Root/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
52static 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
83struct fake_dirent
84{
85    __le32 inode;
86    __le16 rec_len;
87    u8 name_len;
88    u8 file_type;
89};
90
91struct dx_countlimit
92{
93    __le16 limit;
94    __le16 count;
95};
96
97struct 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
109struct 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
127struct dx_node
128{
129    struct fake_dirent fake;
130    struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136    struct buffer_head *bh;
137    struct dx_entry *entries;
138    struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143    u32 hash;
144    u16 offs;
145    u16 size;
146};
147
148static inline unsigned dx_get_block (struct dx_entry *entry);
149static void dx_set_block (struct dx_entry *entry, unsigned value);
150static inline unsigned dx_get_hash (struct dx_entry *entry);
151static void dx_set_hash (struct dx_entry *entry, unsigned value);
152static unsigned dx_get_count (struct dx_entry *entries);
153static unsigned dx_get_limit (struct dx_entry *entries);
154static void dx_set_count (struct dx_entry *entries, unsigned value);
155static void dx_set_limit (struct dx_entry *entries, unsigned value);
156static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157static unsigned dx_node_limit (struct inode *dir);
158static 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);
163static void dx_release (struct dx_frame *frames);
164static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
165            struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168        struct dx_map_entry *offsets, int count);
169static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
170static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172                 struct dx_frame *frame,
173                 struct dx_frame *frames,
174                 __u32 *start_hash);
175static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
176            struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
177            int *err);
178static 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 */
184static inline struct ext3_dir_entry_2 *
185ext3_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
196static inline unsigned dx_get_block (struct dx_entry *entry)
197{
198    return le32_to_cpu(entry->block) & 0x00ffffff;
199}
200
201static inline void dx_set_block (struct dx_entry *entry, unsigned value)
202{
203    entry->block = cpu_to_le32(value);
204}
205
206static inline unsigned dx_get_hash (struct dx_entry *entry)
207{
208    return le32_to_cpu(entry->hash);
209}
210
211static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
212{
213    entry->hash = cpu_to_le32(value);
214}
215
216static inline unsigned dx_get_count (struct dx_entry *entries)
217{
218    return le16_to_cpu(((struct dx_countlimit *) entries)->count);
219}
220
221static inline unsigned dx_get_limit (struct dx_entry *entries)
222{
223    return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
224}
225
226static inline void dx_set_count (struct dx_entry *entries, unsigned value)
227{
228    ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
229}
230
231static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
232{
233    ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
234}
235
236static 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
243static 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
253static 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
264struct stats
265{
266    unsigned names;
267    unsigned space;
268    unsigned bcount;
269};
270
271static 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
301struct 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 */
341static struct dx_frame *
342dx_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    }
463fail2:
464    while (frame >= frame_in) {
465        brelse(frame->bh);
466        frame--;
467    }
468fail:
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
476static 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 */
503static 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 */
566static 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 */
620int 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;
697errout:
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 */
711static 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 */
736static 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
764static 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
778static 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 */
791static 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 */
804static 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 */
853static 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    const u8 *name = entry->name;
862    int ra_max = 0; /* Number of bh's in the readahead
863                   buffer, bh_use[] */
864    int ra_ptr = 0; /* Current index into readahead
865                   buffer */
866    int num = 0;
867    int nblocks, i, err;
868    int namelen;
869
870    *res_dir = NULL;
871    sb = dir->i_sb;
872    namelen = entry->len;
873    if (namelen > EXT3_NAME_LEN)
874        return NULL;
875    if ((namelen <= 2) && (name[0] == '.') &&
876        (name[1] == '.' || name[1] == 0)) {
877        /*
878         * "." or ".." will only be in the first block
879         * NFS may look up ".."; "." should be handled by the VFS
880         */
881        block = start = 0;
882        nblocks = 1;
883        goto restart;
884    }
885    if (is_dx(dir)) {
886        bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
887        /*
888         * On success, or if the error was file not found,
889         * return. Otherwise, fall back to doing a search the
890         * old fashioned way.
891         */
892        if (bh || (err != ERR_BAD_DX_DIR))
893            return bh;
894        dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
895    }
896    nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
897    start = EXT3_I(dir)->i_dir_start_lookup;
898    if (start >= nblocks)
899        start = 0;
900    block = start;
901restart:
902    do {
903        /*
904         * We deal with the read-ahead logic here.
905         */
906        if (ra_ptr >= ra_max) {
907            /* Refill the readahead buffer */
908            ra_ptr = 0;
909            b = block;
910            for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
911                /*
912                 * Terminate if we reach the end of the
913                 * directory and must wrap, or if our
914                 * search has finished at this block.
915                 */
916                if (b >= nblocks || (num && block == start)) {
917                    bh_use[ra_max] = NULL;
918                    break;
919                }
920                num++;
921                bh = ext3_getblk(NULL, dir, b++, 0, &err);
922                bh_use[ra_max] = bh;
923                if (bh)
924                    ll_rw_block(READ_META, 1, &bh);
925            }
926        }
927        if ((bh = bh_use[ra_ptr++]) == NULL)
928            goto next;
929        wait_on_buffer(bh);
930        if (!buffer_uptodate(bh)) {
931            /* read error, skip block & hope for the best */
932            ext3_error(sb, __func__, "reading directory #%lu "
933                   "offset %lu", dir->i_ino, block);
934            brelse(bh);
935            goto next;
936        }
937        i = search_dirblock(bh, dir, entry,
938                block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
939        if (i == 1) {
940            EXT3_I(dir)->i_dir_start_lookup = block;
941            ret = bh;
942            goto cleanup_and_exit;
943        } else {
944            brelse(bh);
945            if (i < 0)
946                goto cleanup_and_exit;
947        }
948    next:
949        if (++block >= nblocks)
950            block = 0;
951    } while (block != start);
952
953    /*
954     * If the directory has grown while we were searching, then
955     * search the last part of the directory before giving up.
