Merge branches 'release' and 'ppc-workaround' into release
[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
864         *res_dir = NULL;
865         sb = dir->i_sb;
866         namelen = dentry->d_name.len;
867         if (namelen > EXT3_NAME_LEN)
868                 return NULL;
869         if (is_dx(dir)) {
870                 bh = ext3_dx_find_entry(dentry, res_dir, &err);
871                 /*
872                  * On success, or if the error was file not found,
873                  * return.  Otherwise, fall back to doing a search the
874                  * old fashioned way.
875                  */
876                 if (bh || (err != ERR_BAD_DX_DIR))
877                         return bh;
878                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
879         }
880         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
881         start = EXT3_I(dir)->i_dir_start_lookup;
882         if (start >= nblocks)
883                 start = 0;
884         block = start;
885 restart:
886         do {
887                 /*
888                  * We deal with the read-ahead logic here.
889                  */
890                 if (ra_ptr >= ra_max) {
891                         /* Refill the readahead buffer */
892                         ra_ptr = 0;
893                         b = block;
894                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
895                                 /*
896                                  * Terminate if we reach the end of the
897                                  * directory and must wrap, or if our
898                                  * search has finished at this block.
899                                  */
900                                 if (b >= nblocks || (num && block == start)) {
901                                         bh_use[ra_max] = NULL;
902                                         break;
903                                 }
904                                 num++;
905                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
906                                 bh_use[ra_max] = bh;
907                                 if (bh)
908                                         ll_rw_block(READ_META, 1, &bh);
909                         }
910                 }
911                 if ((bh = bh_use[ra_ptr++]) == NULL)
912                         goto next;
913                 wait_on_buffer(bh);
914                 if (!buffer_uptodate(bh)) {
915                         /* read error, skip block & hope for the best */
916                         ext3_error(sb, __FUNCTION__, "reading directory #%lu "
917                                    "offset %lu", dir->i_ino, block);
918                         brelse(bh);
919                         goto next;
920                 }
921                 i = search_dirblock(bh, dir, dentry,
922                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
923                 if (i == 1) {
924                         EXT3_I(dir)->i_dir_start_lookup = block;
925                         ret = bh;
926                         goto cleanup_and_exit;
927                 } else {
928                         brelse(bh);
929                         if (i < 0)
930                                 goto cleanup_and_exit;
931                 }
932         next:
933                 if (++block >= nblocks)
934                         block = 0;
935         } while (block != start);
936
937         /*
938          * If the directory has grown while we were searching, then
939          * search the last part of the directory before giving up.
940          */
941         block = nblocks;
942         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
943         if (block < nblocks) {
944                 start = 0;
945                 goto restart;
946         }
947
948 cleanup_and_exit:
949         /* Clean up the read-ahead blocks */
950         for (; ra_ptr < ra_max; ra_ptr++)
951                 brelse (bh_use[ra_ptr]);
952         return ret;
953 }
954
955 static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
956                        struct ext3_dir_entry_2 **res_dir, int *err)
957 {
958         struct super_block * sb;
959         struct dx_hash_info     hinfo;
960         u32 hash;
961         struct dx_frame frames[2], *frame;
962         struct ext3_dir_entry_2 *de, *top;
963         struct buffer_head *bh;
964         unsigned long block;
965         int retval;
966         int namelen = dentry->d_name.len;
967         const u8 *name = dentry->d_name.name;
968         struct inode *dir = dentry->d_parent->d_inode;
969
970         sb = dir->i_sb;
971         /* NFS may look up ".." - look at dx_root directory block */
972         if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
973                 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
974                         return NULL;
975         } else {
976                 frame = frames;
977                 frame->bh = NULL;                       /* for dx_release() */
978                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
979                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
980         }
981         hash = hinfo.hash;
982         do {
983                 block = dx_get_block(frame->at);
984                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
985                         goto errout;
986                 de = (struct ext3_dir_entry_2 *) bh->b_data;
987                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
988                                        EXT3_DIR_REC_LEN(0));
989                 for (; de < top; de = ext3_next_entry(de))
990                 if (ext3_match (namelen, name, de)) {
991                         if (!ext3_check_dir_entry("ext3_find_entry",
992                                                   dir, de, bh,
993                                   (block<<EXT3_BLOCK_SIZE_BITS(sb))
994                                           +((char *)de - bh->b_data))) {
995                                 brelse (bh);
996                                 *err = ERR_BAD_DX_DIR;
997                                 goto errout;
998                         }
999                         *res_dir = de;
1000                         dx_release (frames);
1001                         return bh;
1002                 }
1003                 brelse (bh);
1004                 /* Check to see if we should continue to search */
1005                 retval = ext3_htree_next_block(dir, hash, frame,
1006                                                frames, NULL);
1007                 if (retval < 0) {
1008                         ext3_warning(sb, __FUNCTION__,
1009                              "error reading index page in directory #%lu",
1010                              dir->i_ino);
1011                         *err = retval;
1012                         goto errout;
1013                 }
1014         } while (retval == 1);
1015
1016         *err = -ENOENT;
1017 errout:
1018         dxtrace(printk("%s not found\n", name));
1019         dx_release (frames);
1020         return NULL;
1021 }
1022
1023 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1024 {
1025         struct inode * inode;
1026         struct ext3_dir_entry_2 * de;
1027         struct buffer_head * bh;
1028
1029         if (dentry->d_name.len > EXT3_NAME_LEN)
1030                 return ERR_PTR(-ENAMETOOLONG);
1031
1032         bh = ext3_find_entry(dentry, &de);
1033         inode = NULL;
1034         if (bh) {
1035                 unsigned long ino = le32_to_cpu(de->inode);
1036                 brelse (bh);
1037                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1038                         ext3_error(dir->i_sb, "ext3_lookup",
1039                                    "bad inode number: %lu", ino);
1040                         inode = NULL;
1041                 } else
1042                         inode = iget(dir->i_sb, ino);
1043
1044                 if (!inode)
1045                         return ERR_PTR(-EACCES);
1046
1047                 if (is_bad_inode(inode)) {
1048                         iput(inode);
1049                         return ERR_PTR(-ENOENT);
1050                 }
1051         }
1052         return d_splice_alias(inode, dentry);
1053 }
1054
1055
1056 struct dentry *ext3_get_parent(struct dentry *child)
1057 {
1058         unsigned long ino;
1059         struct dentry *parent;
1060         struct inode *inode;
1061         struct dentry dotdot;
1062         struct ext3_dir_entry_2 * de;
1063         struct buffer_head *bh;
1064
1065         dotdot.d_name.name = "..";
1066         dotdot.d_name.len = 2;
1067         dotdot.d_parent = child; /* confusing, isn't it! */
1068
1069         bh = ext3_find_entry(&dotdot, &de);
1070         inode = NULL;
1071         if (!bh)
1072                 return ERR_PTR(-ENOENT);
1073         ino = le32_to_cpu(de->inode);
1074         brelse(bh);
1075
1076         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1077                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1078                            "bad inode number: %lu", ino);
1079                 inode = NULL;
1080         } else
1081                 inode = iget(child->d_inode->i_sb, ino);
1082
1083         if (!inode)
1084                 return ERR_PTR(-EACCES);
1085
1086         if (is_bad_inode(inode)) {
1087                 iput(inode);
1088                 return ERR_PTR(-ENOENT);
1089         }
1090
1091         parent = d_alloc_anon(inode);
1092         if (!parent) {
1093                 iput(inode);
1094                 parent = ERR_PTR(-ENOMEM);
1095         }
1096         return parent;
1097 }
1098
1099 #define S_SHIFT 12
1100 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1101         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1102         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1103         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1104         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1105         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1106         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1107         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1108 };
1109
1110 static inline void ext3_set_de_type(struct super_block *sb,
1111                                 struct ext3_dir_entry_2 *de,
1112                                 umode_t mode) {
1113         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1114                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1115 }
1116
1117 /*
1118  * Move count entries from end of map between two memory locations.
