1 --- ./fs/ext3/super.c 2002/03/05 06:18:59 2.1
2 +++ ./fs/ext3/super.c 2002/03/05 06:26:56
4 "EXT3 Check option not supported\n");
7 + else if (!strcmp (this_char, "index"))
8 +#ifdef CONFIG_EXT3_INDEX
9 + set_opt (*mount_options, INDEX);
11 + printk("EXT3 index option not supported\n");
13 else if (!strcmp (this_char, "debug"))
14 set_opt (*mount_options, DEBUG);
15 else if (!strcmp (this_char, "errors")) {
16 @@ -702,6 +708,12 @@ static int ext3_setup_super(struct super
17 es->s_mtime = cpu_to_le32(CURRENT_TIME);
18 ext3_update_dynamic_rev(sb);
19 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
21 + if (test_opt(sb, INDEX))
22 + EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX);
23 + else if (EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
24 + set_opt (EXT3_SB(sb)->s_mount_opt, INDEX);
26 ext3_commit_super (sb, es, 1);
27 if (test_opt (sb, DEBUG))
29 --- ./fs/ext3/namei.c 2002/03/05 06:18:59 2.1
30 +++ ./fs/ext3/namei.c 2002/03/06 00:13:18
32 * David S. Miller (davem@caip.rutgers.edu), 1995
33 * Directory entry file type support and forward compatibility hooks
34 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
35 + * Hash Tree Directory indexing (c)
36 + * Daniel Phillips, 2001
37 + * Hash Tree Directory indexing porting
38 + * Christopher Li, 2002
43 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
44 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
46 +static struct buffer_head *ext3_append(handle_t *handle,
47 + struct inode *inode,
48 + u32 *block, int *err)
50 + struct buffer_head *bh;
52 + *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
54 + if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
55 + inode->i_size += inode->i_sb->s_blocksize;
56 + EXT3_I(inode)->i_disksize = inode->i_size;
57 + ext3_journal_get_write_access(handle,bh);
63 +#define assert(test) J_ASSERT(test)
67 +#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
70 +typedef struct { u32 v; } le_u32;
71 +typedef struct { u16 v; } le_u16;
73 +#define dxtrace_on(command) command
74 +#define dxtrace_off(command)
75 +#define dxtrace dxtrace_off
98 + * dx_root_info is laid out so that if it should somehow get overlaid by a
99 + * dirent the two low bits of the hash version will be zero. Therefore, the
100 + * hash version mod 4 should never be 0. Sincerely, the paranoia department.
105 + struct fake_dirent dot;
107 + struct fake_dirent dotdot;
108 + char dotdot_name[4];
109 + struct dx_root_info
111 + le_u32 reserved_zero;
112 + u8 hash_version; /* 0 now, 1 at release */
113 + u8 info_length; /* 8 */
114 + u8 indirect_levels;
118 + struct dx_entry entries[0];
123 + struct fake_dirent fake;
124 + struct dx_entry entries[0];
130 + struct buffer_head *bh;
131 + struct dx_entry *entries;
132 + struct dx_entry *at;
141 +typedef struct ext3_dir_entry_2 ext3_dirent;
142 +static inline unsigned dx_get_block (struct dx_entry *entry);
143 +static void dx_set_block (struct dx_entry *entry, unsigned value);
144 +static inline unsigned dx_get_hash (struct dx_entry *entry);
145 +static void dx_set_hash (struct dx_entry *entry, unsigned value);
146 +static unsigned dx_get_count (struct dx_entry *entries);
147 +static unsigned dx_get_limit (struct dx_entry *entries);
148 +static void dx_set_count (struct dx_entry *entries, unsigned value);
149 +static void dx_set_limit (struct dx_entry *entries, unsigned value);
150 +static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
151 +static unsigned dx_node_limit (struct inode *dir);
152 +static unsigned dx_hack_hash (const u8 *name, int len);
153 +static struct dx_frame *dx_probe (struct inode *dir, u32 hash, struct dx_frame *frame);
154 +static void dx_release (struct dx_frame *frames);
155 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[]);
156 +static void dx_sort_map(struct dx_map_entry *map, unsigned count);
157 +static ext3_dirent *dx_copy_dirents (char *from, char *to,
158 + struct dx_map_entry *map, int count);
159 +static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
162 +#ifdef CONFIG_EXT3_INDEX
164 + * Future: use high four bits of block for coalesce-on-delete flags
165 + * Mask them off for now.
