]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/ext3/namei.c
Merge branch 'linus' into core/softlockup
[linux-2.6-omap-h63xx.git] / 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         bh = ext3_bread(handle, inode, *block, 1, err);
61         if (bh) {
62                 inode->i_size += inode->i_sb->s_blocksize;
63                 EXT3_I(inode)->i_disksize = inode->i_size;
64                 *err = ext3_journal_get_write_access(handle, bh);
65                 if (*err) {
66                         brelse(bh);
67                         bh = NULL;
68                 }
69         }
70         return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #ifdef DX_DEBUG
78 #define dxtrace(command) command
79 #else
80 #define dxtrace(command)
81 #endif
82
83 struct fake_dirent
84 {
85         __le32 inode;
86         __le16 rec_len;
87         u8 name_len;
88         u8 file_type;
89 };
90
91 struct dx_countlimit
92 {
93         __le16 limit;
94         __le16 count;
95 };
96
97 struct dx_entry
98 {
99         __le32 hash;
100         __le32 block;
101 };
102
103 /*
104  * dx_root_info is laid out so that if it should somehow get overlaid by a
105  * dirent the two low bits of the hash version will be zero.  Therefore, the
106  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
107  */
108
109 struct dx_root
110 {
111         struct fake_dirent dot;
112         char dot_name[4];
113         struct fake_dirent dotdot;
114         char dotdot_name[4];
115         struct dx_root_info
116         {
117                 __le32 reserved_zero;
118                 u8 hash_version;
119                 u8 info_length; /* 8 */
120                 u8 indirect_levels;
121                 u8 unused_flags;
122         }
123         info;
124         struct dx_entry entries[0];
125 };
126
127 struct dx_node
128 {
129         struct fake_dirent fake;
130         struct dx_entry entries[0];
131 };
132
133
134 struct dx_frame
135 {
136         struct buffer_head *bh;
137         struct dx_entry *entries;
138         struct dx_entry *at;
139 };
140
141 struct dx_map_entry
142 {
143         u32 hash;
144         u16 offs;
145         u16 size;
146 };
147
148 static inline unsigned dx_get_block (struct dx_entry *entry);
149 static void dx_set_block (struct dx_entry *entry, unsigned value);
150 static inline unsigned dx_get_hash (struct dx_entry *entry);
151 static void dx_set_hash (struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count (struct dx_entry *entries);
153 static unsigned dx_get_limit (struct dx_entry *entries);
154 static void dx_set_count (struct dx_entry *entries, unsigned value);
155 static void dx_set_limit (struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit (struct inode *dir);
158 static struct dx_frame *dx_probe(struct qstr *entry,
159                                  struct inode *dir,
160                                  struct dx_hash_info *hinfo,
161                                  struct dx_frame *frame,
162                                  int *err);
163 static void dx_release (struct dx_frame *frames);
164 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
165                         struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168                 struct dx_map_entry *offsets, int count);
169 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
170 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172                                  struct dx_frame *frame,
173                                  struct dx_frame *frames,
174                                  __u32 *start_hash);
175 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
176                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
177                         int *err);
178 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
179                              struct inode *inode);
180
181 /*
182  * p is at least 6 bytes before the end of page
183  */
184 static inline struct ext3_dir_entry_2 *
185 ext3_next_entry(struct ext3_dir_entry_2 *p)
186 {
187         return (struct ext3_dir_entry_2 *)((char *)p +
188                 ext3_rec_len_from_disk(p->rec_len));
189 }
190
191 /*
192  * Future: use high four bits of block for coalesce-on-delete flags
193  * Mask them off for now.
194  */
195
196 static inline unsigned dx_get_block (struct dx_entry *entry)
197 {
198         return le32_to_cpu(entry->block) & 0x00ffffff;
199 }
200
201 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
202 {
203         entry->block = cpu_to_le32(value);
204 }
205
206 static inline unsigned dx_get_hash (struct dx_entry *entry)
207 {
208         return le32_to_cpu(entry->hash);
209 }
210
211 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
212 {
213         entry->hash = cpu_to_le32(value);
214 }
215
216 static inline unsigned dx_get_count (struct dx_entry *entries)
217 {
218         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
219 }
220
221 static inline unsigned dx_get_limit (struct dx_entry *entries)
222 {
223         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
224 }
225
226 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
227 {
228         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
229 }
230
231 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
232 {
233         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
234 }
235
236 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
237 {
238         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
239                 EXT3_DIR_REC_LEN(2) - infosize;
240         return entry_space / sizeof(struct dx_entry);
241 }
242
243 static inline unsigned dx_node_limit (struct inode *dir)
244 {
245         unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
246         return entry_space / sizeof(struct dx_entry);
247 }
248
249 /*
250  * Debug
251  */
252 #ifdef DX_DEBUG
253 static void dx_show_index (char * label, struct dx_entry *entries)
254 {
255         int i, n = dx_get_count (entries);
256         printk("%s index ", label);
257         for (i = 0; i < n; i++)
258         {
259                 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
260         }
261         printk("\n");
262 }
263
264 struct stats
265 {
266         unsigned names;
267         unsigned space;
268         unsigned bcount;
269 };
270
271 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
272                                  int size, int show_names)
273 {
274         unsigned names = 0, space = 0;
275         char *base = (char *) de;
276         struct dx_hash_info h = *hinfo;
277
278         printk("names: ");
279         while ((char *) de < base + size)
280         {
281                 if (de->inode)
282                 {
283                         if (show_names)
284                         {
285                                 int len = de->name_len;
286                                 char *name = de->name;
287                                 while (len--) printk("%c", *name++);
288                                 ext3fs_dirhash(de->name, de->name_len, &h);
289                                 printk(":%x.%u ", h.hash,
290                                        ((char *) de - base));
291                         }
292                         space += EXT3_DIR_REC_LEN(de->name_len);
293                         names++;
294                 }
295                 de = ext3_next_entry(de);
296         }
297         printk("(%i)\n", names);
298         return (struct stats) { names, space, 1 };
299 }
300
301 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
302                              struct dx_entry *entries, int levels)
303 {
304         unsigned blocksize = dir->i_sb->s_blocksize;
305         unsigned count = dx_get_count (entries), names = 0, space = 0, i;
306         unsigned bcount = 0;
307         struct buffer_head *bh;
308         int err;
309         printk("%i indexed blocks...\n", count);
310         for (i = 0; i < count; i++, entries++)
311         {
312                 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
313                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
314                 struct stats stats;
315                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
316                 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
317                 stats = levels?
318                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
319                    dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
320                 names += stats.names;
321                 space += stats.space;
322                 bcount += stats.bcount;
323                 brelse (bh);
324         }
325         if (bcount)
326                 printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
327                         names, space/bcount,(space/bcount)*100/blocksize);
328         return (struct stats) { names, space, bcount};
329 }
330 #endif /* DX_DEBUG */
331
332 /*
333  * Probe for a directory leaf block to search.
334  *
335  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
336  * error in the directory index, and the caller should fall back to
337  * searching the directory normally.  The callers of dx_probe **MUST**
338  * check for this error code, and make sure it never gets reflected
339  * back to userspace.
340  */
341 static struct dx_frame *
342 dx_probe(struct qstr *entry, struct inode *dir,
343          struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
344 {
345         unsigned count, indirect;
346         struct dx_entry *at, *entries, *p, *q, *m;
347         struct dx_root *root;
348         struct buffer_head *bh;
349         struct dx_frame *frame = frame_in;
350         u32 hash;
351
352         frame->bh = NULL;
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, __func__,
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         if (hinfo->hash_version <= DX_HASH_TEA)
368                 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
369         hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
370         if (entry)
371                 ext3fs_dirhash(entry->name, entry->len, hinfo);
372         hash = hinfo->hash;
373
374         if (root->info.unused_flags & 1) {
375                 ext3_warning(dir->i_sb, __func__,
376                              "Unimplemented inode hash flags: %#06x",
377                              root->info.unused_flags);
378                 brelse(bh);
379                 *err = ERR_BAD_DX_DIR;
380                 goto fail;
381         }
382
383         if ((indirect = root->info.indirect_levels) > 1) {
384                 ext3_warning(dir->i_sb, __func__,
385                              "Unimplemented inode hash depth: %#06x",
386                              root->info.indirect_levels);
387                 brelse(bh);
388                 *err = ERR_BAD_DX_DIR;
389                 goto fail;
390         }
391
392         entries = (struct dx_entry *) (((char *)&root->info) +
393                                        root->info.info_length);
394
395         if (dx_get_limit(entries) != dx_root_limit(dir,
396                                                    root->info.info_length)) {
397                 ext3_warning(dir->i_sb, __func__,
398                              "dx entry: limit != root limit");
399                 brelse(bh);
400                 *err = ERR_BAD_DX_DIR;
401                 goto fail;
402         }
403
404         dxtrace (printk("Look up %x", hash));
405         while (1)
406         {
407                 count = dx_get_count(entries);
408                 if (!count || count > dx_get_limit(entries)) {
409                         ext3_warning(dir->i_sb, __func__,
410                                      "dx entry: no count or count > limit");
411                         brelse(bh);
412                         *err = ERR_BAD_DX_DIR;
413                         goto fail2;
414                 }
415
416                 p = entries + 1;
417                 q = entries + count - 1;
418                 while (p <= q)
419                 {
420                         m = p + (q - p)/2;
421                         dxtrace(printk("."));
422                         if (dx_get_hash(m) > hash)
423                                 q = m - 1;
424                         else
425                                 p = m + 1;
426                 }
427
428                 if (0) // linear search cross check
429                 {
430                         unsigned n = count - 1;
431                         at = entries;
432                         while (n--)
433                         {
434                                 dxtrace(printk(","));
435                                 if (dx_get_hash(++at) > hash)
436                                 {
437                                         at--;
438                                         break;
439                                 }
440                         }
441                         assert (at == p - 1);
442                 }
443
444                 at = p - 1;
445                 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446                 frame->bh = bh;
447                 frame->entries = entries;
448                 frame->at = at;
449                 if (!indirect--) return frame;
450                 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
451                         goto fail2;
452                 at = entries = ((struct dx_node *) bh->b_data)->entries;
453                 if (dx_get_limit(entries) != dx_node_limit (dir)) {
454                         ext3_warning(dir->i_sb, __func__,
455                                      "dx entry: limit != node limit");
456                         brelse(bh);
457                         *err = ERR_BAD_DX_DIR;
458                         goto fail2;
459                 }
460                 frame++;
461                 frame->bh = NULL;
462         }
463 fail2:
464         while (frame >= frame_in) {
465                 brelse(frame->bh);
466                 frame--;
467         }
468 fail:
469         if (*err == ERR_BAD_DX_DIR)
470                 ext3_warning(dir->i_sb, __func__,
471                              "Corrupt dir inode %ld, running e2fsck is "
472                              "recommended.", dir->i_ino);
473         return NULL;
474 }
475
476 static void dx_release (struct dx_frame *frames)
477 {
478         if (frames[0].bh == NULL)
479                 return;
480
481         if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482                 brelse(frames[1].bh);
483         brelse(frames[0].bh);
484 }
485
486 /*
487  * This function increments the frame pointer to search the next leaf
488  * block, and reads in the necessary intervening nodes if the search
489  * should be necessary.  Whether or not the search is necessary is
490  * controlled by the hash parameter.  If the hash value is even, then
491  * the search is only continued if the next block starts with that
492  * hash value.  This is used if we are searching for a specific file.