956     */
957    block = nblocks;
958    nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
959    if (block < nblocks) {
960        start = 0;
961        goto restart;
962    }
963
964cleanup_and_exit:
965    /* Clean up the read-ahead blocks */
966    for (; ra_ptr < ra_max; ra_ptr++)
967        brelse (bh_use[ra_ptr]);
968    return ret;
969}
970
971static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
972            struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
973            int *err)
974{
975    struct super_block *sb = dir->i_sb;
976    struct dx_hash_info hinfo;
977    struct dx_frame frames[2], *frame;
978    struct buffer_head *bh;
979    unsigned long block;
980    int retval;
981
982    if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
983        return NULL;
984    do {
985        block = dx_get_block(frame->at);
986        if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
987            goto errout;
988
989        retval = search_dirblock(bh, dir, entry,
990                     block << EXT3_BLOCK_SIZE_BITS(sb),
991                     res_dir);
992        if (retval == 1) {
993            dx_release(frames);
994            return bh;
995        }
996        brelse(bh);
997        if (retval == -1) {
998            *err = ERR_BAD_DX_DIR;
999            goto errout;
1000        }
1001
1002        /* Check to see if we should continue to search */
1003        retval = ext3_htree_next_block(dir, hinfo.hash, frame,
1004                           frames, NULL);
1005        if (retval < 0) {
1006            ext3_warning(sb, __func__,
1007                 "error reading index page in directory #%lu",
1008                 dir->i_ino);
1009            *err = retval;
1010            goto errout;
1011        }
1012    } while (retval == 1);
1013
1014    *err = -ENOENT;
1015errout:
1016    dxtrace(printk("%s not found\n", name));
1017    dx_release (frames);
1018    return NULL;
1019}
1020
1021static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1022{
1023    struct inode * inode;
1024    struct ext3_dir_entry_2 * de;
1025    struct buffer_head * bh;
1026
1027    if (dentry->d_name.len > EXT3_NAME_LEN)
1028        return ERR_PTR(-ENAMETOOLONG);
1029
1030    bh = ext3_find_entry(dir, &dentry->d_name, &de);
1031    inode = NULL;
1032    if (bh) {
1033        unsigned long ino = le32_to_cpu(de->inode);
1034        brelse (bh);
1035        if (!ext3_valid_inum(dir->i_sb, ino)) {
1036            ext3_error(dir->i_sb, "ext3_lookup",
1037                   "bad inode number: %lu", ino);
1038            return ERR_PTR(-EIO);
1039        }
1040        inode = ext3_iget(dir->i_sb, ino);
1041        if (IS_ERR(inode)) {
1042            if (PTR_ERR(inode) == -ESTALE) {
1043                ext3_error(dir->i_sb, __func__,
1044                        "deleted inode referenced: %lu",
1045                        ino);
1046                return ERR_PTR(-EIO);
1047            } else {
1048                return ERR_CAST(inode);
1049            }
1050        }
1051    }
1052    return d_splice_alias(inode, dentry);
1053}
1054
1055
1056struct dentry *ext3_get_parent(struct dentry *child)
1057{
1058    unsigned long ino;
1059    struct qstr dotdot = {.name = "..", .len = 2};
1060    struct ext3_dir_entry_2 * de;
1061    struct buffer_head *bh;
1062
1063    bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1064    if (!bh)
1065        return ERR_PTR(-ENOENT);
1066    ino = le32_to_cpu(de->inode);
1067    brelse(bh);
1068
1069    if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1070        ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1071               "bad inode number: %lu", ino);
1072        return ERR_PTR(-EIO);
1073    }
1074
1075    return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1076}
1077
1078#define S_SHIFT 12
1079static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1080    [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1081    [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1082    [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1083    [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1084    [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1085    [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1086    [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1087};
1088
1089static inline void ext3_set_de_type(struct super_block *sb,
1090                struct ext3_dir_entry_2 *de,
1091                umode_t mode) {
1092    if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1093        de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1094}
1095
1096/*
1097 * Move count entries from end of map between two memory locations.
1098 * Returns pointer to last entry moved.
1099 */
1100static struct ext3_dir_entry_2 *
1101dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1102{
1103    unsigned rec_len = 0;
1104
1105    while (count--) {
1106        struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1107        rec_len = EXT3_DIR_REC_LEN(de->name_len);
1108        memcpy (to, de, rec_len);
1109        ((struct ext3_dir_entry_2 *) to)->rec_len =
1110                ext3_rec_len_to_disk(rec_len);
1111        de->inode = 0;
1112        map++;
1113        to += rec_len;
1114    }
1115    return (struct ext3_dir_entry_2 *) (to - rec_len);
1116}
1117
1118/*
1119 * Compact each dir entry in the range to the minimal rec_len.
1120 * Returns pointer to last entry in range.
1121 */
1122static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1123{
1124    struct ext3_dir_entry_2 *next, *to, *prev;
1125    struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1126    unsigned rec_len = 0;
1127
1128    prev = to = de;
1129    while ((char *)de < base + blocksize) {
1130        next = ext3_next_entry(de);
1131        if (de->inode && de->name_len) {
1132            rec_len = EXT3_DIR_REC_LEN(de->name_len);
1133            if (de > to)
1134                memmove(to, de, rec_len);
1135            to->rec_len = ext3_rec_len_to_disk(rec_len);
1136            prev = to;
1137            to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1138        }
1139        de = next;
1140    }
1141    return prev;
1142}
1143
1144/*
1145 * Split a full leaf block to make room for a new dir entry.
1146 * Allocate a new block, and move entries so that they are approx. equally full.
1147 * Returns pointer to de in block into which the new entry will be inserted.
1148 */
1149static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1150            struct buffer_head **bh,struct dx_frame *frame,
1151            struct dx_hash_info *hinfo, int *error)
1152{
1153    unsigned blocksize = dir->i_sb->s_blocksize;
1154    unsigned count, continued;
1155    struct buffer_head *bh2;
1156    u32 newblock;
1157    u32 hash2;
1158    struct dx_map_entry *map;
1159    char *data1 = (*bh)->b_data, *data2;
1160    unsigned split, move, size;
1161    struct ext3_dir_entry_2 *de = NULL, *de2;
1162    int err = 0, i;
1163
1164    bh2 = ext3_append (handle, dir, &newblock, &err);
1165    if (!(bh2)) {
1166        brelse(*bh);
1167        *bh = NULL;
1168        goto errout;
1169    }
1170
1171    BUFFER_TRACE(*bh, "get_write_access");
1172    err = ext3_journal_get_write_access(handle, *bh);
1173    if (err)
1174        goto journal_error;
1175
1176    BUFFER_TRACE(frame->bh, "get_write_access");
1177    err = ext3_journal_get_write_access(handle, frame->bh);
1178    if (err)
1179        goto journal_error;
1180
1181    data2 = bh2->b_data;
1182
1183    /* create map in the end of data2 block */
1184    map = (struct dx_map_entry *) (data2 + blocksize);
1185    count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1186                 blocksize, hinfo, map);
1187    map -= count;
1188    dx_sort_map (map, count);
1189    /* Split the existing block in the middle, size-wise */
1190    size = 0;
1191    move = 0;
1192    for (i = count-1; i >= 0; i--) {
1193        /* is more than half of this entry in 2nd half of the block? */
1194        if (size + map[i].size/2 > blocksize/2)
1195            break;
1196        size += map[i].size;
1197        move++;
1198    }
1199    /* map index at which we will split */
1200    split = count - move;
1201    hash2 = map[split].hash;
1202    continued = hash2 == map[split - 1].hash;
1203    dxtrace(printk("Split block %i at %x, %i/%i\n",
1204        dx_get_block(frame->at), hash2, split, count-split));
1205
1206    /* Fancy dance to stay within two buffers */
1207    de2 = dx_move_dirents(data1, data2, map + split, count - split);
1208    de = dx_pack_dirents(data1,blocksize);
1209    de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1210    de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1211    dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1212    dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1213
1214    /* Which block gets the new entry? */
1215    if (hinfo->hash >= hash2)
1216    {
1217        swap(*bh, bh2);
1218        de = de2;
1219    }
1220    dx_insert_block (frame, hash2 + continued, newblock);
1221    err = ext3_journal_dirty_metadata (handle, bh2);
1222    if (err)
1223        goto journal_error;
1224    err = ext3_journal_dirty_metadata (handle, frame->bh);
1225    if (err)
1226        goto journal_error;
1227    brelse (bh2);
1228    dxtrace(dx_show_index ("frame", frame->entries));
1229    return de;
1230
1231journal_error:
1232    brelse(*bh);
1233    brelse(bh2);
1234    *bh = NULL;
1235    ext3_std_error(dir->i_sb, err);
1236errout:
1237    *error = err;
1238    return NULL;
1239}
1240
1241
1242/*
1243 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1244 * it points to a directory entry which is guaranteed to be large
1245 * enough for new directory entry. If de is NULL, then
1246 * add_dirent_to_buf will attempt search the directory block for
1247 * space. It will return -ENOSPC if no space is available, and -EIO
1248 * and -EEXIST if directory entry already exists.