168 +static inline unsigned dx_get_block (struct dx_entry *entry)
170 + return le32_to_cpu(entry->block.v) & 0x00ffffff;
173 +static inline void dx_set_block (struct dx_entry *entry, unsigned value)
175 + entry->block.v = cpu_to_le32(value);
178 +static inline unsigned dx_get_hash (struct dx_entry *entry)
180 + return le32_to_cpu(entry->hash.v);
183 +static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
185 + entry->hash.v = cpu_to_le32(value);
188 +static inline unsigned dx_get_count (struct dx_entry *entries)
190 + return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
193 +static inline unsigned dx_get_limit (struct dx_entry *entries)
195 + return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
198 +static inline void dx_set_count (struct dx_entry *entries, unsigned value)
200 + ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
203 +static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
205 + ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
208 +static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
210 + unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
211 + EXT3_DIR_REC_LEN(2) - infosize;
212 + return 0? 20: entry_space / sizeof(struct dx_entry);
215 +static inline unsigned dx_node_limit (struct inode *dir)
217 + unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
218 + return 0? 22: entry_space / sizeof(struct dx_entry);
221 +/* Hash function - not bad, but still looking for an ideal default */
223 +static unsigned dx_hack_hash (const u8 *name, int len)
225 + u32 hash0 = 0x12a3fe2d, hash1 = 0x37abe8f9;
228 + u32 hash = hash1 + (hash0 ^ (*name++ * 7152373));
229 + if (hash & 0x80000000) hash -= 0x7fffffff;
236 +#define dx_hash(s,n) (dx_hack_hash(s,n) << 1)
241 +static void dx_show_index (char * label, struct dx_entry *entries)
243 + int i, n = dx_get_count (entries);
244 + printk("%s index ", label);
245 + for (i = 0; i < n; i++)
247 + printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
259 +static struct stats dx_show_leaf (ext3_dirent *de, int size, int show_names)
261 + unsigned names = 0, space = 0;
262 + char *base = (char *) de;
264 + while ((char *) de < base + size)
270 + int len = de->name_len;
271 + char *name = de->name;
272 + while (len--) printk("%c", *name++);
273 + printk(":%x.%u ", dx_hash (de->name, de->name_len), ((char *) de - base));
275 + space += EXT3_DIR_REC_LEN(de->name_len);
278 + de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
280 + printk("(%i)\n", names);
281 + return (struct stats) { names, space, 1 };
284 +struct stats dx_show_entries (struct inode *dir, struct dx_entry *entries, int levels)
286 + unsigned blocksize = dir->i_sb->s_blocksize;
287 + unsigned count = dx_get_count (entries), names = 0, space = 0, i;
288 + unsigned bcount = 0;
289 + struct buffer_head *bh;
291 + printk("%i indexed blocks...\n", count);
292 + for (i = 0; i < count; i++, entries++)
294 + u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
295 + u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
296 + struct stats stats;
297 + printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
298 + if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
300 + dx_show_entries (dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
301 + dx_show_leaf ((ext3_dirent *) bh->b_data, blocksize, 0);
302 + names += stats.names;
303 + space += stats.space;
304 + bcount += stats.bcount;
308 + printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
309 + names, space/bcount,(space/bcount)*100/blocksize);