493  *
494  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
495  *
496  * This function returns 1 if the caller should continue to search,
497  * or 0 if it should not.  If there is an error reading one of the
498  * index blocks, it will a negative error code.
499  *
500  * If start_hash is non-null, it will be filled in with the starting
501  * hash of the next page.
502  */
503 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
504                                  struct dx_frame *frame,
505                                  struct dx_frame *frames,
506                                  __u32 *start_hash)
507 {
508         struct dx_frame *p;
509         struct buffer_head *bh;
510         int err, num_frames = 0;
511         __u32 bhash;
512
513         p = frame;
514         /*
515          * Find the next leaf page by incrementing the frame pointer.
516          * If we run out of entries in the interior node, loop around and
517          * increment pointer in the parent node.  When we break out of
518          * this loop, num_frames indicates the number of interior
519          * nodes need to be read.
520          */
521         while (1) {
522                 if (++(p->at) < p->entries + dx_get_count(p->entries))
523                         break;
524                 if (p == frames)
525                         return 0;
526                 num_frames++;
527                 p--;
528         }
529
530         /*
531          * If the hash is 1, then continue only if the next page has a
532          * continuation hash of any value.  This is used for readdir
533          * handling.  Otherwise, check to see if the hash matches the
534          * desired contiuation hash.  If it doesn't, return since
535          * there's no point to read in the successive index pages.
536          */
537         bhash = dx_get_hash(p->at);
538         if (start_hash)
539                 *start_hash = bhash;
540         if ((hash & 1) == 0) {
541                 if ((bhash & ~1) != hash)
542                         return 0;
543         }
544         /*
545          * If the hash is HASH_NB_ALWAYS, we always go to the next
546          * block so no check is necessary
547          */
548         while (num_frames--) {
549                 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
550                                       0, &err)))
551                         return err; /* Failure */
552                 p++;
553                 brelse (p->bh);
554                 p->bh = bh;
555                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
556         }
557         return 1;
558 }
559
560
561 /*
562  * This function fills a red-black tree with information from a
563  * directory block.  It returns the number directory entries loaded
564  * into the tree.  If there is an error it is returned in err.
565  */
566 static int htree_dirblock_to_tree(struct file *dir_file,
567                                   struct inode *dir, int block,
568                                   struct dx_hash_info *hinfo,
569                                   __u32 start_hash, __u32 start_minor_hash)
570 {
571         struct buffer_head *bh;
572         struct ext3_dir_entry_2 *de, *top;
573         int err, count = 0;
574
575         dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
576         if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
577                 return err;
578
579         de = (struct ext3_dir_entry_2 *) bh->b_data;
580         top = (struct ext3_dir_entry_2 *) ((char *) de +
581                                            dir->i_sb->s_blocksize -
582                                            EXT3_DIR_REC_LEN(0));
583         for (; de < top; de = ext3_next_entry(de)) {
584                 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
585                                         (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
586                                                 +((char *)de - bh->b_data))) {
587                         /* On error, skip the f_pos to the next block. */
588                         dir_file->f_pos = (dir_file->f_pos |
589                                         (dir->i_sb->s_blocksize - 1)) + 1;
590                         brelse (bh);
591                         return count;
592                 }
593                 ext3fs_dirhash(de->name, de->name_len, hinfo);
594                 if ((hinfo->hash < start_hash) ||
595                     ((hinfo->hash == start_hash) &&
596                      (hinfo->minor_hash < start_minor_hash)))
597                         continue;
598                 if (de->inode == 0)
599                         continue;
600                 if ((err = ext3_htree_store_dirent(dir_file,
601                                    hinfo->hash, hinfo->minor_hash, de)) != 0) {
602                         brelse(bh);
603                         return err;
604                 }
605                 count++;
606         }
607         brelse(bh);
608         return count;
609 }
610
611
612 /*
613  * This function fills a red-black tree with information from a
614  * directory.  We start scanning the directory in hash order, starting
615  * at start_hash and start_minor_hash.
616  *
617  * This function returns the number of entries inserted into the tree,
618  * or a negative error code.
619  */
620 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
621                          __u32 start_minor_hash, __u32 *next_hash)
622 {
623         struct dx_hash_info hinfo;
624         struct ext3_dir_entry_2 *de;
625         struct dx_frame frames[2], *frame;
626         struct inode *dir;
627         int block, err;
628         int count = 0;
629         int ret;
630         __u32 hashval;
631
632         dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
633                        start_minor_hash));
634         dir = dir_file->f_path.dentry->d_inode;
635         if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
636                 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
637                 if (hinfo.hash_version <= DX_HASH_TEA)
638                         hinfo.hash_version +=
639                                 EXT3_SB(dir->i_sb)->s_hash_unsigned;
640                 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
641                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642                                                start_hash, start_minor_hash);
643                 *next_hash = ~0;
644                 return count;
645         }
646         hinfo.hash = start_hash;
647         hinfo.minor_hash = 0;
648         frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
649         if (!frame)
650                 return err;
651
652         /* Add '.' and '..' from the htree header */
653         if (!start_hash && !start_minor_hash) {
654                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655                 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
656                         goto errout;
657                 count++;
658         }
659         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660                 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
661                 de = ext3_next_entry(de);
662                 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
663                         goto errout;
664                 count++;
665         }
666
667         while (1) {
668                 block = dx_get_block(frame->at);
669                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670                                              start_hash, start_minor_hash);
671                 if (ret < 0) {
672                         err = ret;
673                         goto errout;
674                 }
675                 count += ret;
676                 hashval = ~0;
677                 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
678                                             frame, frames, &hashval);
679                 *next_hash = hashval;
680                 if (ret < 0) {
681                         err = ret;
682                         goto errout;
683                 }
684                 /*
685                  * Stop if:  (a) there are no more entries, or
686                  * (b) we have inserted at least one entry and the
687                  * next hash value is not a continuation
688                  */
689                 if ((ret == 0) ||
690                     (count && ((hashval & 1) == 0)))
691                         break;
692         }
693         dx_release(frames);
694         dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
695                        count, *next_hash));
696         return count;
697 errout:
698         dx_release(frames);
699         return (err);
700 }
701
702
703 /*
704  * Directory block splitting, compacting
705  */
706
707 /*
708  * Create map of hash values, offsets, and sizes, stored at end of block.
709  * Returns number of entries mapped.
710  */
711 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
712                 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
713 {
714         int count = 0;
715         char *base = (char *) de;
716         struct dx_hash_info h = *hinfo;
717
718         while ((char *) de < base + blocksize)
719         {
720                 if (de->name_len && de->inode) {
721                         ext3fs_dirhash(de->name, de->name_len, &h);
722                         map_tail--;
723                         map_tail->hash = h.hash;
724                         map_tail->offs = (u16) ((char *) de - base);
725                         map_tail->size = le16_to_cpu(de->rec_len);
726                         count++;
727                         cond_resched();
728                 }
729                 /* XXX: do we need to check rec_len == 0 case? -Chris */
730                 de = ext3_next_entry(de);
731         }
732         return count;
733 }
734
735 /* Sort map by hash value */
736 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737 {
738         struct dx_map_entry *p, *q, *top = map + count - 1;
739         int more;
740         /* Combsort until bubble sort doesn't suck */
741         while (count > 2)
742         {
743                 count = count*10/13;
744                 if (count - 9 < 2) /* 9, 10 -> 11 */
745                         count = 11;
746                 for (p = top, q = p - count; q >= map; p--, q--)
747                         if (p->hash < q->hash)
748                                 swap(*p, *q);
749         }
750         /* Garden variety bubble sort */
751         do {
752                 more = 0;
753                 q = top;
754                 while (q-- > map)
755                 {
756                         if (q[1].hash >= q[0].hash)
757                                 continue;
758                         swap(*(q+1), *q);
759                         more = 1;
760                 }
761         } while(more);
762 }
763
764 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
765 {
766         struct dx_entry *entries = frame->entries;
767         struct dx_entry *old = frame->at, *new = old + 1;
768         int count = dx_get_count(entries);
769
770         assert(count < dx_get_limit(entries));
771         assert(old < entries + count);
772         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773         dx_set_hash(new, hash);
774         dx_set_block(new, block);
775         dx_set_count(entries, count + 1);
776 }
777
778 static void ext3_update_dx_flag(struct inode *inode)
779 {
780         if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
781                                      EXT3_FEATURE_COMPAT_DIR_INDEX))
782                 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
783 }
784
785 /*
786  * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
787  *
788  * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789  * `de != NULL' is guaranteed by caller.
