Fix 64KB blocksize in ext3 directories
[linux-2.6] / fs / ext3 / namei.c
1 /*
2  *  linux/fs/ext3/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/jbd.h>
30 #include <linux/time.h>
31 #include <linux/ext3_fs.h>
32 #include <linux/ext3_jbd.h>
33 #include <linux/fcntl.h>
34 #include <linux/stat.h>
35 #include <linux/string.h>
36 #include <linux/quotaops.h>
37 #include <linux/buffer_head.h>
38 #include <linux/bio.h>
39
40 #include "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext3_append(handle_t *handle,
53                                         struct inode *inode,
54                                         u32 *block, int *err)
55 {
56         struct buffer_head *bh;
57
58         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60         if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61                 inode->i_size += inode->i_sb->s_blocksize;
62                 EXT3_I(inode)->i_disksize = inode->i_size;
63                 ext3_journal_get_write_access(handle,bh);
64         }
65         return bh;
66 }
67
68 #ifndef assert
69 #define assert(test) J_ASSERT(test)
70 #endif
71
72 #ifndef swap
73 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84         __le32 inode;
85         __le16 rec_len;
86         u8 name_len;
87         u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92         __le16 limit;
93         __le16 count;
94 };
95
96 struct dx_entry
97 {
98         __le32 hash;
99         __le32 block;
100 };
101
102 /*
103  * dx_root_info is laid out so that if it should somehow get overlaid by a
104  * dirent the two low bits of the hash version will be zero.  Therefore, the
105  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
106  */
107
108 struct dx_root
109 {
110         struct fake_dirent dot;
111         char dot_name[4];
112         struct fake_dirent dotdot;
113         char dotdot_name[4];
114         struct dx_root_info
115         {
116                 __le32 reserved_zero;
117                 u8 hash_version;
118                 u8 info_length; /* 8 */
119                 u8 indirect_levels;
120                 u8 unused_flags;
121         }
122         info;
123         struct dx_entry entries[0];
124 };
125
126 struct dx_node
127 {
128         struct fake_dirent fake;
129         struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135         struct buffer_head *bh;
136         struct dx_entry *entries;
137         struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142         u32 hash;
143         u16 offs;
144         u16 size;
145 };
146
147 static inline unsigned dx_get_block (struct dx_entry *entry);
148 static void dx_set_block (struct dx_entry *entry, unsigned value);
149 static inline unsigned dx_get_hash (struct dx_entry *entry);
150 static void dx_set_hash (struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count (struct dx_entry *entries);
152 static unsigned dx_get_limit (struct dx_entry *entries);
153 static void dx_set_count (struct dx_entry *entries, unsigned value);
154 static void dx_set_limit (struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit (struct inode *dir);
157 static struct dx_frame *dx_probe(struct dentry *dentry,
158                                  struct inode *dir,
159                                  struct dx_hash_info *hinfo,
160                                  struct dx_frame *frame,
161                                  int *err);
162 static void dx_release (struct dx_frame *frames);
163 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167                 struct dx_map_entry *offsets, int count);
168 static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171                                  struct dx_frame *frame,
172                                  struct dx_frame *frames,
173                                  __u32 *start_hash);
174 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175                        struct ext3_dir_entry_2 **res_dir, int *err);
176 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177                              struct inode *inode);
178
179 /*
180  * p is at least 6 bytes before the end of page
181  */
182 static inline struct ext3_dir_entry_2 *
183 ext3_next_entry(struct ext3_dir_entry_2 *p)
184 {
185         return (struct ext3_dir_entry_2 *)((char *)p +
186                 ext3_rec_len_from_disk(p->rec_len));
187 }
188
189 /*
190  * Future: use high four bits of block for coalesce-on-delete flags
191  * Mask them off for now.
192  */
193
194 static inline unsigned dx_get_block (struct dx_entry *entry)
195 {
196         return le32_to_cpu(entry->block) & 0x00ffffff;
197 }
198
199 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
200 {
201         entry->block = cpu_to_le32(value);
202 }
203
204 static inline unsigned dx_get_hash (struct dx_entry *entry)
205 {
206         return le32_to_cpu(entry->hash);
207 }
208
209 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
210 {
211         entry->hash = cpu_to_le32(value);
212 }
213
214 static inline unsigned dx_get_count (struct dx_entry *entries)
215 {
216         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
217 }
218
219 static inline unsigned dx_get_limit (struct dx_entry *entries)
220 {
221         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
222 }
223
224 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
225 {
226         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
227 }
228
229 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
230 {
231         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
232 }
233
234 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
235 {
236         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
237                 EXT3_DIR_REC_LEN(2) - infosize;
238         return 0? 20: entry_space / sizeof(struct dx_entry);
239 }
240
241 static inline unsigned dx_node_limit (struct inode *dir)
242 {
243         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
244         return 0? 22: entry_space / sizeof(struct dx_entry);
245 }
246
247 /*
248  * Debug
249  */
250 #ifdef DX_DEBUG
251 static void dx_show_index (char * label, struct dx_entry *entries)
252 {
253         int i, n = dx_get_count (entries);
254         printk("%s index ", label);
255         for (i = 0; i < n; i++)
256         {
257                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
258         }
259         printk("\n");
260 }
261
262 struct stats
263 {
264         unsigned names;
265         unsigned space;
266         unsigned bcount;
267 };
268
269 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
270                                  int size, int show_names)
271 {
272         unsigned names = 0, space = 0;
273         char *base = (char *) de;
274         struct dx_hash_info h = *hinfo;
275
276         printk("names: ");
277         while ((char *) de < base + size)
278         {
279                 if (de->inode)
280                 {
281                         if (show_names)
282                         {
283                                 int len = de->name_len;
284                                 char *name = de->name;
285                                 while (len--) printk("%c", *name++);
286                                 ext3fs_dirhash(de->name, de->name_len, &h);
287                                 printk(":%x.%u ", h.hash,
288                                        ((char *) de - base));
289                         }
290                         space += EXT3_DIR_REC_LEN(de->name_len);
291                         names++;
292                 }
293                 de = ext3_next_entry(de);
294         }
295         printk("(%i)\n", names);
296         return (struct stats) { names, space, 1 };
297 }
298
299 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
300                              struct dx_entry *entries, int levels)
301 {
302         unsigned blocksize = dir->i_sb->s_blocksize;
303         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
304         unsigned bcount = 0;
305         struct buffer_head *bh;
306         int err;
307         printk("%i indexed blocks...\n", count);
308         for (i = 0; i < count; i++, entries++)
309         {
310                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
311                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
312                 struct stats stats;
313                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
314                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
315                 stats = levels?
316                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
317                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
318                 names += stats.names;
319                 space += stats.space;
320                 bcount += stats.bcount;
321                 brelse (bh);
322         }
323         if (bcount)
324                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
325                         names, space/bcount,(space/bcount)*100/blocksize);
326         return (struct stats) { names, space, bcount};
327 }
328 #endif /* DX_DEBUG */
329
330 /*
331  * Probe for a directory leaf block to search.
332  *
333  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
334  * error in the directory index, and the caller should fall back to
335  * searching the directory normally.  The callers of dx_probe **MUST**
336  * check for this error code, and make sure it never gets reflected
337  * back to userspace.
338  */
339 static struct dx_frame *
340 dx_probe(struct dentry *dentry, struct inode *dir,
341          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
342 {
343         unsigned count, indirect;
344         struct dx_entry *at, *entries, *p, *q, *m;
345         struct dx_root *root;
346         struct buffer_head *bh;
347         struct dx_frame *frame = frame_in;
348         u32 hash;
349
350         frame->bh = NULL;
351         if (dentry)
352                 dir = dentry->d_parent->d_inode;
353         if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
354                 goto fail;
355         root = (struct dx_root *) bh->b_data;
356         if (root->info.hash_version != DX_HASH_TEA &&
357             root->info.hash_version != DX_HASH_HALF_MD4 &&
358             root->info.hash_version != DX_HASH_LEGACY) {
359                 ext3_warning(dir->i_sb, __FUNCTION__,
360                              "Unrecognised inode hash code %d",
361                              root->info.hash_version);
362                 brelse(bh);
363                 *err = ERR_BAD_DX_DIR;
364                 goto fail;
365         }
366         hinfo->hash_version = root->info.hash_version;
367         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
368         if (dentry)
369                 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
370         hash = hinfo->hash;
371
372         if (root->info.unused_flags & 1) {
373                 ext3_warning(dir->i_sb, __FUNCTION__,
374                              "Unimplemented inode hash flags: %#06x",
375                              root->info.unused_flags);
376                 brelse(bh);
377                 *err = ERR_BAD_DX_DIR;
378                 goto fail;
379         }
380
381         if ((indirect = root->info.indirect_levels) > 1) {
382                 ext3_warning(dir->i_sb, __FUNCTION__,
383                              "Unimplemented inode hash depth: %#06x",
384                              root->info.indirect_levels);
385                 brelse(bh);
386                 *err = ERR_BAD_DX_DIR;
387                 goto fail;
388         }
389
390         entries = (struct dx_entry *) (((char *)&root->info) +
391                                        root->info.info_length);
392
393         if (dx_get_limit(entries) != dx_root_limit(dir,
394                                                    root->info.info_length)) {
395                 ext3_warning(dir->i_sb, __FUNCTION__,
396                              "dx entry: limit != root limit");
397                 brelse(bh);
398                 *err = ERR_BAD_DX_DIR;
399                 goto fail;
400         }
401
402         dxtrace (printk("Look up %x", hash));
403         while (1)
404         {
405                 count = dx_get_count(entries);
406                 if (!count || count > dx_get_limit(entries)) {
407                         ext3_warning(dir->i_sb, __FUNCTION__,
408                                      "dx entry: no count or count > limit");
409                         brelse(bh);
410                         *err = ERR_BAD_DX_DIR;
411                         goto fail2;
412                 }
413
414                 p = entries + 1;
415                 q = entries + count - 1;
416                 while (p <= q)
417                 {
418                         m = p + (q - p)/2;
419                         dxtrace(printk("."));
420                         if (dx_get_hash(m) > hash)
421                                 q = m - 1;
422                         else
423                                 p = m + 1;
424                 }
425
426                 if (0) // linear search cross check
427                 {
428                         unsigned n = count - 1;
429                         at = entries;
430                         while (n--)
431                         {
432                                 dxtrace(printk(","));
433                                 if (dx_get_hash(++at) > hash)
434                                 {
435                                         at--;
436                                         break;
437                                 }
438                         }
439                         assert (at == p - 1);
440                 }
441
442                 at = p - 1;
443                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
444                 frame->bh = bh;
445                 frame->entries = entries;
446                 frame->at = at;
447                 if (!indirect--) return frame;
448                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
449                         goto fail2;
450                 at = entries = ((struct dx_node *) bh->b_data)->entries;
451                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
452                         ext3_warning(dir->i_sb, __FUNCTION__,
453                                      "dx entry: limit != node limit");
454                         brelse(bh);
455                         *err = ERR_BAD_DX_DIR;
456                         goto fail2;
457                 }
458                 frame++;
459                 frame->bh = NULL;
460         }
461 fail2:
462         while (frame >= frame_in) {
463                 brelse(frame->bh);
464                 frame--;
465         }
466 fail:
467         if (*err == ERR_BAD_DX_DIR)
468                 ext3_warning(dir->i_sb, __FUNCTION__,
469                              "Corrupt dir inode %ld, running e2fsck is "
470                              "recommended.", dir->i_ino);
471         return NULL;
472 }
473
474 static void dx_release (struct dx_frame *frames)
475 {
476         if (frames[0].bh == NULL)
477                 return;
478
479         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
480                 brelse(frames[1].bh);
481         brelse(frames[0].bh);
482 }
483
484 /*
485  * This function increments the frame pointer to search the next leaf
486  * block, and reads in the necessary intervening nodes if the search
487  * should be necessary.  Whether or not the search is necessary is
488  * controlled by the hash parameter.  If the hash value is even, then
489  * the search is only continued if the next block starts with that
490  * hash value.  This is used if we are searching for a specific file.
491  *
492  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
493  *
494  * This function returns 1 if the caller should continue to search,
495  * or 0 if it should not.  If there is an error reading one of the
496  * index blocks, it will a negative error code.
497  *
498  * If start_hash is non-null, it will be filled in with the starting
499  * hash of the next page.
500  */
501 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
502                                  struct dx_frame *frame,
503                                  struct dx_frame *frames,
504                                  __u32 *start_hash)
505 {
506         struct dx_frame *p;
507         struct buffer_head *bh;
508         int err, num_frames = 0;
509         __u32 bhash;
510
511         p = frame;
512         /*
513          * Find the next leaf page by incrementing the frame pointer.
514          * If we run out of entries in the interior node, loop around and
515          * increment pointer in the parent node.  When we break out of
516          * this loop, num_frames indicates the number of interior
517          * nodes need to be read.
518          */
519         while (1) {
520                 if (++(p->at) < p->entries + dx_get_count(p->entries))
521                         break;
522                 if (p == frames)
523                         return 0;
524                 num_frames++;
525                 p--;
526         }
527
528         /*
529          * If the hash is 1, then continue only if the next page has a
530          * continuation hash of any value.  This is used for readdir
531          * handling.  Otherwise, check to see if the hash matches the
532          * desired contiuation hash.  If it doesn't, return since
533          * there's no point to read in the successive index pages.
534          */
535         bhash = dx_get_hash(p->at);
536         if (start_hash)
537                 *start_hash = bhash;
538         if ((hash & 1) == 0) {
539                 if ((bhash & ~1) != hash)
540                         return 0;
541         }
542         /*
543          * If the hash is HASH_NB_ALWAYS, we always go to the next
544          * block so no check is necessary
545          */
546         while (num_frames--) {
547                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
548                                       0, &err)))
549                         return err; /* Failure */
550                 p++;
551                 brelse (p->bh);
552                 p->bh = bh;
553                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
554         }
555         return 1;
556 }
557
558
559 /*
560  * This function fills a red-black tree with information from a
561  * directory block.  It returns the number directory entries loaded
562  * into the tree.  If there is an error it is returned in err.
563  */
564 static int htree_dirblock_to_tree(struct file *dir_file,
565                                   struct inode *dir, int block,
566                                   struct dx_hash_info *hinfo,
567                                   __u32 start_hash, __u32 start_minor_hash)
568 {
569         struct buffer_head *bh;
570         struct ext3_dir_entry_2 *de, *top;
571         int err, count = 0;
572
573         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
574         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
575                 return err;
576
577         de = (struct ext3_dir_entry_2 *) bh->b_data;
578         top = (struct ext3_dir_entry_2 *) ((char *) de +
579                                            dir->i_sb->s_blocksize -
580                                            EXT3_DIR_REC_LEN(0));
581         for (; de < top; de = ext3_next_entry(de)) {
582                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
583                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
584                                                 +((char *)de - bh->b_data))) {
585                         /* On error, skip the f_pos to the next block. */
586                         dir_file->f_pos = (dir_file->f_pos |
587                                         (dir->i_sb->s_blocksize - 1)) + 1;
588                         brelse (bh);
589                         return count;
590                 }
591                 ext3fs_dirhash(de->name, de->name_len, hinfo);
592                 if ((hinfo->hash < start_hash) ||
593                     ((hinfo->hash == start_hash) &&
594                      (hinfo->minor_hash < start_minor_hash)))
595                         continue;
596                 if (de->inode == 0)
597                         continue;
598                 if ((err = ext3_htree_store_dirent(dir_file,
599                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
600                         brelse(bh);
601                         return err;
602                 }
603                 count++;
604         }
605         brelse(bh);
606         return count;
607 }
608
609
610 /*
611  * This function fills a red-black tree with information from a
612  * directory.  We start scanning the directory in hash order, starting
613  * at start_hash and start_minor_hash.
614  *
615  * This function returns the number of entries inserted into the tree,
616  * or a negative error code.
617  */
618 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
619                          __u32 start_minor_hash, __u32 *next_hash)
620 {
621         struct dx_hash_info hinfo;
622         struct ext3_dir_entry_2 *de;
623         struct dx_frame frames[2], *frame;
624         struct inode *dir;
625         int block, err;
626         int count = 0;
627         int ret;
628         __u32 hashval;
629
630         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
631                        start_minor_hash));
632         dir = dir_file->f_path.dentry->d_inode;
633         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
634                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
635                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
636                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
637                                                start_hash, start_minor_hash);
638                 *next_hash = ~0;
639                 return count;
640         }
641         hinfo.hash = start_hash;
642         hinfo.minor_hash = 0;
643         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
644         if (!frame)
645                 return err;
646
647         /* Add '.' and '..' from the htree header */
648         if (!start_hash && !start_minor_hash) {
649                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
650                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
651                         goto errout;
652                 count++;
653         }
654         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
655                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
656                 de = ext3_next_entry(de);
657                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
658                         goto errout;
659                 count++;
660         }
661
662         while (1) {
663                 block = dx_get_block(frame->at);
664                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
665                                              start_hash, start_minor_hash);
666                 if (ret < 0) {
667                         err = ret;
668                         goto errout;
669                 }
670                 count += ret;
671                 hashval = ~0;
672                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
673                                             frame, frames, &hashval);
674                 *next_hash = hashval;
675                 if (ret < 0) {
676                         err = ret;
677                         goto errout;
678                 }
679                 /*
680                  * Stop if:  (a) there are no more entries, or
681                  * (b) we have inserted at least one entry and the
682                  * next hash value is not a continuation
683                  */
684                 if ((ret == 0) ||
685                     (count && ((hashval & 1) == 0)))
686                         break;
687         }
688         dx_release(frames);
689         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
690                        count, *next_hash));
691         return count;
692 errout:
693         dx_release(frames);
694         return (err);
695 }
696
697
698 /*
699  * Directory block splitting, compacting
700  */
701
702 /*
703  * Create map of hash values, offsets, and sizes, stored at end of block.
704  * Returns number of entries mapped.
705  */
706 static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
707                         struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
708 {
709         int count = 0;
710         char *base = (char *) de;
711         struct dx_hash_info h = *hinfo;
712
713         while ((char *) de < base + size)
714         {
715                 if (de->name_len && de->inode) {
716                         ext3fs_dirhash(de->name, de->name_len, &h);
717                         map_tail--;
718                         map_tail->hash = h.hash;
719                         map_tail->offs = (u16) ((char *) de - base);
720                         map_tail->size = le16_to_cpu(de->rec_len);
721                         count++;
722                         cond_resched();
723                 }
724                 /* XXX: do we need to check rec_len == 0 case? -Chris */
725                 de = ext3_next_entry(de);
726         }
727         return count;
728 }
729
730 /* Sort map by hash value */
731 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
732 {
733         struct dx_map_entry *p, *q, *top = map + count - 1;
734         int more;
735         /* Combsort until bubble sort doesn't suck */
736         while (count > 2)
737         {
738                 count = count*10/13;
739                 if (count - 9 < 2) /* 9, 10 -> 11 */
740                         count = 11;
741                 for (p = top, q = p - count; q >= map; p--, q--)
742                         if (p->hash < q->hash)
743                                 swap(*p, *q);
744         }
745         /* Garden variety bubble sort */
746         do {
747                 more = 0;
748                 q = top;
749                 while (q-- > map)
750                 {
751                         if (q[1].hash >= q[0].hash)
752                                 continue;
753                         swap(*(q+1), *q);
754                         more = 1;
755                 }
756         } while(more);
757 }
758
759 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
760 {
761         struct dx_entry *entries = frame->entries;
762         struct dx_entry *old = frame->at, *new = old + 1;
763         int count = dx_get_count(entries);
764
765         assert(count < dx_get_limit(entries));
766         assert(old < entries + count);
767         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
768         dx_set_hash(new, hash);
769         dx_set_block(new, block);
770         dx_set_count(entries, count + 1);
771 }
772
773 static void ext3_update_dx_flag(struct inode *inode)
774 {
775         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
776                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
777                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
778 }
779
780 /*
781  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
782  *
783  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
784  * `de != NULL' is guaranteed by caller.
785  */
786 static inline int ext3_match (int len, const char * const name,
787                               struct ext3_dir_entry_2 * de)
788 {
789         if (len != de->name_len)
790                 return 0;
791         if (!de->inode)
792                 return 0;
793         return !memcmp(name, de->name, len);
794 }
795
796 /*
797  * Returns 0 if not found, -1 on failure, and 1 on success
798  */
799 static inline int search_dirblock(struct buffer_head * bh,
800                                   struct inode *dir,
801                                   struct dentry *dentry,
802                                   unsigned long offset,
803                                   struct ext3_dir_entry_2 ** res_dir)
804 {
805         struct ext3_dir_entry_2 * de;
806         char * dlimit;
807         int de_len;
808         const char *name = dentry->d_name.name;
809         int namelen = dentry->d_name.len;
810
811         de = (struct ext3_dir_entry_2 *) bh->b_data;
812         dlimit = bh->b_data + dir->i_sb->s_blocksize;
813         while ((char *) de < dlimit) {
814                 /* this code is executed quadratically often */
815                 /* do minimal checking `by hand' */
816
817                 if ((char *) de + namelen <= dlimit &&
818                     ext3_match (namelen, name, de)) {
819                         /* found a match - just to be sure, do a full check */
820                         if (!ext3_check_dir_entry("ext3_find_entry",
821                                                   dir, de, bh, offset))
822                                 return -1;
823                         *res_dir = de;
824                         return 1;
825                 }
826                 /* prevent looping on a bad block */
827                 de_len = ext3_rec_len_from_disk(de->rec_len);
828                 if (de_len <= 0)
829                         return -1;
830                 offset += de_len;
831                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
832         }
833         return 0;
834 }
835
836
837 /*
838  *      ext3_find_entry()
839  *
840  * finds an entry in the specified directory with the wanted name. It
841  * returns the cache buffer in which the entry was found, and the entry
842  * itself (as a parameter - res_dir). It does NOT read the inode of the
843  * entry - you'll have to do that yourself if you want to.
844  *
845  * The returned buffer_head has ->b_count elevated.  The caller is expected
846  * to brelse() it when appropriate.
847  */
848 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
849                                         struct ext3_dir_entry_2 ** res_dir)
850 {
851         struct super_block * sb;
852         struct buffer_head * bh_use[NAMEI_RA_SIZE];
853         struct buffer_head * bh, *ret = NULL;
854         unsigned long start, block, b;
855         int ra_max = 0;         /* Number of bh's in the readahead
856                                    buffer, bh_use[] */
857         int ra_ptr = 0;         /* Current index into readahead
858                                    buffer */
859         int num = 0;
860         int nblocks, i, err;
861         struct inode *dir = dentry->d_parent->d_inode;
862         int namelen;
863         const u8 *name;
864         unsigned blocksize;
865
866         *res_dir = NULL;
867         sb = dir->i_sb;
868         blocksize = sb->s_blocksize;
869         namelen = dentry->d_name.len;
870         name = dentry->d_name.name;
871         if (namelen > EXT3_NAME_LEN)
872                 return NULL;
873         if (is_dx(dir)) {
874                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
875                 /*
876                  * On success, or if the error was file not found,
877                  * return.  Otherwise, fall back to doing a search the
878                  * old fashioned way.
879                  */
880                 if (bh || (err != ERR_BAD_DX_DIR))
881                         return bh;
882                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
883         }
884         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
885         start = EXT3_I(dir)->i_dir_start_lookup;
886         if (start >= nblocks)
887                 start = 0;
888         block = start;
889 restart:
890         do {
891                 /*
892                  * We deal with the read-ahead logic here.
893                  */
894                 if (ra_ptr >= ra_max) {
895                         /* Refill the readahead buffer */
896                         ra_ptr = 0;
897                         b = block;
898                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
899                                 /*
900                                  * Terminate if we reach the end of the
901                                  * directory and must wrap, or if our
902                                  * search has finished at this block.
903                                  */
904                                 if (b >= nblocks || (num && block == start)) {
905                                         bh_use[ra_max] = NULL;
906                                         break;
907                                 }
908                                 num++;
909                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
910                                 bh_use[ra_max] = bh;
911                                 if (bh)
912                                         ll_rw_block(READ_META, 1, &bh);
913                         }
914                 }
915                 if ((bh = bh_use[ra_ptr++]) == NULL)
916                         goto next;
917                 wait_on_buffer(bh);
918                 if (!buffer_uptodate(bh)) {
919                         /* read error, skip block & hope for the best */
920                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
921                                    "offset %lu", dir->i_ino, block);
922                         brelse(bh);
923                         goto next;
924                 }
925                 i = search_dirblock(bh, dir, dentry,
926                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
927                 if (i == 1) {
928                         EXT3_I(dir)->i_dir_start_lookup = block;
929                         ret = bh;
930                         goto cleanup_and_exit;
931                 } else {
932                         brelse(bh);
933                         if (i < 0)
934                                 goto cleanup_and_exit;
935                 }
936         next:
937                 if (++block >= nblocks)
938                         block = 0;
939         } while (block != start);
940
941         /*
942          * If the directory has grown while we were searching, then
943          * search the last part of the directory before giving up.
944          */
945         block = nblocks;
946         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
947         if (block < nblocks) {
948                 start = 0;
949                 goto restart;
950         }
951
952 cleanup_and_exit:
953         /* Clean up the read-ahead blocks */
954         for (; ra_ptr < ra_max; ra_ptr++)
955                 brelse (bh_use[ra_ptr]);
956         return ret;
957 }
958
959 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
960                        struct ext3_dir_entry_2 **res_dir, int *err)
961 {
962         struct super_block * sb;
963         struct dx_hash_info     hinfo;
964         u32 hash;
965         struct dx_frame frames[2], *frame;
966         struct ext3_dir_entry_2 *de, *top;
967         struct buffer_head *bh;
968         unsigned long block;
969         int retval;
970         int namelen = dentry->d_name.len;
971         const u8 *name = dentry->d_name.name;
972         struct inode *dir = dentry->d_parent->d_inode;
973
974         sb = dir->i_sb;
975         /* NFS may look up ".." - look at dx_root directory block */
976         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
977                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
978                         return NULL;
979         } else {
980                 frame = frames;
981                 frame->bh = NULL;                       /* for dx_release() */
982                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
983                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
984         }
985         hash = hinfo.hash;
986         do {
987                 block = dx_get_block(frame->at);
988                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
989                         goto errout;
990                 de = (struct ext3_dir_entry_2 *) bh->b_data;
991                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
992                                        EXT3_DIR_REC_LEN(0));
993                 for (; de < top; de = ext3_next_entry(de))
994                 if (ext3_match (namelen, name, de)) {
995                         if (!ext3_check_dir_entry("ext3_find_entry",
996                                                   dir, de, bh,
997                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
998                                           +((char *)de - bh->b_data))) {
999                                 brelse (bh);
1000                                 *err = ERR_BAD_DX_DIR;
1001                                 goto errout;
1002                         }
1003                         *res_dir = de;
1004                         dx_release (frames);
1005                         return bh;
1006                 }
1007                 brelse (bh);
1008                 /* Check to see if we should continue to search */
1009                 retval = ext3_htree_next_block(dir, hash, frame,
1010                                                frames, NULL);
1011                 if (retval < 0) {
1012                         ext3_warning(sb, __FUNCTION__,
1013                              "error reading index page in directory #%lu",
1014                              dir->i_ino);
1015                         *err = retval;
1016                         goto errout;
1017                 }
1018         } while (retval == 1);
1019
1020         *err = -ENOENT;
1021 errout:
1022         dxtrace(printk("%s not found\n", name));
1023         dx_release (frames);
1024         return NULL;
1025 }
1026
1027 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1028 {
1029         struct inode * inode;
1030         struct ext3_dir_entry_2 * de;
1031         struct buffer_head * bh;
1032
1033         if (dentry->d_name.len > EXT3_NAME_LEN)
1034                 return ERR_PTR(-ENAMETOOLONG);
1035
1036         bh = ext3_find_entry(dentry, &de);
1037         inode = NULL;
1038         if (bh) {
1039                 unsigned long ino = le32_to_cpu(de->inode);
1040                 brelse (bh);
1041                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1042                         ext3_error(dir->i_sb, "ext3_lookup",
1043                                    "bad inode number: %lu", ino);
1044                         inode = NULL;
1045                 } else
1046                         inode = iget(dir->i_sb, ino);
1047
1048                 if (!inode)
1049                         return ERR_PTR(-EACCES);
1050
1051                 if (is_bad_inode(inode)) {
1052                         iput(inode);
1053                         return ERR_PTR(-ENOENT);
1054                 }
1055         }
1056         return d_splice_alias(inode, dentry);
1057 }
1058
1059
1060 struct dentry *ext3_get_parent(struct dentry *child)
1061 {
1062         unsigned long ino;
1063         struct dentry *parent;
1064         struct inode *inode;
1065         struct dentry dotdot;
1066         struct ext3_dir_entry_2 * de;
1067         struct buffer_head *bh;
1068
1069         dotdot.d_name.name = "..";
1070         dotdot.d_name.len = 2;
1071         dotdot.d_parent = child; /* confusing, isn't it! */
1072
1073         bh = ext3_find_entry(&dotdot, &de);
1074         inode = NULL;
1075         if (!bh)
1076                 return ERR_PTR(-ENOENT);
1077         ino = le32_to_cpu(de->inode);
1078         brelse(bh);
1079
1080         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1081                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1082                            "bad inode number: %lu", ino);
1083                 inode = NULL;
1084         } else
1085                 inode = iget(child->d_inode->i_sb, ino);
1086
1087         if (!inode)
1088                 return ERR_PTR(-EACCES);
1089
1090         if (is_bad_inode(inode)) {
1091                 iput(inode);
1092                 return ERR_PTR(-ENOENT);
1093         }
1094
1095         parent = d_alloc_anon(inode);
1096         if (!parent) {
1097                 iput(inode);
1098                 parent = ERR_PTR(-ENOMEM);
1099         }
1100         return parent;
1101 }
1102
1103 #define S_SHIFT 12
1104 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1105         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1106         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1107         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1108         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1109         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1110         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1111         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1112 };
1113
1114 static inline void ext3_set_de_type(struct super_block *sb,
1115                                 struct ext3_dir_entry_2 *de,
1116                                 umode_t mode) {
1117         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1118                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1119 }
1120
1121 /*
1122  * Move count entries from end of map between two memory locations.
1123  * Returns pointer to last entry moved.
1124  */
1125 static struct ext3_dir_entry_2 *
1126 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1127 {
1128         unsigned rec_len = 0;
1129
1130         while (count--) {
1131                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1132                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1133                 memcpy (to, de, rec_len);
1134                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1135                                 ext3_rec_len_to_disk(rec_len);
1136                 de->inode = 0;
1137                 map++;
1138                 to += rec_len;
1139         }
1140         return (struct ext3_dir_entry_2 *) (to - rec_len);
1141 }
1142
1143 /*
1144  * Compact each dir entry in the range to the minimal rec_len.
1145  * Returns pointer to last entry in range.
1146  */
1147 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1148 {
1149         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1150         unsigned rec_len = 0;
1151
1152         prev = to = de;
1153         while ((char*)de < base + size) {
1154                 next = ext3_next_entry(de);
1155                 if (de->inode && de->name_len) {
1156                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1157                         if (de > to)
1158                                 memmove(to, de, rec_len);
1159                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1160                         prev = to;
1161                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1162                 }
1163                 de = next;
1164         }
1165         return prev;
1166 }
1167
1168 /*
1169  * Split a full leaf block to make room for a new dir entry.
1170  * Allocate a new block, and move entries so that they are approx. equally full.
1171  * Returns pointer to de in block into which the new entry will be inserted.
1172  */
1173 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1174                         struct buffer_head **bh,struct dx_frame *frame,
1175                         struct dx_hash_info *hinfo, int *error)
1176 {
1177         unsigned blocksize = dir->i_sb->s_blocksize;
1178         unsigned count, continued;
1179         struct buffer_head *bh2;
1180         u32 newblock;
1181         u32 hash2;
1182         struct dx_map_entry *map;
1183         char *data1 = (*bh)->b_data, *data2;
1184         unsigned split, move, size, i;
1185         struct ext3_dir_entry_2 *de = NULL, *de2;
1186         int     err = 0;
1187
1188         bh2 = ext3_append (handle, dir, &newblock, &err);
1189         if (!(bh2)) {
1190                 brelse(*bh);
1191                 *bh = NULL;
1192                 goto errout;
1193         }
1194
1195         BUFFER_TRACE(*bh, "get_write_access");
1196         err = ext3_journal_get_write_access(handle, *bh);
1197         if (err)
1198                 goto journal_error;
1199
1200         BUFFER_TRACE(frame->bh, "get_write_access");
1201         err = ext3_journal_get_write_access(handle, frame->bh);
1202         if (err)
1203                 goto journal_error;
1204
1205         data2 = bh2->b_data;
1206
1207         /* create map in the end of data2 block */
1208         map = (struct dx_map_entry *) (data2 + blocksize);
1209         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1210                              blocksize, hinfo, map);
1211         map -= count;
1212         dx_sort_map (map, count);
1213         /* Split the existing block in the middle, size-wise */
1214         size = 0;
1215         move = 0;
1216         for (i = count-1; i >= 0; i--) {
1217                 /* is more than half of this entry in 2nd half of the block? */
1218                 if (size + map[i].size/2 > blocksize/2)
1219                         break;
1220                 size += map[i].size;
1221                 move++;
1222         }
1223         /* map index at which we will split */
1224         split = count - move;
1225         hash2 = map[split].hash;
1226         continued = hash2 == map[split - 1].hash;
1227         dxtrace(printk("Split block %i at %x, %i/%i\n",
1228                 dx_get_block(frame->at), hash2, split, count-split));
1229
1230         /* Fancy dance to stay within two buffers */
1231         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1232         de = dx_pack_dirents(data1,blocksize);
1233         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1234         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1235         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1236         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1237
1238         /* Which block gets the new entry? */
1239         if (hinfo->hash >= hash2)
1240         {
1241                 swap(*bh, bh2);
1242                 de = de2;
1243         }
1244         dx_insert_block (frame, hash2 + continued, newblock);
1245         err = ext3_journal_dirty_metadata (handle, bh2);
1246         if (err)
1247                 goto journal_error;
1248         err = ext3_journal_dirty_metadata (handle, frame->bh);
1249         if (err)
1250                 goto journal_error;
1251         brelse (bh2);
1252         dxtrace(dx_show_index ("frame", frame->entries));
1253         return de;
1254
1255 journal_error:
1256         brelse(*bh);
1257         brelse(bh2);
1258         *bh = NULL;
1259         ext3_std_error(dir->i_sb, err);
1260 errout:
1261         *error = err;
1262         return NULL;
1263 }
1264
1265
1266 /*
1267  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1268  * it points to a directory entry which is guaranteed to be large
1269  * enough for new directory entry.  If de is NULL, then
1270  * add_dirent_to_buf will attempt search the directory block for
1271  * space.  It will return -ENOSPC if no space is available, and -EIO
1272  * and -EEXIST if directory entry already exists.
1273  *
1274  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1275  * all other cases bh is released.
1276  */
1277 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1278                              struct inode *inode, struct ext3_dir_entry_2 *de,
1279                              struct buffer_head * bh)
1280 {
1281         struct inode    *dir = dentry->d_parent->d_inode;
1282         const char      *name = dentry->d_name.name;
1283         int             namelen = dentry->d_name.len;
1284         unsigned long   offset = 0;
1285         unsigned short  reclen;
1286         int             nlen, rlen, err;
1287         char            *top;
1288
1289         reclen = EXT3_DIR_REC_LEN(namelen);
1290         if (!de) {
1291                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1292                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1293                 while ((char *) de <= top) {
1294                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1295                                                   bh, offset)) {
1296                                 brelse (bh);
1297                                 return -EIO;
1298                         }
1299                         if (ext3_match (namelen, name, de)) {
1300                                 brelse (bh);
1301                                 return -EEXIST;
1302                         }
1303                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1304                         rlen = ext3_rec_len_from_disk(de->rec_len);
1305                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1306                                 break;
1307                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1308                         offset += rlen;
1309                 }
1310                 if ((char *) de > top)
1311                         return -ENOSPC;
1312         }
1313         BUFFER_TRACE(bh, "get_write_access");
1314         err = ext3_journal_get_write_access(handle, bh);
1315         if (err) {
1316                 ext3_std_error(dir->i_sb, err);
1317                 brelse(bh);
1318                 return err;
1319         }
1320
1321         /* By now the buffer is marked for journaling */
1322         nlen = EXT3_DIR_REC_LEN(de->name_len);
1323         rlen = ext3_rec_len_from_disk(de->rec_len);
1324         if (de->inode) {
1325                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1326                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1327                 de->rec_len = ext3_rec_len_to_disk(nlen);
1328                 de = de1;
1329         }
1330         de->file_type = EXT3_FT_UNKNOWN;
1331         if (inode) {
1332                 de->inode = cpu_to_le32(inode->i_ino);
1333                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1334         } else
1335                 de->inode = 0;
1336         de->name_len = namelen;
1337         memcpy (de->name, name, namelen);
1338         /*
1339          * XXX shouldn't update any times until successful
1340          * completion of syscall, but too many callers depend
1341          * on this.
1342          *
1343          * XXX similarly, too many callers depend on
1344          * ext3_new_inode() setting the times, but error
1345          * recovery deletes the inode, so the worst that can
1346          * happen is that the times are slightly out of date
1347          * and/or different from the directory change time.
1348          */
1349         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1350         ext3_update_dx_flag(dir);
1351         dir->i_version++;
1352         ext3_mark_inode_dirty(handle, dir);
1353         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1354         err = ext3_journal_dirty_metadata(handle, bh);
1355         if (err)
1356                 ext3_std_error(dir->i_sb, err);
1357         brelse(bh);
1358         return 0;
1359 }
1360
1361 /*
1362  * This converts a one block unindexed directory to a 3 block indexed
1363  * directory, and adds the dentry to the indexed directory.
1364  */
1365 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1366                             struct inode *inode, struct buffer_head *bh)
1367 {
1368         struct inode    *dir = dentry->d_parent->d_inode;
1369         const char      *name = dentry->d_name.name;
1370         int             namelen = dentry->d_name.len;
1371         struct buffer_head *bh2;
1372         struct dx_root  *root;
1373         struct dx_frame frames[2], *frame;
1374         struct dx_entry *entries;
1375         struct ext3_dir_entry_2 *de, *de2;
1376         char            *data1, *top;
1377         unsigned        len;
1378         int             retval;
1379         unsigned        blocksize;
1380         struct dx_hash_info hinfo;
1381         u32             block;
1382         struct fake_dirent *fde;
1383
1384         blocksize =  dir->i_sb->s_blocksize;
1385         dxtrace(printk("Creating index\n"));
1386         retval = ext3_journal_get_write_access(handle, bh);
1387         if (retval) {
1388                 ext3_std_error(dir->i_sb, retval);
1389                 brelse(bh);
1390                 return retval;
1391         }
1392         root = (struct dx_root *) bh->b_data;
1393
1394         bh2 = ext3_append (handle, dir, &block, &retval);
1395         if (!(bh2)) {
1396                 brelse(bh);
1397                 return retval;
1398         }
1399         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1400         data1 = bh2->b_data;
1401
1402         /* The 0th block becomes the root, move the dirents out */
1403         fde = &root->dotdot;
1404         de = (struct ext3_dir_entry_2 *)((char *)fde +
1405                         ext3_rec_len_from_disk(fde->rec_len));
1406         len = ((char *) root) + blocksize - (char *) de;
1407         memcpy (data1, de, len);
1408         de = (struct ext3_dir_entry_2 *) data1;
1409         top = data1 + len;
1410         while ((char *)(de2 = ext3_next_entry(de)) < top)
1411                 de = de2;
1412         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1413         /* Initialize the root; the dot dirents already exist */
1414         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1415         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1416         memset (&root->info, 0, sizeof(root->info));
1417         root->info.info_length = sizeof(root->info);
1418         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1419         entries = root->entries;
1420         dx_set_block (entries, 1);
1421         dx_set_count (entries, 1);
1422         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1423
1424         /* Initialize as for dx_probe */
1425         hinfo.hash_version = root->info.hash_version;
1426         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1427         ext3fs_dirhash(name, namelen, &hinfo);
1428         frame = frames;
1429         frame->entries = entries;
1430         frame->at = entries;
1431         frame->bh = bh;
1432         bh = bh2;
1433         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1434         dx_release (frames);
1435         if (!(de))
1436                 return retval;
1437
1438         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1439 }
1440
1441 /*
1442  *      ext3_add_entry()
1443  *
1444  * adds a file entry to the specified directory, using the same
1445  * semantics as ext3_find_entry(). It returns NULL if it failed.
1446  *
1447  * NOTE!! The inode part of 'de' is left at 0 - which means you
1448  * may not sleep between calling this and putting something into
1449  * the entry, as someone else might have used it while you slept.
1450  */
1451 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1452         struct inode *inode)
1453 {
1454         struct inode *dir = dentry->d_parent->d_inode;
1455         unsigned long offset;
1456         struct buffer_head * bh;
1457         struct ext3_dir_entry_2 *de;
1458         struct super_block * sb;
1459         int     retval;
1460         int     dx_fallback=0;
1461         unsigned blocksize;
1462         u32 block, blocks;
1463
1464         sb = dir->i_sb;
1465         blocksize = sb->s_blocksize;
1466         if (!dentry->d_name.len)
1467                 return -EINVAL;
1468         if (is_dx(dir)) {
1469                 retval = ext3_dx_add_entry(handle, dentry, inode);
1470                 if (!retval || (retval != ERR_BAD_DX_DIR))
1471                         return retval;
1472                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1473                 dx_fallback++;
1474                 ext3_mark_inode_dirty(handle, dir);
1475         }
1476         blocks = dir->i_size >> sb->s_blocksize_bits;
1477         for (block = 0, offset = 0; block < blocks; block++) {
1478                 bh = ext3_bread(handle, dir, block, 0, &retval);
1479                 if(!bh)
1480                         return retval;
1481                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1482                 if (retval != -ENOSPC)
1483                         return retval;
1484
1485                 if (blocks == 1 && !dx_fallback &&
1486                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1487                         return make_indexed_dir(handle, dentry, inode, bh);
1488                 brelse(bh);
1489         }
1490         bh = ext3_append(handle, dir, &block, &retval);
1491         if (!bh)
1492                 return retval;
1493         de = (struct ext3_dir_entry_2 *) bh->b_data;
1494         de->inode = 0;
1495         de->rec_len = ext3_rec_len_to_disk(blocksize);
1496         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1497 }
1498
1499 /*
1500  * Returns 0 for success, or a negative error value
1501  */
1502 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1503                              struct inode *inode)
1504 {
1505         struct dx_frame frames[2], *frame;
1506         struct dx_entry *entries, *at;
1507         struct dx_hash_info hinfo;
1508         struct buffer_head * bh;
1509         struct inode *dir = dentry->d_parent->d_inode;
1510         struct super_block * sb = dir->i_sb;
1511         struct ext3_dir_entry_2 *de;
1512         int err;
1513
1514         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1515         if (!frame)
1516                 return err;
1517         entries = frame->entries;
1518         at = frame->at;
1519
1520         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1521                 goto cleanup;
1522
1523         BUFFER_TRACE(bh, "get_write_access");
1524         err = ext3_journal_get_write_access(handle, bh);
1525         if (err)
1526                 goto journal_error;
1527
1528         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1529         if (err != -ENOSPC) {
1530                 bh = NULL;
1531                 goto cleanup;
1532         }
1533
1534         /* Block full, should compress but for now just split */
1535         dxtrace(printk("using %u of %u node entries\n",
1536                        dx_get_count(entries), dx_get_limit(entries)));
1537         /* Need to split index? */
1538         if (dx_get_count(entries) == dx_get_limit(entries)) {
1539                 u32 newblock;
1540                 unsigned icount = dx_get_count(entries);
1541                 int levels = frame - frames;
1542                 struct dx_entry *entries2;
1543                 struct dx_node *node2;
1544                 struct buffer_head *bh2;
1545
1546                 if (levels && (dx_get_count(frames->entries) ==
1547                                dx_get_limit(frames->entries))) {
1548                         ext3_warning(sb, __FUNCTION__,
1549                                      "Directory index full!");
1550                         err = -ENOSPC;
1551                         goto cleanup;
1552                 }
1553                 bh2 = ext3_append (handle, dir, &newblock, &err);
1554                 if (!(bh2))
1555                         goto cleanup;
1556                 node2 = (struct dx_node *)(bh2->b_data);
1557                 entries2 = node2->entries;
1558                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1559                 node2->fake.inode = 0;
1560                 BUFFER_TRACE(frame->bh, "get_write_access");
1561                 err = ext3_journal_get_write_access(handle, frame->bh);
1562                 if (err)
1563                         goto journal_error;
1564                 if (levels) {
1565                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1566                         unsigned hash2 = dx_get_hash(entries + icount1);
1567                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1568
1569                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1570                         err = ext3_journal_get_write_access(handle,
1571                                                              frames[0].bh);
1572                         if (err)
1573                                 goto journal_error;
1574
1575                         memcpy ((char *) entries2, (char *) (entries + icount1),
1576                                 icount2 * sizeof(struct dx_entry));
1577                         dx_set_count (entries, icount1);
1578                         dx_set_count (entries2, icount2);
1579                         dx_set_limit (entries2, dx_node_limit(dir));
1580
1581                         /* Which index block gets the new entry? */
1582                         if (at - entries >= icount1) {
1583                                 frame->at = at = at - entries - icount1 + entries2;
1584                                 frame->entries = entries = entries2;
1585                                 swap(frame->bh, bh2);
1586                         }
1587                         dx_insert_block (frames + 0, hash2, newblock);
1588                         dxtrace(dx_show_index ("node", frames[1].entries));
1589                         dxtrace(dx_show_index ("node",
1590                                ((struct dx_node *) bh2->b_data)->entries));
1591                         err = ext3_journal_dirty_metadata(handle, bh2);
1592                         if (err)
1593                                 goto journal_error;
1594                         brelse (bh2);
1595                 } else {
1596                         dxtrace(printk("Creating second level index...\n"));
1597                         memcpy((char *) entries2, (char *) entries,
1598                                icount * sizeof(struct dx_entry));
1599                         dx_set_limit(entries2, dx_node_limit(dir));
1600
1601                         /* Set up root */
1602                         dx_set_count(entries, 1);
1603                         dx_set_block(entries + 0, newblock);
1604                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1605
1606                         /* Add new access path frame */
1607                         frame = frames + 1;
1608                         frame->at = at = at - entries + entries2;
1609                         frame->entries = entries = entries2;
1610                         frame->bh = bh2;
1611                         err = ext3_journal_get_write_access(handle,
1612                                                              frame->bh);
1613                         if (err)
1614                                 goto journal_error;
1615                 }
1616                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1617         }
1618         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1619         if (!de)
1620                 goto cleanup;
1621         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1622         bh = NULL;
1623         goto cleanup;
1624
1625 journal_error:
1626         ext3_std_error(dir->i_sb, err);
1627 cleanup:
1628         if (bh)
1629                 brelse(bh);
1630         dx_release(frames);
1631         return err;
1632 }
1633
1634 /*
1635  * ext3_delete_entry deletes a directory entry by merging it with the
1636  * previous entry
1637  */
1638 static int ext3_delete_entry (handle_t *handle,
1639                               struct inode * dir,
1640                               struct ext3_dir_entry_2 * de_del,
1641                               struct buffer_head * bh)
1642 {
1643         struct ext3_dir_entry_2 * de, * pde;
1644         int i;
1645
1646         i = 0;
1647         pde = NULL;
1648         de = (struct ext3_dir_entry_2 *) bh->b_data;
1649         while (i < bh->b_size) {
1650                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1651                         return -EIO;
1652                 if (de == de_del)  {
1653                         BUFFER_TRACE(bh, "get_write_access");
1654                         ext3_journal_get_write_access(handle, bh);
1655                         if (pde)
1656                                 pde->rec_len = ext3_rec_len_to_disk(
1657                                         ext3_rec_len_from_disk(pde->rec_len) +
1658                                         ext3_rec_len_from_disk(de->rec_len));
1659                         else
1660                                 de->inode = 0;
1661                         dir->i_version++;
1662                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1663                         ext3_journal_dirty_metadata(handle, bh);
1664                         return 0;
1665                 }
1666                 i += ext3_rec_len_from_disk(de->rec_len);
1667                 pde = de;
1668                 de = ext3_next_entry(de);
1669         }
1670         return -ENOENT;
1671 }
1672
1673 static int ext3_add_nondir(handle_t *handle,
1674                 struct dentry *dentry, struct inode *inode)
1675 {
1676         int err = ext3_add_entry(handle, dentry, inode);
1677         if (!err) {
1678                 ext3_mark_inode_dirty(handle, inode);
1679                 d_instantiate(dentry, inode);
1680                 return 0;
1681         }
1682         drop_nlink(inode);
1683         iput(inode);
1684         return err;
1685 }
1686
1687 /*
1688  * By the time this is called, we already have created
1689  * the directory cache entry for the new file, but it
1690  * is so far negative - it has no inode.
1691  *
1692  * If the create succeeds, we fill in the inode information
1693  * with d_instantiate().
1694  */
1695 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1696                 struct nameidata *nd)
1697 {
1698         handle_t *handle;
1699         struct inode * inode;
1700         int err, retries = 0;
1701
1702 retry:
1703         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1704                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1705                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1706         if (IS_ERR(handle))
1707                 return PTR_ERR(handle);
1708
1709         if (IS_DIRSYNC(dir))
1710                 handle->h_sync = 1;
1711
1712         inode = ext3_new_inode (handle, dir, mode);
1713         err = PTR_ERR(inode);
1714         if (!IS_ERR(inode)) {
1715                 inode->i_op = &ext3_file_inode_operations;
1716                 inode->i_fop = &ext3_file_operations;
1717                 ext3_set_aops(inode);
1718                 err = ext3_add_nondir(handle, dentry, inode);
1719         }
1720         ext3_journal_stop(handle);
1721         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1722                 goto retry;
1723         return err;
1724 }
1725
1726 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1727                         int mode, dev_t rdev)
1728 {
1729         handle_t *handle;
1730         struct inode *inode;
1731         int err, retries = 0;
1732
1733         if (!new_valid_dev(rdev))
1734                 return -EINVAL;
1735
1736 retry:
1737         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1738                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1739                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1740         if (IS_ERR(handle))
1741                 return PTR_ERR(handle);
1742
1743         if (IS_DIRSYNC(dir))
1744                 handle->h_sync = 1;
1745
1746         inode = ext3_new_inode (handle, dir, mode);
1747         err = PTR_ERR(inode);
1748         if (!IS_ERR(inode)) {
1749                 init_special_inode(inode, inode->i_mode, rdev);
1750 #ifdef CONFIG_EXT3_FS_XATTR
1751                 inode->i_op = &ext3_special_inode_operations;
1752 #endif
1753                 err = ext3_add_nondir(handle, dentry, inode);
1754         }
1755         ext3_journal_stop(handle);
1756         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1757                 goto retry;
1758         return err;
1759 }
1760
1761 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1762 {
1763         handle_t *handle;
1764         struct inode * inode;
1765         struct buffer_head * dir_block;
1766         struct ext3_dir_entry_2 * de;
1767         int err, retries = 0;
1768
1769         if (dir->i_nlink >= EXT3_LINK_MAX)
1770                 return -EMLINK;
1771
1772 retry:
1773         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1774                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1775                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1776         if (IS_ERR(handle))
1777                 return PTR_ERR(handle);
1778
1779         if (IS_DIRSYNC(dir))
1780                 handle->h_sync = 1;
1781
1782         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1783         err = PTR_ERR(inode);
1784         if (IS_ERR(inode))
1785                 goto out_stop;
1786
1787         inode->i_op = &ext3_dir_inode_operations;
1788         inode->i_fop = &ext3_dir_operations;
1789         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1790         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1791         if (!dir_block) {
1792                 drop_nlink(inode); /* is this nlink == 0? */
1793                 ext3_mark_inode_dirty(handle, inode);
1794                 iput (inode);
1795                 goto out_stop;
1796         }
1797         BUFFER_TRACE(dir_block, "get_write_access");
1798         ext3_journal_get_write_access(handle, dir_block);
1799         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1800         de->inode = cpu_to_le32(inode->i_ino);
1801         de->name_len = 1;
1802         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1803         strcpy (de->name, ".");
1804         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1805         de = ext3_next_entry(de);
1806         de->inode = cpu_to_le32(dir->i_ino);
1807         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1808                                         EXT3_DIR_REC_LEN(1));
1809         de->name_len = 2;
1810         strcpy (de->name, "..");
1811         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1812         inode->i_nlink = 2;
1813         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1814         ext3_journal_dirty_metadata(handle, dir_block);
1815         brelse (dir_block);
1816         ext3_mark_inode_dirty(handle, inode);
1817         err = ext3_add_entry (handle, dentry, inode);
1818         if (err) {
1819                 inode->i_nlink = 0;
1820                 ext3_mark_inode_dirty(handle, inode);
1821                 iput (inode);
1822                 goto out_stop;
1823         }
1824         inc_nlink(dir);
1825         ext3_update_dx_flag(dir);
1826         ext3_mark_inode_dirty(handle, dir);
1827         d_instantiate(dentry, inode);
1828 out_stop:
1829         ext3_journal_stop(handle);
1830         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1831                 goto retry;
1832         return err;
1833 }
1834
1835 /*
1836  * routine to check that the specified directory is empty (for rmdir)
1837  */
1838 static int empty_dir (struct inode * inode)
1839 {
1840         unsigned long offset;
1841         struct buffer_head * bh;
1842         struct ext3_dir_entry_2 * de, * de1;
1843         struct super_block * sb;
1844         int err = 0;
1845
1846         sb = inode->i_sb;
1847         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1848             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1849                 if (err)
1850                         ext3_error(inode->i_sb, __FUNCTION__,
1851                                    "error %d reading directory #%lu offset 0",
1852                                    err, inode->i_ino);
1853                 else
1854                         ext3_warning(inode->i_sb, __FUNCTION__,
1855                                      "bad directory (dir #%lu) - no data block",
1856                                      inode->i_ino);
1857                 return 1;
1858         }
1859         de = (struct ext3_dir_entry_2 *) bh->b_data;
1860         de1 = ext3_next_entry(de);
1861         if (le32_to_cpu(de->inode) != inode->i_ino ||
1862                         !le32_to_cpu(de1->inode) ||
1863                         strcmp (".", de->name) ||
1864                         strcmp ("..", de1->name)) {
1865                 ext3_warning (inode->i_sb, "empty_dir",
1866                               "bad directory (dir #%lu) - no `.' or `..'",
1867                               inode->i_ino);
1868                 brelse (bh);
1869                 return 1;
1870         }
1871         offset = ext3_rec_len_from_disk(de->rec_len) +
1872                         ext3_rec_len_from_disk(de1->rec_len);
1873         de = ext3_next_entry(de1);
1874         while (offset < inode->i_size ) {
1875                 if (!bh ||
1876                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1877                         err = 0;
1878                         brelse (bh);
1879                         bh = ext3_bread (NULL, inode,
1880                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1881                         if (!bh) {
1882                                 if (err)
1883                                         ext3_error(sb, __FUNCTION__,
1884                                                    "error %d reading directory"
1885                                                    " #%lu offset %lu",
1886                                                    err, inode->i_ino, offset);
1887                                 offset += sb->s_blocksize;
1888                                 continue;
1889                         }
1890                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1891                 }
1892                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1893                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1894                                                          sb->s_blocksize);
1895                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1896                         continue;
1897                 }
1898                 if (le32_to_cpu(de->inode)) {
1899                         brelse (bh);
1900                         return 0;
1901                 }
1902                 offset += ext3_rec_len_from_disk(de->rec_len);
1903                 de = ext3_next_entry(de);
1904         }
1905         brelse (bh);
1906         return 1;
1907 }
1908
1909 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1910  * such inodes, starting at the superblock, in case we crash before the
1911  * file is closed/deleted, or in case the inode truncate spans multiple
1912  * transactions and the last transaction is not recovered after a crash.
1913  *
1914  * At filesystem recovery time, we walk this list deleting unlinked
1915  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1916  */
1917 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1918 {
1919         struct super_block *sb = inode->i_sb;
1920         struct ext3_iloc iloc;
1921         int err = 0, rc;
1922
1923         lock_super(sb);
1924         if (!list_empty(&EXT3_I(inode)->i_orphan))
1925                 goto out_unlock;
1926
1927         /* Orphan handling is only valid for files with data blocks
1928          * being truncated, or files being unlinked. */
1929
1930         /* @@@ FIXME: Observation from aviro:
1931          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1932          * here (on lock_super()), so race with ext3_link() which might bump
1933          * ->i_nlink. For, say it, character device. Not a regular file,
1934          * not a directory, not a symlink and ->i_nlink > 0.
1935          */
1936         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1937                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1938
1939         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1940         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1941         if (err)
1942                 goto out_unlock;
1943
1944         err = ext3_reserve_inode_write(handle, inode, &iloc);
1945         if (err)
1946                 goto out_unlock;
1947
1948         /* Insert this inode at the head of the on-disk orphan list... */
1949         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1950         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1951         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1952         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1953         if (!err)
1954                 err = rc;
1955
1956         /* Only add to the head of the in-memory list if all the
1957          * previous operations succeeded.  If the orphan_add is going to
1958          * fail (possibly taking the journal offline), we can't risk
1959          * leaving the inode on the orphan list: stray orphan-list
1960          * entries can cause panics at unmount time.
1961          *
1962          * This is safe: on error we're going to ignore the orphan list
1963          * anyway on the next recovery. */
1964         if (!err)
1965                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1966
1967         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1968         jbd_debug(4, "orphan inode %lu will point to %d\n",
1969                         inode->i_ino, NEXT_ORPHAN(inode));
1970 out_unlock:
1971         unlock_super(sb);
1972         ext3_std_error(inode->i_sb, err);
1973         return err;
1974 }
1975
1976 /*
1977  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1978  * of such inodes stored on disk, because it is finally being cleaned up.
1979  */
1980 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1981 {
1982         struct list_head *prev;
1983         struct ext3_inode_info *ei = EXT3_I(inode);
1984         struct ext3_sb_info *sbi;
1985         unsigned long ino_next;
1986         struct ext3_iloc iloc;
1987         int err = 0;
1988
1989         lock_super(inode->i_sb);
1990         if (list_empty(&ei->i_orphan)) {
1991                 unlock_super(inode->i_sb);
1992                 return 0;
1993         }
1994
1995         ino_next = NEXT_ORPHAN(inode);
1996         prev = ei->i_orphan.prev;
1997         sbi = EXT3_SB(inode->i_sb);
1998
1999         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2000
2001         list_del_init(&ei->i_orphan);
2002
2003         /* If we're on an error path, we may not have a valid
2004          * transaction handle with which to update the orphan list on
2005          * disk, but we still need to remove the inode from the linked
2006          * list in memory. */
2007         if (!handle)
2008                 goto out;
2009
2010         err = ext3_reserve_inode_write(handle, inode, &iloc);
2011         if (err)
2012                 goto out_err;
2013
2014         if (prev == &sbi->s_orphan) {
2015                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2016                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2017                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2018                 if (err)
2019                         goto out_brelse;
2020                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2021                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2022         } else {
2023                 struct ext3_iloc iloc2;
2024                 struct inode *i_prev =
2025                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2026
2027                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2028                           i_prev->i_ino, ino_next);
2029                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2030                 if (err)
2031                         goto out_brelse;
2032                 NEXT_ORPHAN(i_prev) = ino_next;
2033                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2034         }
2035         if (err)
2036                 goto out_brelse;
2037         NEXT_ORPHAN(inode) = 0;
2038         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2039
2040 out_err:
2041         ext3_std_error(inode->i_sb, err);
2042 out:
2043         unlock_super(inode->i_sb);
2044         return err;
2045
2046 out_brelse:
2047         brelse(iloc.bh);
2048         goto out_err;
2049 }
2050
2051 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2052 {
2053         int retval;
2054         struct inode * inode;
2055         struct buffer_head * bh;
2056         struct ext3_dir_entry_2 * de;
2057         handle_t *handle;
2058
2059         /* Initialize quotas before so that eventual writes go in
2060          * separate transaction */
2061         DQUOT_INIT(dentry->d_inode);
2062         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2063         if (IS_ERR(handle))
2064                 return PTR_ERR(handle);
2065
2066         retval = -ENOENT;
2067         bh = ext3_find_entry (dentry, &de);
2068         if (!bh)
2069                 goto end_rmdir;
2070
2071         if (IS_DIRSYNC(dir))
2072                 handle->h_sync = 1;
2073
2074         inode = dentry->d_inode;
2075
2076         retval = -EIO;
2077         if (le32_to_cpu(de->inode) != inode->i_ino)
2078                 goto end_rmdir;
2079
2080         retval = -ENOTEMPTY;
2081         if (!empty_dir (inode))
2082                 goto end_rmdir;
2083
2084         retval = ext3_delete_entry(handle, dir, de, bh);
2085         if (retval)
2086                 goto end_rmdir;
2087         if (inode->i_nlink != 2)
2088                 ext3_warning (inode->i_sb, "ext3_rmdir",
2089                               "empty directory has nlink!=2 (%d)",
2090                               inode->i_nlink);
2091         inode->i_version++;
2092         clear_nlink(inode);
2093         /* There's no need to set i_disksize: the fact that i_nlink is
2094          * zero will ensure that the right thing happens during any
2095          * recovery. */
2096         inode->i_size = 0;
2097         ext3_orphan_add(handle, inode);
2098         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2099         ext3_mark_inode_dirty(handle, inode);
2100         drop_nlink(dir);
2101         ext3_update_dx_flag(dir);
2102         ext3_mark_inode_dirty(handle, dir);
2103
2104 end_rmdir:
2105         ext3_journal_stop(handle);
2106         brelse (bh);
2107         return retval;
2108 }
2109
2110 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2111 {
2112         int retval;
2113         struct inode * inode;
2114         struct buffer_head * bh;
2115         struct ext3_dir_entry_2 * de;
2116         handle_t *handle;
2117
2118         /* Initialize quotas before so that eventual writes go
2119          * in separate transaction */
2120         DQUOT_INIT(dentry->d_inode);
2121         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2122         if (IS_ERR(handle))
2123                 return PTR_ERR(handle);
2124
2125         if (IS_DIRSYNC(dir))
2126                 handle->h_sync = 1;
2127
2128         retval = -ENOENT;
2129         bh = ext3_find_entry (dentry, &de);
2130         if (!bh)
2131                 goto end_unlink;
2132
2133         inode = dentry->d_inode;
2134
2135         retval = -EIO;
2136         if (le32_to_cpu(de->inode) != inode->i_ino)
2137                 goto end_unlink;
2138
2139         if (!inode->i_nlink) {
2140                 ext3_warning (inode->i_sb, "ext3_unlink",
2141                               "Deleting nonexistent file (%lu), %d",
2142                               inode->i_ino, inode->i_nlink);
2143                 inode->i_nlink = 1;
2144         }
2145         retval = ext3_delete_entry(handle, dir, de, bh);
2146         if (retval)
2147                 goto end_unlink;
2148         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2149         ext3_update_dx_flag(dir);
2150         ext3_mark_inode_dirty(handle, dir);
2151         drop_nlink(inode);
2152         if (!inode->i_nlink)
2153                 ext3_orphan_add(handle, inode);
2154         inode->i_ctime = dir->i_ctime;
2155         ext3_mark_inode_dirty(handle, inode);
2156         retval = 0;
2157
2158 end_unlink:
2159         ext3_journal_stop(handle);
2160         brelse (bh);
2161         return retval;
2162 }
2163
2164 static int ext3_symlink (struct inode * dir,
2165                 struct dentry *dentry, const char * symname)
2166 {
2167         handle_t *handle;
2168         struct inode * inode;
2169         int l, err, retries = 0;
2170
2171         l = strlen(symname)+1;
2172         if (l > dir->i_sb->s_blocksize)
2173                 return -ENAMETOOLONG;
2174
2175 retry:
2176         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2177                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2178                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2179         if (IS_ERR(handle))
2180                 return PTR_ERR(handle);
2181
2182         if (IS_DIRSYNC(dir))
2183                 handle->h_sync = 1;
2184
2185         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2186         err = PTR_ERR(inode);
2187         if (IS_ERR(inode))
2188                 goto out_stop;
2189
2190         if (l > sizeof (EXT3_I(inode)->i_data)) {
2191                 inode->i_op = &ext3_symlink_inode_operations;
2192                 ext3_set_aops(inode);
2193                 /*
2194                  * page_symlink() calls into ext3_prepare/commit_write.
2195                  * We have a transaction open.  All is sweetness.  It also sets
2196                  * i_size in generic_commit_write().
2197                  */
2198                 err = __page_symlink(inode, symname, l,
2199                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2200                 if (err) {
2201                         drop_nlink(inode);
2202                         ext3_mark_inode_dirty(handle, inode);
2203                         iput (inode);
2204                         goto out_stop;
2205                 }
2206         } else {
2207                 inode->i_op = &ext3_fast_symlink_inode_operations;
2208                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2209                 inode->i_size = l-1;
2210         }
2211         EXT3_I(inode)->i_disksize = inode->i_size;
2212         err = ext3_add_nondir(handle, dentry, inode);
2213 out_stop:
2214         ext3_journal_stop(handle);
2215         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2216                 goto retry;
2217         return err;
2218 }
2219
2220 static int ext3_link (struct dentry * old_dentry,
2221                 struct inode * dir, struct dentry *dentry)
2222 {
2223         handle_t *handle;
2224         struct inode *inode = old_dentry->d_inode;
2225         int err, retries = 0;
2226
2227         if (inode->i_nlink >= EXT3_LINK_MAX)
2228                 return -EMLINK;
2229         /*
2230          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2231          * otherwise has the potential to corrupt the orphan inode list.
2232          */
2233         if (inode->i_nlink == 0)
2234                 return -ENOENT;
2235
2236 retry:
2237         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2238                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2239         if (IS_ERR(handle))
2240                 return PTR_ERR(handle);
2241
2242         if (IS_DIRSYNC(dir))
2243                 handle->h_sync = 1;
2244
2245         inode->i_ctime = CURRENT_TIME_SEC;
2246         inc_nlink(inode);
2247         atomic_inc(&inode->i_count);
2248
2249         err = ext3_add_nondir(handle, dentry, inode);
2250         ext3_journal_stop(handle);
2251         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2252                 goto retry;
2253         return err;
2254 }
2255
2256 #define PARENT_INO(buffer) \
2257         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2258
2259 /*
2260  * Anybody can rename anything with this: the permission checks are left to the
2261  * higher-level routines.
2262  */
2263 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2264                            struct inode * new_dir,struct dentry *new_dentry)
2265 {
2266         handle_t *handle;
2267         struct inode * old_inode, * new_inode;
2268         struct buffer_head * old_bh, * new_bh, * dir_bh;
2269         struct ext3_dir_entry_2 * old_de, * new_de;
2270         int retval;
2271
2272         old_bh = new_bh = dir_bh = NULL;
2273
2274         /* Initialize quotas before so that eventual writes go
2275          * in separate transaction */
2276         if (new_dentry->d_inode)
2277                 DQUOT_INIT(new_dentry->d_inode);
2278         handle = ext3_journal_start(old_dir, 2 *
2279                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2280                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2281         if (IS_ERR(handle))
2282                 return PTR_ERR(handle);
2283
2284         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2285                 handle->h_sync = 1;
2286
2287         old_bh = ext3_find_entry (old_dentry, &old_de);
2288         /*
2289          *  Check for inode number is _not_ due to possible IO errors.
2290          *  We might rmdir the source, keep it as pwd of some process
2291          *  and merrily kill the link to whatever was created under the
2292          *  same name. Goodbye sticky bit ;-<
2293          */
2294         old_inode = old_dentry->d_inode;
2295         retval = -ENOENT;
2296         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2297                 goto end_rename;
2298
2299         new_inode = new_dentry->d_inode;
2300         new_bh = ext3_find_entry (new_dentry, &new_de);
2301         if (new_bh) {
2302                 if (!new_inode) {
2303                         brelse (new_bh);
2304                         new_bh = NULL;
2305                 }
2306         }
2307         if (S_ISDIR(old_inode->i_mode)) {
2308                 if (new_inode) {
2309                         retval = -ENOTEMPTY;
2310                         if (!empty_dir (new_inode))
2311                                 goto end_rename;
2312                 }
2313                 retval = -EIO;
2314                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2315                 if (!dir_bh)
2316                         goto end_rename;
2317                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2318                         goto end_rename;
2319                 retval = -EMLINK;
2320                 if (!new_inode && new_dir!=old_dir &&
2321                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2322                         goto end_rename;
2323         }
2324         if (!new_bh) {
2325                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2326                 if (retval)
2327                         goto end_rename;
2328         } else {
2329                 BUFFER_TRACE(new_bh, "get write access");
2330                 ext3_journal_get_write_access(handle, new_bh);
2331                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2332                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2333                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2334                         new_de->file_type = old_de->file_type;
2335                 new_dir->i_version++;
2336                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2337                 ext3_journal_dirty_metadata(handle, new_bh);
2338                 brelse(new_bh);
2339                 new_bh = NULL;
2340         }
2341
2342         /*
2343          * Like most other Unix systems, set the ctime for inodes on a
2344          * rename.
2345          */
2346         old_inode->i_ctime = CURRENT_TIME_SEC;
2347         ext3_mark_inode_dirty(handle, old_inode);
2348
2349         /*
2350          * ok, that's it
2351          */
2352         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2353             old_de->name_len != old_dentry->d_name.len ||
2354             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2355             (retval = ext3_delete_entry(handle, old_dir,
2356                                         old_de, old_bh)) == -ENOENT) {
2357                 /* old_de could have moved from under us during htree split, so
2358                  * make sure that we are deleting the right entry.  We might
2359                  * also be pointing to a stale entry in the unused part of
2360                  * old_bh so just checking inum and the name isn't enough. */
2361                 struct buffer_head *old_bh2;
2362                 struct ext3_dir_entry_2 *old_de2;
2363
2364                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2365                 if (old_bh2) {
2366                         retval = ext3_delete_entry(handle, old_dir,
2367                                                    old_de2, old_bh2);
2368                         brelse(old_bh2);
2369                 }
2370         }
2371         if (retval) {
2372                 ext3_warning(old_dir->i_sb, "ext3_rename",
2373                                 "Deleting old file (%lu), %d, error=%d",
2374                                 old_dir->i_ino, old_dir->i_nlink, retval);
2375         }
2376
2377         if (new_inode) {
2378                 drop_nlink(new_inode);
2379                 new_inode->i_ctime = CURRENT_TIME_SEC;
2380         }
2381         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2382         ext3_update_dx_flag(old_dir);
2383         if (dir_bh) {
2384                 BUFFER_TRACE(dir_bh, "get_write_access");
2385                 ext3_journal_get_write_access(handle, dir_bh);
2386                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2387                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2388                 ext3_journal_dirty_metadata(handle, dir_bh);
2389                 drop_nlink(old_dir);
2390                 if (new_inode) {
2391                         drop_nlink(new_inode);
2392                 } else {
2393                         inc_nlink(new_dir);
2394                         ext3_update_dx_flag(new_dir);
2395                         ext3_mark_inode_dirty(handle, new_dir);
2396                 }
2397         }
2398         ext3_mark_inode_dirty(handle, old_dir);
2399         if (new_inode) {
2400                 ext3_mark_inode_dirty(handle, new_inode);
2401                 if (!new_inode->i_nlink)
2402                         ext3_orphan_add(handle, new_inode);
2403         }
2404         retval = 0;
2405
2406 end_rename:
2407         brelse (dir_bh);
2408         brelse (old_bh);
2409         brelse (new_bh);
2410         ext3_journal_stop(handle);
2411         return retval;
2412 }
2413
2414 /*
2415  * directories can handle most operations...
2416  */
2417 const struct inode_operations ext3_dir_inode_operations = {
2418         .create         = ext3_create,
2419         .lookup         = ext3_lookup,
2420         .link           = ext3_link,
2421         .unlink         = ext3_unlink,
2422         .symlink        = ext3_symlink,
2423         .mkdir          = ext3_mkdir,
2424         .rmdir          = ext3_rmdir,
2425         .mknod          = ext3_mknod,
2426         .rename         = ext3_rename,
2427         .setattr        = ext3_setattr,
2428 #ifdef CONFIG_EXT3_FS_XATTR
2429         .setxattr       = generic_setxattr,
2430         .getxattr       = generic_getxattr,
2431         .listxattr      = ext3_listxattr,
2432         .removexattr    = generic_removexattr,
2433 #endif
2434         .permission     = ext3_permission,
2435 };
2436
2437 const struct inode_operations ext3_special_inode_operations = {
2438         .setattr        = ext3_setattr,
2439 #ifdef CONFIG_EXT3_FS_XATTR
2440         .setxattr       = generic_setxattr,
2441         .getxattr       = generic_getxattr,
2442         .listxattr      = ext3_listxattr,
2443         .removexattr    = generic_removexattr,
2444 #endif
2445         .permission     = ext3_permission,
2446 };