310 + return (struct stats) { names, space, bcount};
314 + * Probe for a directory leaf block to search
317 +static struct dx_frame *
318 +dx_probe(struct inode *dir, u32 hash, struct dx_frame *frame_in)
320 + unsigned count, indirect;
321 + struct dx_entry *at, *entries, *p, *q, *m;
322 + struct dx_root *root;
323 + struct buffer_head *bh;
324 + struct dx_frame *frame = frame_in;
328 + if (!(bh = ext3_bread(NULL, dir, 0, 0, &err)))
330 + root = (struct dx_root *) bh->b_data;
331 + if (root->info.hash_version > 0 || root->info.unused_flags & 1) {
335 + if ((indirect = root->info.indirect_levels) > 1) {
339 + entries = (struct dx_entry *) (((char *) &root->info) + root->info.info_length);
340 + assert (dx_get_limit(entries) == dx_root_limit(dir, root->info.info_length));
341 + dxtrace (printk("Look up %x", hash));
344 + count = dx_get_count(entries);
345 + assert (count && count <= dx_get_limit(entries));
347 + q = entries + count - 1;
351 + dxtrace(printk("."));
352 + if (dx_get_hash(m) > hash)
358 + if (0) // linear search cross check
360 + unsigned n = count - 1;
364 + dxtrace(printk(","));
365 + if (dx_get_hash(++at) > hash)
371 + assert (at == p - 1);
375 + dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
377 + frame->entries = entries;
379 + if (!indirect--) return frame;
380 + if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0,&err)))
382 + at = entries = ((struct dx_node *) bh->b_data)->entries;
383 + assert (dx_get_limit(entries) == dx_node_limit (dir));
387 + while (frame >= frame_in) {
395 +static void dx_release (struct dx_frame *frames)
397 + if (frames[0].bh == NULL)
400 + if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
401 + brelse (frames[1].bh);
402 + brelse (frames[0].bh);
406 + * Directory block splitting, compacting
409 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[])
412 + char *base = (char *) de;
413 + while ((char *) de < base + size)
415 + map[count].hash = dx_hash (de->name, de->name_len);
416 + map[count].offs = (u32) ((char *) de - base);
417 + de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
423 +static void dx_sort_map (struct dx_map_entry *map, unsigned count)
425 + struct dx_map_entry *p, *q, *top = map + count - 1;
427 + /* Combsort until bubble sort doesn't suck */
430 + count = count*10/13;
431 + if (count - 9 < 2) /* 9, 10 -> 11 */
433 + for (p = top, q = p - count; q >= map; p--, q--)
434 + if (p->hash < q->hash)
437 + /* Garden variety bubble sort */
443 + if (q[1].hash >= q[0].hash)
451 +static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
453 + struct dx_entry *entries = frame->entries;
454 + struct dx_entry *old = frame->at, *new = old + 1;
455 + int count = dx_get_count(entries);
457 + assert(count < dx_get_limit(entries));
458 + assert(old < entries + count);
459 + memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
460 + dx_set_hash(new, hash);
461 + dx_set_block(new, block);
462 + dx_set_count(entries, count + 1);
466 +static void ext3_update_dx_flag(struct inode *inode)
468 + if (!test_opt(inode->i_sb, INDEX))
469 + EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
473 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
479 + * p is at least 6 bytes before the end of page
481 +static inline ext3_dirent *ext3_next_entry(ext3_dirent *p)
483 + return (ext3_dirent *)((char*)p + le16_to_cpu(p->rec_len));
490 * finds an entry in the specified directory with the wanted name. It
492 * The returned buffer_head has ->b_count elevated. The caller is expected
493 * to brelse() it when appropriate.