790  */
791 static inline int ext3_match (int len, const char * const name,
792                               struct ext3_dir_entry_2 * de)
793 {
794         if (len != de->name_len)
795                 return 0;
796         if (!de->inode)
797                 return 0;
798         return !memcmp(name, de->name, len);
799 }
800
801 /*
802  * Returns 0 if not found, -1 on failure, and 1 on success
803  */
804 static inline int search_dirblock(struct buffer_head * bh,
805                                   struct inode *dir,
806                                   struct qstr *child,
807                                   unsigned long offset,
808                                   struct ext3_dir_entry_2 ** res_dir)
809 {
810         struct ext3_dir_entry_2 * de;
811         char * dlimit;
812         int de_len;
813         const char *name = child->name;
814         int namelen = child->len;
815
816         de = (struct ext3_dir_entry_2 *) bh->b_data;
817         dlimit = bh->b_data + dir->i_sb->s_blocksize;
818         while ((char *) de < dlimit) {
819                 /* this code is executed quadratically often */
820                 /* do minimal checking `by hand' */
821
822                 if ((char *) de + namelen <= dlimit &&
823                     ext3_match (namelen, name, de)) {
824                         /* found a match - just to be sure, do a full check */
825                         if (!ext3_check_dir_entry("ext3_find_entry",
826                                                   dir, de, bh, offset))
827                                 return -1;
828                         *res_dir = de;
829                         return 1;
830                 }
831                 /* prevent looping on a bad block */
832                 de_len = ext3_rec_len_from_disk(de->rec_len);
833                 if (de_len <= 0)
834                         return -1;
835                 offset += de_len;
836                 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
837         }
838         return 0;
839 }
840
841
842 /*
843  *      ext3_find_entry()
844  *
845  * finds an entry in the specified directory with the wanted name. It
846  * returns the cache buffer in which the entry was found, and the entry
847  * itself (as a parameter - res_dir). It does NOT read the inode of the
848  * entry - you'll have to do that yourself if you want to.
849  *
850  * The returned buffer_head has ->b_count elevated.  The caller is expected
851  * to brelse() it when appropriate.
852  */
853 static struct buffer_head *ext3_find_entry(struct inode *dir,
854                                         struct qstr *entry,
855                                         struct ext3_dir_entry_2 **res_dir)
856 {
857         struct super_block * sb;
858         struct buffer_head * bh_use[NAMEI_RA_SIZE];
859         struct buffer_head * bh, *ret = NULL;
860         unsigned long start, block, b;
861         int ra_max = 0;         /* Number of bh's in the readahead
862                                    buffer, bh_use[] */
863         int ra_ptr = 0;         /* Current index into readahead
864                                    buffer */
865         int num = 0;
866         int nblocks, i, err;
867         int namelen;
868
869         *res_dir = NULL;
870         sb = dir->i_sb;
871         namelen = entry->len;
872         if (namelen > EXT3_NAME_LEN)
873                 return NULL;
874         if (is_dx(dir)) {
875                 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
876                 /*
877                  * On success, or if the error was file not found,
878                  * return.  Otherwise, fall back to doing a search the
879                  * old fashioned way.
880                  */
881                 if (bh || (err != ERR_BAD_DX_DIR))
882                         return bh;
883                 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
884         }
885         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886         start = EXT3_I(dir)->i_dir_start_lookup;
887         if (start >= nblocks)
888                 start = 0;
889         block = start;
890 restart:
891         do {
892                 /*
893                  * We deal with the read-ahead logic here.
894                  */
895                 if (ra_ptr >= ra_max) {
896                         /* Refill the readahead buffer */
897                         ra_ptr = 0;
898                         b = block;
899                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900                                 /*
901                                  * Terminate if we reach the end of the
902                                  * directory and must wrap, or if our
903                                  * search has finished at this block.
904                                  */
905                                 if (b >= nblocks || (num && block == start)) {
906                                         bh_use[ra_max] = NULL;
907                                         break;
908                                 }
909                                 num++;
910                                 bh = ext3_getblk(NULL, dir, b++, 0, &err);
911                                 bh_use[ra_max] = bh;
912                                 if (bh)
913                                         ll_rw_block(READ_META, 1, &bh);
914                         }
915                 }
916                 if ((bh = bh_use[ra_ptr++]) == NULL)
917                         goto next;
918                 wait_on_buffer(bh);
919                 if (!buffer_uptodate(bh)) {
920                         /* read error, skip block & hope for the best */
921                         ext3_error(sb, __func__, "reading directory #%lu "
922                                    "offset %lu", dir->i_ino, block);
923                         brelse(bh);
924                         goto next;
925                 }
926                 i = search_dirblock(bh, dir, entry,
927                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
928                 if (i == 1) {
929                         EXT3_I(dir)->i_dir_start_lookup = block;
930                         ret = bh;
931                         goto cleanup_and_exit;
932                 } else {
933                         brelse(bh);
934                         if (i < 0)
935                                 goto cleanup_and_exit;
936                 }
937         next:
938                 if (++block >= nblocks)
939                         block = 0;
940         } while (block != start);
941
942         /*
943          * If the directory has grown while we were searching, then
944          * search the last part of the directory before giving up.
945          */
946         block = nblocks;
947         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
948         if (block < nblocks) {
949                 start = 0;
950                 goto restart;
951         }
952
953 cleanup_and_exit:
954         /* Clean up the read-ahead blocks */
955         for (; ra_ptr < ra_max; ra_ptr++)
956                 brelse (bh_use[ra_ptr]);
957         return ret;
958 }
959
960 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
961                         struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
962                         int *err)
963 {
964         struct super_block * sb;
965         struct dx_hash_info     hinfo;
966         u32 hash;
967         struct dx_frame frames[2], *frame;
968         struct ext3_dir_entry_2 *de, *top;
969         struct buffer_head *bh;
970         unsigned long block;
971         int retval;
972         int namelen = entry->len;
973         const u8 *name = entry->name;
974
975         sb = dir->i_sb;
976         /* NFS may look up ".." - look at dx_root directory block */
977         if (namelen > 2 || name[0] != '.'|| (namelen == 2 && name[1] != '.')) {
978                 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
979                         return NULL;
980         } else {
981                 frame = frames;
982                 frame->bh = NULL;                       /* for dx_release() */
983                 frame->at = (struct dx_entry *)frames;  /* hack for zero entry*/
984                 dx_set_block(frame->at, 0);             /* dx_root block is 0 */
985         }
986         hash = hinfo.hash;
987         do {
988                 block = dx_get_block(frame->at);
989                 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
990                         goto errout;
991                 de = (struct ext3_dir_entry_2 *) bh->b_data;
992                 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
993                                        EXT3_DIR_REC_LEN(0));
994                 for (; de < top; de = ext3_next_entry(de)) {
995                         int off = (block << EXT3_BLOCK_SIZE_BITS(sb))
996                                   + ((char *) de - bh->b_data);
997
998                         if (!ext3_check_dir_entry(__func__, dir, de, bh, off)) {
999                                 brelse(bh);
1000                                 *err = ERR_BAD_DX_DIR;
1001                                 goto errout;
1002                         }
1003
1004                         if (ext3_match(namelen, name, de)) {
1005                                 *res_dir = de;
1006                                 dx_release(frames);
1007                                 return bh;
1008                         }
1009                 }
1010                 brelse (bh);
1011                 /* Check to see if we should continue to search */
1012                 retval = ext3_htree_next_block(dir, hash, frame,
1013                                                frames, NULL);
1014                 if (retval < 0) {
1015                         ext3_warning(sb, __func__,
1016                              "error reading index page in directory #%lu",
1017                              dir->i_ino);
1018                         *err = retval;
1019                         goto errout;
1020                 }
1021         } while (retval == 1);
1022
1023         *err = -ENOENT;
1024 errout:
1025         dxtrace(printk("%s not found\n", name));
1026         dx_release (frames);
1027         return NULL;
1028 }
1029
1030 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1031 {
1032         struct inode * inode;
1033         struct ext3_dir_entry_2 * de;
1034         struct buffer_head * bh;
1035
1036         if (dentry->d_name.len > EXT3_NAME_LEN)
1037                 return ERR_PTR(-ENAMETOOLONG);
1038
1039         bh = ext3_find_entry(dir, &dentry->d_name, &de);
1040         inode = NULL;
1041         if (bh) {
1042                 unsigned long ino = le32_to_cpu(de->inode);
1043                 brelse (bh);
1044                 if (!ext3_valid_inum(dir->i_sb, ino)) {
1045                         ext3_error(dir->i_sb, "ext3_lookup",
1046                                    "bad inode number: %lu", ino);
1047                         return ERR_PTR(-EIO);
1048                 }
1049                 inode = ext3_iget(dir->i_sb, ino);
1050                 if (unlikely(IS_ERR(inode))) {
1051                         if (PTR_ERR(inode) == -ESTALE) {
1052                                 ext3_error(dir->i_sb, __func__,
1053                                                 "deleted inode referenced: %lu",
1054                                                 ino);
1055                                 return ERR_PTR(-EIO);
1056                         } else {
1057                                 return ERR_CAST(inode);
1058                         }
1059                 }
1060         }
1061         return d_splice_alias(inode, dentry);
1062 }
1063
1064
1065 struct dentry *ext3_get_parent(struct dentry *child)
1066 {
1067         unsigned long ino;
1068         struct qstr dotdot = {.name = "..", .len = 2};
1069         struct ext3_dir_entry_2 * de;
1070         struct buffer_head *bh;
1071
1072         bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1073         if (!bh)
1074                 return ERR_PTR(-ENOENT);
1075         ino = le32_to_cpu(de->inode);
1076         brelse(bh);
1077
1078         if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1079                 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1080                            "bad inode number: %lu", ino);
1081                 return ERR_PTR(-EIO);
1082         }
1083
1084         return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1085 }
1086
1087 #define S_SHIFT 12
1088 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1089         [S_IFREG >> S_SHIFT]    = EXT3_FT_REG_FILE,
1090         [S_IFDIR >> S_SHIFT]    = EXT3_FT_DIR,
1091         [S_IFCHR >> S_SHIFT]    = EXT3_FT_CHRDEV,
1092         [S_IFBLK >> S_SHIFT]    = EXT3_FT_BLKDEV,
1093         [S_IFIFO >> S_SHIFT]    = EXT3_FT_FIFO,
1094         [S_IFSOCK >> S_SHIFT]   = EXT3_FT_SOCK,
1095         [S_IFLNK >> S_SHIFT]    = EXT3_FT_SYMLINK,
1096 };
1097
1098 static inline void ext3_set_de_type(struct super_block *sb,
1099                                 struct ext3_dir_entry_2 *de,
1100                                 umode_t mode) {
1101         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1102                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1103 }
1104
1105 /*
1106  * Move count entries from end of map between two memory locations.
