so that they are approx. equally full.
1713  * Returns pointer to de in block into which the new entry will be inserted.
1714  */
1715 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1716                         struct buffer_head **bh,struct dx_frame *frame,
1717                         struct dx_hash_info *hinfo)
1718 {
1719         unsigned blocksize = dir->i_sb->s_blocksize;
1720         unsigned count, continued;
1721         struct buffer_head *bh2;
1722         ext4_lblk_t newblock;
1723         u32 hash2;
1724         struct dx_map_entry *map;
1725         char *data1 = (*bh)->b_data, *data2;
1726         unsigned split, move, size;
1727         struct ext4_dir_entry_2 *de = NULL, *de2;
1728         struct ext4_dir_entry_tail *t;
1729         int     csum_size = 0;
1730         int     err = 0, i;
1731
1732         if (ext4_has_metadata_csum(dir->i_sb))
1733                 csum_size = sizeof(struct ext4_dir_entry_tail);
1734
1735         bh2 = ext4_append(handle, dir, &newblock);
1736         if (IS_ERR(bh2)) {
1737                 brelse(*bh);
1738                 *bh = NULL;
1739                 return (struct ext4_dir_entry_2 *) bh2;
1740         }
1741
1742         BUFFER_TRACE(*bh, "get_write_access");
1743         err = ext4_journal_get_write_access(handle, *bh);
1744         if (err)
1745                 goto journal_error;
1746
1747         BUFFER_TRACE(frame->bh, "get_write_access");
1748         err = ext4_journal_get_write_access(handle, frame->bh);
1749         if (err)
1750                 goto journal_error;
1751
1752         data2 = bh2->b_data;
1753
1754         /* create map in the end of data2 block */
1755         map = (struct dx_map_entry *) (data2 + blocksize);
1756         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1757                              blocksize, hinfo, map);
1758         map -= count;
1759         dx_sort_map(map, count);
1760         /* Split the existing block in the middle, size-wise */
1761         size = 0;
1762         move = 0;
1763         for (i = count-1; i >= 0; i--) {
1764                 /* is more than half of this entry in 2nd half of the block? */
1765                 if (size + map[i].size/2 > blocksize/2)
1766                         break;
1767                 size += map[i].size;
1768                 move++;
1769         }
1770         /* map index at which we will split */
1771         split = count - move;
1772         hash2 = map[split].hash;
1773         continued = hash2 == map[split - 1].hash;
1774         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1775                         (unsigned long)dx_get_block(frame->at),
1776                                         hash2, split, count-split));
1777
1778         /* Fancy dance to stay within two buffers */
1779         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1780                               blocksize);
1781         de = dx_pack_dirents(data1, blocksize);
1782         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1783                                            (char *) de,
1784                                            blocksize);
1785         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1786                                             (char *) de2,
1787                                             blocksize);
1788         if (csum_size) {
1789                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1790                 initialize_dirent_tail(t, blocksize);
1791
1792                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1793                 initialize_dirent_tail(t, blocksize);
1794         }
1795
1796         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1797                         blocksize, 1));
1798         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1799                         blocksize, 1));
1800
1801         /* Which block gets the new entry? */
1802         if (hinfo->hash >= hash2) {
1803                 swap(*bh, bh2);
1804                 de = de2;
1805         }
1806         dx_insert_block(frame, hash2 + continued, newblock);
1807         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1808         if (err)
1809                 goto journal_error;
1810         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1811         if (err)
1812                 goto journal_error;
1813         brelse(bh2);
1814         dxtrace(dx_show_index("frame", frame->entries));
1815         return de;
1816
1817 journal_error:
1818         brelse(*bh);
1819         brelse(bh2);
1820         *bh = NULL;
1821         ext4_std_error(dir->i_sb, err);
1822         return ERR_PTR(err);
1823 }
1824
1825 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1826                       struct buffer_head *bh,
1827                       void *buf, int buf_size,
1828                       struct ext4_filename *fname,
1829                       struct ext4_dir_entry_2 **dest_de)
1830 {
1831         struct ext4_dir_entry_2 *de;
1832         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1833         int nlen, rlen;
1834         unsigned int offset = 0;
1835         char *top;
1836         int res;
1837
1838         de = (struct ext4_dir_entry_2 *)buf;
1839         top = buf + buf_size - reclen;
1840         while ((char *) de <= top) {
1841                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1842                                          buf, buf_size, offset)) {
1843                         res = -EFSCORRUPTED;
1844                         goto return_result;
1845                 }
1846                 /* Provide crypto context and crypto buffer to ext4 match */
1847                 res = ext4_match(fname, de);
1848                 if (res < 0)
1849                         goto return_result;
1850                 if (res > 0) {
1851                         res = -EEXIST;
1852                         goto return_result;
1853                 }
1854                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1855                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1856                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1857                         break;
1858                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1859                 offset += rlen;
1860         }
1861
1862         if ((char *) de > top)
1863                 res = -ENOSPC;
1864         else {
1865                 *dest_de = de;
1866                 res = 0;
1867         }
1868 return_result:
1869         return res;
1870 }
1871
1872 void ext4_insert_dentry(struct inode *inode,
1873                         struct ext4_dir_entry_2 *de,
1874                         int buf_size,
1875                         struct ext4_filename *fname)
1876 {
1877
1878         int nlen, rlen;
1879
1880         nlen = EXT4_DIR_REC_LEN(de->name_len);
1881         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1882         if (de->inode) {
1883                 struct ext4_dir_entry_2 *de1 =
1884                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1885                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1886                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1887                 de = de1;
1888         }
1889         de->file_type = EXT4_FT_UNKNOWN;
1890         de->inode = cpu_to_le32(inode->i_ino);
1891         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1892         de->name_len = fname_len(fname);
1893         memcpy(de->name, fname_name(fname), fname_len(fname));
1894 }
1895
1896 /*
1897  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1898  * it points to a directory entry which is guaranteed to be large
1899  * enough for new directory entry.  If de is NULL, then
1900  * add_dirent_to_buf will attempt search the directory block for
1901  * space.  It will return -ENOSPC if no space is available, and -EIO
1902  * and -EEXIST if directory entry already exists.