497 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
498 struct ext3_dir_entry_2 ** res_dir)
500 @@ -119,10 +564,76 @@
503 struct inode *dir = dentry->d_parent->d_inode;
506 + unsigned blocksize;
507 + ext3_dirent *de, *top;
511 + blocksize = sb->s_blocksize;
512 + namelen = dentry->d_name.len;
513 + name = dentry->d_name.name;
514 + if (namelen > EXT3_NAME_LEN)
516 + if (ext3_dx && is_dx(dir)) {
517 + u32 hash = dx_hash (name, namelen);
518 + struct dx_frame frames[2], *frame;
519 + if (!(frame = dx_probe (dir, hash, frames)))
522 + block = dx_get_block(frame->at);
523 + if (!(bh = ext3_bread (NULL,dir, block, 0, &err)))
525 + de = (ext3_dirent *) bh->b_data;
526 + top = (ext3_dirent *) ((char *) de + blocksize -
527 + EXT3_DIR_REC_LEN(0));
528 + for (; de < top; de = ext3_next_entry(de))
529 + if (ext3_match (namelen, name, de)) {
530 + if (!ext3_check_dir_entry("ext3_find_entry",
532 + (block<<EXT3_BLOCK_SIZE_BITS(sb))
533 + +((char *)de - bh->b_data))) {
541 + /* Same hash continues in next block? Search on. */
542 + if (++(frame->at) == frame->entries + dx_get_count(frame->entries))
544 + struct buffer_head *bh2;
545 + if (frame == frames)
547 + if (++(frames->at) == frames->entries + dx_get_count(frames->entries))
549 + /* should omit read if not continued */
550 + if (!(bh2 = ext3_bread (NULL, dir,
551 + dx_get_block(frames->at),
554 + brelse (frame->bh);
556 + frame->at = frame->entries = ((struct dx_node *) bh2->b_data)->entries;
557 + /* Subtle: the 0th entry has the count, find the hash in frame above */
558 + if ((dx_get_hash(frames->at) & -2) == hash)
562 + if ((dx_get_hash(frame->at) & -2) == hash)
565 + dxtrace(printk("%s not found\n", name));
566 + dx_release (frames);
569 + dx_release (frames);
574 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
575 start = dir->u.ext3_i.i_dir_start_lookup;
576 if (start >= nblocks)
578 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
581 +static ext3_dirent *
582 +dx_copy_dirents (char *from, char *to, struct dx_map_entry *map, int count)
584 + unsigned rec_len = 0;
587 + ext3_dirent *de = (ext3_dirent *) (from + map->offs);
588 + rec_len = EXT3_DIR_REC_LEN(de->name_len);
589 + memcpy (to, de, rec_len);
590 + ((ext3_dirent *) to)->rec_len = rec_len;
594 + return (ext3_dirent *) (to - rec_len);
597 +#ifdef CONFIG_EXT3_INDEX
598 +static ext3_dirent *do_split(handle_t *handle, struct inode *dir,
599 + struct buffer_head **bh,struct dx_frame *frame,
600 + u32 hash, int *error)
602 + unsigned blocksize = dir->i_sb->s_blocksize;
603 + unsigned count, continued;
604 + struct buffer_head *bh2;
606 + unsigned MAX_DX_MAP = PAGE_CACHE_SIZE/EXT3_DIR_REC_LEN(1) + 1;
608 + struct dx_map_entry map[MAX_DX_MAP];
609 + char *data1 = (*bh)->b_data, *data2, *data3;
611 + ext3_dirent *de, *de2;
613 + bh2 = ext3_append (handle, dir, &newblock, error);
618 + return (ext3_dirent *)bh2;
621 + BUFFER_TRACE(*bh, "get_write_access");
622 + ext3_journal_get_write_access(handle, *bh);
623 + BUFFER_TRACE(frame->bh, "get_write_access");
624 + ext3_journal_get_write_access(handle, frame->bh);
626 + data2 = bh2->b_data;
628 + count = dx_make_map ((ext3_dirent *) data1, blocksize, map);
629 + split = count/2; // need to adjust to actual middle
630 + dx_sort_map (map, count);
631 + hash2 = map[split].hash;
632 + continued = hash2 == map[split - 1].hash;
633 + dxtrace(printk("Split block %i at %x, %i/%i\n",
634 + dx_get_block(frame->at), hash2, split, count-split));
636 + /* Fancy dance to stay within two buffers */
637 + de2 = dx_copy_dirents (data1, data2, map + split, count - split);
638 + data3 = (char *) de2 + de2->rec_len;
639 + de = dx_copy_dirents (data1, data3, map, split);