1119  * Returns pointer to last entry moved.
1120  */
1121 static struct ext3_dir_entry_2 *
1122 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1123 {
1124         unsigned rec_len = 0;
1125
1126         while (count--) {
1127                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1128                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1129                 memcpy (to, de, rec_len);
1130                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1131                                 ext3_rec_len_to_disk(rec_len);
1132                 de->inode = 0;
1133                 map++;
1134                 to += rec_len;
1135         }
1136         return (struct ext3_dir_entry_2 *) (to - rec_len);
1137 }
1138
1139 /*
1140  * Compact each dir entry in the range to the minimal rec_len.
1141  * Returns pointer to last entry in range.
1142  */
1143 static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1144 {
1145         struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1146         unsigned rec_len = 0;
1147
1148         prev = to = de;
1149         while ((char*)de < base + size) {
1150                 next = ext3_next_entry(de);
1151                 if (de->inode && de->name_len) {
1152                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1153                         if (de > to)
1154                                 memmove(to, de, rec_len);
1155                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1156                         prev = to;
1157                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1158                 }
1159                 de = next;
1160         }
1161         return prev;
1162 }
1163
1164 /*
1165  * Split a full leaf block to make room for a new dir entry.
1166  * Allocate a new block, and move entries so that they are approx. equally full.
1167  * Returns pointer to de in block into which the new entry will be inserted.
1168  */
1169 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1170                         struct buffer_head **bh,struct dx_frame *frame,
1171                         struct dx_hash_info *hinfo, int *error)
1172 {
1173         unsigned blocksize = dir->i_sb->s_blocksize;
1174         unsigned count, continued;
1175         struct buffer_head *bh2;
1176         u32 newblock;
1177         u32 hash2;
1178         struct dx_map_entry *map;
1179         char *data1 = (*bh)->b_data, *data2;
1180         unsigned split, move, size, i;
1181         struct ext3_dir_entry_2 *de = NULL, *de2;
1182         int     err = 0;
1183
1184         bh2 = ext3_append (handle, dir, &newblock, &err);
1185         if (!(bh2)) {
1186                 brelse(*bh);
1187                 *bh = NULL;
1188                 goto errout;
1189         }
1190
1191         BUFFER_TRACE(*bh, "get_write_access");
1192         err = ext3_journal_get_write_access(handle, *bh);
1193         if (err)
1194                 goto journal_error;
1195
1196         BUFFER_TRACE(frame->bh, "get_write_access");
1197         err = ext3_journal_get_write_access(handle, frame->bh);
1198         if (err)
1199                 goto journal_error;
1200
1201         data2 = bh2->b_data;
1202
1203         /* create map in the end of data2 block */
1204         map = (struct dx_map_entry *) (data2 + blocksize);
1205         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1206                              blocksize, hinfo, map);
1207         map -= count;
1208         dx_sort_map (map, count);
1209         /* Split the existing block in the middle, size-wise */
1210         size = 0;
1211         move = 0;
1212         for (i = count-1; i >= 0; i--) {
1213                 /* is more than half of this entry in 2nd half of the block? */
1214                 if (size + map[i].size/2 > blocksize/2)
1215                         break;
1216                 size += map[i].size;
1217                 move++;
1218         }
1219         /* map index at which we will split */
1220         split = count - move;
1221         hash2 = map[split].hash;
1222         continued = hash2 == map[split - 1].hash;
1223         dxtrace(printk("Split block %i at %x, %i/%i\n",
1224                 dx_get_block(frame->at), hash2, split, count-split));
1225
1226         /* Fancy dance to stay within two buffers */
1227         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1228         de = dx_pack_dirents(data1,blocksize);
1229         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1230         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1231         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1232         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1233
1234         /* Which block gets the new entry? */
1235         if (hinfo->hash >= hash2)
1236         {
1237                 swap(*bh, bh2);
1238                 de = de2;
1239         }
1240         dx_insert_block (frame, hash2 + continued, newblock);
1241         err = ext3_journal_dirty_metadata (handle, bh2);
1242         if (err)
1243                 goto journal_error;
1244         err = ext3_journal_dirty_metadata (handle, frame->bh);
1245         if (err)
1246                 goto journal_error;
1247         brelse (bh2);
1248         dxtrace(dx_show_index ("frame", frame->entries));
1249         return de;
1250
1251 journal_error:
1252         brelse(*bh);
1253         brelse(bh2);
1254         *bh = NULL;
1255         ext3_std_error(dir->i_sb, err);
1256 errout:
1257         *error = err;
1258         return NULL;
1259 }
1260
1261
1262 /*
1263  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1264  * it points to a directory entry which is guaranteed to be large
1265  * enough for new directory entry.  If de is NULL, then
1266  * add_dirent_to_buf will attempt search the directory block for
1267  * space.  It will return -ENOSPC if no space is available, and -EIO
1268  * and -EEXIST if directory entry already exists.
1269  *
1270  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1271  * all other cases bh is released.