1249 *
1250 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1251 * all other cases bh is released.
1252 */
1253static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1254                 struct inode *inode, struct ext3_dir_entry_2 *de,
1255                 struct buffer_head * bh)
1256{
1257    struct inode *dir = dentry->d_parent->d_inode;
1258    const char *name = dentry->d_name.name;
1259    int namelen = dentry->d_name.len;
1260    unsigned long offset = 0;
1261    unsigned short reclen;
1262    int nlen, rlen, err;
1263    char *top;
1264
1265    reclen = EXT3_DIR_REC_LEN(namelen);
1266    if (!de) {
1267        de = (struct ext3_dir_entry_2 *)bh->b_data;
1268        top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1269        while ((char *) de <= top) {
1270            if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1271                          bh, offset)) {
1272                brelse (bh);
1273                return -EIO;
1274            }
1275            if (ext3_match (namelen, name, de)) {
1276                brelse (bh);
1277                return -EEXIST;
1278            }
1279            nlen = EXT3_DIR_REC_LEN(de->name_len);
1280            rlen = ext3_rec_len_from_disk(de->rec_len);
1281            if ((de->inode? rlen - nlen: rlen) >= reclen)
1282                break;
1283            de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1284            offset += rlen;
1285        }
1286        if ((char *) de > top)
1287            return -ENOSPC;
1288    }
1289    BUFFER_TRACE(bh, "get_write_access");
1290    err = ext3_journal_get_write_access(handle, bh);
1291    if (err) {
1292        ext3_std_error(dir->i_sb, err);
1293        brelse(bh);
1294        return err;
1295    }
1296
1297    /* By now the buffer is marked for journaling */
1298    nlen = EXT3_DIR_REC_LEN(de->name_len);
1299    rlen = ext3_rec_len_from_disk(de->rec_len);
1300    if (de->inode) {
1301        struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1302        de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1303        de->rec_len = ext3_rec_len_to_disk(nlen);
1304        de = de1;
1305    }
1306    de->file_type = EXT3_FT_UNKNOWN;
1307    if (inode) {
1308        de->inode = cpu_to_le32(inode->i_ino);
1309        ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1310    } else
1311        de->inode = 0;
1312    de->name_len = namelen;
1313    memcpy (de->name, name, namelen);
1314    /*
1315     * XXX shouldn't update any times until successful
1316     * completion of syscall, but too many callers depend
1317     * on this.
1318     *
1319     * XXX similarly, too many callers depend on
1320     * ext3_new_inode() setting the times, but error
1321     * recovery deletes the inode, so the worst that can
1322     * happen is that the times are slightly out of date
1323     * and/or different from the directory change time.
1324     */
1325    dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1326    ext3_update_dx_flag(dir);
1327    dir->i_version++;
1328    ext3_mark_inode_dirty(handle, dir);
1329    BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1330    err = ext3_journal_dirty_metadata(handle, bh);
1331    if (err)
1332        ext3_std_error(dir->i_sb, err);
1333    brelse(bh);
1334    return 0;
1335}
1336
1337/*
1338 * This converts a one block unindexed directory to a 3 block indexed
1339 * directory, and adds the dentry to the indexed directory.
1340 */
1341static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1342                struct inode *inode, struct buffer_head *bh)
1343{
1344    struct inode *dir = dentry->d_parent->d_inode;
1345    const char *name = dentry->d_name.name;
1346    int namelen = dentry->d_name.len;
1347    struct buffer_head *bh2;
1348    struct dx_root *root;
1349    struct dx_frame frames[2], *frame;
1350    struct dx_entry *entries;
1351    struct ext3_dir_entry_2 *de, *de2;
1352    char *data1, *top;
1353    unsigned len;
1354    int retval;
1355    unsigned blocksize;
1356    struct dx_hash_info hinfo;
1357    u32 block;
1358    struct fake_dirent *fde;
1359
1360    blocksize = dir->i_sb->s_blocksize;
1361    dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1362    retval = ext3_journal_get_write_access(handle, bh);
1363    if (retval) {
1364        ext3_std_error(dir->i_sb, retval);
1365        brelse(bh);
1366        return retval;
1367    }
1368    root = (struct dx_root *) bh->b_data;
1369
1370    /* The 0th block becomes the root, move the dirents out */
1371    fde = &root->dotdot;
1372    de = (struct ext3_dir_entry_2 *)((char *)fde +
1373            ext3_rec_len_from_disk(fde->rec_len));
1374    if ((char *) de >= (((char *) root) + blocksize)) {
1375        ext3_error(dir->i_sb, __func__,
1376               "invalid rec_len for '..' in inode %lu",
1377               dir->i_ino);
1378        brelse(bh);
1379        return -EIO;
1380    }
1381    len = ((char *) root) + blocksize - (char *) de;
1382
1383    bh2 = ext3_append (handle, dir, &block, &retval);
1384    if (!(bh2)) {
1385        brelse(bh);
1386        return retval;
1387    }
1388    EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1389    data1 = bh2->b_data;
1390
1391    memcpy (data1, de, len);
1392    de = (struct ext3_dir_entry_2 *) data1;
1393    top = data1 + len;
1394    while ((char *)(de2 = ext3_next_entry(de)) < top)
1395        de = de2;
1396    de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1397    /* Initialize the root; the dot dirents already exist */
1398    de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1399    de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1400    memset (&root->info, 0, sizeof(root->info));
1401    root->info.info_length = sizeof(root->info);
1402    root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1403    entries = root->entries;
1404    dx_set_block (entries, 1);
1405    dx_set_count (entries, 1);
1406    dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1407
1408    /* Initialize as for dx_probe */
1409    hinfo.hash_version = root->info.hash_version;
1410    if (hinfo.hash_version <= DX_HASH_TEA)
1411        hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1412    hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1413    ext3fs_dirhash(name, namelen, &hinfo);
1414    frame = frames;
1415    frame->entries = entries;
1416    frame->at = entries;
1417    frame->bh = bh;
1418    bh = bh2;
1419    /*
1420     * Mark buffers dirty here so that if do_split() fails we write a
1421     * consistent set of buffers to disk.