640 + memcpy(data1, data3, (char *) de + de->rec_len - data3);
641 + de = (ext3_dirent *) ((char *) de - data3 + data1); // relocate de
642 + de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
643 + de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
644 + dxtrace(dx_show_leaf ((ext3_dirent *) data1, blocksize, 1));
645 + dxtrace(dx_show_leaf ((ext3_dirent *) data2, blocksize, 1));
647 + /* Which block gets the new entry? */
653 + dx_insert_block (frame, hash2 + continued, newblock);
654 + ext3_journal_dirty_metadata (handle, bh2);
656 + ext3_journal_dirty_metadata (handle, frame->bh);
657 + dxtrace(dx_show_index ("frame", frame->entries));
668 * AKPM: the journalling code here looks wrong on the error paths
671 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
674 @@ -258,117 +852,281 @@
675 const char *name = dentry->d_name.name;
676 int namelen = dentry->d_name.len;
677 unsigned long offset;
678 - unsigned short rec_len;
679 struct buffer_head * bh;
680 - struct ext3_dir_entry_2 * de, * de1;
681 - struct super_block * sb;
683 + struct super_block * sb = dir->i_sb;
685 + unsigned short reclen = EXT3_DIR_REC_LEN(namelen);
688 + unsigned blocksize = sb->s_blocksize;
689 + unsigned nlen, rlen;
695 - bh = ext3_bread (handle, dir, 0, 0, &retval);
698 - rec_len = EXT3_DIR_REC_LEN(namelen);
700 - de = (struct ext3_dir_entry_2 *) bh->b_data;
702 - if ((char *)de >= sb->s_blocksize + bh->b_data) {
705 - bh = ext3_bread (handle, dir,
706 - offset >> EXT3_BLOCK_SIZE_BITS(sb), 1, &retval);
709 - if (dir->i_size <= offset) {
710 - if (dir->i_size == 0) {
713 + if (ext3_dx && is_dx(dir)) {
714 + struct dx_frame frames[2], *frame;
715 + struct dx_entry *entries, *at;
719 + hash = dx_hash(name, namelen);
720 + /* FIXME: do something if dx_probe() fails here */
721 + frame = dx_probe(dir, hash, frames);
722 + entries = frame->entries;
725 + if (!(bh = ext3_bread(handle,dir, dx_get_block(at), 0,&retval)))
728 + BUFFER_TRACE(bh, "get_write_access");
729 + ext3_journal_get_write_access(handle, bh);
731 + data1 = bh->b_data;
732 + de = (ext3_dirent *) data1;
733 + top = data1 + (0? 200: blocksize);
734 + while ((char *) de < top)
736 + /* FIXME: check EEXIST and dir */
737 + nlen = EXT3_DIR_REC_LEN(de->name_len);
738 + rlen = le16_to_cpu(de->rec_len);
739 + if ((de->inode? rlen - nlen: rlen) >= reclen)
741 + de = (ext3_dirent *) ((char *) de + rlen);
743 + /* Block full, should compress but for now just split */
744 + dxtrace(printk("using %u of %u node entries\n",
745 + dx_get_count(entries), dx_get_limit(entries)));
746 + /* Need to split index? */
747 + if (dx_get_count(entries) == dx_get_limit(entries))
750 + unsigned icount = dx_get_count(entries);
751 + int levels = frame - frames;
752 + struct dx_entry *entries2;
753 + struct dx_node *node2;
754 + struct buffer_head *bh2;
755 + if (levels && dx_get_count(frames->entries) == dx_get_limit(frames->entries))
757 + bh2 = ext3_append (handle, dir, &newblock, &retval);
760 + node2 = (struct dx_node *)(bh2->b_data);
761 + entries2 = node2->entries;
762 + node2->fake.rec_len = cpu_to_le16(blocksize);
763 + node2->fake.inode = 0;
764 + BUFFER_TRACE(frame->bh, "get_write_access");
765 + ext3_journal_get_write_access(handle, frame->bh);
768 + unsigned icount1 = icount/2, icount2 = icount - icount1;
769 + unsigned hash2 = dx_get_hash(entries + icount1);
770 + dxtrace(printk("Split index %i/%i\n", icount1, icount2));
772 + BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
773 + ext3_journal_get_write_access(handle, frames[0].bh);
775 + memcpy ((char *) entries2, (char *) (entries + icount1),
776 + icount2 * sizeof(struct dx_entry));
777 + dx_set_count (entries, icount1);
778 + dx_set_count (entries2, icount2);