1903  */
1904 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1905                              struct inode *dir,
1906                              struct inode *inode, struct ext4_dir_entry_2 *de,
1907                              struct buffer_head *bh)
1908 {
1909         unsigned int    blocksize = dir->i_sb->s_blocksize;
1910         int             csum_size = 0;
1911         int             err;
1912
1913         if (ext4_has_metadata_csum(inode->i_sb))
1914                 csum_size = sizeof(struct ext4_dir_entry_tail);
1915
1916         if (!de) {
1917                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1918                                         blocksize - csum_size, fname, &de);
1919                 if (err)
1920                         return err;
1921         }
1922         BUFFER_TRACE(bh, "get_write_access");
1923         err = ext4_journal_get_write_access(handle, bh);
1924         if (err) {
1925                 ext4_std_error(dir->i_sb, err);
1926                 return err;
1927         }
1928
1929         /* By now the buffer is marked for journaling */
1930         ext4_insert_dentry(inode, de, blocksize, fname);
1931
1932         /*
1933          * XXX shouldn't update any times until successful
1934          * completion of syscall, but too many callers depend
1935          * on this.
1936          *
1937          * XXX similarly, too many callers depend on
1938          * ext4_new_inode() setting the times, but error
1939          * recovery deletes the inode, so the worst that can
1940          * happen is that the times are slightly out of date
1941          * and/or different from the directory change time.
1942          */
1943         dir->i_mtime = dir->i_ctime = current_time(dir);
1944         ext4_update_dx_flag(dir);
1945         dir->i_version++;
1946         ext4_mark_inode_dirty(handle, dir);
1947         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1948         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1949         if (err)
1950                 ext4_std_error(dir->i_sb, err);
1951         return 0;
1952 }
1953
1954 /*
1955  * This converts a one block unindexed directory to a 3 block indexed
1956  * directory, and adds the dentry to the indexed directory.
1957  */
1958 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1959                             struct inode *dir,
1960                             struct inode *inode, struct buffer_head *bh)
1961 {
1962         struct buffer_head *bh2;
1963         struct dx_root  *root;
1964         struct dx_frame frames[2], *frame;
1965         struct dx_entry *entries;
1966         struct ext4_dir_entry_2 *de, *de2;
1967         struct ext4_dir_entry_tail *t;
1968         char            *data1, *top;
1969         unsigned        len;
1970         int             retval;
1971         unsigned        blocksize;
1972         ext4_lblk_t  block;
1973         struct fake_dirent *fde;
1974         int csum_size = 0;
1975
1976         if (ext4_has_metadata_csum(inode->i_sb))
1977                 csum_size = sizeof(struct ext4_dir_entry_tail);
1978
1979         blocksize =  dir->i_sb->s_blocksize;
1980         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1981         BUFFER_TRACE(bh, "get_write_access");
1982         retval = ext4_journal_get_write_access(handle, bh);
1983         if (retval) {
1984                 ext4_std_error(dir->i_sb, retval);
1985                 brelse(bh);
1986                 return retval;
1987         }
1988         root = (struct dx_root *) bh->b_data;
1989
1990         /* The 0th block becomes the root, move the dirents out */
1991         fde = &root->dotdot;
1992         de = (struct ext4_dir_entry_2 *)((char *)fde +
1993                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1994         if ((char *) de >= (((char *) root) + blocksize)) {
1995                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1996                 brelse(bh);
1997                 return -EFSCORRUPTED;
1998         }
1999         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2000
2001         /* Allocate new block for the 0th block's dirents */
2002         bh2 = ext4_append(handle, dir, &block);
2003         if (IS_ERR(bh2)) {
2004                 brelse(bh);
2005                 return PTR_ERR(bh2);
2006         }
2007         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2008         data1 = bh2->b_data;
2009
2010         memcpy (data1, de, len);
2011         de = (struct ext4_dir_entry_2 *) data1;
2012         top = data1 + len;
2013         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2014                 de = de2;
2015         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2016                                            (char *) de,
2017                                            blocksize);
2018
2019         if (csum_size) {
2020                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2021                 initialize_dirent_tail(t, blocksize);
2022         }
2023
2024         /* Initialize the root; the dot dirents already exist */
2025         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2026         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2027                                            blocksize);
2028         memset (&root->info, 0, sizeof(root->info));
2029         root->info.info_length = sizeof(root->info);
2030         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2031         entries = root->entries;
2032         dx_set_block(entries, 1);
2033         dx_set_count(entries, 1);
2034         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2035
2036         /* Initialize as for dx_probe */
2037         fname->hinfo.hash_version = root->info.hash_version;
2038         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2039                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2040         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2041         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2042
2043         memset(frames, 0, sizeof(frames));
2044         frame = frames;
2045         frame->entries = entries;
2046         frame->at = entries;
2047         frame->bh = bh;
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, bh2);
2053         if (retval)
2054                 goto out_frames;        
2055
2056         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2057         if (IS_ERR(de)) {
2058                 retval = PTR_ERR(de);
2059                 goto out_frames;
2060         }
2061
2062         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2063 out_frames:
2064         /*
2065          * Even if the block split failed, we have to properly write
2066          * out all the changes we did so far. Otherwise we can end up
2067          * with corrupted filesystem.