1422     */
1423    ext3_journal_dirty_metadata(handle, frame->bh);
1424    ext3_journal_dirty_metadata(handle, bh);
1425    de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1426    if (!de) {
1427        ext3_mark_inode_dirty(handle, dir);
1428        dx_release(frames);
1429        return retval;
1430    }
1431    dx_release(frames);
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 */
1446static 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    struct buffer_head * bh;
1451    struct ext3_dir_entry_2 *de;
1452    struct super_block * sb;
1453    int retval;
1454    int dx_fallback=0;
1455    unsigned blocksize;
1456    u32 block, blocks;
1457
1458    sb = dir->i_sb;
1459    blocksize = sb->s_blocksize;
1460    if (!dentry->d_name.len)
1461        return -EINVAL;
1462    if (is_dx(dir)) {
1463        retval = ext3_dx_add_entry(handle, dentry, inode);
1464        if (!retval || (retval != ERR_BAD_DX_DIR))
1465            return retval;
1466        EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1467        dx_fallback++;
1468        ext3_mark_inode_dirty(handle, dir);
1469    }
1470    blocks = dir->i_size >> sb->s_blocksize_bits;
1471    for (block = 0; block < blocks; block++) {
1472        bh = ext3_bread(handle, dir, block, 0, &retval);
1473        if(!bh)
1474            return retval;
1475        retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1476        if (retval != -ENOSPC)
1477            return retval;
1478
1479        if (blocks == 1 && !dx_fallback &&
1480            EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1481            return make_indexed_dir(handle, dentry, inode, bh);
1482        brelse(bh);
1483    }
1484    bh = ext3_append(handle, dir, &block, &retval);
1485    if (!bh)
1486        return retval;
1487    de = (struct ext3_dir_entry_2 *) bh->b_data;
1488    de->inode = 0;
1489    de->rec_len = ext3_rec_len_to_disk(blocksize);
1490    return add_dirent_to_buf(handle, dentry, inode, de, bh);
1491}
1492
1493/*
1494 * Returns 0 for success, or a negative error value
1495 */
1496static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1497                 struct inode *inode)
1498{
1499    struct dx_frame frames[2], *frame;
1500    struct dx_entry *entries, *at;
1501    struct dx_hash_info hinfo;
1502    struct buffer_head * bh;
1503    struct inode *dir = dentry->d_parent->d_inode;
1504    struct super_block * sb = dir->i_sb;
1505    struct ext3_dir_entry_2 *de;
1506    int err;
1507
1508    frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1509    if (!frame)
1510        return err;
1511    entries = frame->entries;
1512    at = frame->at;
1513
1514    if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1515        goto cleanup;
1516
1517    BUFFER_TRACE(bh, "get_write_access");
1518    err = ext3_journal_get_write_access(handle, bh);
1519    if (err)
1520        goto journal_error;
1521
1522    err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1523    if (err != -ENOSPC) {
1524        bh = NULL;
1525        goto cleanup;
1526    }
1527
1528    /* Block full, should compress but for now just split */
1529    dxtrace(printk("using %u of %u node entries\n",
1530               dx_get_count(entries), dx_get_limit(entries)));
1531    /* Need to split index? */
1532    if (dx_get_count(entries) == dx_get_limit(entries)) {
1533        u32 newblock;
1534        unsigned icount = dx_get_count(entries);
1535        int levels = frame - frames;
1536        struct dx_entry *entries2;
1537        struct dx_node *node2;
1538        struct buffer_head *bh2;
1539
1540        if (levels && (dx_get_count(frames->entries) ==
1541                   dx_get_limit(frames->entries))) {
1542            ext3_warning(sb, __func__,
1543                     "Directory index full!");
1544            err = -ENOSPC;
1545            goto cleanup;
1546        }
1547        bh2 = ext3_append (handle, dir, &newblock, &err);
1548        if (!(bh2))
1549            goto cleanup;
1550        node2 = (struct dx_node *)(bh2->b_data);
1551        entries2 = node2->entries;
1552        memset(&node2->fake, 0, sizeof(struct fake_dirent));
1553        node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1554        BUFFER_TRACE(frame->bh, "get_write_access");
1555        err = ext3_journal_get_write_access(handle, frame->bh);
1556        if (err)
1557            goto journal_error;
1558        if (levels) {
1559            unsigned icount1 = icount/2, icount2 = icount - icount1;
1560            unsigned hash2 = dx_get_hash(entries + icount1);
1561            dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1562
1563            BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1564            err = ext3_journal_get_write_access(handle,
1565                                 frames[0].bh);
1566            if (err)
1567                goto journal_error;
1568
1569            memcpy ((char *) entries2, (char *) (entries + icount1),
1570                icount2 * sizeof(struct dx_entry));
1571            dx_set_count (entries, icount1);
1572            dx_set_count (entries2, icount2);
1573            dx_set_limit (entries2, dx_node_limit(dir));
1574
1575            /* Which index block gets the new entry? */
1576            if (at - entries >= icount1) {
1577                frame->at = at = at - entries - icount1 + entries2;
1578                frame->entries = entries = entries2;
1579                swap(frame->bh, bh2);
1580            }
1581            dx_insert_block (frames + 0, hash2, newblock);
1582            dxtrace(dx_show_index ("node", frames[1].entries));
1583            dxtrace(dx_show_index ("node",
1584                   ((struct dx_node *) bh2->b_data)->entries));
1585            err = ext3_journal_dirty_metadata(handle, bh2);
1586            if (err)
1587                goto journal_error;
1588            brelse (bh2);
1589        } else {
1590            dxtrace(printk("Creating second level index...\n"));
1591            memcpy((char *) entries2, (char *) entries,
1592                   icount * sizeof(struct dx_entry));
1593            dx_set_limit(entries2, dx_node_limit(dir));
1594
1595            /* Set up root */
1596            dx_set_count(entries, 1);
1597            dx_set_block(entries + 0, newblock);
1598            ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1599
1600            /* Add new access path frame */
1601            frame = frames + 1;
1602            frame->at = at = at - entries + entries2;
1603            frame->entries = entries = entries2;
1604            frame->bh = bh2;
1605            err = ext3_journal_get_write_access(handle,
1606                                 frame->bh);
1607            if (err)
1608                goto journal_error;
1609        }
1610        err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1611        if (err)
1612            goto journal_error;
1613    }
1614    de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1615    if (!de)
1616        goto cleanup;
1617    err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1618    bh = NULL;
1619    goto cleanup;
1620
1621journal_error:
1622    ext3_std_error(dir->i_sb, err);
1623cleanup:
1624    if (bh)
1625        brelse(bh);
1626    dx_release(frames);
1627    return err;
1628}
1629
1630/*
1631 * ext3_delete_entry deletes a directory entry by merging it with the
1632 * previous entry
1633 */
1634static int ext3_delete_entry (handle_t *handle,
1635                  struct inode * dir,
1636                  struct ext3_dir_entry_2 * de_del,
1637                  struct buffer_head * bh)
1638{
1639    struct ext3_dir_entry_2 * de, * pde;
1640    int i;
1641
1642    i = 0;
1643    pde = NULL;
1644    de = (struct ext3_dir_entry_2 *) bh->b_data;
1645    while (i < bh->b_size) {
1646        if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1647            return -EIO;
1648        if (de == de_del) {
1649            int err;
1650
1651            BUFFER_TRACE(bh, "get_write_access");
1652            err = ext3_journal_get_write_access(handle, bh);
1653            if (err)
1654                goto journal_error;
1655
1656            if (pde)
1657                pde->rec_len = ext3_rec_len_to_disk(
1658                    ext3_rec_len_from_disk(pde->rec_len) +
1659                    ext3_rec_len_from_disk(de->rec_len));
1660            else
1661                de->inode = 0;
1662            dir->i_version++;
1663            BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1664            err = ext3_journal_dirty_metadata(handle, bh);
1665            if (err) {
1666journal_error:
1667                ext3_std_error(dir->i_sb, err);
1668                return err;
1669            }
1670            return 0;
1671        }
1672        i += ext3_rec_len_from_disk(de->rec_len);
1673        pde = de;
1674        de = ext3_next_entry(de);
1675    }
1676    return -ENOENT;
1677}
1678
1679static int ext3_add_nondir(handle_t *handle,
1680        struct dentry *dentry, struct inode *inode)
1681{
1682    int err = ext3_add_entry(handle, dentry, inode);
1683    if (!err) {
1684        ext3_mark_inode_dirty(handle, inode);
1685        d_instantiate(dentry, inode);
1686        unlock_new_inode(inode);
1687        return 0;
1688    }
1689    drop_nlink(inode);
1690    unlock_new_inode(inode);
1691    iput(inode);
1692    return err;
1693}
1694
1695/*
1696 * By the time this is called, we already have created
1697 * the directory cache entry for the new file, but it
1698 * is so far negative - it has no inode.