779 + dx_set_limit (entries2, dx_node_limit(dir));
781 + /* Which index block gets the new entry? */
782 + if (at - entries >= icount1) {
783 + frame->at = at = at - entries - icount1 + entries2;
784 + frame->entries = entries = entries2;
785 + swap(frame->bh, bh2);
788 - ext3_debug ("creating next block\n");
790 - BUFFER_TRACE(bh, "get_write_access");
791 - ext3_journal_get_write_access(handle, bh);
792 - de = (struct ext3_dir_entry_2 *) bh->b_data;
794 - de->rec_len = le16_to_cpu(sb->s_blocksize);
795 - dir->u.ext3_i.i_disksize =
796 - dir->i_size = offset + sb->s_blocksize;
797 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
798 - ext3_mark_inode_dirty(handle, dir);
799 + dx_insert_block (frames + 0, hash2, newblock);
800 + dxtrace(dx_show_index ("node", frames[1].entries));
801 + dxtrace(dx_show_index ("node",
802 + ((struct dx_node *) bh2->b_data)->entries));
803 + ext3_journal_dirty_metadata(handle, bh2);
807 - ext3_debug ("skipping to next block\n");
809 - de = (struct ext3_dir_entry_2 *) bh->b_data;
810 + dxtrace(printk("Creating second level index...\n"));
811 + memcpy((char *) entries2, (char *) entries,
812 + icount * sizeof(struct dx_entry));
813 + dx_set_limit(entries2, dx_node_limit(dir));
816 + dx_set_count(entries, 1);
817 + dx_set_block(entries + 0, newblock);
818 + ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
820 + /* Add new access path frame */
821 + frame = frames + 1;
822 + frame->at = at = at - entries + entries2;
823 + frame->entries = entries = entries2;
825 + ext3_journal_get_write_access(handle, frame->bh);
827 + ext3_journal_dirty_metadata(handle, frames[0].bh);
829 - if (!ext3_check_dir_entry ("ext3_add_entry", dir, de, bh,
834 - if (ext3_match (namelen, name, de)) {
835 + de = do_split(handle, dir, &bh, frame, hash, &retval);
836 + dx_release (frames);
839 + nlen = EXT3_DIR_REC_LEN(de->name_len);
840 + rlen = le16_to_cpu(de->rec_len);
844 + dx_release (frames);
848 + ext3_warning(sb, __FUNCTION__, "Directory index full!\n");
853 + dx_release (frames);
857 + blocks = dir->i_size >> sb->s_blocksize_bits;
858 + for (block = 0, offset = 0; block < blocks; block++) {
859 + bh = ext3_bread(handle, dir, block, 0, &retval);
862 + de = (ext3_dirent *)bh->b_data;
863 + top = bh->b_data + blocksize - reclen;
864 + while ((char *) de <= top) {
865 + if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
870 + if (ext3_match (namelen, name, de)) {
874 - if ((le32_to_cpu(de->inode) == 0 &&
875 - le16_to_cpu(de->rec_len) >= rec_len) ||
876 - (le16_to_cpu(de->rec_len) >=
877 - EXT3_DIR_REC_LEN(de->name_len) + rec_len)) {
878 - BUFFER_TRACE(bh, "get_write_access");
879 - ext3_journal_get_write_access(handle, bh);
880 - /* By now the buffer is marked for journaling */
881 - offset += le16_to_cpu(de->rec_len);
882 - if (le32_to_cpu(de->inode)) {
883 - de1 = (struct ext3_dir_entry_2 *) ((char *) de +
884 - EXT3_DIR_REC_LEN(de->name_len));
886 - cpu_to_le16(le16_to_cpu(de->rec_len) -
887 - EXT3_DIR_REC_LEN(de->name_len));
888 - de->rec_len = cpu_to_le16(
889 - EXT3_DIR_REC_LEN(de->name_len));
892 - de->file_type = EXT3_FT_UNKNOWN;
894 - de->inode = cpu_to_le32(inode->i_ino);
895 - ext3_set_de_type(dir->i_sb, de, inode->i_mode);
898 - de->name_len = namelen;
899 - memcpy (de->name, name, namelen);
901 - * XXX shouldn't update any times until successful
902 - * completion of syscall, but too many callers depend
905 - * XXX similarly, too many callers depend on
906 - * ext3_new_inode() setting the times, but error
907 - * recovery deletes the inode, so the worst that can
908 - * happen is that the times are slightly out of date
909 - * and/or different from the directory change time.