1272  */
1273 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1274                              struct inode *inode, struct ext3_dir_entry_2 *de,
1275                              struct buffer_head * bh)
1276 {
1277         struct inode    *dir = dentry->d_parent->d_inode;
1278         const char      *name = dentry->d_name.name;
1279         int             namelen = dentry->d_name.len;
1280         unsigned long   offset = 0;
1281         unsigned short  reclen;
1282         int             nlen, rlen, err;
1283         char            *top;
1284
1285         reclen = EXT3_DIR_REC_LEN(namelen);
1286         if (!de) {
1287                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1288                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1289                 while ((char *) de <= top) {
1290                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1291                                                   bh, offset)) {
1292                                 brelse (bh);
1293                                 return -EIO;
1294                         }
1295                         if (ext3_match (namelen, name, de)) {
1296                                 brelse (bh);
1297                                 return -EEXIST;
1298                         }
1299                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1300                         rlen = ext3_rec_len_from_disk(de->rec_len);
1301                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1302                                 break;
1303                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1304                         offset += rlen;
1305                 }
1306                 if ((char *) de > top)
1307                         return -ENOSPC;
1308         }
1309         BUFFER_TRACE(bh, "get_write_access");
1310         err = ext3_journal_get_write_access(handle, bh);
1311         if (err) {
1312                 ext3_std_error(dir->i_sb, err);
1313                 brelse(bh);
1314                 return err;
1315         }
1316
1317         /* By now the buffer is marked for journaling */
1318         nlen = EXT3_DIR_REC_LEN(de->name_len);
1319         rlen = ext3_rec_len_from_disk(de->rec_len);
1320         if (de->inode) {
1321                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1322                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1323                 de->rec_len = ext3_rec_len_to_disk(nlen);
1324                 de = de1;
1325         }
1326         de->file_type = EXT3_FT_UNKNOWN;
1327         if (inode) {
1328                 de->inode = cpu_to_le32(inode->i_ino);
1329                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1330         } else
1331                 de->inode = 0;
1332         de->name_len = namelen;
1333         memcpy (de->name, name, namelen);
1334         /*
1335          * XXX shouldn't update any times until successful
1336          * completion of syscall, but too many callers depend
1337          * on this.
1338          *
1339          * XXX similarly, too many callers depend on
1340          * ext3_new_inode() setting the times, but error
1341          * recovery deletes the inode, so the worst that can
1342          * happen is that the times are slightly out of date
1343          * and/or different from the directory change time.
1344          */
1345         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1346         ext3_update_dx_flag(dir);
1347         dir->i_version++;
1348         ext3_mark_inode_dirty(handle, dir);
1349         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1350         err = ext3_journal_dirty_metadata(handle, bh);
1351         if (err)
1352                 ext3_std_error(dir->i_sb, err);
1353         brelse(bh);
1354         return 0;
1355 }
1356
1357 /*
1358  * This converts a one block unindexed directory to a 3 block indexed
1359  * directory, and adds the dentry to the indexed directory.
1360  */
1361 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1362                             struct inode *inode, struct buffer_head *bh)
1363 {
1364         struct inode    *dir = dentry->d_parent->d_inode;
1365         const char      *name = dentry->d_name.name;
1366         int             namelen = dentry->d_name.len;
1367         struct buffer_head *bh2;
1368         struct dx_root  *root;
1369         struct dx_frame frames[2], *frame;
1370         struct dx_entry *entries;
1371         struct ext3_dir_entry_2 *de, *de2;
1372         char            *data1, *top;
1373         unsigned        len;
1374         int             retval;
1375         unsigned        blocksize;
1376         struct dx_hash_info hinfo;
1377         u32             block;
1378         struct fake_dirent *fde;
1379
1380         blocksize =  dir->i_sb->s_blocksize;
1381         dxtrace(printk("Creating index\n"));
1382         retval = ext3_journal_get_write_access(handle, bh);
1383         if (retval) {
1384                 ext3_std_error(dir->i_sb, retval);
1385                 brelse(bh);
1386                 return retval;
1387         }
1388         root = (struct dx_root *) bh->b_data;
1389
1390         bh2 = ext3_append (handle, dir, &block, &retval);
1391         if (!(bh2)) {
1392                 brelse(bh);
1393                 return retval;
1394         }
1395         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1396         data1 = bh2->b_data;
1397
1398         /* The 0th block becomes the root, move the dirents out */
1399         fde = &root->dotdot;
1400         de = (struct ext3_dir_entry_2 *)((char *)fde +
1401                         ext3_rec_len_from_disk(fde->rec_len));
1402         len = ((char *) root) + blocksize - (char *) de;
1403         memcpy (data1, de, len);
1404         de = (struct ext3_dir_entry_2 *) data1;
1405         top = data1 + len;
1406         while ((char *)(de2 = ext3_next_entry(de)) < top)
1407                 de = de2;
1408         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1409         /* Initialize the root; the dot dirents already exist */
1410         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1411         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1412         memset (&root->info, 0, sizeof(root->info));
1413         root->info.info_length = sizeof(root->info);
1414         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1415         entries = root->entries;
1416         dx_set_block (entries, 1);
1417         dx_set_count (entries, 1);
1418         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1419
1420         /* Initialize as for dx_probe */
1421         hinfo.hash_version = root->info.hash_version;
1422         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1423         ext3fs_dirhash(name, namelen, &hinfo);
1424         frame = frames;
1425         frame->entries = entries;
1426         frame->at = entries;
1427         frame->bh = bh;
1428         bh = bh2;
1429         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1430         dx_release (frames);
1431         if (!(de))
1432                 return retval;
1433
1434         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1435 }
1436
1437 /*
1438  *      ext3_add_entry()
1439  *
1440  * adds a file entry to the specified directory, using the same
1441  * semantics as ext3_find_entry(). It returns NULL if it failed.
1442  *
1443  * NOTE!! The inode part of 'de' is left at 0 - which means you
1444  * may not sleep between calling this and putting something into
1445  * the entry, as someone else might have used it while you slept.