1699 *
1700 * If the create succeeds, we fill in the inode information
1701 * with d_instantiate().
1702 */
1703static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1704        struct nameidata *nd)
1705{
1706    handle_t *handle;
1707    struct inode * inode;
1708    int err, retries = 0;
1709
1710    dquot_initialize(dir);
1711
1712retry:
1713    handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1714                    EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1715                    EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1716    if (IS_ERR(handle))
1717        return PTR_ERR(handle);
1718
1719    if (IS_DIRSYNC(dir))
1720        handle->h_sync = 1;
1721
1722    inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1723    err = PTR_ERR(inode);
1724    if (!IS_ERR(inode)) {
1725        inode->i_op = &ext3_file_inode_operations;
1726        inode->i_fop = &ext3_file_operations;
1727        ext3_set_aops(inode);
1728        err = ext3_add_nondir(handle, dentry, inode);
1729    }
1730    ext3_journal_stop(handle);
1731    if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1732        goto retry;
1733    return err;
1734}
1735
1736static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1737            int mode, dev_t rdev)
1738{
1739    handle_t *handle;
1740    struct inode *inode;
1741    int err, retries = 0;
1742
1743    if (!new_valid_dev(rdev))
1744        return -EINVAL;
1745
1746    dquot_initialize(dir);
1747
1748retry:
1749    handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1750                    EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1751                    EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1752    if (IS_ERR(handle))
1753        return PTR_ERR(handle);
1754
1755    if (IS_DIRSYNC(dir))
1756        handle->h_sync = 1;
1757
1758    inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1759    err = PTR_ERR(inode);
1760    if (!IS_ERR(inode)) {
1761        init_special_inode(inode, inode->i_mode, rdev);
1762#ifdef CONFIG_EXT3_FS_XATTR
1763        inode->i_op = &ext3_special_inode_operations;
1764#endif
1765        err = ext3_add_nondir(handle, dentry, inode);
1766    }
1767    ext3_journal_stop(handle);
1768    if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1769        goto retry;
1770    return err;
1771}
1772
1773static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1774{
1775    handle_t *handle;
1776    struct inode * inode;
1777    struct buffer_head * dir_block = NULL;
1778    struct ext3_dir_entry_2 * de;
1779    int err, retries = 0;
1780
1781    if (dir->i_nlink >= EXT3_LINK_MAX)
1782        return -EMLINK;
1783
1784    dquot_initialize(dir);
1785
1786retry:
1787    handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1788                    EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1789                    EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1790    if (IS_ERR(handle))
1791        return PTR_ERR(handle);
1792
1793    if (IS_DIRSYNC(dir))
1794        handle->h_sync = 1;
1795
1796    inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1797    err = PTR_ERR(inode);
1798    if (IS_ERR(inode))
1799        goto out_stop;
1800
1801    inode->i_op = &ext3_dir_inode_operations;
1802    inode->i_fop = &ext3_dir_operations;
1803    inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1804    dir_block = ext3_bread (handle, inode, 0, 1, &err);
1805    if (!dir_block)
1806        goto out_clear_inode;
1807
1808    BUFFER_TRACE(dir_block, "get_write_access");
1809    err = ext3_journal_get_write_access(handle, dir_block);
1810    if (err)
1811        goto out_clear_inode;
1812
1813    de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1814    de->inode = cpu_to_le32(inode->i_ino);
1815    de->name_len = 1;
1816    de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1817    strcpy (de->name, ".");
1818    ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1819    de = ext3_next_entry(de);
1820    de->inode = cpu_to_le32(dir->i_ino);
1821    de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1822                    EXT3_DIR_REC_LEN(1));
1823    de->name_len = 2;
1824    strcpy (de->name, "..");
1825    ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1826    inode->i_nlink = 2;
1827    BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1828    err = ext3_journal_dirty_metadata(handle, dir_block);
1829    if (err)
1830        goto out_clear_inode;
1831
1832    err = ext3_mark_inode_dirty(handle, inode);
1833    if (!err)
1834        err = ext3_add_entry (handle, dentry, inode);
1835
1836    if (err) {
1837out_clear_inode:
1838        inode->i_nlink = 0;
1839        unlock_new_inode(inode);
1840        ext3_mark_inode_dirty(handle, inode);
1841        iput (inode);
1842        goto out_stop;
1843    }
1844    inc_nlink(dir);
1845    ext3_update_dx_flag(dir);
1846    err = ext3_mark_inode_dirty(handle, dir);
1847    if (err)
1848        goto out_clear_inode;
1849
1850    d_instantiate(dentry, inode);
1851    unlock_new_inode(inode);
1852out_stop:
1853    brelse(dir_block);
1854    ext3_journal_stop(handle);
1855    if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1856        goto retry;
1857    return err;
1858}
1859
1860/*
1861 * routine to check that the specified directory is empty (for rmdir)
1862 */
1863static int empty_dir (struct inode * inode)
1864{
1865    unsigned long offset;
1866    struct buffer_head * bh;
1867    struct ext3_dir_entry_2 * de, * de1;
1868    struct super_block * sb;
1869    int err = 0;
1870
1871    sb = inode->i_sb;
1872    if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1873        !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1874        if (err)
1875            ext3_error(inode->i_sb, __func__,
1876                   "error %d reading directory #%lu offset 0",
1877                   err, inode->i_ino);
1878        else
1879            ext3_warning(inode->i_sb, __func__,
1880                     "bad directory (dir #%lu) - no data block",
1881                     inode->i_ino);
1882        return 1;
1883    }
1884    de = (struct ext3_dir_entry_2 *) bh->b_data;
1885    de1 = ext3_next_entry(de);
1886    if (le32_to_cpu(de->inode) != inode->i_ino ||
1887            !le32_to_cpu(de1->inode) ||
1888            strcmp (".", de->name) ||
1889            strcmp ("..", de1->name)) {
1890        ext3_warning (inode->i_sb, "empty_dir",
1891                  "bad directory (dir #%lu) - no `.' or `..'",
1892                  inode->i_ino);
1893        brelse (bh);
1894        return 1;
1895    }
1896    offset = ext3_rec_len_from_disk(de->rec_len) +
1897            ext3_rec_len_from_disk(de1->rec_len);
1898    de = ext3_next_entry(de1);
1899    while (offset < inode->i_size ) {
1900        if (!bh ||
1901            (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1902            err = 0;
1903            brelse (bh);
1904            bh = ext3_bread (NULL, inode,
1905                offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1906            if (!bh) {
1907                if (err)
1908                    ext3_error(sb, __func__,
1909                           "error %d reading directory"
1910                           " #%lu offset %lu",
1911                           err, inode->i_ino, offset);
1912                offset += sb->s_blocksize;
1913                continue;
1914            }
1915            de = (struct ext3_dir_entry_2 *) bh->b_data;
1916        }
1917        if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1918            de = (struct ext3_dir_entry_2 *)(bh->b_data +
1919                             sb->s_blocksize);
1920            offset = (offset | (sb->s_blocksize - 1)) + 1;
1921            continue;
1922        }
1923        if (le32_to_cpu(de->inode)) {
1924            brelse (bh);
1925            return 0;
1926        }
1927        offset += ext3_rec_len_from_disk(de->rec_len);
1928        de = ext3_next_entry(de);
1929    }
1930    brelse (bh);
1931    return 1;
1932}
1933
1934/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1935 * such inodes, starting at the superblock, in case we crash before the
1936 * file is closed/deleted, or in case the inode truncate spans multiple
1937 * transactions and the last transaction is not recovered after a crash.