911 - dir->i_mtime = dir->i_ctime = CURRENT_TIME;
912 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
913 - ext3_mark_inode_dirty(handle, dir);
914 - dir->i_version = ++event;
915 - BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
916 - ext3_journal_dirty_metadata(handle, bh);
917 + nlen = EXT3_DIR_REC_LEN(de->name_len);
918 + rlen = le16_to_cpu(de->rec_len);
919 + if ((de->inode? rlen - nlen: rlen) >= reclen)
921 + de = (ext3_dirent *)((char *)de + rlen);
924 + if (ext3_dx && blocks == 1 && test_opt(sb, INDEX))
925 + goto dx_make_index;
928 + bh = ext3_append(handle, dir, &block, &retval);
931 + de = (ext3_dirent *) bh->b_data;
933 + de->rec_len = cpu_to_le16(rlen = blocksize);
938 + BUFFER_TRACE(bh, "get_write_access");
939 + ext3_journal_get_write_access(handle, bh);
940 + /* By now the buffer is marked for journaling */
942 + ext3_dirent *de1 = (ext3_dirent *)((char *)de + nlen);
943 + de1->rec_len = cpu_to_le16(rlen - nlen);
944 + de->rec_len = cpu_to_le16(nlen);
947 + de->file_type = EXT3_FT_UNKNOWN;
949 + de->inode = cpu_to_le32(inode->i_ino);
950 + ext3_set_de_type(dir->i_sb, de, inode->i_mode);
953 + de->name_len = namelen;
954 + memcpy (de->name, name, namelen);
956 + * XXX shouldn't update any times until successful
957 + * completion of syscall, but too many callers depend
960 + * XXX similarly, too many callers depend on
961 + * ext3_new_inode() setting the times, but error
962 + * recovery deletes the inode, so the worst that can
963 + * happen is that the times are slightly out of date
964 + * and/or different from the directory change time.
966 + dir->i_mtime = dir->i_ctime = CURRENT_TIME;
967 + ext3_update_dx_flag(dir);
968 + dir->i_version = ++event;
969 + ext3_mark_inode_dirty(handle, dir);
970 + BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
971 + ext3_journal_dirty_metadata(handle, bh);
977 + struct buffer_head *bh2;
978 + struct dx_root *root;
979 + struct dx_frame frames[2], *frame;
980 + struct dx_entry *entries;
986 + dxtrace(printk("Creating index\n"));
987 + ext3_journal_get_write_access(handle, bh);
988 + root = (struct dx_root *) bh->b_data;
990 + EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
991 + bh2 = ext3_append (handle, dir, &block, &retval);
998 - offset += le16_to_cpu(de->rec_len);
999 - de = (struct ext3_dir_entry_2 *)
1000 - ((char *) de + le16_to_cpu(de->rec_len));
1001 + data1 = bh2->b_data;
1003 + /* The 0th block becomes the root, move the dirents out */
1004 + de = (ext3_dirent *) &root->info;
1005 + len = ((char *) root) + blocksize - (char *) de;
1006 + memcpy (data1, de, len);
1007 + de = (ext3_dirent *) data1;
1008 + top = data1 + len;
1009 + while (((char *) de2=(char*)de+le16_to_cpu(de->rec_len)) < top)
1011 + de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1012 + /* Initialize the root; the dot dirents already exist */
1013 + de = (ext3_dirent *) (&root->dotdot);
1014 + de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1015 + memset (&root->info, 0, sizeof(root->info));
1016 + root->info.info_length = sizeof(root->info);
1017 + entries = root->entries;
1018 + dx_set_block (entries, 1);
1019 + dx_set_count (entries, 1);
1020 + dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1022 + /* Initialize as for dx_probe */
1023 + hash = dx_hash (name, namelen);
1025 + frame->entries = entries;
1026 + frame->at = entries;
1029 + de = do_split(handle,dir, &bh, frame, hash, &retval);
1030 + dx_release (frames);
1033 + nlen = EXT3_DIR_REC_LEN(de->name_len);
1034 + rlen = le16_to_cpu(de->rec_len);
1047 * ext3_delete_entry deletes a directory entry by merging it with the
1049 @@ -451,7 +1212,8 @@
1050 struct inode * inode;
1053 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1054 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1055 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1057 return PTR_ERR(handle);
1059 @@ -478,7 +1240,8 @@
1060 struct inode *inode;