1107  * Returns pointer to last entry moved.
1108  */
1109 static struct ext3_dir_entry_2 *
1110 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1111 {
1112         unsigned rec_len = 0;
1113
1114         while (count--) {
1115                 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1116                 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1117                 memcpy (to, de, rec_len);
1118                 ((struct ext3_dir_entry_2 *) to)->rec_len =
1119                                 ext3_rec_len_to_disk(rec_len);
1120                 de->inode = 0;
1121                 map++;
1122                 to += rec_len;
1123         }
1124         return (struct ext3_dir_entry_2 *) (to - rec_len);
1125 }
1126
1127 /*
1128  * Compact each dir entry in the range to the minimal rec_len.
1129  * Returns pointer to last entry in range.
1130  */
1131 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1132 {
1133         struct ext3_dir_entry_2 *next, *to, *prev;
1134         struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1135         unsigned rec_len = 0;
1136
1137         prev = to = de;
1138         while ((char *)de < base + blocksize) {
1139                 next = ext3_next_entry(de);
1140                 if (de->inode && de->name_len) {
1141                         rec_len = EXT3_DIR_REC_LEN(de->name_len);
1142                         if (de > to)
1143                                 memmove(to, de, rec_len);
1144                         to->rec_len = ext3_rec_len_to_disk(rec_len);
1145                         prev = to;
1146                         to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1147                 }
1148                 de = next;
1149         }
1150         return prev;
1151 }
1152
1153 /*
1154  * Split a full leaf block to make room for a new dir entry.
1155  * Allocate a new block, and move entries so that they are approx. equally full.
1156  * Returns pointer to de in block into which the new entry will be inserted.
1157  */
1158 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1159                         struct buffer_head **bh,struct dx_frame *frame,
1160                         struct dx_hash_info *hinfo, int *error)
1161 {
1162         unsigned blocksize = dir->i_sb->s_blocksize;
1163         unsigned count, continued;
1164         struct buffer_head *bh2;
1165         u32 newblock;
1166         u32 hash2;
1167         struct dx_map_entry *map;
1168         char *data1 = (*bh)->b_data, *data2;
1169         unsigned split, move, size;
1170         struct ext3_dir_entry_2 *de = NULL, *de2;
1171         int     err = 0, i;
1172
1173         bh2 = ext3_append (handle, dir, &newblock, &err);
1174         if (!(bh2)) {
1175                 brelse(*bh);
1176                 *bh = NULL;
1177                 goto errout;
1178         }
1179
1180         BUFFER_TRACE(*bh, "get_write_access");
1181         err = ext3_journal_get_write_access(handle, *bh);
1182         if (err)
1183                 goto journal_error;
1184
1185         BUFFER_TRACE(frame->bh, "get_write_access");
1186         err = ext3_journal_get_write_access(handle, frame->bh);
1187         if (err)
1188                 goto journal_error;
1189
1190         data2 = bh2->b_data;
1191
1192         /* create map in the end of data2 block */
1193         map = (struct dx_map_entry *) (data2 + blocksize);
1194         count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1195                              blocksize, hinfo, map);
1196         map -= count;
1197         dx_sort_map (map, count);
1198         /* Split the existing block in the middle, size-wise */
1199         size = 0;
1200         move = 0;
1201         for (i = count-1; i >= 0; i--) {
1202                 /* is more than half of this entry in 2nd half of the block? */
1203                 if (size + map[i].size/2 > blocksize/2)
1204                         break;
1205                 size += map[i].size;
1206                 move++;
1207         }
1208         /* map index at which we will split */
1209         split = count - move;
1210         hash2 = map[split].hash;
1211         continued = hash2 == map[split - 1].hash;
1212         dxtrace(printk("Split block %i at %x, %i/%i\n",
1213                 dx_get_block(frame->at), hash2, split, count-split));
1214
1215         /* Fancy dance to stay within two buffers */
1216         de2 = dx_move_dirents(data1, data2, map + split, count - split);
1217         de = dx_pack_dirents(data1,blocksize);
1218         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1219         de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1220         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1221         dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1222
1223         /* Which block gets the new entry? */
1224         if (hinfo->hash >= hash2)
1225         {
1226                 swap(*bh, bh2);
1227                 de = de2;
1228         }
1229         dx_insert_block (frame, hash2 + continued, newblock);
1230         err = ext3_journal_dirty_metadata (handle, bh2);
1231         if (err)
1232                 goto journal_error;
1233         err = ext3_journal_dirty_metadata (handle, frame->bh);
1234         if (err)
1235                 goto journal_error;
1236         brelse (bh2);
1237         dxtrace(dx_show_index ("frame", frame->entries));
1238         return de;
1239
1240 journal_error:
1241         brelse(*bh);
1242         brelse(bh2);
1243         *bh = NULL;
1244         ext3_std_error(dir->i_sb, err);
1245 errout:
1246         *error = err;
1247         return NULL;
1248 }
1249
1250
1251 /*
1252  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1253  * it points to a directory entry which is guaranteed to be large
1254  * enough for new directory entry.  If de is NULL, then
1255  * add_dirent_to_buf will attempt search the directory block for
1256  * space.  It will return -ENOSPC if no space is available, and -EIO
1257  * and -EEXIST if directory entry already exists.
1258  *
1259  * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1260  * all other cases bh is released.
1261  */
1262 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1263                              struct inode *inode, struct ext3_dir_entry_2 *de,
1264                              struct buffer_head * bh)
1265 {
1266         struct inode    *dir = dentry->d_parent->d_inode;
1267         const char      *name = dentry->d_name.name;
1268         int             namelen = dentry->d_name.len;
1269         unsigned long   offset = 0;
1270         unsigned short  reclen;
1271         int             nlen, rlen, err;
1272         char            *top;
1273
1274         reclen = EXT3_DIR_REC_LEN(namelen);
1275         if (!de) {
1276                 de = (struct ext3_dir_entry_2 *)bh->b_data;
1277                 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1278                 while ((char *) de <= top) {
1279                         if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1280                                                   bh, offset)) {
1281                                 brelse (bh);
1282                                 return -EIO;
1283                         }
1284                         if (ext3_match (namelen, name, de)) {
1285                                 brelse (bh);
1286                                 return -EEXIST;
1287                         }
1288                         nlen = EXT3_DIR_REC_LEN(de->name_len);
1289                         rlen = ext3_rec_len_from_disk(de->rec_len);
1290                         if ((de->inode? rlen - nlen: rlen) >= reclen)
1291                                 break;
1292                         de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1293                         offset += rlen;
1294                 }
1295                 if ((char *) de > top)
1296                         return -ENOSPC;
1297         }
1298         BUFFER_TRACE(bh, "get_write_access");
1299         err = ext3_journal_get_write_access(handle, bh);
1300         if (err) {
1301                 ext3_std_error(dir->i_sb, err);
1302                 brelse(bh);
1303                 return err;
1304         }
1305
1306         /* By now the buffer is marked for journaling */
1307         nlen = EXT3_DIR_REC_LEN(de->name_len);
1308         rlen = ext3_rec_len_from_disk(de->rec_len);
1309         if (de->inode) {
1310                 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1311                 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1312                 de->rec_len = ext3_rec_len_to_disk(nlen);
1313                 de = de1;
1314         }
1315         de->file_type = EXT3_FT_UNKNOWN;
1316         if (inode) {
1317                 de->inode = cpu_to_le32(inode->i_ino);
1318                 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1319         } else
1320                 de->inode = 0;
1321         de->name_len = namelen;
1322         memcpy (de->name, name, namelen);
1323         /*
1324          * XXX shouldn't update any times until successful
1325          * completion of syscall, but too many callers depend
1326          * on this.
1327          *
1328          * XXX similarly, too many callers depend on
1329          * ext3_new_inode() setting the times, but error
1330          * recovery deletes the inode, so the worst that can
1331          * happen is that the times are slightly out of date
1332          * and/or different from the directory change time.