1446  */
1447 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1448         struct inode *inode)
1449 {
1450         struct inode *dir = dentry->d_parent->d_inode;
1451         unsigned long offset;
1452         struct buffer_head * bh;
1453         struct ext3_dir_entry_2 *de;
1454         struct super_block * sb;
1455         int     retval;
1456         int     dx_fallback=0;
1457         unsigned blocksize;
1458         u32 block, blocks;
1459
1460         sb = dir->i_sb;
1461         blocksize = sb->s_blocksize;
1462         if (!dentry->d_name.len)
1463                 return -EINVAL;
1464         if (is_dx(dir)) {
1465                 retval = ext3_dx_add_entry(handle, dentry, inode);
1466                 if (!retval || (retval != ERR_BAD_DX_DIR))
1467                         return retval;
1468                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1469                 dx_fallback++;
1470                 ext3_mark_inode_dirty(handle, dir);
1471         }
1472         blocks = dir->i_size >> sb->s_blocksize_bits;
1473         for (block = 0, offset = 0; block < blocks; block++) {
1474                 bh = ext3_bread(handle, dir, block, 0, &retval);
1475                 if(!bh)
1476                         return retval;
1477                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1478                 if (retval != -ENOSPC)
1479                         return retval;
1480
1481                 if (blocks == 1 && !dx_fallback &&
1482                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1483                         return make_indexed_dir(handle, dentry, inode, bh);
1484                 brelse(bh);
1485         }
1486         bh = ext3_append(handle, dir, &block, &retval);
1487         if (!bh)
1488                 return retval;
1489         de = (struct ext3_dir_entry_2 *) bh->b_data;
1490         de->inode = 0;
1491         de->rec_len = ext3_rec_len_to_disk(blocksize);
1492         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1493 }
1494
1495 /*
1496  * Returns 0 for success, or a negative error value
1497  */
1498 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1499                              struct inode *inode)
1500 {
1501         struct dx_frame frames[2], *frame;
1502         struct dx_entry *entries, *at;
1503         struct dx_hash_info hinfo;
1504         struct buffer_head * bh;
1505         struct inode *dir = dentry->d_parent->d_inode;
1506         struct super_block * sb = dir->i_sb;
1507         struct ext3_dir_entry_2 *de;
1508         int err;
1509
1510         frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1511         if (!frame)
1512                 return err;
1513         entries = frame->entries;
1514         at = frame->at;
1515
1516         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1517                 goto cleanup;
1518
1519         BUFFER_TRACE(bh, "get_write_access");
1520         err = ext3_journal_get_write_access(handle, bh);
1521         if (err)
1522                 goto journal_error;
1523
1524         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1525         if (err != -ENOSPC) {
1526                 bh = NULL;
1527                 goto cleanup;
1528         }
1529
1530         /* Block full, should compress but for now just split */
1531         dxtrace(printk("using %u of %u node entries\n",
1532                        dx_get_count(entries), dx_get_limit(entries)));
1533         /* Need to split index? */
1534         if (dx_get_count(entries) == dx_get_limit(entries)) {
1535                 u32 newblock;
1536                 unsigned icount = dx_get_count(entries);
1537                 int levels = frame - frames;
1538                 struct dx_entry *entries2;
1539                 struct dx_node *node2;
1540                 struct buffer_head *bh2;
1541
1542                 if (levels && (dx_get_count(frames->entries) ==
1543                                dx_get_limit(frames->entries))) {
1544                         ext3_warning(sb, __FUNCTION__,
1545                                      "Directory index full!");
1546                         err = -ENOSPC;
1547                         goto cleanup;
1548                 }
1549                 bh2 = ext3_append (handle, dir, &newblock, &err);
1550                 if (!(bh2))
1551                         goto cleanup;
1552                 node2 = (struct dx_node *)(bh2->b_data);
1553                 entries2 = node2->entries;
1554                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1555                 node2->fake.inode = 0;
1556                 BUFFER_TRACE(frame->bh, "get_write_access");
1557                 err = ext3_journal_get_write_access(handle, frame->bh);
1558                 if (err)
1559                         goto journal_error;
1560                 if (levels) {
1561                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1562                         unsigned hash2 = dx_get_hash(entries + icount1);
1563                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1564
1565                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1566                         err = ext3_journal_get_write_access(handle,
1567                                                              frames[0].bh);
1568                         if (err)
1569                                 goto journal_error;
1570
1571                         memcpy ((char *) entries2, (char *) (entries + icount1),
1572                                 icount2 * sizeof(struct dx_entry));
1573                         dx_set_count (entries, icount1);
1574                         dx_set_count (entries2, icount2);
1575                         dx_set_limit (entries2, dx_node_limit(dir));
1576
1577                         /* Which index block gets the new entry? */
1578                         if (at - entries >= icount1) {
1579                                 frame->at = at = at - entries - icount1 + entries2;
1580                                 frame->entries = entries = entries2;
1581                                 swap(frame->bh, bh2);
1582                         }
1583                         dx_insert_block (frames + 0, hash2, newblock);
1584                         dxtrace(dx_show_index ("node", frames[1].entries));
1585                         dxtrace(dx_show_index ("node",
1586                                ((struct dx_node *) bh2->b_data)->entries));
1587                         err = ext3_journal_dirty_metadata(handle, bh2);
1588                         if (err)
1589                                 goto journal_error;
1590                         brelse (bh2);
1591                 } else {
1592                         dxtrace(printk("Creating second level index...\n"));
1593                         memcpy((char *) entries2, (char *) entries,
1594                                icount * sizeof(struct dx_entry));
1595                         dx_set_limit(entries2, dx_node_limit(dir));
1596
1597                         /* Set up root */
1598                         dx_set_count(entries, 1);
1599                         dx_set_block(entries + 0, newblock);
1600                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1601
1602                         /* Add new access path frame */
1603                         frame = frames + 1;
1604                         frame->at = at = at - entries + entries2;
1605                         frame->entries = entries = entries2;
1606                         frame->bh = bh2;
1607                         err = ext3_journal_get_write_access(handle,
1608                                                              frame->bh);
1609                         if (err)
1610                                 goto journal_error;
1611                 }
1612                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1613         }
1614         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1615         if (!de)
1616                 goto cleanup;
1617         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1618         bh = NULL;
1619         goto cleanup;
1620
1621 journal_error:
1622         ext3_std_error(dir->i_sb, err);
1623 cleanup:
1624         if (bh)
1625                 brelse(bh);
1626         dx_release(frames);
1627         return err;
1628 }
1629
1630 /*
1631  * ext3_delete_entry deletes a directory entry by merging it with the
1632  * previous entry
1633  */
1634 static int ext3_delete_entry (handle_t *handle,
1635                               struct inode * dir,
1636                               struct ext3_dir_entry_2 * de_del,
1637                               struct buffer_head * bh)
1638 {
1639         struct ext3_dir_entry_2 * de, * pde;
1640         int i;
1641
1642         i = 0;
1643         pde = NULL;
1644         de = (struct ext3_dir_entry_2 *) bh->b_data;
1645         while (i < bh->b_size) {
1646                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1647                         return -EIO;
1648                 if (de == de_del)  {
1649                         BUFFER_TRACE(bh, "get_write_access");
1650                         ext3_journal_get_write_access(handle, bh);
1651                         if (pde)
1652                                 pde->rec_len = ext3_rec_len_to_disk(
1653                                         ext3_rec_len_from_disk(pde->rec_len) +
1654                                         ext3_rec_len_from_disk(de->rec_len));
1655                         else
1656                                 de->inode = 0;
1657                         dir->i_version++;
1658                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1659                         ext3_journal_dirty_metadata(handle, bh);
1660                         return 0;
1661                 }
1662                 i += ext3_rec_len_from_disk(de->rec_len);
1663                 pde = de;
1664                 de = ext3_next_entry(de);
1665         }
1666         return -ENOENT;
1667 }
1668
1669 static int ext3_add_nondir(handle_t *handle,
1670                 struct dentry *dentry, struct inode *inode)
1671 {
1672         int err = ext3_add_entry(handle, dentry, inode);
1673         if (!err) {
1674                 ext3_mark_inode_dirty(handle, inode);
1675                 d_instantiate(dentry, inode);
1676                 return 0;
1677         }
1678         drop_nlink(inode);
1679         iput(inode);
1680         return err;
1681 }
1682
1683 /*
1684  * By the time this is called, we already have created
1685  * the directory cache entry for the new file, but it
1686  * is so far negative - it has no inode.