1938 *
1939 * At filesystem recovery time, we walk this list deleting unlinked
1940 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1941 */
1942int ext3_orphan_add(handle_t *handle, struct inode *inode)
1943{
1944    struct super_block *sb = inode->i_sb;
1945    struct ext3_iloc iloc;
1946    int err = 0, rc;
1947
1948    mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1949    if (!list_empty(&EXT3_I(inode)->i_orphan))
1950        goto out_unlock;
1951
1952    /* Orphan handling is only valid for files with data blocks
1953     * being truncated, or files being unlinked. */
1954
1955    /* @@@ FIXME: Observation from aviro:
1956     * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1957     * here (on s_orphan_lock), so race with ext3_link() which might bump
1958     * ->i_nlink. For, say it, character device. Not a regular file,
1959     * not a directory, not a symlink and ->i_nlink > 0.
1960     *
1961     * tytso, 4/25/2009: I'm not sure how that could happen;
1962     * shouldn't the fs core protect us from these sort of
1963     * unlink()/link() races?
1964     */
1965    J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1966        S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1967
1968    BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1969    err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1970    if (err)
1971        goto out_unlock;
1972
1973    err = ext3_reserve_inode_write(handle, inode, &iloc);
1974    if (err)
1975        goto out_unlock;
1976
1977    /* Insert this inode at the head of the on-disk orphan list... */
1978    NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1979    EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1980    err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1981    rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1982    if (!err)
1983        err = rc;
1984
1985    /* Only add to the head of the in-memory list if all the
1986     * previous operations succeeded. If the orphan_add is going to
1987     * fail (possibly taking the journal offline), we can't risk
1988     * leaving the inode on the orphan list: stray orphan-list
1989     * entries can cause panics at unmount time.
1990     *
1991     * This is safe: on error we're going to ignore the orphan list
1992     * anyway on the next recovery. */
1993    if (!err)
1994        list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1995
1996    jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1997    jbd_debug(4, "orphan inode %lu will point to %d\n",
1998            inode->i_ino, NEXT_ORPHAN(inode));
1999out_unlock:
2000    mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
2001    ext3_std_error(inode->i_sb, err);
2002    return err;
2003}
2004
2005/*
2006 * ext3_orphan_del() removes an unlinked or truncated inode from the list
2007 * of such inodes stored on disk, because it is finally being cleaned up.
2008 */
2009int ext3_orphan_del(handle_t *handle, struct inode *inode)
2010{
2011    struct list_head *prev;
2012    struct ext3_inode_info *ei = EXT3_I(inode);
2013    struct ext3_sb_info *sbi;
2014    unsigned long ino_next;
2015    struct ext3_iloc iloc;
2016    int err = 0;
2017
2018    mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2019    if (list_empty(&ei->i_orphan))
2020        goto out;
2021
2022    ino_next = NEXT_ORPHAN(inode);
2023    prev = ei->i_orphan.prev;
2024    sbi = EXT3_SB(inode->i_sb);
2025
2026    jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2027
2028    list_del_init(&ei->i_orphan);
2029
2030    /* If we're on an error path, we may not have a valid
2031     * transaction handle with which to update the orphan list on
2032     * disk, but we still need to remove the inode from the linked
2033     * list in memory. */
2034    if (!handle)
2035        goto out;
2036
2037    err = ext3_reserve_inode_write(handle, inode, &iloc);
2038    if (err)
2039        goto out_err;
2040
2041    if (prev == &sbi->s_orphan) {
2042        jbd_debug(4, "superblock will point to %lu\n", ino_next);
2043        BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2044        err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2045        if (err)
2046            goto out_brelse;
2047        sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2048        err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2049    } else {
2050        struct ext3_iloc iloc2;
2051        struct inode *i_prev =
2052            &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2053
2054        jbd_debug(4, "orphan inode %lu will point to %lu\n",
2055              i_prev->i_ino, ino_next);
2056        err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2057        if (err)
2058            goto out_brelse;
2059        NEXT_ORPHAN(i_prev) = ino_next;
2060        err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2061    }
2062    if (err)
2063        goto out_brelse;
2064    NEXT_ORPHAN(inode) = 0;
2065    err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2066
2067out_err:
2068    ext3_std_error(inode->i_sb, err);
2069out:
2070    mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2071    return err;
2072
2073out_brelse:
2074    brelse(iloc.bh);
2075    goto out_err;
2076}
2077
2078static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2079{
2080    int retval;
2081    struct inode * inode;
2082    struct buffer_head * bh;
2083    struct ext3_dir_entry_2 * de;
2084    handle_t *handle;
2085
2086    /* Initialize quotas before so that eventual writes go in
2087     * separate transaction */
2088    dquot_initialize(dir);
2089    dquot_initialize(dentry->d_inode);
2090
2091    handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2092    if (IS_ERR(handle))
2093        return PTR_ERR(handle);
2094
2095    retval = -ENOENT;
2096    bh = ext3_find_entry(dir, &dentry->d_name, &de);
2097    if (!bh)
2098        goto end_rmdir;
2099
2100    if (IS_DIRSYNC(dir))
2101        handle->h_sync = 1;
2102
2103    inode = dentry->d_inode;
2104
2105    retval = -EIO;
2106    if (le32_to_cpu(de->inode) != inode->i_ino)
2107        goto end_rmdir;
2108
2109    retval = -ENOTEMPTY;
2110    if (!empty_dir (inode))
2111        goto end_rmdir;
2112
2113    retval = ext3_delete_entry(handle, dir, de, bh);
2114    if (retval)
2115        goto end_rmdir;
2116    if (inode->i_nlink != 2)
2117        ext3_warning (inode->i_sb, "ext3_rmdir",
2118                  "empty directory has nlink!=2 (%d)",
2119                  inode->i_nlink);
2120    inode->i_version++;
2121    clear_nlink(inode);
2122    /* There's no need to set i_disksize: the fact that i_nlink is
2123     * zero will ensure that the right thing happens during any
2124     * recovery. */
2125    inode->i_size = 0;
2126    ext3_orphan_add(handle, inode);
2127    inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2128    ext3_mark_inode_dirty(handle, inode);
2129    drop_nlink(dir);
2130    ext3_update_dx_flag(dir);
2131    ext3_mark_inode_dirty(handle, dir);