1333          */
1334         dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1335         ext3_update_dx_flag(dir);
1336         dir->i_version++;
1337         ext3_mark_inode_dirty(handle, dir);
1338         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1339         err = ext3_journal_dirty_metadata(handle, bh);
1340         if (err)
1341                 ext3_std_error(dir->i_sb, err);
1342         brelse(bh);
1343         return 0;
1344 }
1345
1346 /*
1347  * This converts a one block unindexed directory to a 3 block indexed
1348  * directory, and adds the dentry to the indexed directory.
1349  */
1350 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1351                             struct inode *inode, struct buffer_head *bh)
1352 {
1353         struct inode    *dir = dentry->d_parent->d_inode;
1354         const char      *name = dentry->d_name.name;
1355         int             namelen = dentry->d_name.len;
1356         struct buffer_head *bh2;
1357         struct dx_root  *root;
1358         struct dx_frame frames[2], *frame;
1359         struct dx_entry *entries;
1360         struct ext3_dir_entry_2 *de, *de2;
1361         char            *data1, *top;
1362         unsigned        len;
1363         int             retval;
1364         unsigned        blocksize;
1365         struct dx_hash_info hinfo;
1366         u32             block;
1367         struct fake_dirent *fde;
1368
1369         blocksize =  dir->i_sb->s_blocksize;
1370         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1371         retval = ext3_journal_get_write_access(handle, bh);
1372         if (retval) {
1373                 ext3_std_error(dir->i_sb, retval);
1374                 brelse(bh);
1375                 return retval;
1376         }
1377         root = (struct dx_root *) bh->b_data;
1378
1379         /* The 0th block becomes the root, move the dirents out */
1380         fde = &root->dotdot;
1381         de = (struct ext3_dir_entry_2 *)((char *)fde +
1382                         ext3_rec_len_from_disk(fde->rec_len));
1383         if ((char *) de >= (((char *) root) + blocksize)) {
1384                 ext3_error(dir->i_sb, __func__,
1385                            "invalid rec_len for '..' in inode %lu",
1386                            dir->i_ino);
1387                 brelse(bh);
1388                 return -EIO;
1389         }
1390         len = ((char *) root) + blocksize - (char *) de;
1391
1392         bh2 = ext3_append (handle, dir, &block, &retval);
1393         if (!(bh2)) {
1394                 brelse(bh);
1395                 return retval;
1396         }
1397         EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1398         data1 = bh2->b_data;
1399
1400         memcpy (data1, de, len);
1401         de = (struct ext3_dir_entry_2 *) data1;
1402         top = data1 + len;
1403         while ((char *)(de2 = ext3_next_entry(de)) < top)
1404                 de = de2;
1405         de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1406         /* Initialize the root; the dot dirents already exist */
1407         de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1408         de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1409         memset (&root->info, 0, sizeof(root->info));
1410         root->info.info_length = sizeof(root->info);
1411         root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1412         entries = root->entries;
1413         dx_set_block (entries, 1);
1414         dx_set_count (entries, 1);
1415         dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1416
1417         /* Initialize as for dx_probe */
1418         hinfo.hash_version = root->info.hash_version;
1419         if (hinfo.hash_version <= DX_HASH_TEA)
1420                 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1421         hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1422         ext3fs_dirhash(name, namelen, &hinfo);
1423         frame = frames;
1424         frame->entries = entries;
1425         frame->at = entries;
1426         frame->bh = bh;
1427         bh = bh2;
1428         de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1429         dx_release (frames);
1430         if (!(de))
1431                 return retval;
1432
1433         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1434 }
1435
1436 /*
1437  *      ext3_add_entry()
1438  *
1439  * adds a file entry to the specified directory, using the same
1440  * semantics as ext3_find_entry(). It returns NULL if it failed.
1441  *
1442  * NOTE!! The inode part of 'de' is left at 0 - which means you
1443  * may not sleep between calling this and putting something into
1444  * the entry, as someone else might have used it while you slept.
1445  */
1446 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1447         struct inode *inode)
1448 {
1449         struct inode *dir = dentry->d_parent->d_inode;
1450         unsigned long offset;
1451         struct buffer_head * bh;
1452         struct ext3_dir_entry_2 *de;
1453         struct super_block * sb;
1454         int     retval;
1455         int     dx_fallback=0;
1456         unsigned blocksize;
1457         u32 block, blocks;
1458
1459         sb = dir->i_sb;
1460         blocksize = sb->s_blocksize;
1461         if (!dentry->d_name.len)
1462                 return -EINVAL;
1463         if (is_dx(dir)) {
1464                 retval = ext3_dx_add_entry(handle, dentry, inode);
1465                 if (!retval || (retval != ERR_BAD_DX_DIR))
1466                         return retval;
1467                 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1468                 dx_fallback++;
1469                 ext3_mark_inode_dirty(handle, dir);
1470         }
1471         blocks = dir->i_size >> sb->s_blocksize_bits;
1472         for (block = 0, offset = 0; block < blocks; block++) {
1473                 bh = ext3_bread(handle, dir, block, 0, &retval);
1474                 if(!bh)
1475                         return retval;
1476                 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477                 if (retval != -ENOSPC)
1478                         return retval;
1479
1480                 if (blocks == 1 && !dx_fallback &&
1481                     EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1482                         return make_indexed_dir(handle, dentry, inode, bh);
1483                 brelse(bh);
1484         }
1485         bh = ext3_append(handle, dir, &block, &retval);
1486         if (!bh)
1487                 return retval;
1488         de = (struct ext3_dir_entry_2 *) bh->b_data;
1489         de->inode = 0;
1490         de->rec_len = ext3_rec_len_to_disk(blocksize);
1491         return add_dirent_to_buf(handle, dentry, inode, de, bh);
1492 }
1493
1494 /*
1495  * Returns 0 for success, or a negative error value
1496  */
1497 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1498                              struct inode *inode)
1499 {
1500         struct dx_frame frames[2], *frame;
1501         struct dx_entry *entries, *at;
1502         struct dx_hash_info hinfo;
1503         struct buffer_head * bh;
1504         struct inode *dir = dentry->d_parent->d_inode;
1505         struct super_block * sb = dir->i_sb;
1506         struct ext3_dir_entry_2 *de;
1507         int err;
1508
1509         frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1510         if (!frame)
1511                 return err;
1512         entries = frame->entries;
1513         at = frame->at;
1514
1515         if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1516                 goto cleanup;
1517
1518         BUFFER_TRACE(bh, "get_write_access");
1519         err = ext3_journal_get_write_access(handle, bh);
1520         if (err)
1521                 goto journal_error;
1522
1523         err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1524         if (err != -ENOSPC) {
1525                 bh = NULL;
1526                 goto cleanup;
1527         }
1528
1529         /* Block full, should compress but for now just split */
1530         dxtrace(printk("using %u of %u node entries\n",
1531                        dx_get_count(entries), dx_get_limit(entries)));
1532         /* Need to split index? */
1533         if (dx_get_count(entries) == dx_get_limit(entries)) {
1534                 u32 newblock;
1535                 unsigned icount = dx_get_count(entries);
1536                 int levels = frame - frames;
1537                 struct dx_entry *entries2;
1538                 struct dx_node *node2;
1539                 struct buffer_head *bh2;
1540
1541                 if (levels && (dx_get_count(frames->entries) ==
1542                                dx_get_limit(frames->entries))) {
1543                         ext3_warning(sb, __func__,
1544                                      "Directory index full!");
1545                         err = -ENOSPC;
1546                         goto cleanup;
1547                 }
1548                 bh2 = ext3_append (handle, dir, &newblock, &err);
1549                 if (!(bh2))
1550                         goto cleanup;
1551                 node2 = (struct dx_node *)(bh2->b_data);
1552                 entries2 = node2->entries;
1553                 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1554                 node2->fake.inode = 0;
1555                 BUFFER_TRACE(frame->bh, "get_write_access");
1556                 err = ext3_journal_get_write_access(handle, frame->bh);
1557                 if (err)
1558                         goto journal_error;
1559                 if (levels) {
1560                         unsigned icount1 = icount/2, icount2 = icount - icount1;
1561                         unsigned hash2 = dx_get_hash(entries + icount1);
1562                         dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1563
1564                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1565                         err = ext3_journal_get_write_access(handle,
1566                                                              frames[0].bh);
1567                         if (err)
1568                                 goto journal_error;
1569
1570                         memcpy ((char *) entries2, (char *) (entries + icount1),
1571                                 icount2 * sizeof(struct dx_entry));
1572                         dx_set_count (entries, icount1);
1573                         dx_set_count (entries2, icount2);
1574                         dx_set_limit (entries2, dx_node_limit(dir));
1575
1576                         /* Which index block gets the new entry? */
1577                         if (at - entries >= icount1) {
1578                                 frame->at = at = at - entries - icount1 + entries2;
1579                                 frame->entries = entries = entries2;
1580                                 swap(frame->bh, bh2);
1581                         }
1582                         dx_insert_block (frames + 0, hash2, newblock);
1583                         dxtrace(dx_show_index ("node", frames[1].entries));
1584                         dxtrace(dx_show_index ("node",
1585                                ((struct dx_node *) bh2->b_data)->entries));
1586                         err = ext3_journal_dirty_metadata(handle, bh2);
1587                         if (err)
1588                                 goto journal_error;
1589                         brelse (bh2);
1590                 } else {
1591                         dxtrace(printk("Creating second level index...\n"));
1592                         memcpy((char *) entries2, (char *) entries,
1593                                icount * sizeof(struct dx_entry));
1594                         dx_set_limit(entries2, dx_node_limit(dir));
1595
1596                         /* Set up root */
1597                         dx_set_count(entries, 1);
1598                         dx_set_block(entries + 0, newblock);
1599                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1600
1601                         /* Add new access path frame */
1602                         frame = frames + 1;
1603                         frame->at = at = at - entries + entries2;
1604                         frame->entries = entries = entries2;
1605                         frame->bh = bh2;
1606                         err = ext3_journal_get_write_access(handle,
1607                                                              frame->bh);
1608                         if (err)