1687  *
1688  * If the create succeeds, we fill in the inode information
1689  * with d_instantiate().
1690  */
1691 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1692                 struct nameidata *nd)
1693 {
1694         handle_t *handle;
1695         struct inode * inode;
1696         int err, retries = 0;
1697
1698 retry:
1699         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1700                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1701                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1702         if (IS_ERR(handle))
1703                 return PTR_ERR(handle);
1704
1705         if (IS_DIRSYNC(dir))
1706                 handle->h_sync = 1;
1707
1708         inode = ext3_new_inode (handle, dir, mode);
1709         err = PTR_ERR(inode);
1710         if (!IS_ERR(inode)) {
1711                 inode->i_op = &ext3_file_inode_operations;
1712                 inode->i_fop = &ext3_file_operations;
1713                 ext3_set_aops(inode);
1714                 err = ext3_add_nondir(handle, dentry, inode);
1715         }
1716         ext3_journal_stop(handle);
1717         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1718                 goto retry;
1719         return err;
1720 }
1721
1722 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1723                         int mode, dev_t rdev)
1724 {
1725         handle_t *handle;
1726         struct inode *inode;
1727         int err, retries = 0;
1728
1729         if (!new_valid_dev(rdev))
1730                 return -EINVAL;
1731
1732 retry:
1733         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1734                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1735                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1736         if (IS_ERR(handle))
1737                 return PTR_ERR(handle);
1738
1739         if (IS_DIRSYNC(dir))
1740                 handle->h_sync = 1;
1741
1742         inode = ext3_new_inode (handle, dir, mode);
1743         err = PTR_ERR(inode);
1744         if (!IS_ERR(inode)) {
1745                 init_special_inode(inode, inode->i_mode, rdev);
1746 #ifdef CONFIG_EXT3_FS_XATTR
1747                 inode->i_op = &ext3_special_inode_operations;
1748 #endif
1749                 err = ext3_add_nondir(handle, dentry, inode);
1750         }
1751         ext3_journal_stop(handle);
1752         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1753                 goto retry;
1754         return err;
1755 }
1756
1757 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1758 {
1759         handle_t *handle;
1760         struct inode * inode;
1761         struct buffer_head * dir_block;
1762         struct ext3_dir_entry_2 * de;
1763         int err, retries = 0;
1764
1765         if (dir->i_nlink >= EXT3_LINK_MAX)
1766                 return -EMLINK;
1767
1768 retry:
1769         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1770                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1771                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1772         if (IS_ERR(handle))
1773                 return PTR_ERR(handle);
1774
1775         if (IS_DIRSYNC(dir))
1776                 handle->h_sync = 1;
1777
1778         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1779         err = PTR_ERR(inode);
1780         if (IS_ERR(inode))
1781                 goto out_stop;
1782
1783         inode->i_op = &ext3_dir_inode_operations;
1784         inode->i_fop = &ext3_dir_operations;
1785         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1786         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1787         if (!dir_block) {
1788                 drop_nlink(inode); /* is this nlink == 0? */
1789                 ext3_mark_inode_dirty(handle, inode);
1790                 iput (inode);
1791                 goto out_stop;
1792         }
1793         BUFFER_TRACE(dir_block, "get_write_access");
1794         ext3_journal_get_write_access(handle, dir_block);
1795         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1796         de->inode = cpu_to_le32(inode->i_ino);
1797         de->name_len = 1;
1798         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1799         strcpy (de->name, ".");
1800         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1801         de = ext3_next_entry(de);
1802         de->inode = cpu_to_le32(dir->i_ino);
1803         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1804                                         EXT3_DIR_REC_LEN(1));
1805         de->name_len = 2;
1806         strcpy (de->name, "..");
1807         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1808         inode->i_nlink = 2;
1809         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1810         ext3_journal_dirty_metadata(handle, dir_block);
1811         brelse (dir_block);
1812         ext3_mark_inode_dirty(handle, inode);
1813         err = ext3_add_entry (handle, dentry, inode);
1814         if (err) {
1815                 inode->i_nlink = 0;
1816                 ext3_mark_inode_dirty(handle, inode);
1817                 iput (inode);
1818                 goto out_stop;
1819         }
1820         inc_nlink(dir);
1821         ext3_update_dx_flag(dir);
1822         ext3_mark_inode_dirty(handle, dir);
1823         d_instantiate(dentry, inode);
1824 out_stop:
1825         ext3_journal_stop(handle);
1826         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1827                 goto retry;
1828         return err;
1829 }
1830
1831 /*
1832  * routine to check that the specified directory is empty (for rmdir)
1833  */
1834 static int empty_dir (struct inode * inode)
1835 {
1836         unsigned long offset;
1837         struct buffer_head * bh;
1838         struct ext3_dir_entry_2 * de, * de1;
1839         struct super_block * sb;
1840         int err = 0;
1841
1842         sb = inode->i_sb;
1843         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1844             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1845                 if (err)
1846                         ext3_error(inode->i_sb, __FUNCTION__,
1847                                    "error %d reading directory #%lu offset 0",
1848                                    err, inode->i_ino);
1849                 else
1850                         ext3_warning(inode->i_sb, __FUNCTION__,
1851                                      "bad directory (dir #%lu) - no data block",
1852                                      inode->i_ino);
1853                 return 1;
1854         }
1855         de = (struct ext3_dir_entry_2 *) bh->b_data;
1856         de1 = ext3_next_entry(de);
1857         if (le32_to_cpu(de->inode) != inode->i_ino ||
1858                         !le32_to_cpu(de1->inode) ||
1859                         strcmp (".", de->name) ||
1860                         strcmp ("..", de1->name)) {
1861                 ext3_warning (inode->i_sb, "empty_dir",
1862                               "bad directory (dir #%lu) - no `.' or `..'",
1863                               inode->i_ino);
1864                 brelse (bh);
1865                 return 1;
1866         }
1867         offset = ext3_rec_len_from_disk(de->rec_len) +
1868                         ext3_rec_len_from_disk(de1->rec_len);
1869         de = ext3_next_entry(de1);
1870         while (offset < inode->i_size ) {
1871                 if (!bh ||
1872                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1873                         err = 0;