2068          */
2069         if (retval)
2070                 ext4_mark_inode_dirty(handle, dir);
2071         dx_release(frames);
2072         brelse(bh2);
2073         return retval;
2074 }
2075
2076 /*
2077  *      ext4_add_entry()
2078  *
2079  * adds a file entry to the specified directory, using the same
2080  * semantics as ext4_find_entry(). It returns NULL if it failed.
2081  *
2082  * NOTE!! The inode part of 'de' is left at 0 - which means you
2083  * may not sleep between calling this and putting something into
2084  * the entry, as someone else might have used it while you slept.
2085  */
2086 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2087                           struct inode *inode)
2088 {
2089         struct inode *dir = d_inode(dentry->d_parent);
2090         struct buffer_head *bh = NULL;
2091         struct ext4_dir_entry_2 *de;
2092         struct ext4_dir_entry_tail *t;
2093         struct super_block *sb;
2094         struct ext4_filename fname;
2095         int     retval;
2096         int     dx_fallback=0;
2097         unsigned blocksize;
2098         ext4_lblk_t block, blocks;
2099         int     csum_size = 0;
2100
2101         if (ext4_has_metadata_csum(inode->i_sb))
2102                 csum_size = sizeof(struct ext4_dir_entry_tail);
2103
2104         sb = dir->i_sb;
2105         blocksize = sb->s_blocksize;
2106         if (!dentry->d_name.len)
2107                 return -EINVAL;
2108
2109         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2110         if (retval)
2111                 return retval;
2112
2113         if (ext4_has_inline_data(dir)) {
2114                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2115                 if (retval < 0)
2116                         goto out;
2117                 if (retval == 1) {
2118                         retval = 0;
2119                         goto out;
2120                 }
2121         }
2122
2123         if (is_dx(dir)) {
2124                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2125                 if (!retval || (retval != ERR_BAD_DX_DIR))
2126                         goto out;
2127                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2128                 dx_fallback++;
2129                 ext4_mark_inode_dirty(handle, dir);
2130         }
2131         blocks = dir->i_size >> sb->s_blocksize_bits;
2132         for (block = 0; block < blocks; block++) {
2133                 bh = ext4_read_dirblock(dir, block, DIRENT);
2134                 if (IS_ERR(bh)) {
2135                         retval = PTR_ERR(bh);
2136                         bh = NULL;
2137                         goto out;
2138                 }
2139                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2140                                            NULL, bh);
2141                 if (retval != -ENOSPC)
2142                         goto out;
2143
2144                 if (blocks == 1 && !dx_fallback &&
2145                     ext4_has_feature_dir_index(sb)) {
2146                         retval = make_indexed_dir(handle, &fname, dir,
2147                                                   inode, bh);
2148                         bh = NULL; /* make_indexed_dir releases bh */
2149                         goto out;
2150                 }
2151                 brelse(bh);
2152         }
2153         bh = ext4_append(handle, dir, &block);
2154         if (IS_ERR(bh)) {
2155                 retval = PTR_ERR(bh);
2156                 bh = NULL;
2157                 goto out;
2158         }
2159         de = (struct ext4_dir_entry_2 *) bh->b_data;
2160         de->inode = 0;
2161         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2162
2163         if (csum_size) {
2164                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2165                 initialize_dirent_tail(t, blocksize);
2166         }
2167
2168         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2169 out:
2170         ext4_fname_free_filename(&fname);
2171         brelse(bh);
2172         if (retval == 0)
2173                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2174         return retval;
2175 }
2176
2177 /*
2178  * Returns 0 for success, or a negative error value
2179  */
2180 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2181                              struct inode *dir, struct inode *inode)
2182 {
2183         struct dx_frame frames[2], *frame;
2184         struct dx_entry *entries, *at;
2185         struct buffer_head *bh;
2186         struct super_block *sb = dir->i_sb;
2187         struct ext4_dir_entry_2 *de;
2188         int err;
2189
2190         frame = dx_probe(fname, dir, NULL, frames);
2191         if (IS_ERR(frame))
2192                 return PTR_ERR(frame);
2193         entries = frame->entries;
2194         at = frame->at;
2195         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2196         if (IS_ERR(bh)) {
2197                 err = PTR_ERR(bh);
2198                 bh = NULL;
2199                 goto cleanup;
2200         }
2201
2202         BUFFER_TRACE(bh, "get_write_access");
2203         err = ext4_journal_get_write_access(handle, bh);
2204         if (err)
2205                 goto journal_error;
2206
2207         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2208         if (err != -ENOSPC)
2209                 goto cleanup;
2210
2211         /* Block full, should compress but for now just split */
2212         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2213                        dx_get_count(entries), dx_get_limit(entries)));
2214         /* Need to split index? */
2215         if (dx_get_count(entries) == dx_get_limit(entries)) {
2216                 ext4_lblk_t newblock;
2217                 unsigned icount = dx_get_count(entries);
2218                 int levels = frame - frames;
2219                 struct dx_entry *entries2;
2220                 struct dx_node *node2;
2221                 struct buffer_head *bh2;
2222
2223                 if (levels && (dx_get_count(frames->entries) ==
2224                                dx_get_limit(frames->entries))) {
2225                         ext4_warning_inode(dir, "Directory index full!");
2226                         err = -ENOSPC;
2227                         goto cleanup;
2228                 }
2229                 bh2 = ext4_append(handle, dir, &newblock);
2230                 if (IS_ERR(bh2)) {
2231                         err = PTR_ERR(bh2);
2232                         goto cleanup;
2233                 }
2234                 node2 = (struct dx_node *)(bh2->b_data);
2235                 entries2 = node2->entries;
2236                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2237                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2238                                                            sb->s_blocksize);
2239                 BUFFER_TRACE(frame->bh, "get_write_access");
2240                 err = ext4_journal_get_write_access(handle, frame->bh);
2241                 if (err)
2242                         goto journal_error;
2243                 if (levels) {
2244                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2245                         unsigned hash2 = dx_get_hash(entries + icount1);