1609                                 goto journal_error;
1610                 }
1611                 ext3_journal_dirty_metadata(handle, frames[0].bh);
1612         }
1613         de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1614         if (!de)
1615                 goto cleanup;
1616         err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1617         bh = NULL;
1618         goto cleanup;
1619
1620 journal_error:
1621         ext3_std_error(dir->i_sb, err);
1622 cleanup:
1623         if (bh)
1624                 brelse(bh);
1625         dx_release(frames);
1626         return err;
1627 }
1628
1629 /*
1630  * ext3_delete_entry deletes a directory entry by merging it with the
1631  * previous entry
1632  */
1633 static int ext3_delete_entry (handle_t *handle,
1634                               struct inode * dir,
1635                               struct ext3_dir_entry_2 * de_del,
1636                               struct buffer_head * bh)
1637 {
1638         struct ext3_dir_entry_2 * de, * pde;
1639         int i;
1640
1641         i = 0;
1642         pde = NULL;
1643         de = (struct ext3_dir_entry_2 *) bh->b_data;
1644         while (i < bh->b_size) {
1645                 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1646                         return -EIO;
1647                 if (de == de_del)  {
1648                         BUFFER_TRACE(bh, "get_write_access");
1649                         ext3_journal_get_write_access(handle, bh);
1650                         if (pde)
1651                                 pde->rec_len = ext3_rec_len_to_disk(
1652                                         ext3_rec_len_from_disk(pde->rec_len) +
1653                                         ext3_rec_len_from_disk(de->rec_len));
1654                         else
1655                                 de->inode = 0;
1656                         dir->i_version++;
1657                         BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1658                         ext3_journal_dirty_metadata(handle, bh);
1659                         return 0;
1660                 }
1661                 i += ext3_rec_len_from_disk(de->rec_len);
1662                 pde = de;
1663                 de = ext3_next_entry(de);
1664         }
1665         return -ENOENT;
1666 }
1667
1668 static int ext3_add_nondir(handle_t *handle,
1669                 struct dentry *dentry, struct inode *inode)
1670 {
1671         int err = ext3_add_entry(handle, dentry, inode);
1672         if (!err) {
1673                 ext3_mark_inode_dirty(handle, inode);
1674                 d_instantiate(dentry, inode);
1675                 unlock_new_inode(inode);
1676                 return 0;
1677         }
1678         drop_nlink(inode);
1679         unlock_new_inode(inode);
1680         iput(inode);
1681         return err;
1682 }
1683
1684 /*
1685  * By the time this is called, we already have created
1686  * the directory cache entry for the new file, but it
1687  * is so far negative - it has no inode.
1688  *
1689  * If the create succeeds, we fill in the inode information
1690  * with d_instantiate().
1691  */
1692 static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1693                 struct nameidata *nd)
1694 {
1695         handle_t *handle;
1696         struct inode * inode;
1697         int err, retries = 0;
1698
1699 retry:
1700         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1701                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1702                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1703         if (IS_ERR(handle))
1704                 return PTR_ERR(handle);
1705
1706         if (IS_DIRSYNC(dir))
1707                 handle->h_sync = 1;
1708
1709         inode = ext3_new_inode (handle, dir, mode);
1710         err = PTR_ERR(inode);
1711         if (!IS_ERR(inode)) {
1712                 inode->i_op = &ext3_file_inode_operations;
1713                 inode->i_fop = &ext3_file_operations;
1714                 ext3_set_aops(inode);
1715                 err = ext3_add_nondir(handle, dentry, inode);
1716         }
1717         ext3_journal_stop(handle);
1718         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1719                 goto retry;
1720         return err;
1721 }
1722
1723 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1724                         int mode, dev_t rdev)
1725 {
1726         handle_t *handle;
1727         struct inode *inode;
1728         int err, retries = 0;
1729
1730         if (!new_valid_dev(rdev))
1731                 return -EINVAL;
1732
1733 retry:
1734         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1735                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1736                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1737         if (IS_ERR(handle))
1738                 return PTR_ERR(handle);
1739
1740         if (IS_DIRSYNC(dir))
1741                 handle->h_sync = 1;
1742
1743         inode = ext3_new_inode (handle, dir, mode);
1744         err = PTR_ERR(inode);
1745         if (!IS_ERR(inode)) {
1746                 init_special_inode(inode, inode->i_mode, rdev);
1747 #ifdef CONFIG_EXT3_FS_XATTR
1748                 inode->i_op = &ext3_special_inode_operations;
1749 #endif
1750                 err = ext3_add_nondir(handle, dentry, inode);
1751         }
1752         ext3_journal_stop(handle);
1753         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1754                 goto retry;
1755         return err;
1756 }
1757
1758 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1759 {
1760         handle_t *handle;
1761         struct inode * inode;
1762         struct buffer_head * dir_block;
1763         struct ext3_dir_entry_2 * de;
1764         int err, retries = 0;
1765
1766         if (dir->i_nlink >= EXT3_LINK_MAX)
1767                 return -EMLINK;
1768
1769 retry:
1770         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1771                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1772                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
1773         if (IS_ERR(handle))
1774                 return PTR_ERR(handle);
1775
1776         if (IS_DIRSYNC(dir))
1777                 handle->h_sync = 1;
1778
1779         inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1780         err = PTR_ERR(inode);
1781         if (IS_ERR(inode))
1782                 goto out_stop;
1783
1784         inode->i_op = &ext3_dir_inode_operations;
1785         inode->i_fop = &ext3_dir_operations;
1786         inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1787         dir_block = ext3_bread (handle, inode, 0, 1, &err);
1788         if (!dir_block) {
1789                 drop_nlink(inode); /* is this nlink == 0? */
1790                 unlock_new_inode(inode);
1791                 ext3_mark_inode_dirty(handle, inode);
1792                 iput (inode);
1793                 goto out_stop;
1794         }
1795         BUFFER_TRACE(dir_block, "get_write_access");
1796         ext3_journal_get_write_access(handle, dir_block);
1797         de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1798         de->inode = cpu_to_le32(inode->i_ino);
1799         de->name_len = 1;
1800         de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1801         strcpy (de->name, ".");
1802         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1803         de = ext3_next_entry(de);
1804         de->inode = cpu_to_le32(dir->i_ino);
1805         de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1806                                         EXT3_DIR_REC_LEN(1));
1807         de->name_len = 2;
1808         strcpy (de->name, "..");
1809         ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1810         inode->i_nlink = 2;
1811         BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1812         ext3_journal_dirty_metadata(handle, dir_block);
1813         brelse (dir_block);
1814         ext3_mark_inode_dirty(handle, inode);
1815         err = ext3_add_entry (handle, dentry, inode);
1816         if (err) {
1817                 inode->i_nlink = 0;
1818                 unlock_new_inode(inode);
1819                 ext3_mark_inode_dirty(handle, inode);
1820                 iput (inode);
1821                 goto out_stop;
1822         }
1823         inc_nlink(dir);
1824         ext3_update_dx_flag(dir);
1825         ext3_mark_inode_dirty(handle, dir);
1826         d_instantiate(dentry, inode);
1827         unlock_new_inode(inode);
1828 out_stop:
1829         ext3_journal_stop(handle);
1830         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1831                 goto retry;
1832         return err;
1833 }
1834
1835 /*
1836  * routine to check that the specified directory is empty (for rmdir)
1837  */
1838 static int empty_dir (struct inode * inode)
1839 {
1840         unsigned long offset;
1841         struct buffer_head * bh;
1842         struct ext3_dir_entry_2 * de, * de1;
1843         struct super_block * sb;
1844         int err = 0;
1845
1846         sb = inode->i_sb;
1847         if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1848             !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1849                 if (err)
1850                         ext3_error(inode->i_sb, __func__,
1851                                    "error %d reading directory #%lu offset 0",
1852                                    err, inode->i_ino);
1853                 else
1854                         ext3_warning(inode->i_sb, __func__,
1855                                      "bad directory (dir #%lu) - no data block",
1856                                      inode->i_ino);
1857                 return 1;
1858         }
1859         de = (struct ext3_dir_entry_2 *) bh->b_data;
1860         de1 = ext3_next_entry(de);
1861         if (le32_to_cpu(de->inode) != inode->i_ino ||
1862                         !le32_to_cpu(de1->inode) ||
1863                         strcmp (".", de->name) ||
1864                         strcmp ("..", de1->name)) {
1865                 ext3_warning (inode->i_sb, "empty_dir",
1866                               "bad directory (dir #%lu) - no `.' or `..'",
1867                               inode->i_ino);
1868                 brelse (bh);
1869                 return 1;
1870         }
1871         offset = ext3_rec_len_from_disk(de->rec_len) +
1872                         ext3_rec_len_from_disk(de1->rec_len);
1873         de = ext3_next_entry(de1);
1874         while (offset < inode->i_size ) {
1875                 if (!bh ||
1876                         (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1877                         err = 0;
1878                         brelse (bh);
1879                         bh = ext3_bread (NULL, inode,
1880                                 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1881                         if (!bh) {
1882                                 if (err)
1883                                         ext3_error(sb, __func__,
1884                                                    "error %d reading directory"
1885                                                    " #%lu offset %lu",
1886                                                    err, inode->i_ino, offset);
1887                                 offset += sb->s_blocksize;
1888                                 continue;
1889                         }
1890                         de = (struct ext3_dir_entry_2 *) bh->b_data;
1891                 }
1892                 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1893                         de = (struct ext3_dir_entry_2 *)(bh->b_data +
1894                                                          sb->s_blocksize);
1895                         offset = (offset | (sb->s_blocksize - 1)) + 1;
1896                         continue;
1897                 }
1898                 if (le32_to_cpu(de->inode)) {
1899                         brelse (bh);
1900                         return 0;
1901                 }
1902                 offset += ext3_rec_len_from_disk(de->rec_len);
1903                 de = ext3_next_entry(de);
1904         }
1905         brelse (bh);
1906         return 1;
1907 }
1908
1909 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1910  * such inodes, starting at the superblock, in case we crash before the
1911  * file is closed/deleted, or in case the inode truncate spans multiple
1912  * transactions and the last transaction is not recovered after a crash.