1063 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1064 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1065 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1067 return PTR_ERR(handle);
1069 @@ -507,7 +1270,8 @@
1070 if (dir->i_nlink >= EXT3_LINK_MAX)
1073 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1074 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1075 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1077 return PTR_ERR(handle);
1079 @@ -550,7 +1320,7 @@
1083 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1084 + ext3_update_dx_flag(dir);
1085 ext3_mark_inode_dirty(handle, dir);
1086 d_instantiate(dentry, inode);
1088 @@ -832,7 +1596,7 @@
1089 ext3_mark_inode_dirty(handle, inode);
1091 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1092 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1093 + ext3_update_dx_flag(dir);
1094 ext3_mark_inode_dirty(handle, dir);
1097 @@ -878,7 +1642,7 @@
1100 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1101 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1102 + ext3_update_dx_flag(dir);
1103 ext3_mark_inode_dirty(handle, dir);
1105 if (!inode->i_nlink)
1106 @@ -904,7 +1668,8 @@
1107 if (l > dir->i_sb->s_blocksize)
1108 return -ENAMETOOLONG;
1110 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 5);
1111 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1112 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5);
1114 return PTR_ERR(handle);
1116 @@ -959,7 +1724,8 @@
1117 if (inode->i_nlink >= EXT3_LINK_MAX)
1120 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS);
1121 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1122 + EXT3_INDEX_EXTRA_TRANS_BLOCKS);
1124 return PTR_ERR(handle);
1126 @@ -995,7 +1761,8 @@
1128 old_bh = new_bh = dir_bh = NULL;
1130 - handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS + 2);
1131 + handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
1132 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
1134 return PTR_ERR(handle);
1136 @@ -1077,7 +1844,7 @@
1137 new_inode->i_ctime = CURRENT_TIME;
1139 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1140 - old_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1141 + ext3_update_dx_flag(old_dir);
1143 BUFFER_TRACE(dir_bh, "get_write_access");
1144 ext3_journal_get_write_access(handle, dir_bh);
1145 @@ -1089,7 +1856,7 @@
1146 new_inode->i_nlink--;
1149 - new_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1150 + ext3_update_dx_flag(new_dir);
1151 ext3_mark_inode_dirty(handle, new_dir);
1154 --- ./include/linux/ext3_fs.h 2002/03/05 06:18:59 2.1
1155 +++ ./include/linux/ext3_fs.h 2002/03/05 06:26:56
1157 #define EXT3_MOUNT_WRITEBACK_DATA 0x0C00 /* No data ordering */
1158 #define EXT3_MOUNT_UPDATE_JOURNAL 0x1000 /* Update the journal format */
1159 #define EXT3_MOUNT_NO_UID32 0x2000 /* Disable 32-bit UIDs */
1160 +#define EXT3_MOUNT_INDEX 0x4000 /* Enable directory index */
1162 /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
1163 #ifndef _LINUX_EXT2_FS_H
1164 @@ -575,6 +576,24 @@
1165 #define EXT3_DIR_ROUND (EXT3_DIR_PAD - 1)
1166 #define EXT3_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT3_DIR_ROUND) & \
1169 + * Hash Tree Directory indexing
1170 + * (c) Daniel Phillips, 2001
1173 +#define CONFIG_EXT3_INDEX
1175 +#ifdef CONFIG_EXT3_INDEX
1176 + enum {ext3_dx = 1};
1177 + #define is_dx(dir) (EXT3_I(dir)->i_flags & EXT3_INDEX_FL)
1178 +#define EXT3_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT3_LINK_MAX)
1179 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
1181 + enum {ext3_dx = 0};
1182 + #define is_dx(dir) 0
1183 +#define EXT3_DIR_LINK_MAX(dir) ((dir)->i_nlink >= EXT3_LINK_MAX)
1184 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2)
1189 --- ./include/linux/ext3_jbd.h 2002/03/05 06:18:59 2.1
1190 +++ ./include/linux/ext3_jbd.h 2002/03/05 06:33:54
1193 #define EXT3_RESERVE_TRANS_BLOCKS 12
1195 +#define EXT3_INDEX_EXTRA_TRANS_BLOCKS 8
1198 ext3_mark_iloc_dirty(handle_t *handle,
1199 struct inode *inode,