2246                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2247                                        icount1, icount2));
2248
2249                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2250                         err = ext4_journal_get_write_access(handle,
2251                                                              frames[0].bh);
2252                         if (err)
2253                                 goto journal_error;
2254
2255                         memcpy((char *) entries2, (char *) (entries + icount1),
2256                                icount2 * sizeof(struct dx_entry));
2257                         dx_set_count(entries, icount1);
2258                         dx_set_count(entries2, icount2);
2259                         dx_set_limit(entries2, dx_node_limit(dir));
2260
2261                         /* Which index block gets the new entry? */
2262                         if (at - entries >= icount1) {
2263                                 frame->at = at = at - entries - icount1 + entries2;
2264                                 frame->entries = entries = entries2;
2265                                 swap(frame->bh, bh2);
2266                         }
2267                         dx_insert_block(frames + 0, hash2, newblock);
2268                         dxtrace(dx_show_index("node", frames[1].entries));
2269                         dxtrace(dx_show_index("node",
2270                                ((struct dx_node *) bh2->b_data)->entries));
2271                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2272                         if (err)
2273                                 goto journal_error;
2274                         brelse (bh2);
2275                 } else {
2276                         dxtrace(printk(KERN_DEBUG
2277                                        "Creating second level index...\n"));
2278                         memcpy((char *) entries2, (char *) entries,
2279                                icount * sizeof(struct dx_entry));
2280                         dx_set_limit(entries2, dx_node_limit(dir));
2281
2282                         /* Set up root */
2283                         dx_set_count(entries, 1);
2284                         dx_set_block(entries + 0, newblock);
2285                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2286
2287                         /* Add new access path frame */
2288                         frame = frames + 1;
2289                         frame->at = at = at - entries + entries2;
2290                         frame->entries = entries = entries2;
2291                         frame->bh = bh2;
2292                         err = ext4_journal_get_write_access(handle,
2293                                                              frame->bh);
2294                         if (err)
2295                                 goto journal_error;
2296                 }
2297                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2298                 if (err) {
2299                         ext4_std_error(inode->i_sb, err);
2300                         goto cleanup;
2301                 }
2302         }
2303         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2304         if (IS_ERR(de)) {
2305                 err = PTR_ERR(de);
2306                 goto cleanup;
2307         }
2308         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2309         goto cleanup;
2310
2311 journal_error:
2312         ext4_std_error(dir->i_sb, err);
2313 cleanup:
2314         brelse(bh);
2315         dx_release(frames);
2316         return err;
2317 }
2318
2319 /*
2320  * ext4_generic_delete_entry deletes a directory entry by merging it
2321  * with the previous entry
2322  */
2323 int ext4_generic_delete_entry(handle_t *handle,
2324                               struct inode *dir,
2325                               struct ext4_dir_entry_2 *de_del,
2326                               struct buffer_head *bh,
2327                               void *entry_buf,
2328                               int buf_size,
2329                               int csum_size)
2330 {
2331         struct ext4_dir_entry_2 *de, *pde;
2332         unsigned int blocksize = dir->i_sb->s_blocksize;
2333         int i;
2334
2335         i = 0;
2336         pde = NULL;
2337         de = (struct ext4_dir_entry_2 *)entry_buf;
2338         while (i < buf_size - csum_size) {
2339                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2340                                          bh->b_data, bh->b_size, i))
2341                         return -EFSCORRUPTED;
2342                 if (de == de_del)  {
2343                         if (pde)
2344                                 pde->rec_len = ext4_rec_len_to_disk(
2345                                         ext4_rec_len_from_disk(pde->rec_len,
2346                                                                blocksize) +
2347                                         ext4_rec_len_from_disk(de->rec_len,
2348                                                                blocksize),
2349                                         blocksize);
2350                         else
2351                                 de->inode = 0;
2352                         dir->i_version++;
2353                         return 0;
2354                 }
2355                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2356                 pde = de;
2357                 de = ext4_next_entry(de, blocksize);
2358         }
2359         return -ENOENT;
2360 }
2361
2362 static int ext4_delete_entry(handle_t *handle,
2363                              struct inode *dir,
2364                              struct ext4_dir_entry_2 *de_del,
2365                              struct buffer_head *bh)
2366 {
2367         int err, csum_size = 0;
2368
2369         if (ext4_has_inline_data(dir)) {
2370                 int has_inline_data = 1;
2371                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2372                                                &has_inline_data);
2373                 if (has_inline_data)
2374                         return err;
2375         }
2376
2377         if (ext4_has_metadata_csum(dir->i_sb))
2378                 csum_size = sizeof(struct ext4_dir_entry_tail);
2379
2380         BUFFER_TRACE(bh, "get_write_access");
2381         err = ext4_journal_get_write_access(handle, bh);
2382         if (unlikely(err))
2383                 goto out;
2384
2385         err = ext4_generic_delete_entry(handle, dir, de_del,
2386                                         bh, bh->b_data,
2387                                         dir->i_sb->s_blocksize, csum_size);
2388         if (err)
2389                 goto out;
2390
2391         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2392         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2393         if (unlikely(err))
2394                 goto out;
2395
2396         return 0;
2397 out:
2398         if (err != -ENOENT)
2399                 ext4_std_error(dir->i_sb, err);
2400         return err;
2401 }
2402
2403 /*
2404  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2405  * since this indicates that nlinks count was previously 1.
2406  */
2407 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2408 {
2409         inc_nlink(inode);
2410         if (is_dx(inode) && inode->i_nlink > 1) {
2411                 /* limit is 16-bit i_links_count */
2412                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2413                         set_nlink(inode, 1);
2414                         ext4_set_feature_dir_nlink(inode->i_sb);
2415                 }
2416         }
2417 }
2418
2419 /*
2420  * If a directory had nlink == 1, then we should let it be 1. This indicates
2421  * directory has >EXT4_LINK_MAX subdirs.
2422  */
2423 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2424 {
2425         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2426                 drop_nlink(inode);
2427 }
2428
2429
2430 static int ext4_add_nondir(handle_t *handle,
2431                 struct dentry *dentry, struct inode *inode)
2432 {
2433         int err = ext4_add_entry(handle, dentry, inode);
2434         if (!err) {
2435                 ext4_mark_inode_dirty(handle, inode);
2436                 unlock_new_inode(inode);
2437                 d_instantiate(dentry, inode);
2438                 return 0;
2439         }
2440         drop_nlink(inode);
2441         unlock_new_inode(inode);
2442         iput(inode);
2443         return err;
2444 }
2445
2446 /*
2447  * By the time this is called, we already have created
2448  * the directory cache entry for the new file, but it
2449  * is so far negative - it has no inode.
2450  *
2451  * If the create succeeds, we fill in the inode information
2452  * with d_instantiate().
2453  */
2454 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2455                        bool excl)
2456 {
2457         handle_t *handle;
2458         struct inode *inode;
2459         int err, credits, retries = 0;
2460
2461         err = dquot_initialize(dir);
2462         if (err)
2463                 return err;
2464
2465         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2466                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2467 retry:
2468         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2469                                             NULL, EXT4_HT_DIR, credits);
2470         handle = ext4_journal_current_handle();
2471         err = PTR_ERR(inode);
2472         if (!IS_ERR(inode)) {
2473                 inode->i_op = &ext4_file_inode_operations;
2474                 inode->i_fop = &ext4_file_operations;
2475                 ext4_set_aops(inode);
2476                 err = ext4_add_nondir(handle, dentry, inode);
2477                 if (!err && IS_DIRSYNC(dir))
2478                         ext4_handle_sync(handle);
2479         }
2480         if (handle)
2481                 ext4_journal_stop(handle);
2482         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2483                 goto retry;
2484         return err;
2485 }
2486
2487 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2488                       umode_t mode, dev_t rdev)
2489 {
2490         handle_t *handle;
2491         struct inode *inode;
2492         int err, credits, retries = 0;
2493
2494         err = dquot_initialize(dir);
2495         if (err)
2496                 return err;
2497
2498         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2499                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2500 retry:
2501         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2502                                             NULL, EXT4_HT_DIR, credits);
2503         handle = ext4_journal_current_handle();
2504         err = PTR_ERR(inode);
2505         if (!IS_ERR(inode)) {
2506                 init_special_inode(inode, inode->i_mode, rdev);
2507                 inode->i_op = &ext4_special_inode_operations;
2508                 err = ext4_add_nondir(handle, dentry, inode);
2509                 if (!err && IS_DIRSYNC(dir))
2510                         ext4_handle_sync(handle);
2511         }
2512         if (handle)
2513                 ext4_journal_stop(handle);
2514         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2515                 goto retry;
2516         return err;
2517 }
2518
2519 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2520 {
2521         handle_t *handle;
2522         struct inode *inode;
2523         int err, retries = 0;
2524
2525         err = dquot_initialize(dir);
2526         if (err)
2527                 return err;
2528
2529 retry:
2530         inode = ext4_new_inode_start_handle(dir, mode,
2531                                             NULL, 0, NULL,
2532                                             EXT4_HT_DIR,
2533                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2534                           4 + EXT4_XATTR_TRANS_BLOCKS);
2535         handle = ext4_journal_current_handle();
2536         err = PTR_ERR(inode);
2537         if (!IS_ERR(inode)) {
2538                 inode->i_op = &ext4_file_inode_operations;
2539                 inode->i_fop = &ext4_file_operations;
2540                 ext4_set_aops(inode);
2541                 d_tmpfile(dentry, inode);
2542                 err = ext4_orphan_add(handle, inode);
2543                 if (err)
2544                         goto err_unlock_inode;
2545                 mark_inode_dirty(inode);
2546                 unlock_new_inode(inode);
2547         }
2548         if (handle)
2549                 ext4_journal_stop(handle);
2550         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2551                 goto retry;
2552         return err;
2553 err_unlock_inode:
2554         ext4_journal_stop(handle);
2555         unlock_new_inode(inode);
2556         return err;
2557 }
2558
2559 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2560                           struct ext4_dir_entry_2 *de,
2561                           int blocksize, int csum_size,
2562                           unsigned int parent_ino, int dotdot_real_len)
2563 {
2564         de->inode = cpu_to_le32(inode->i_ino);
2565         de->name_len = 1;
2566         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2567                                            blocksize);
2568         strcpy(de->name, ".");
2569         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2570
2571         de = ext4_next_entry(de, blocksize);
2572         de->inode = cpu_to_le32(parent_ino);
2573         de->name_len = 2;
2574         if (!dotdot_real_len)
2575                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2576                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2577                                         blocksize);
2578         else
2579                 de->rec_len = ext4_rec_len_to_disk(
2580                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2581         strcpy(de->name, "..");
2582         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2583
2584         return ext4_next_entry(de, blocksize);
2585 }
2586
2587 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2588                              struct inode *inode)
2589 {
2590         struct buffer_head *dir_block = NULL;
2591         struct ext4_dir_entry_2 *de;
2592         struct ext4_dir_entry_tail *t;
2593         ext4_lblk_t block = 0;
2594         unsigned int blocksize = dir->i_sb->s_blocksize;
2595         int csum_size = 0;
2596         int err;
2597
2598         if (ext4_has_metadata_csum(dir->i_sb))
2599                 csum_size = sizeof(struct ext4_dir_entry_tail);
2600
2601         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2602                 err = ext4_try_create_inline_dir(handle, dir, inode);
2603                 if (err < 0 && err != -ENOSPC)
2604                         goto out;
2605                 if (!err)
2606                         goto out;
2607         }
2608
2609         inode->i_size = 0;
2610         dir_block = ext4_append(handle, inode, &block);
2611         if (IS_ERR(dir_block))
2612                 return PTR_ERR(dir_block);
2613         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2614         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2615         set_nlink(inode, 2);
2616         if (csum_size) {
2617                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2618                 initialize_dirent_tail(t, blocksize);
2619         }
2620
2621         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2622         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2623         if (err)
2624                 goto out;
2625         set_buffer_verified(dir_block);
2626 out:
2627         brelse(dir_block);
2628         return err;
2629 }
2630
2631 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2632 {
2633         handle_t *handle;
2634         struct inode *inode;
2635         int err, credits, retries = 0;
2636
2637         if (EXT4_DIR_LINK_MAX(dir))
2638                 return -EMLINK;
2639
2640         err = dquot_initialize(dir);
2641         if (err)
2642                 return err;
2643
2644         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2645                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2646 retry:
2647         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2648                                             &dentry->d_name,
2649                                             0, NULL, EXT4_HT_DIR, credits);
2650         handle = ext4_journal_current_handle();
2651         err = PTR_ERR(inode);
2652         if (IS_ERR(inode))
2653                 goto out_stop;
2654
2655         inode->i_op = &ext4_dir_inode_operations;
2656         inode->i_fop = &ext4_dir_operations;
2657         err = ext4_init_new_dir(handle, dir, inode);
2658         if (err)
2659                 goto out_clear_inode;
2660         err = ext4_mark_inode_dirty(handle, inode);
2661         if (!err)
2662                 err = ext4_add_entry(handle, dentry, inode);
2663         if (err) {
2664 out_clear_inode:
2665                 clear_nlink(inode);
2666                 unlock_new_inode(inode);
2667                 ext4_mark_inode_dirty(handle, inode);
2668                 iput(inode);
2669                 goto out_stop;
2670         }
2671         ext4_inc_count(handle, dir);
2672         ext4_update_dx_flag(dir);
2673         err = ext4_mark_inode_dirty(handle, dir);
2674         if (err)
2675                 goto out_clear_inode;
2676         unlock_new_inode(inode);
2677         d_instantiate(dentry, inode);
2678         if (IS_DIRSYNC(dir))
2679                 ext4_handle_sync(handle);
2680
2681 out_stop:
2682         if (handle)
2683                 ext4_journal_stop(handle);
2684         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2685                 goto retry;
2686         return err;
2687 }
2688
2689 /*
2690  * routine to check that the specified directory is empty (for rmdir)
2691  */
2692 bool ext4_empty_dir(struct inode *inode)
2693 {
2694         unsigned int offset;
2695         struct buffer_head *bh;
2696         struct ext4_dir_entry_2 *de, *de1;
2697         struct super_block *sb;
2698
2699         if (ext4_has_inline_data(inode)) {
2700                 int has_inline_data = 1;
2701                 int ret;
2702
2703                 ret = empty_inline_dir(inode, &has_inline_data);
2704                 if (has_inline_data)
2705                         return ret;
2706         }
2707
2708         sb = inode->i_sb;
2709         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2710                 EXT4_ERROR_INODE(inode, "invalid size");
2711                 return true;
2712         }
2713         bh = ext4_read_dirblock(inode, 0, EITHER);
2714         if (IS_ERR(bh))
2715                 return true;
2716
2717         de = (struct ext4_dir_entry_2 *) bh->b_data;
2718         de1 = ext4_next_entry(de, sb->s_blocksize);
2719         if (le32_to_cpu(de->inode) != inode->i_ino ||
2720                         le32_to_cpu(de1->inode) == 0 ||
2721                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2722                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2723                 brelse(bh);
2724                 return true;
2725         }
2726         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2727                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2728         de = ext4_next_entry(de1, sb->s_blocksize);
2729         while (offset < inode->i_size) {
2730                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2731                         unsigned int lblock;
2732                         brelse(bh);
2733                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2734                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2735                         if (IS_ERR(bh))
2736                                 return true;
2737                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2738                 }
2739                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2740                                          bh->b_data, bh->b_size, offset)) {
2741                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2742                                                          sb->s_blocksize);
2743                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2744                         continue;
2745                 }
2746                 if (le32_to_cpu(de->inode)) {
2747                         brelse(bh);
2748                         return false;
2749                 }
2750                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2751                 de = ext4_next_entry(de, sb->s_blocksize);
2752         }
2753         brelse(bh);
2754         return true;
2755 }
2756
2757 /*
2758  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2759  * such inodes, starting at the superblock, in case we crash before the
2760  * file is closed/deleted, or in case the inode truncate spans multiple
2761  * transactions and the last transaction is not recovered after a crash.
2762  *
2763  * At filesystem recovery time, we walk this list deleting unlinked
2764  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2765  *
2766  * Orphan list manipulation functions must be called under i_mutex unless
2767  * we are just creating the inode or deleting it.
2768  */
2769 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2770 {
2771         struct super_block *sb = inode->i_sb;
2772         struct ext4_sb_info *sbi = EXT4_SB(sb);
2773         struct ext4_iloc iloc;
2774         int err = 0, rc;
2775         bool dirty = false;
2776
2777         if (!sbi->s_journal || is_bad_inode(inode))
2778                 return 0;
2779
2780         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2781                      !inode_is_locked(inode));
2782         /*
2783          * Exit early if inode already is on orphan list. This is a big speedup
2784          * since we don't have to contend on the global s_orphan_lock.
2785          */
2786         if (!list_empty(&EXT4_I(inode)->i_orphan))
2787                 return 0;
2788
2789         /*
2790          * Orphan handling is only valid for files with data blocks
2791          * being truncated, or files being unlinked. Note that we either
2792          * hold i_mutex, or the inode can not be referenced from outside,
2793          * so i_nlink should not be bumped due to race
2794          */
2795         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2796                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2797
2798         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2799         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2800         if (err)
2801                 goto out;
2802
2803         err = ext4_reserve_inode_write(handle, inode, &iloc);
2804         if (err)
2805                 goto out;
2806
2807         mutex_lock(&sbi->s_orphan_lock);
2808         /*
2809          * Due to previous errors inode may be already a part of on-disk
2810          * orphan list. If so skip on-disk list modification.
2811          */
2812         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2813             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2814                 /* Insert this inode at the head of the on-disk orphan list */
2815                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2816                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2817                 dirty = true;
2818         }
2819         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2820         mutex_unlock(&sbi->s_orphan_lock);
2821
2822         if (dirty) {
2823                 err = ext4_handle_dirty_super(handle, sb);
2824                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2825                 if (!err)
2826                         err = rc;
2827                 if (err) {
2828                         /*
2829                          * We have to remove inode from in-memory list if
2830                          * addition to on disk orphan list failed. Stray orphan
2831                          * list entries can cause panics at unmount time.
2832                          */
2833                         mutex_lock(&sbi->s_orphan_lock);
2834                         list_del_init(&EXT4_I(inode)->i_orphan);
2835                         mutex_unlock(&sbi->s_orphan_lock);
2836                 }
2837         }
2838         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2839         jbd_debug(4, "orphan inode %lu will point to %d\n",
2840                         inode->i_ino, NEXT_ORPHAN(inode));
2841 out:
2842         ext4_std_error(sb, err);
2843         return err;
2844 }
2845
2846 /*
2847  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2848  * of such inodes stored on disk, because it is finally being cleaned up.
2849  */
2850 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2851 {
2852         struct list_head *prev;
2853         struct ext4_inode_info *ei = EXT4_I(inode);
2854         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2855         __u32 ino_next;
2856         struct ext4_iloc iloc;
2857         int err = 0;
2858
2859         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2860                 return 0;
2861
2862         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2863                      !inode_is_locked(inode));
2864         /* Do this quick check before taking global s_orphan_lock. */
2865         if (list_empty(&ei->i_orphan))
2866                 return 0;
2867
2868         if (handle) {
2869                 /* Grab inode buffer early before taking global s_orphan_lock */
2870                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2871         }
2872
2873         mutex_lock(&sbi->s_orphan_lock);
2874         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2875
2876         prev = ei->i_orphan.prev;
2877         list_del_init(&ei->i_orphan);
2878
2879         /* If we're on an error path, we may not have a valid
2880          * transaction handle with which to update the orphan list on
2881          * disk, but we still need to remove the inode from the linked
2882          * list in memory. */
2883         if (!handle || err) {
2884                 mutex_unlock(&sbi->s_orphan_lock);
2885                 goto out_err;
2886         }
2887
2888         ino_next = NEXT_ORPHAN(inode);
2889         if (prev == &sbi->s_orphan) {
2890                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2891                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2892                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2893                 if (err) {
2894                         mutex_unlock(&sbi->s_orphan_lock);
2895                         goto out_brelse;
2896                 }
2897                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2898                 mutex_unlock(&sbi->s_orphan_lock);
2899                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2900         } else {
2901                 struct ext4_iloc iloc2;
2902                 struct inode *i_prev =
2903                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2904
2905                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2906                           i_prev->i_ino, ino_next);
2907                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2908                 if (err) {
2909                         mutex_unlock(&sbi->s_orphan_lock);
2910                         goto out_brelse;
2911                 }
2912                 NEXT_ORPHAN(i_prev) = ino_next;
2913                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2914                 mutex_unlock(&sbi->s_orphan_lock);
2915         }
2916         if (err)
2917                 goto out_brelse;
2918         NEXT_ORPHAN(inode) = 0;
2919         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2920 out_err:
2921         ext4_std_error(inode->i_sb, err);
2922         return err;
2923
2924 out_brelse:
2925         brelse(iloc.bh);
2926         goto out_err;
2927 }
2928
2929 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2930 {
2931         int retval;
2932         struct inode *inode;
2933         struct buffer_head *bh;
2934         struct ext4_dir_entry_2 *de;
2935         handle_t *handle = NULL;
2936
2937         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2938                 return -EIO;
2939
2940         /* Initialize quotas before so that eventual writes go in
2941          * separate transaction */
2942         retval = dquot_initialize(dir);
2943         if (retval)
2944                 return retval;
2945         retval = dquot_initialize(d_inode(dentry));
2946         if (retval)
2947                 return retval;
2948
2949         retval = -ENOENT;
2950         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2951         if (IS_ERR(bh))
2952                 return PTR_ERR(bh);
2953         if (!bh)
2954                 goto end_rmdir;
2955
2956         inode = d_inode(dentry);
2957
2958         retval = -EFSCORRUPTED;
2959         if (le32_to_cpu(de->inode) != inode->i_ino)
2960                 goto end_rmdir;
2961
2962         retval = -ENOTEMPTY;
2963         if (!ext4_empty_dir(inode))
2964                 goto end_rmdir;
2965
2966         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2967                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2968         if (IS_ERR(handle)) {
2969                 retval = PTR_ERR(handle);
2970                 handle = NULL;
2971                 goto end_rmdir;
2972         }
2973
2974         if (IS_DIRSYNC(dir))
2975                 ext4_handle_sync(handle);
2976
2977         retval = ext4_delete_entry(handle, dir, de, bh);
2978         if (retval)
2979                 goto end_rmdir;
2980         if (!EXT4_DIR_LINK_EMPTY(inode))
2981                 ext4_warning_inode(inode,
2982                              "empty directory '%.*s' has too many links (%u)",
2983                              dentry->d_name.len, dentry->d_name.name,
2984                              inode->i_nlink);
2985         inode->i_version++;
2986         clear_nlink(inode);
2987         /* There's no need to set i_disksize: the fact that i_nlink is
2988          * zero will ensure that the right thing happens during any
2989          * recovery. */
2990         inode->i_size = 0;
2991         ext4_orphan_add(handle, inode);
2992         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2993         ext4_mark_inode_dirty(handle, inode);
2994         ext4_dec_count(handle, dir);
2995         ext4_update_dx_flag(dir);
2996         ext4_mark_inode_dirty(handle, dir);
2997
2998 end_rmdir:
2999         brelse(bh);
3000         if (handle)
3001                 ext4_journal_stop(handle);
3002         return retval;
3003 }
3004
3005 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3006 {
3007         int retval;
3008         struct inode *inode;
3009         struct buffer_head *bh;
3010         struct ext4_dir_entry_2 *de;
3011         handle_t *handle = NULL;
3012
3013         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3014                 return -EIO;
3015
3016         trace_ext4_unlink_enter(dir, dentry);
3017         /* Initialize quotas before so that eventual writes go
3018          * in separate transaction */
3019         retval = dquot_initialize(dir);
3020         if (retval)
3021                 return retval;
3022         retval = dquot_initialize(d_inode(dentry));
3023         if (retval)
3024                 return retval;
3025
3026         retval = -ENOENT;
3027         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3028         if (IS_ERR(bh))
3029                 return PTR_ERR(bh);
3030         if (!bh)
3031                 goto end_unlink;
3032
3033         inode = d_inode(dentry);
3034
3035         retval = -EFSCORRUPTED;
3036         if (le32_to_cpu(de->inode) != inode->i_ino)
3037                 goto end_unlink;
3038
3039         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3040                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3041         if (IS_ERR(handle)) {
3042                 retval = PTR_ERR(handle);
3043                 handle = NULL;
3044                 goto end_unlink;
3045         }
3046
3047         if (IS_DIRSYNC(dir))
3048                 ext4_handle_sync(handle);
3049
3050         if (inode->i_nlink == 0) {
3051                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3052                                    dentry->d_name.len, dentry->d_name.name);
3053                 set_nlink(inode, 1);
3054         }
3055         retval = ext4_delete_entry(handle, dir, de, bh);
3056         if (retval)
3057                 goto end_unlink;
3058         dir->i_ctime = dir->i_mtime = current_time(dir);
3059         ext4_update_dx_flag(dir);
3060         ext4_mark_inode_dirty(handle, dir);
3061         drop_nlink(inode);
3062         if (!inode->i_nlink)
3063                 ext4_orphan_add(handle, inode);
3064         inode->i_ctime = current_time(inode);
3065         ext4_mark_inode_dirty(handle, inode);
3066
3067 end_unlink:
3068         brelse(bh);
3069         if (handle)
3070                 ext4_journal_stop(handle);
3071         trace_ext4_unlink_exit(dentry, retval);
3072         return retval;
3073 }
3074
3075 static int ext4_symlink(struct inode *dir,
3076                         struct dentry *dentry, const char *symname)
3077 {
3078         handle_t *handle;
3079         struct inode *inode;
3080         int err, len = strlen(symname);
3081         int credits;
3082         bool encryption_required;
3083         struct fscrypt_str disk_link;
3084         struct fscrypt_symlink_data *sd = NULL;
3085
3086         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3087                 return -EIO;
3088
3089         disk_link.len = len + 1;
3090         disk_link.name = (char *) symname;
3091
3092         encryption_required = (ext4_encrypted_inode(dir) ||
3093                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3094         if (encryption_required) {
3095                 err = fscrypt_get_encryption_info(dir);
3096                 if (err)
3097                         return err;
3098                 if (!fscrypt_has_encryption_key(dir))
3099                         return -ENOKEY;
3100                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3101                                  sizeof(struct fscrypt_symlink_data));
3102                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3103                 if (!sd)
3104                         return -ENOMEM;
3105         }
3106
3107         if (disk_link.len > dir->i_sb->s_blocksize) {
3108                 err = -ENAMETOOLONG;
3109                 goto err_free_sd;
3110         }
3111
3112         err = dquot_initialize(dir);
3113         if (err)
3114                 goto err_free_sd;
3115
3116         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3117                 /*
3118                  * For non-fast symlinks, we just allocate inode and put it on
3119                  * orphan list in the first transaction => we need bitmap,
3120                  * group descriptor, sb, inode block, quota blocks, and
3121                  * possibly selinux xattr blocks.
3122                  */
3123                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3124                           EXT4_XATTR_TRANS_BLOCKS;
3125         } else {
3126                 /*
3127                  * Fast symlink. We have to add entry to directory
3128                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3129                  * allocate new inode (bitmap, group descriptor, inode block,
3130                  * quota blocks, sb is already counted in previous macros).
3131                  */
3132                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3133                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3134         }
3135
3136         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3137                                             &dentry->d_name, 0, NULL,
3138                                             EXT4_HT_DIR, credits);
3139         handle = ext4_journal_current_handle();
3140         if (IS_ERR(inode)) {
3141                 if (handle)
3142                         ext4_journal_stop(handle);
3143                 err = PTR_ERR(inode);
3144                 goto err_free_sd;