2132
2133end_rmdir:
2134    ext3_journal_stop(handle);
2135    brelse (bh);
2136    return retval;
2137}
2138
2139static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2140{
2141    int retval;
2142    struct inode * inode;
2143    struct buffer_head * bh;
2144    struct ext3_dir_entry_2 * de;
2145    handle_t *handle;
2146
2147    /* Initialize quotas before so that eventual writes go
2148     * in separate transaction */
2149    dquot_initialize(dir);
2150    dquot_initialize(dentry->d_inode);
2151
2152    handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2153    if (IS_ERR(handle))
2154        return PTR_ERR(handle);
2155
2156    if (IS_DIRSYNC(dir))
2157        handle->h_sync = 1;
2158
2159    retval = -ENOENT;
2160    bh = ext3_find_entry(dir, &dentry->d_name, &de);
2161    if (!bh)
2162        goto end_unlink;
2163
2164    inode = dentry->d_inode;
2165
2166    retval = -EIO;
2167    if (le32_to_cpu(de->inode) != inode->i_ino)
2168        goto end_unlink;
2169
2170    if (!inode->i_nlink) {
2171        ext3_warning (inode->i_sb, "ext3_unlink",
2172                  "Deleting nonexistent file (%lu), %d",
2173                  inode->i_ino, inode->i_nlink);
2174        inode->i_nlink = 1;
2175    }
2176    retval = ext3_delete_entry(handle, dir, de, bh);
2177    if (retval)
2178        goto end_unlink;
2179    dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2180    ext3_update_dx_flag(dir);
2181    ext3_mark_inode_dirty(handle, dir);
2182    drop_nlink(inode);
2183    if (!inode->i_nlink)
2184        ext3_orphan_add(handle, inode);
2185    inode->i_ctime = dir->i_ctime;
2186    ext3_mark_inode_dirty(handle, inode);
2187    retval = 0;
2188
2189end_unlink:
2190    ext3_journal_stop(handle);
2191    brelse (bh);
2192    return retval;
2193}
2194
2195static int ext3_symlink (struct inode * dir,
2196        struct dentry *dentry, const char * symname)
2197{
2198    handle_t *handle;
2199    struct inode * inode;
2200    int l, err, retries = 0;
2201    int credits;
2202
2203    l = strlen(symname)+1;
2204    if (l > dir->i_sb->s_blocksize)
2205        return -ENAMETOOLONG;
2206
2207    dquot_initialize(dir);
2208
2209    if (l > EXT3_N_BLOCKS * 4) {
2210        /*
2211         * For non-fast symlinks, we just allocate inode and put it on
2212         * orphan list in the first transaction => we need bitmap,
2213         * group descriptor, sb, inode block, quota blocks.
2214         */
2215        credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2216    } else {
2217        /*
2218         * Fast symlink. We have to add entry to directory
2219         * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2220         * allocate new inode (bitmap, group descriptor, inode block,
2221         * quota blocks, sb is already counted in previous macros).
2222         */
2223        credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2224              EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2225              EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2226    }
2227retry:
2228    handle = ext3_journal_start(dir, credits);
2229    if (IS_ERR(handle))
2230        return PTR_ERR(handle);
2231
2232    if (IS_DIRSYNC(dir))
2233        handle->h_sync = 1;
2234
2235    inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2236    err = PTR_ERR(inode);
2237    if (IS_ERR(inode))
2238        goto out_stop;
2239
2240    if (l > EXT3_N_BLOCKS * 4) {
2241        inode->i_op = &ext3_symlink_inode_operations;
2242        ext3_set_aops(inode);
2243        /*
2244         * We cannot call page_symlink() with transaction started
2245         * because it calls into ext3_write_begin() which acquires page
2246         * lock which ranks below transaction start (and it can also
2247         * wait for journal commit if we are running out of space). So
2248         * we have to stop transaction now and restart it when symlink
2249         * contents is written.
2250         *
2251         * To keep fs consistent in case of crash, we have to put inode
2252         * to orphan list in the mean time.
2253         */
2254        drop_nlink(inode);
2255        err = ext3_orphan_add(handle, inode);
2256        ext3_journal_stop(handle);
2257        if (err)
2258            goto err_drop_inode;
2259        err = __page_symlink(inode, symname, l, 1);
2260        if (err)
2261            goto err_drop_inode;
2262        /*
2263         * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2264         * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2265         */
2266        handle = ext3_journal_start(dir,
2267                EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2268                EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2269        if (IS_ERR(handle)) {
2270            err = PTR_ERR(handle);
2271            goto err_drop_inode;
2272        }
2273        inc_nlink(inode);
2274        err = ext3_orphan_del(handle, inode);
2275        if (err) {
2276            ext3_journal_stop(handle);
2277            drop_nlink(inode);
2278            goto err_drop_inode;
2279        }
2280    } else {
2281        inode->i_op = &ext3_fast_symlink_inode_operations;
2282        memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2283        inode->i_size = l-1;
2284    }
2285    EXT3_I(inode)->i_disksize = inode->i_size;
2286    err = ext3_add_nondir(handle, dentry, inode);
2287out_stop:
2288    ext3_journal_stop(handle);
2289    if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2290        goto retry;
2291    return err;
2292err_drop_inode:
2293    unlock_new_inode(inode);
2294    iput(inode);
2295    return err;
2296}
2297
2298static int ext3_link (struct dentry * old_dentry,
2299        struct inode * dir, struct dentry *dentry)
2300{
2301    handle_t *handle;
2302    struct inode *inode = old_dentry->d_inode;
2303    int err, retries = 0;
2304
2305    if (inode->i_nlink >= EXT3_LINK_MAX)
2306        return -EMLINK;
2307
2308    dquot_initialize(dir);
2309
2310retry:
2311    handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2312                    EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2313    if (IS_ERR(handle))
2314        return PTR_ERR(handle);
2315
2316    if (IS_DIRSYNC(dir))
2317        handle->h_sync = 1;
2318
2319    inode->i_ctime = CURRENT_TIME_SEC;
2320    inc_nlink(inode);
2321    ihold(inode);
2322
2323    err = ext3_add_entry(handle, dentry, inode);
2324    if (!err) {
2325        ext3_mark_inode_dirty(handle, inode);
2326        d_instantiate(dentry, inode);
2327    } else {
2328        drop_nlink(inode);
2329        iput(inode);
2330    }
2331    ext3_journal_stop(handle);
2332    if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2333        goto retry;
2334    return err;
2335}
2336
2337#define PARENT_INO(buffer) \
2338    (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2339
2340/*
2341 * Anybody can rename anything with this: the permission checks are left to the
2342 * higher-level routines.