1913  *
1914  * At filesystem recovery time, we walk this list deleting unlinked
1915  * inodes and truncating linked inodes in ext3_orphan_cleanup().
1916  */
1917 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1918 {
1919         struct super_block *sb = inode->i_sb;
1920         struct ext3_iloc iloc;
1921         int err = 0, rc;
1922
1923         lock_super(sb);
1924         if (!list_empty(&EXT3_I(inode)->i_orphan))
1925                 goto out_unlock;
1926
1927         /* Orphan handling is only valid for files with data blocks
1928          * being truncated, or files being unlinked. */
1929
1930         /* @@@ FIXME: Observation from aviro:
1931          * I think I can trigger J_ASSERT in ext3_orphan_add().  We block
1932          * here (on lock_super()), so race with ext3_link() which might bump
1933          * ->i_nlink. For, say it, character device. Not a regular file,
1934          * not a directory, not a symlink and ->i_nlink > 0.
1935          */
1936         J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1937                 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1938
1939         BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1940         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1941         if (err)
1942                 goto out_unlock;
1943
1944         err = ext3_reserve_inode_write(handle, inode, &iloc);
1945         if (err)
1946                 goto out_unlock;
1947
1948         /* Insert this inode at the head of the on-disk orphan list... */
1949         NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1950         EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1951         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1952         rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1953         if (!err)
1954                 err = rc;
1955
1956         /* Only add to the head of the in-memory list if all the
1957          * previous operations succeeded.  If the orphan_add is going to
1958          * fail (possibly taking the journal offline), we can't risk
1959          * leaving the inode on the orphan list: stray orphan-list
1960          * entries can cause panics at unmount time.
1961          *
1962          * This is safe: on error we're going to ignore the orphan list
1963          * anyway on the next recovery. */
1964         if (!err)
1965                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1966
1967         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1968         jbd_debug(4, "orphan inode %lu will point to %d\n",
1969                         inode->i_ino, NEXT_ORPHAN(inode));
1970 out_unlock:
1971         unlock_super(sb);
1972         ext3_std_error(inode->i_sb, err);
1973         return err;
1974 }
1975
1976 /*
1977  * ext3_orphan_del() removes an unlinked or truncated inode from the list
1978  * of such inodes stored on disk, because it is finally being cleaned up.
1979  */
1980 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1981 {
1982         struct list_head *prev;
1983         struct ext3_inode_info *ei = EXT3_I(inode);
1984         struct ext3_sb_info *sbi;
1985         unsigned long ino_next;
1986         struct ext3_iloc iloc;
1987         int err = 0;
1988
1989         lock_super(inode->i_sb);
1990         if (list_empty(&ei->i_orphan)) {
1991                 unlock_super(inode->i_sb);
1992                 return 0;
1993         }
1994
1995         ino_next = NEXT_ORPHAN(inode);
1996         prev = ei->i_orphan.prev;
1997         sbi = EXT3_SB(inode->i_sb);
1998
1999         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2000
2001         list_del_init(&ei->i_orphan);
2002
2003         /* If we're on an error path, we may not have a valid
2004          * transaction handle with which to update the orphan list on
2005          * disk, but we still need to remove the inode from the linked
2006          * list in memory. */
2007         if (!handle)
2008                 goto out;
2009
2010         err = ext3_reserve_inode_write(handle, inode, &iloc);
2011         if (err)
2012                 goto out_err;
2013
2014         if (prev == &sbi->s_orphan) {
2015                 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2016                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2017                 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2018                 if (err)
2019                         goto out_brelse;
2020                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2021                 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2022         } else {
2023                 struct ext3_iloc iloc2;
2024                 struct inode *i_prev =
2025                         &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2026
2027                 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2028                           i_prev->i_ino, ino_next);
2029                 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2030                 if (err)
2031                         goto out_brelse;
2032                 NEXT_ORPHAN(i_prev) = ino_next;
2033                 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2034         }
2035         if (err)
2036                 goto out_brelse;
2037         NEXT_ORPHAN(inode) = 0;
2038         err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2039
2040 out_err:
2041         ext3_std_error(inode->i_sb, err);
2042 out:
2043         unlock_super(inode->i_sb);
2044         return err;
2045
2046 out_brelse:
2047         brelse(iloc.bh);
2048         goto out_err;
2049 }
2050
2051 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2052 {
2053         int retval;
2054         struct inode * inode;
2055         struct buffer_head * bh;
2056         struct ext3_dir_entry_2 * de;
2057         handle_t *handle;
2058
2059         /* Initialize quotas before so that eventual writes go in
2060          * separate transaction */
2061         vfs_dq_init(dentry->d_inode);
2062         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2063         if (IS_ERR(handle))
2064                 return PTR_ERR(handle);
2065
2066         retval = -ENOENT;
2067         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2068         if (!bh)
2069                 goto end_rmdir;
2070
2071         if (IS_DIRSYNC(dir))
2072                 handle->h_sync = 1;
2073
2074         inode = dentry->d_inode;
2075
2076         retval = -EIO;
2077         if (le32_to_cpu(de->inode) != inode->i_ino)
2078                 goto end_rmdir;
2079
2080         retval = -ENOTEMPTY;
2081         if (!empty_dir (inode))
2082                 goto end_rmdir;
2083
2084         retval = ext3_delete_entry(handle, dir, de, bh);
2085         if (retval)
2086                 goto end_rmdir;
2087         if (inode->i_nlink != 2)
2088                 ext3_warning (inode->i_sb, "ext3_rmdir",
2089                               "empty directory has nlink!=2 (%d)",
2090                               inode->i_nlink);
2091         inode->i_version++;
2092         clear_nlink(inode);
2093         /* There's no need to set i_disksize: the fact that i_nlink is
2094          * zero will ensure that the right thing happens during any
2095          * recovery. */
2096         inode->i_size = 0;
2097         ext3_orphan_add(handle, inode);
2098         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2099         ext3_mark_inode_dirty(handle, inode);
2100         drop_nlink(dir);
2101         ext3_update_dx_flag(dir);
2102         ext3_mark_inode_dirty(handle, dir);
2103
2104 end_rmdir:
2105         ext3_journal_stop(handle);
2106         brelse (bh);
2107         return retval;
2108 }
2109
2110 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2111 {
2112         int retval;
2113         struct inode * inode;
2114         struct buffer_head * bh;
2115         struct ext3_dir_entry_2 * de;
2116         handle_t *handle;
2117
2118         /* Initialize quotas before so that eventual writes go
2119          * in separate transaction */
2120         vfs_dq_init(dentry->d_inode);
2121         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2122         if (IS_ERR(handle))
2123                 return PTR_ERR(handle);
2124
2125         if (IS_DIRSYNC(dir))
2126                 handle->h_sync = 1;
2127
2128         retval = -ENOENT;
2129         bh = ext3_find_entry(dir, &dentry->d_name, &de);
2130         if (!bh)
2131                 goto end_unlink;
2132
2133         inode = dentry->d_inode;
2134
2135         retval = -EIO;
2136         if (le32_to_cpu(de->inode) != inode->i_ino)
2137                 goto end_unlink;
2138
2139         if (!inode->i_nlink) {
2140                 ext3_warning (inode->i_sb, "ext3_unlink",
2141                               "Deleting nonexistent file (%lu), %d",
2142                               inode->i_ino, inode->i_nlink);
2143                 inode->i_nlink = 1;
2144         }
2145         retval = ext3_delete_entry(handle, dir, de, bh);
2146         if (retval)
2147                 goto end_unlink;
2148         dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2149         ext3_update_dx_flag(dir);
2150         ext3_mark_inode_dirty(handle, dir);
2151         drop_nlink(inode);
2152         if (!inode->i_nlink)
2153                 ext3_orphan_add(handle, inode);
2154         inode->i_ctime = dir->i_ctime;
2155         ext3_mark_inode_dirty(handle, inode);
2156         retval = 0;
2157
2158 end_unlink:
2159         ext3_journal_stop(handle);
2160         brelse (bh);
2161         return retval;
2162 }
2163
2164 static int ext3_symlink (struct inode * dir,
2165                 struct dentry *dentry, const char * symname)
2166 {
2167         handle_t *handle;
2168         struct inode * inode;
2169         int l, err, retries = 0;
2170
2171         l = strlen(symname)+1;
2172         if (l > dir->i_sb->s_blocksize)
2173                 return -ENAMETOOLONG;
2174
2175 retry:
2176         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2177                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2178                                         2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
2179         if (IS_ERR(handle))
2180                 return PTR_ERR(handle);
2181
2182         if (IS_DIRSYNC(dir))
2183                 handle->h_sync = 1;
2184
2185         inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2186         err = PTR_ERR(inode);
2187         if (IS_ERR(inode))
2188                 goto out_stop;
2189
2190         if (l > sizeof (EXT3_I(inode)->i_data)) {
2191                 inode->i_op = &ext3_symlink_inode_operations;
2192                 ext3_set_aops(inode);
2193                 /*
2194                  * page_symlink() calls into ext3_prepare/commit_write.
2195                  * We have a transaction open.  All is sweetness.  It also sets
2196                  * i_size in generic_commit_write().
2197                  */
2198                 err = __page_symlink(inode, symname, l, 1);
2199                 if (err) {
2200                         drop_nlink(inode);
2201                         unlock_new_inode(inode);
2202                         ext3_mark_inode_dirty(handle, inode);
2203                         iput (inode);
2204                         goto out_stop;
2205                 }
2206         } else {
2207                 inode->i_op = &ext3_fast_symlink_inode_operations;
2208                 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2209                 inode->i_size = l-1;
2210         }
2211         EXT3_I(inode)->i_disksize = inode->i_size;
2212         err = ext3_add_nondir(handle, dentry, inode);
2213 out_stop:
2214         ext3_journal_stop(handle);
2215         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2216                 goto retry;
2217         return err;
2218 }
2219
2220 static int ext3_link (struct dentry * old_dentry,
2221                 struct inode * dir, struct dentry *dentry)
2222 {
2223         handle_t *handle;
2224         struct inode *inode = old_dentry->d_inode;
2225         int err, retries = 0;
2226
2227         if (inode->i_nlink >= EXT3_LINK_MAX)
2228                 return -EMLINK;
2229         /*
2230          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
2231          * otherwise has the potential to corrupt the orphan inode list.
2232          */
2233         if (inode->i_nlink == 0)
2234                 return -ENOENT;
2235
2236 retry:
2237         handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2238                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2239         if (IS_ERR(handle))
2240                 return PTR_ERR(handle);
2241
2242         if (IS_DIRSYNC(dir))
2243                 handle->h_sync = 1;
2244
2245         inode->i_ctime = CURRENT_TIME_SEC;
2246         inc_nlink(inode);
2247         atomic_inc(&inode->i_count);
2248
2249         err = ext3_add_entry(handle, dentry, inode);
2250         if (!err) {
2251                 ext3_mark_inode_dirty(handle, inode);
2252                 d_instantiate(dentry, inode);
2253         } else {
2254                 drop_nlink(inode);
2255                 iput(inode);
2256         }
2257         ext3_journal_stop(handle);
2258         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2259                 goto retry;
2260         return err;
2261 }
2262
2263 #define PARENT_INO(buffer) \
2264         (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2265
2266 /*
2267  * Anybody can rename anything with this: the permission checks are left to the
2268  * higher-level routines.
2269  */
2270 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2271                            struct inode * new_dir,struct dentry *new_dentry)
2272 {
2273         handle_t *handle;
2274         struct inode * old_inode, * new_inode;
2275         struct buffer_head * old_bh, * new_bh, * dir_bh;
2276         struct ext3_dir_entry_2 * old_de, * new_de;
2277         int retval, flush_file = 0;
2278
2279         old_bh = new_bh = dir_bh = NULL;
2280
2281         /* Initialize quotas before so that eventual writes go
2282          * in separate transaction */
2283         if (new_dentry->d_inode)
2284                 vfs_dq_init(new_dentry->d_inode);
2285         handle = ext3_journal_start(old_dir, 2 *
2286                                         EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2287                                         EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2288         if (IS_ERR(handle))
2289                 return PTR_ERR(handle);
2290
2291         if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2292                 handle->h_sync = 1;
2293
2294         old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2295         /*
2296          *  Check for inode number is _not_ due to possible IO errors.
2297          *  We might rmdir the source, keep it as pwd of some process
2298          *  and merrily kill the link to whatever was created under the
2299          *  same name. Goodbye sticky bit ;-<
2300          */
2301         old_inode = old_dentry->d_inode;
2302         retval = -ENOENT;
2303         if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2304                 goto end_rename;
2305
2306         new_inode = new_dentry->d_inode;
2307         new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2308         if (new_bh) {
2309                 if (!new_inode) {
2310                         brelse (new_bh);
2311                         new_bh = NULL;
2312                 }
2313         }
2314         if (S_ISDIR(old_inode->i_mode)) {
2315                 if (new_inode) {
2316                         retval = -ENOTEMPTY;
2317                         if (!empty_dir (new_inode))
2318                                 goto end_rename;
2319                 }
2320                 retval = -EIO;
2321                 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2322                 if (!dir_bh)
2323                         goto end_rename;
2324                 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2325                         goto end_rename;
2326                 retval = -EMLINK;
2327                 if (!new_inode && new_dir!=old_dir &&
2328                                 new_dir->i_nlink >= EXT3_LINK_MAX)
2329                         goto end_rename;
2330         }
2331         if (!new_bh) {
2332                 retval = ext3_add_entry (handle, new_dentry, old_inode);
2333                 if (retval)
2334                         goto end_rename;
2335         } else {
2336                 BUFFER_TRACE(new_bh, "get write access");
2337                 ext3_journal_get_write_access(handle, new_bh);
2338                 new_de->inode = cpu_to_le32(old_inode->i_ino);
2339                 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2340                                               EXT3_FEATURE_INCOMPAT_FILETYPE))
2341                         new_de->file_type = old_de->file_type;
2342                 new_dir->i_version++;
2343                 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2344                 ext3_mark_inode_dirty(handle, new_dir);
2345                 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2346                 ext3_journal_dirty_metadata(handle, new_bh);
2347                 brelse(new_bh);
2348                 new_bh = NULL;
2349         }
2350
2351         /*
2352          * Like most other Unix systems, set the ctime for inodes on a
2353          * rename.
2354          */
2355         old_inode->i_ctime = CURRENT_TIME_SEC;
2356         ext3_mark_inode_dirty(handle, old_inode);
2357
2358         /*
2359          * ok, that's it
2360          */
2361         if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2362             old_de->name_len != old_dentry->d_name.len ||
2363             strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2364             (retval = ext3_delete_entry(handle, old_dir,
2365                                         old_de, old_bh)) == -ENOENT) {
2366                 /* old_de could have moved from under us during htree split, so
2367                  * make sure that we are deleting the right entry.  We might
2368                  * also be pointing to a stale entry in the unused part of
2369                  * old_bh so just checking inum and the name isn't enough. */
2370                 struct buffer_head *old_bh2;
2371                 struct ext3_dir_entry_2 *old_de2;
2372
2373                 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2374                                           &old_de2);
2375                 if (old_bh2) {
2376                         retval = ext3_delete_entry(handle, old_dir,
2377                                                    old_de2, old_bh2);
2378                         brelse(old_bh2);
2379                 }
2380         }
2381         if (retval) {
2382                 ext3_warning(old_dir->i_sb, "ext3_rename",
2383                                 "Deleting old file (%lu), %d, error=%d",
2384                                 old_dir->i_ino, old_dir->i_nlink, retval);
2385         }
2386
2387         if (new_inode) {
2388                 drop_nlink(new_inode);
2389                 new_inode->i_ctime = CURRENT_TIME_SEC;
2390         }
2391         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2392         ext3_update_dx_flag(old_dir);
2393         if (dir_bh) {
2394                 BUFFER_TRACE(dir_bh, "get_write_access");
2395                 ext3_journal_get_write_access(handle, dir_bh);
2396                 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2397                 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2398                 ext3_journal_dirty_metadata(handle, dir_bh);
2399                 drop_nlink(old_dir);
2400                 if (new_inode) {
2401                         drop_nlink(new_inode);
2402                 } else {
2403                         inc_nlink(new_dir);
2404                         ext3_update_dx_flag(new_dir);
2405                         ext3_mark_inode_dirty(handle, new_dir);
2406                 }
2407         }
2408         ext3_mark_inode_dirty(handle, old_dir);
2409         if (new_inode) {
2410                 ext3_mark_inode_dirty(handle, new_inode);
2411                 if (!new_inode->i_nlink)
2412                         ext3_orphan_add(handle, new_inode);
2413                 if (ext3_should_writeback_data(new_inode))
2414                         flush_file = 1;
2415         }
2416         retval = 0;
2417
2418 end_rename:
2419         brelse (dir_bh);
2420         brelse (old_bh);
2421         brelse (new_bh);
2422         ext3_journal_stop(handle);
2423         if (retval == 0 && flush_file)
2424                 filemap_flush(old_inode->i_mapping);
2425         return retval;
2426 }
2427
2428 /*
2429  * directories can handle most operations...
2430  */
2431 const struct inode_operations ext3_dir_inode_operations = {
2432         .create         = ext3_create,
2433         .lookup         = ext3_lookup,
2434         .link           = ext3_link,
2435         .unlink         = ext3_unlink,
2436         .symlink        = ext3_symlink,
2437         .mkdir          = ext3_mkdir,
2438         .rmdir          = ext3_rmdir,
2439         .mknod          = ext3_mknod,
2440         .rename         = ext3_rename,
2441         .setattr        = ext3_setattr,
2442 #ifdef CONFIG_EXT3_FS_XATTR
2443         .setxattr       = generic_setxattr,
2444         .getxattr       = generic_getxattr,
2445         .listxattr      = ext3_listxattr,
2446         .removexattr    = generic_removexattr,
2447 #endif
2448         .permission     = ext3_permission,
2449 };
2450
2451 const struct inode_operations ext3_special_inode_operations = {
2452         .setattr        = ext3_setattr,
2453 #ifdef CONFIG_EXT3_FS_XATTR
2454         .setxattr       = generic_setxattr,
2455         .getxattr       = generic_getxattr,
2456         .listxattr      = ext3_listxattr,
2457         .removexattr    = generic_removexattr,
2458 #endif
2459         .permission     = ext3_permission,
2460 };