1874                         brelse (bh);
1875                         bh = ext3_bread (NULL, inode,
1876                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1877                         if (!bh) {
1878                                 if (err)
1879                                         ext3_error(sb, __FUNCTION__,
1880                                                    "error %d reading directory"
1881                                                    " #%lu offset %lu",
1882                                                    err, inode->i_ino, offset);
1883                                 offset += sb->s_blocksize;
1884                                 continue;
1885                         }
1886                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1887                 }
1888                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1889                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1890                                                          sb->s_blocksize);
1891                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1892                         continue;
1893                 }
1894                 if (le32_to_cpu(de->inode)) {
1895                         brelse (bh);
1896                         return 0;
1897                 }
1898                 offset += ext3_rec_len_from_disk(de->rec_len);
1899                 de = ext3_next_entry(de);
1900         }
1901         brelse (bh);
1902         return 1;
1903 }
1904
1905 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1906  * such inodes, starting at the superblock, in case we crash before the
1907  * file is closed/deleted, or in case the inode truncate spans multiple
1908  * transactions and the last transaction is not recovered after a crash.
1909  *
1910  * At filesystem recovery time, we walk this list deleting unlinked
1911  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1912  */
1913 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1914 {
1915         struct super_block *sb = inode->i_sb;
1916         struct ext3_iloc iloc;
1917         int err = 0, rc;
1918
1919         lock_super(sb);
1920         if (!list_empty(&EXT3_I(inode)->i_orphan))
1921                 goto out_unlock;
1922
1923         /* Orphan handling is only valid for files with data blocks
1924          * being truncated, or files being unlinked. */
1925
1926         /* @@@ FIXME: Observation from aviro:
1927          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1928          * here (on lock_super()), so race with ext3_link() which might bump
1929          * ->i_nlink. For, say it, character device. Not a regular file,
1930          * not a directory, not a symlink and ->i_nlink > 0.
1931          */
1932         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1933                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1934
1935         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1936         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1937         if (err)
1938                 goto out_unlock;
1939
1940         err = ext3_reserve_inode_write(handle, inode, &iloc);
1941         if (err)
1942                 goto out_unlock;
1943
1944         /* Insert this inode at the head of the on-disk orphan list... */
1945         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1946         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1947         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1948         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1949         if (!err)
1950                 err = rc;
1951
1952         /* Only add to the head of the in-memory list if all the
1953          * previous operations succeeded.  If the orphan_add is going to
1954          * fail (possibly taking the journal offline), we can't risk
1955          * leaving the inode on the orphan list: stray orphan-list
1956          * entries can cause panics at unmount time.
1957          *
1958          * This is safe: on error we're going to ignore the orphan list
1959          * anyway on the next recovery. */
1960         if (!err)
1961                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1962
1963         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1964         jbd_debug(4, "orphan inode %lu will point to %d\n",
1965                         inode->i_ino, NEXT_ORPHAN(inode));
1966 out_unlock:
1967         unlock_super(sb);
1968         ext3_std_error(inode->i_sb, err);
1969         return err;
1970 }
1971
1972 /*
1973  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1974  * of such inodes stored on disk, because it is finally being cleaned up.
1975  */
1976 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1977 {
1978         struct list_head *prev;
1979         struct ext3_inode_info *ei = EXT3_I(inode);
1980         struct ext3_sb_info *sbi;
1981         unsigned long ino_next;
1982         struct ext3_iloc iloc;
1983         int err = 0;
1984
1985         lock_super(inode->i_sb);
1986         if (list_empty(&ei->i_orphan)) {
1987                 unlock_super(inode->i_sb);
1988                 return 0;
1989         }
1990
1991         ino_next = NEXT_ORPHAN(inode);
1992         prev = ei->i_orphan.prev;
1993         sbi = EXT3_SB(inode->i_sb);
1994
1995         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1996
1997         list_del_init(&ei->i_orphan);
1998
1999         /* If we're on an error path, we may not have a valid
2000          * transaction handle with which to update the orphan list on
2001          * disk, but we still need to remove the inode from the linked
2002          * list in memory. */
2003         if (!handle)
2004                 goto out;
2005
2006         err = ext3_reserve_inode_write(handle, inode, &iloc);
2007         if (err)
2008                 goto out_err;
2009
2010         if (prev == &sbi->s_orphan) {
2011                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2012                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2013                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2014                 if (err)
2015                         goto out_brelse;
2016                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2017                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2018         } else {
2019                 struct ext3_iloc iloc2;
2020                 struct inode *i_prev =
2021                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2022
2023                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2024                           i_prev->i_ino, ino_next);
2025                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2026                 if (err)
2027                         goto out_brelse;
2028                 NEXT_ORPHAN(i_prev) = ino_next;
2029                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2030         }
2031         if (err)
2032                 goto out_brelse;
2033         NEXT_ORPHAN(inode) = 0;
2034         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2035
2036 out_err:
2037         ext3_std_error(inode->i_sb, err);
2038 out:
2039         unlock_super(inode->i_sb);
2040         return err;
2041
2042 out_brelse:
2043         brelse(iloc.bh);
2044         goto out_err;
2045 }
2046
2047 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2048 {
2049         int retval;
2050         struct inode * inode;
2051         struct buffer_head * bh;
2052         struct ext3_dir_entry_2 * de;
2053         handle_t *handle;
2054
2055         /* Initialize quotas before so that eventual writes go in
2056          * separate transaction */
2057         DQUOT_INIT(dentry->d_inode);
2058         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2059         if (IS_ERR(handle))
2060                 return PTR_ERR(handle);
2061
2062         retval = -ENOENT;
2063         bh = ext3_find_entry (dentry, &de);
2064         if (!bh)
2065                 goto end_rmdir;
2066
2067         if (IS_DIRSYNC(dir))
2068                 handle->h_sync = 1;
2069
2070         inode = dentry->d_inode;
2071
2072         retval = -EIO;
2073         if (le32_to_cpu(de->inode) != inode->i_ino)
2074                 goto end_rmdir;
2075
2076         retval = -ENOTEMPTY;
2077         if (!empty_dir (inode))
2078                 goto end_rmdir;
2079
2080         retval = ext3_delete_entry(handle, dir, de, bh);
2081         if (retval)
2082                 goto end_rmdir;
2083         if (inode->i_nlink != 2)
2084                 ext3_warning (inode->i_sb, "ext3_rmdir",
2085                               "empty directory has nlink!=2 (%d)",
2086                               inode->i_nlink);
2087         inode->i_version++;
2088         clear_nlink(inode);
2089         /* There's no need to set i_disksize: the fact that i_nlink is
2090          * zero will ensure that the right thing happens during any
2091          * recovery. */
2092         inode->i_size = 0;
2093         ext3_orphan_add(handle, inode);
2094         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2095         ext3_mark_inode_dirty(handle, inode);
2096         drop_nlink(dir);
2097         ext3_update_dx_flag(dir);
2098         ext3_mark_inode_dirty(handle, dir);
2099
2100 end_rmdir:
2101         ext3_journal_stop(handle);
2102         brelse (bh);
2103         return retval;
2104 }
2105
2106 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2107 {
2108         int retval;
2109         struct inode * inode;
2110         struct buffer_head * bh;
2111         struct ext3_dir_entry_2 * de;
2112         handle_t *handle;
2113
2114         /* Initialize quotas before so that eventual writes go
2115          * in separate transaction */
2116         DQUOT_INIT(dentry->d_inode);
2117         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2118         if (IS_ERR(handle))
2119                 return PTR_ERR(handle);
2120
2121         if (IS_DIRSYNC(dir))
2122                 handle->h_sync = 1;
2123
2124         retval = -ENOENT;
2125         bh = ext3_find_entry (dentry, &de);
2126         if (!bh)
2127                 goto end_unlink;
2128
2129         inode = dentry->d_inode;
2130
2131         retval = -EIO;
2132         if (le32_to_cpu(de->inode) != inode->i_ino)
2133                 goto end_unlink;
2134
2135         if (!inode->i_nlink) {
2136                 ext3_warning (inode->i_sb, "ext3_unlink",
2137                               "Deleting nonexistent file (%lu), %d",
2138                               inode->i_ino, inode->i_nlink);
2139                 inode->i_nlink = 1;
2140         }
2141         retval = ext3_delete_entry(handle, dir, de, bh);
2142         if (retval)
2143                 goto end_unlink;
2144         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2145         ext3_update_dx_flag(dir);
2146         ext3_mark_inode_dirty(handle, dir);
2147         drop_nlink(inode);
2148         if (!inode->i_nlink)
2149                 ext3_orphan_add(handle, inode);
2150         inode->i_ctime = dir->i_ctime;
2151         ext3_mark_inode_dirty(handle, inode);
2152         retval = 0;
2153
2154 end_unlink:
2155         ext3_journal_stop(handle);
2156         brelse (bh);
2157         return retval;
2158 }
2159
2160 static int ext3_symlink (struct inode * dir,
2161                 struct dentry *dentry, const char * symname)
2162 {
2163         handle_t *handle;
2164         struct inode * inode;
2165         int l, err, retries = 0;
2166
2167         l = strlen(symname)+1;
2168         if (l > dir->i_sb->s_blocksize)
2169                 return -ENAMETOOLONG;
2170
2171 retry:
2172         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2173                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2174                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2175         if (IS_ERR(handle))
2176                 return PTR_ERR(handle);
2177
2178         if (IS_DIRSYNC(dir))
2179                 handle->h_sync = 1;
2180
2181         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2182         err = PTR_ERR(inode);
2183         if (IS_ERR(inode))
2184                 goto out_stop;
2185
2186         if (l > sizeof (EXT3_I(inode)->i_data)) {
2187                 inode->i_op = &ext3_symlink_inode_operations;
2188                 ext3_set_aops(inode);
2189                 /*
2190                  * page_symlink() calls into ext3_prepare/commit_write.
2191                  * We have a transaction open.  All is sweetness.  It also sets
2192                  * i_size in generic_commit_write().
2193                  */
2194                 err = __page_symlink(inode, symname, l,
2195                                 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2196                 if (err) {
2197                         drop_nlink(inode);
2198                         ext3_mark_inode_dirty(handle, inode);
2199                         iput (inode);
2200                         goto out_stop;
2201                 }
2202         } else {
2203                 inode->i_op = &ext3_fast_symlink_inode_operations;
2204                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2205                 inode->i_size = l-1;
2206         }
2207         EXT3_I(inode)->i_disksize = inode->i_size;
2208         err = ext3_add_nondir(handle, dentry, inode);
2209 out_stop:
2210         ext3_journal_stop(handle);
2211         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2212                 goto retry;
2213         return err;
2214 }
2215
2216 static int ext3_link (struct dentry * old_dentry,
2217                 struct inode * dir, struct dentry *dentry)
2218 {
2219         handle_t *handle;
2220         struct inode *inode = old_dentry->d_inode;
2221         int err, retries = 0;
2222
2223         if (inode->i_nlink >= EXT3_LINK_MAX)
2224                 return -EMLINK;
2225         /*
2226          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2227          * otherwise has the potential to corrupt the orphan inode list.
2228          */
2229         if (inode->i_nlink == 0)
2230                 return -ENOENT;
2231
2232 retry:
2233         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2234                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2235         if (IS_ERR(handle))
2236                 return PTR_ERR(handle);
2237
2238         if (IS_DIRSYNC(dir))
2239                 handle->h_sync = 1;
2240
2241         inode->i_ctime = CURRENT_TIME_SEC;
2242         inc_nlink(inode);
2243         atomic_inc(&inode->i_count);
2244
2245         err = ext3_add_nondir(handle, dentry, inode);
2246         ext3_journal_stop(handle);
2247         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2248                 goto retry;
2249         return err;
2250 }
2251
2252 #define PARENT_INO(buffer) \
2253         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2254
2255 /*
2256  * Anybody can rename anything with this: the permission checks are left to the
2257  * higher-level routines.
2258  */
2259 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2260                            struct inode * new_dir,struct dentry *new_dentry)
2261 {
2262         handle_t *handle;
2263         struct inode * old_inode, * new_inode;
2264         struct buffer_head * old_bh, * new_bh, * dir_bh;
2265         struct ext3_dir_entry_2 * old_de, * new_de;
2266         int retval;
2267
2268         old_bh = new_bh = dir_bh = NULL;
2269
2270         /* Initialize quotas before so that eventual writes go
2271          * in separate transaction */
2272         if (new_dentry->d_inode)
2273                 DQUOT_INIT(new_dentry->d_inode);
2274         handle = ext3_journal_start(old_dir, 2 *
2275                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2276                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2277         if (IS_ERR(handle))
2278                 return PTR_ERR(handle);
2279
2280         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2281                 handle->h_sync = 1;
2282
2283         old_bh = ext3_find_entry (old_dentry, &old_de);
2284         /*
2285          *  Check for inode number is _not_ due to possible IO errors.
2286          *  We might rmdir the source, keep it as pwd of some process
2287          *  and merrily kill the link to whatever was created under the
2288          *  same name. Goodbye sticky bit ;-<
2289          */
2290         old_inode = old_dentry->d_inode;
2291         retval = -ENOENT;
2292         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2293                 goto end_rename;
2294
2295         new_inode = new_dentry->d_inode;
2296         new_bh = ext3_find_entry (new_dentry, &new_de);
2297         if (new_bh) {
2298                 if (!new_inode) {
2299                         brelse (new_bh);
2300                         new_bh = NULL;
2301                 }
2302         }
2303         if (S_ISDIR(old_inode->i_mode)) {
2304                 if (new_inode) {
2305                         retval = -ENOTEMPTY;
2306                         if (!empty_dir (new_inode))
2307                                 goto end_rename;
2308                 }
2309                 retval = -EIO;
2310                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2311                 if (!dir_bh)
2312                         goto end_rename;
2313                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2314                         goto end_rename;
2315                 retval = -EMLINK;
2316                 if (!new_inode && new_dir!=old_dir &&
2317                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2318                         goto end_rename;
2319         }
2320         if (!new_bh) {
2321                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2322                 if (retval)
2323                         goto end_rename;
2324         } else {
2325                 BUFFER_TRACE(new_bh, "get write access");
2326                 ext3_journal_get_write_access(handle, new_bh);
2327                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2328                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2329                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2330                         new_de->file_type = old_de->file_type;
2331                 new_dir->i_version++;
2332                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2333                 ext3_journal_dirty_metadata(handle, new_bh);
2334                 brelse(new_bh);
2335                 new_bh = NULL;
2336         }
2337
2338         /*
2339          * Like most other Unix systems, set the ctime for inodes on a
2340          * rename.
2341          */
2342         old_inode->i_ctime = CURRENT_TIME_SEC;
2343         ext3_mark_inode_dirty(handle, old_inode);
2344
2345         /*
2346          * ok, that's it
2347          */
2348         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2349             old_de->name_len != old_dentry->d_name.len ||
2350             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2351             (retval = ext3_delete_entry(handle, old_dir,
2352                                         old_de, old_bh)) == -ENOENT) {
2353                 /* old_de could have moved from under us during htree split, so
2354                  * make sure that we are deleting the right entry.  We might
2355                  * also be pointing to a stale entry in the unused part of
2356                  * old_bh so just checking inum and the name isn't enough. */
2357                 struct buffer_head *old_bh2;
2358                 struct ext3_dir_entry_2 *old_de2;
2359
2360                 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2361                 if (old_bh2) {
2362                         retval = ext3_delete_entry(handle, old_dir,
2363                                                    old_de2, old_bh2);
2364                         brelse(old_bh2);
2365                 }
2366         }
2367         if (retval) {
2368                 ext3_warning(old_dir->i_sb, "ext3_rename",
2369                                 "Deleting old file (%lu), %d, error=%d",
2370                                 old_dir->i_ino, old_dir->i_nlink, retval);
2371         }
2372
2373         if (new_inode) {
2374                 drop_nlink(new_inode);
2375                 new_inode->i_ctime = CURRENT_TIME_SEC;
2376         }
2377         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2378         ext3_update_dx_flag(old_dir);
2379         if (dir_bh) {
2380                 BUFFER_TRACE(dir_bh, "get_write_access");
2381                 ext3_journal_get_write_access(handle, dir_bh);
2382                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2383                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2384                 ext3_journal_dirty_metadata(handle, dir_bh);
2385                 drop_nlink(old_dir);
2386                 if (new_inode) {
2387                         drop_nlink(new_inode);
2388                 } else {
2389                         inc_nlink(new_dir);
2390                         ext3_update_dx_flag(new_dir);
2391                         ext3_mark_inode_dirty(handle, new_dir);
2392                 }
2393         }
2394         ext3_mark_inode_dirty(handle, old_dir);
2395         if (new_inode) {
2396                 ext3_mark_inode_dirty(handle, new_inode);
2397                 if (!new_inode->i_nlink)
2398                         ext3_orphan_add(handle, new_inode);
2399         }
2400         retval = 0;
2401
2402 end_rename:
2403         brelse (dir_bh);
2404         brelse (old_bh);
2405         brelse (new_bh);
2406         ext3_journal_stop(handle);
2407         return retval;
2408 }
2409
2410 /*
2411  * directories can handle most operations...
2412  */
2413 const struct inode_operations ext3_dir_inode_operations = {
2414         .create         = ext3_create,
2415         .lookup         = ext3_lookup,
2416         .link           = ext3_link,
2417         .unlink         = ext3_unlink,
2418         .symlink        = ext3_symlink,
2419         .mkdir          = ext3_mkdir,
2420         .rmdir          = ext3_rmdir,
2421         .mknod          = ext3_mknod,
2422         .rename         = ext3_rename,
2423         .setattr        = ext3_setattr,
2424 #ifdef CONFIG_EXT3_FS_XATTR
2425         .setxattr       = generic_setxattr,
2426         .getxattr       = generic_getxattr,
2427         .listxattr      = ext3_listxattr,
2428         .removexattr    = generic_removexattr,
2429 #endif
2430         .permission     = ext3_permission,
2431 };
2432
2433 const struct inode_operations ext3_special_inode_operations = {
2434         .setattr        = ext3_setattr,
2435 #ifdef CONFIG_EXT3_FS_XATTR
2436         .setxattr       = generic_setxattr,
2437         .getxattr       = generic_getxattr,
2438         .listxattr      = ext3_listxattr,
2439         .removexattr    = generic_removexattr,
2440 #endif
2441         .permission     = ext3_permission,
2442 };