2343 */
2344static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2345               struct inode * new_dir,struct dentry *new_dentry)
2346{
2347    handle_t *handle;
2348    struct inode * old_inode, * new_inode;
2349    struct buffer_head * old_bh, * new_bh, * dir_bh;
2350    struct ext3_dir_entry_2 * old_de, * new_de;
2351    int retval, flush_file = 0;
2352
2353    dquot_initialize(old_dir);
2354    dquot_initialize(new_dir);
2355
2356    old_bh = new_bh = dir_bh = NULL;
2357
2358    /* Initialize quotas before so that eventual writes go
2359     * in separate transaction */
2360    if (new_dentry->d_inode)
2361        dquot_initialize(new_dentry->d_inode);
2362    handle = ext3_journal_start(old_dir, 2 *
2363                    EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2364                    EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2365    if (IS_ERR(handle))
2366        return PTR_ERR(handle);
2367
2368    if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2369        handle->h_sync = 1;
2370
2371    old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2372    /*
2373     * Check for inode number is _not_ due to possible IO errors.
2374     * We might rmdir the source, keep it as pwd of some process
2375     * and merrily kill the link to whatever was created under the
2376     * same name. Goodbye sticky bit ;-<
2377     */
2378    old_inode = old_dentry->d_inode;
2379    retval = -ENOENT;
2380    if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2381        goto end_rename;
2382
2383    new_inode = new_dentry->d_inode;
2384    new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2385    if (new_bh) {
2386        if (!new_inode) {
2387            brelse (new_bh);
2388            new_bh = NULL;
2389        }
2390    }
2391    if (S_ISDIR(old_inode->i_mode)) {
2392        if (new_inode) {
2393            retval = -ENOTEMPTY;
2394            if (!empty_dir (new_inode))
2395                goto end_rename;
2396        }
2397        retval = -EIO;
2398        dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2399        if (!dir_bh)
2400            goto end_rename;
2401        if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2402            goto end_rename;
2403        retval = -EMLINK;
2404        if (!new_inode && new_dir!=old_dir &&
2405                new_dir->i_nlink >= EXT3_LINK_MAX)
2406            goto end_rename;
2407    }
2408    if (!new_bh) {
2409        retval = ext3_add_entry (handle, new_dentry, old_inode);
2410        if (retval)
2411            goto end_rename;
2412    } else {
2413        BUFFER_TRACE(new_bh, "get write access");
2414        retval = ext3_journal_get_write_access(handle, new_bh);
2415        if (retval)
2416            goto journal_error;
2417        new_de->inode = cpu_to_le32(old_inode->i_ino);
2418        if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2419                          EXT3_FEATURE_INCOMPAT_FILETYPE))
2420            new_de->file_type = old_de->file_type;
2421        new_dir->i_version++;
2422        new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2423        ext3_mark_inode_dirty(handle, new_dir);
2424        BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2425        retval = ext3_journal_dirty_metadata(handle, new_bh);
2426        if (retval)
2427            goto journal_error;
2428        brelse(new_bh);
2429        new_bh = NULL;
2430    }
2431
2432    /*
2433     * Like most other Unix systems, set the ctime for inodes on a
2434     * rename.
2435     */
2436    old_inode->i_ctime = CURRENT_TIME_SEC;
2437    ext3_mark_inode_dirty(handle, old_inode);
2438
2439    /*
2440     * ok, that's it
2441     */
2442    if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2443        old_de->name_len != old_dentry->d_name.len ||
2444        strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2445        (retval = ext3_delete_entry(handle, old_dir,
2446                    old_de, old_bh)) == -ENOENT) {
2447        /* old_de could have moved from under us during htree split, so
2448         * make sure that we are deleting the right entry. We might
2449         * also be pointing to a stale entry in the unused part of
2450         * old_bh so just checking inum and the name isn't enough. */
2451        struct buffer_head *old_bh2;
2452        struct ext3_dir_entry_2 *old_de2;
2453
2454        old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2455                      &old_de2);
2456        if (old_bh2) {
2457            retval = ext3_delete_entry(handle, old_dir,
2458                           old_de2, old_bh2);
2459            brelse(old_bh2);
2460        }
2461    }
2462    if (retval) {
2463        ext3_warning(old_dir->i_sb, "ext3_rename",
2464                "Deleting old file (%lu), %d, error=%d",
2465                old_dir->i_ino, old_dir->i_nlink, retval);
2466    }
2467
2468    if (new_inode) {
2469        drop_nlink(new_inode);
2470        new_inode->i_ctime = CURRENT_TIME_SEC;
2471    }
2472    old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2473    ext3_update_dx_flag(old_dir);
2474    if (dir_bh) {
2475        BUFFER_TRACE(dir_bh, "get_write_access");
2476        retval = ext3_journal_get_write_access(handle, dir_bh);
2477        if (retval)
2478            goto journal_error;
2479        PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2480        BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2481        retval = ext3_journal_dirty_metadata(handle, dir_bh);
2482        if (retval) {
2483journal_error:
2484            ext3_std_error(new_dir->i_sb, retval);
2485            goto end_rename;
2486        }
2487        drop_nlink(old_dir);
2488        if (new_inode) {
2489            drop_nlink(new_inode);
2490        } else {
2491            inc_nlink(new_dir);
2492            ext3_update_dx_flag(new_dir);
2493            ext3_mark_inode_dirty(handle, new_dir);
2494        }
2495    }
2496    ext3_mark_inode_dirty(handle, old_dir);
2497    if (new_inode) {
2498        ext3_mark_inode_dirty(handle, new_inode);
2499        if (!new_inode->i_nlink)
2500            ext3_orphan_add(handle, new_inode);
2501        if (ext3_should_writeback_data(new_inode))
2502            flush_file = 1;
2503    }
2504    retval = 0;
2505
2506end_rename:
2507    brelse (dir_bh);
2508    brelse (old_bh);
2509    brelse (new_bh);
2510    ext3_journal_stop(handle);
2511    if (retval == 0 && flush_file)
2512        filemap_flush(old_inode->i_mapping);
2513    return retval;
2514}
2515
2516/*
2517 * directories can handle most operations...
2518 */
2519const struct inode_operations ext3_dir_inode_operations = {
2520    .create = ext3_create,
2521    .lookup = ext3_lookup,
2522    .link = ext3_link,
2523    .unlink = ext3_unlink,
2524    .symlink = ext3_symlink,
2525    .mkdir = ext3_mkdir,
2526    .rmdir = ext3_rmdir,
2527    .mknod = ext3_mknod,
2528    .rename = ext3_rename,
2529    .setattr = ext3_setattr,
2530#ifdef CONFIG_EXT3_FS_XATTR
2531    .setxattr = generic_setxattr,
2532    .getxattr = generic_getxattr,
2533    .listxattr = ext3_listxattr,
2534    .removexattr = generic_removexattr,
2535#endif
2536    .check_acl = ext3_check_acl,
2537};
2538
2539const struct inode_operations ext3_special_inode_operations = {
2540    .setattr = ext3_setattr,
2541#ifdef CONFIG_EXT3_FS_XATTR
2542    .setxattr = generic_setxattr,
2543    .getxattr = generic_getxattr,
2544    .listxattr = ext3_listxattr,
2545    .removexattr = generic_removexattr,
2546#endif
2547    .check_acl = ext3_check_acl,
2548};
2549

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive