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 #include <linux/string.h>
44 #include <linux/locks.h>
45 #include <linux/quotaops.h>
47 +#include <linux/slab.h>
50 * define how far ahead to read directories while searching them.
52 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
53 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
55 +static struct buffer_head *ext3_append(handle_t *handle,
56 + struct inode *inode,
57 + u32 *block, int *err)
59 + struct buffer_head *bh;
61 + *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63 + if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
64 + inode->i_size += inode->i_sb->s_blocksize;
65 + EXT3_I(inode)->i_disksize = inode->i_size;
66 + ext3_journal_get_write_access(handle,bh);
72 +#define assert(test) J_ASSERT(test)
76 +#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79 +typedef struct { u32 v; } le_u32;
80 +typedef struct { u16 v; } le_u16;
82 +#define dxtrace_on(command) command
83 +#define dxtrace_off(command)
84 +#define dxtrace dxtrace_off
107 + * dx_root_info is laid out so that if it should somehow get overlaid by a
108 + * dirent the two low bits of the hash version will be zero. Therefore, the
109 + * hash version mod 4 should never be 0. Sincerely, the paranoia department.
114 + struct fake_dirent dot;
116 + struct fake_dirent dotdot;
117 + char dotdot_name[4];
118 + struct dx_root_info
120 + le_u32 reserved_zero;
121 + u8 hash_version; /* 0 now, 1 at release */
122 + u8 info_length; /* 8 */
123 + u8 indirect_levels;
127 + struct dx_entry entries[0];
132 + struct fake_dirent fake;
133 + struct dx_entry entries[0];
139 + struct buffer_head *bh;
140 + struct dx_entry *entries;
141 + struct dx_entry *at;
150 +typedef struct ext3_dir_entry_2 ext3_dirent;
151 +static inline unsigned dx_get_block (struct dx_entry *entry);
152 +static void dx_set_block (struct dx_entry *entry, unsigned value);
153 +static inline unsigned dx_get_hash (struct dx_entry *entry);
154 +static void dx_set_hash (struct dx_entry *entry, unsigned value);
155 +static unsigned dx_get_count (struct dx_entry *entries);
156 +static unsigned dx_get_limit (struct dx_entry *entries);
157 +static void dx_set_count (struct dx_entry *entries, unsigned value);
158 +static void dx_set_limit (struct dx_entry *entries, unsigned value);
159 +static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
160 +static unsigned dx_node_limit (struct inode *dir);
161 +static unsigned dx_hack_hash (const u8 *name, int len);
162 +static struct dx_frame *dx_probe (struct inode *dir, u32 hash, struct dx_frame *frame);
163 +static void dx_release (struct dx_frame *frames);
164 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[]);
165 +static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 +static ext3_dirent *dx_copy_dirents (char *from, char *to,
167 + struct dx_map_entry *map, int count);
168 +static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171 +#ifdef CONFIG_EXT3_INDEX
173 + * Future: use high four bits of block for coalesce-on-delete flags
174 + * Mask them off for now.
177 +static inline unsigned dx_get_block (struct dx_entry *entry)
179 + return le32_to_cpu(entry->block.v) & 0x00ffffff;
182 +static inline void dx_set_block (struct dx_entry *entry, unsigned value)
184 + entry->block.v = cpu_to_le32(value);
187 +static inline unsigned dx_get_hash (struct dx_entry *entry)
189 + return le32_to_cpu(entry->hash.v);
192 +static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
194 + entry->hash.v = cpu_to_le32(value);
197 +static inline unsigned dx_get_count (struct dx_entry *entries)
199 + return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
202 +static inline unsigned dx_get_limit (struct dx_entry *entries)
204 + return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
207 +static inline void dx_set_count (struct dx_entry *entries, unsigned value)
209 + ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
212 +static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
214 + ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
217 +static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
219 + unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
220 + EXT3_DIR_REC_LEN(2) - infosize;
221 + return 0? 20: entry_space / sizeof(struct dx_entry);
224 +static inline unsigned dx_node_limit (struct inode *dir)
226 + unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
227 + return 0? 22: entry_space / sizeof(struct dx_entry);
230 +/* Hash function - not bad, but still looking for an ideal default */
232 +static unsigned dx_hack_hash (const u8 *name, int len)
234 + u32 hash0 = 0x12a3fe2d, hash1 = 0x37abe8f9;
237 + u32 hash = hash1 + (hash0 ^ (*name++ * 7152373));
238 + if (hash & 0x80000000) hash -= 0x7fffffff;
245 +#define dx_hash(s,n) (dx_hack_hash(s,n) << 1)
250 +static void dx_show_index (char * label, struct dx_entry *entries)
252 + int i, n = dx_get_count (entries);
253 + printk("%s index ", label);
254 + for (i = 0; i < n; i++)
256 + printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
268 +static struct stats dx_show_leaf (ext3_dirent *de, int size, int show_names)
270 + unsigned names = 0, space = 0;
271 + char *base = (char *) de;
273 + while ((char *) de < base + size)
279 + int len = de->name_len;
280 + char *name = de->name;
281 + while (len--) printk("%c", *name++);
282 + printk(":%x.%u ", dx_hash (de->name, de->name_len), ((char *) de - base));
284 + space += EXT3_DIR_REC_LEN(de->name_len);
287 + de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
289 + printk("(%i)\n", names);
290 + return (struct stats) { names, space, 1 };
293 +struct stats dx_show_entries (struct inode *dir, struct dx_entry *entries, int levels)
295 + unsigned blocksize = dir->i_sb->s_blocksize;
296 + unsigned count = dx_get_count (entries), names = 0, space = 0, i;
297 + unsigned bcount = 0;
298 + struct buffer_head *bh;
300 + printk("%i indexed blocks...\n", count);
301 + for (i = 0; i < count; i++, entries++)
303 + u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
304 + u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
305 + struct stats stats;
306 + printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
307 + if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
309 + dx_show_entries (dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
310 + dx_show_leaf ((ext3_dirent *) bh->b_data, blocksize, 0);
311 + names += stats.names;
312 + space += stats.space;
313 + bcount += stats.bcount;
317 + printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
318 + names, space/bcount,(space/bcount)*100/blocksize);
319 + return (struct stats) { names, space, bcount};
323 + * Probe for a directory leaf block to search
326 +static struct dx_frame *
327 +dx_probe(struct inode *dir, u32 hash, struct dx_frame *frame_in)
329 + unsigned count, indirect;
330 + struct dx_entry *at, *entries, *p, *q, *m;
331 + struct dx_root *root;
332 + struct buffer_head *bh;
333 + struct dx_frame *frame = frame_in;
337 + if (!(bh = ext3_bread(NULL, dir, 0, 0, &err)))
339 + root = (struct dx_root *) bh->b_data;
340 + if (root->info.hash_version > 0 || root->info.unused_flags & 1) {
344 + if ((indirect = root->info.indirect_levels) > 1) {
348 + entries = (struct dx_entry *) (((char *) &root->info) + root->info.info_length);
349 + assert (dx_get_limit(entries) == dx_root_limit(dir, root->info.info_length));
350 + dxtrace (printk("Look up %x", hash));
353 + count = dx_get_count(entries);
354 + assert (count && count <= dx_get_limit(entries));
356 + q = entries + count - 1;
360 + dxtrace(printk("."));
361 + if (dx_get_hash(m) > hash)
367 + if (0) // linear search cross check
369 + unsigned n = count - 1;
373 + dxtrace(printk(","));
374 + if (dx_get_hash(++at) > hash)
380 + assert (at == p - 1);
384 + dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
386 + frame->entries = entries;
388 + if (!indirect--) return frame;
389 + if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0,&err)))
391 + at = entries = ((struct dx_node *) bh->b_data)->entries;
392 + assert (dx_get_limit(entries) == dx_node_limit (dir));
396 + while (frame >= frame_in) {
404 +static void dx_release (struct dx_frame *frames)
406 + if (frames[0].bh == NULL)
409 + if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
410 + brelse (frames[1].bh);
411 + brelse (frames[0].bh);
415 + * Directory block splitting, compacting
418 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[])
421 + char *base = (char *) de;
422 + while ((char *) de < base + size)
424 + map[count].hash = dx_hash (de->name, de->name_len);
425 + map[count].offs = (u32) ((char *) de - base);
426 + de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
432 +static void dx_sort_map (struct dx_map_entry *map, unsigned count)
434 + struct dx_map_entry *p, *q, *top = map + count - 1;
436 + /* Combsort until bubble sort doesn't suck */
439 + count = count*10/13;
440 + if (count - 9 < 2) /* 9, 10 -> 11 */
442 + for (p = top, q = p - count; q >= map; p--, q--)
443 + if (p->hash < q->hash)
446 + /* Garden variety bubble sort */
452 + if (q[1].hash >= q[0].hash)
460 +static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
462 + struct dx_entry *entries = frame->entries;
463 + struct dx_entry *old = frame->at, *new = old + 1;
464 + int count = dx_get_count(entries);
466 + assert(count < dx_get_limit(entries));
467 + assert(old < entries + count);
468 + memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
469 + dx_set_hash(new, hash);
470 + dx_set_block(new, block);
471 + dx_set_count(entries, count + 1);
475 +static void ext3_update_dx_flag(struct inode *inode)
477 + if (!test_opt(inode->i_sb, INDEX))
478 + EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
482 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
488 + * p is at least 6 bytes before the end of page
490 +static inline ext3_dirent *ext3_next_entry(ext3_dirent *p)
492 + return (ext3_dirent *)((char*)p + le16_to_cpu(p->rec_len));
499 * finds an entry in the specified directory with the wanted name. It
501 * The returned buffer_head has ->b_count elevated. The caller is expected
502 * to brelse() it when appropriate.
506 static struct buffer_head * ext3_find_entry (struct dentry *dentry,
507 struct ext3_dir_entry_2 ** res_dir)
509 @@ -119,10 +564,76 @@
512 struct inode *dir = dentry->d_parent->d_inode;
515 + unsigned blocksize;
516 + ext3_dirent *de, *top;
520 + blocksize = sb->s_blocksize;
521 + namelen = dentry->d_name.len;
522 + name = dentry->d_name.name;
523 + if (namelen > EXT3_NAME_LEN)
525 + if (ext3_dx && is_dx(dir)) {
526 + u32 hash = dx_hash (name, namelen);
527 + struct dx_frame frames[2], *frame;
528 + if (!(frame = dx_probe (dir, hash, frames)))
531 + block = dx_get_block(frame->at);
532 + if (!(bh = ext3_bread (NULL,dir, block, 0, &err)))
534 + de = (ext3_dirent *) bh->b_data;
535 + top = (ext3_dirent *) ((char *) de + blocksize -
536 + EXT3_DIR_REC_LEN(0));
537 + for (; de < top; de = ext3_next_entry(de))
538 + if (ext3_match (namelen, name, de)) {
539 + if (!ext3_check_dir_entry("ext3_find_entry",
541 + (block<<EXT3_BLOCK_SIZE_BITS(sb))
542 + +((char *)de - bh->b_data))) {
550 + /* Same hash continues in next block? Search on. */
551 + if (++(frame->at) == frame->entries + dx_get_count(frame->entries))
553 + struct buffer_head *bh2;
554 + if (frame == frames)
556 + if (++(frames->at) == frames->entries + dx_get_count(frames->entries))
558 + /* should omit read if not continued */
559 + if (!(bh2 = ext3_bread (NULL, dir,
560 + dx_get_block(frames->at),
563 + brelse (frame->bh);
565 + frame->at = frame->entries = ((struct dx_node *) bh2->b_data)->entries;
566 + /* Subtle: the 0th entry has the count, find the hash in frame above */
567 + if ((dx_get_hash(frames->at) & -2) == hash)
571 + if ((dx_get_hash(frame->at) & -2) == hash)
574 + dxtrace(printk("%s not found\n", name));
575 + dx_release (frames);
578 + dx_release (frames);
583 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
584 start = dir->u.ext3_i.i_dir_start_lookup;
585 if (start >= nblocks)
587 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
590 +static ext3_dirent *
591 +dx_copy_dirents (char *from, char *to, struct dx_map_entry *map, int count)
593 + unsigned rec_len = 0;
596 + ext3_dirent *de = (ext3_dirent *) (from + map->offs);
597 + rec_len = EXT3_DIR_REC_LEN(de->name_len);
598 + memcpy (to, de, rec_len);
599 + ((ext3_dirent *) to)->rec_len = rec_len;
603 + return (ext3_dirent *) (to - rec_len);
606 +#ifdef CONFIG_EXT3_INDEX
607 +static ext3_dirent *do_split(handle_t *handle, struct inode *dir,
608 + struct buffer_head **bh,struct dx_frame *frame,
609 + u32 hash, int *error)
611 + unsigned blocksize = dir->i_sb->s_blocksize;
612 + unsigned count, continued;
613 + struct buffer_head *bh2;
615 + unsigned MAX_DX_MAP = PAGE_CACHE_SIZE/EXT3_DIR_REC_LEN(1) + 1;
617 + struct dx_map_entry *map;
618 + char *data1 = (*bh)->b_data, *data2, *data3;
620 + ext3_dirent *de, *de2;
622 + bh2 = ext3_append (handle, dir, &newblock, error);
627 + return (ext3_dirent *)bh2;
630 + BUFFER_TRACE(*bh, "get_write_access");
631 + ext3_journal_get_write_access(handle, *bh);
632 + BUFFER_TRACE(frame->bh, "get_write_access");
633 + ext3_journal_get_write_access(handle, frame->bh);
635 + data2 = bh2->b_data;
637 + map = kmalloc(sizeof(*map) * MAX_DX_MAP, GFP_KERNEL);
639 + panic("no memory for do_split\n");
640 + count = dx_make_map ((ext3_dirent *) data1, blocksize, map);
641 + split = count/2; // need to adjust to actual middle
642 + dx_sort_map (map, count);
643 + hash2 = map[split].hash;
644 + continued = hash2 == map[split - 1].hash;
645 + dxtrace(printk("Split block %i at %x, %i/%i\n",
646 + dx_get_block(frame->at), hash2, split, count-split));
648 + /* Fancy dance to stay within two buffers */
649 + de2 = dx_copy_dirents (data1, data2, map + split, count - split);
650 + data3 = (char *) de2 + de2->rec_len;
651 + de = dx_copy_dirents (data1, data3, map, split);
652 + memcpy(data1, data3, (char *) de + de->rec_len - data3);
653 + de = (ext3_dirent *) ((char *) de - data3 + data1); // relocate de
654 + de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
655 + de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
656 + dxtrace(dx_show_leaf ((ext3_dirent *) data1, blocksize, 1));
657 + dxtrace(dx_show_leaf ((ext3_dirent *) data2, blocksize, 1));
659 + /* Which block gets the new entry? */
665 + dx_insert_block (frame, hash2 + continued, newblock);
666 + ext3_journal_dirty_metadata (handle, bh2);
668 + ext3_journal_dirty_metadata (handle, frame->bh);
669 + dxtrace(dx_show_index ("frame", frame->entries));
681 * AKPM: the journalling code here looks wrong on the error paths
684 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
687 @@ -258,117 +852,281 @@
688 const char *name = dentry->d_name.name;
689 int namelen = dentry->d_name.len;
690 unsigned long offset;
691 - unsigned short rec_len;
692 struct buffer_head * bh;
693 - struct ext3_dir_entry_2 * de, * de1;
694 - struct super_block * sb;
696 + struct super_block * sb = dir->i_sb;
698 + unsigned short reclen = EXT3_DIR_REC_LEN(namelen);
701 + unsigned blocksize = sb->s_blocksize;
702 + unsigned nlen, rlen;
708 - bh = ext3_bread (handle, dir, 0, 0, &retval);
711 - rec_len = EXT3_DIR_REC_LEN(namelen);
713 - de = (struct ext3_dir_entry_2 *) bh->b_data;
715 - if ((char *)de >= sb->s_blocksize + bh->b_data) {
718 - bh = ext3_bread (handle, dir,
719 - offset >> EXT3_BLOCK_SIZE_BITS(sb), 1, &retval);
722 - if (dir->i_size <= offset) {
723 - if (dir->i_size == 0) {
726 + if (ext3_dx && is_dx(dir)) {
727 + struct dx_frame frames[2], *frame;
728 + struct dx_entry *entries, *at;
732 + hash = dx_hash(name, namelen);
733 + /* FIXME: do something if dx_probe() fails here */
734 + frame = dx_probe(dir, hash, frames);
735 + entries = frame->entries;
738 + if (!(bh = ext3_bread(handle,dir, dx_get_block(at), 0,&retval)))
741 + BUFFER_TRACE(bh, "get_write_access");
742 + ext3_journal_get_write_access(handle, bh);
744 + data1 = bh->b_data;
745 + de = (ext3_dirent *) data1;
746 + top = data1 + (0? 200: blocksize);
747 + while ((char *) de < top)
749 + /* FIXME: check EEXIST and dir */
750 + nlen = EXT3_DIR_REC_LEN(de->name_len);
751 + rlen = le16_to_cpu(de->rec_len);
752 + if ((de->inode? rlen - nlen: rlen) >= reclen)
754 + de = (ext3_dirent *) ((char *) de + rlen);
756 + /* Block full, should compress but for now just split */
757 + dxtrace(printk("using %u of %u node entries\n",
758 + dx_get_count(entries), dx_get_limit(entries)));
759 + /* Need to split index? */
760 + if (dx_get_count(entries) == dx_get_limit(entries))
763 + unsigned icount = dx_get_count(entries);
764 + int levels = frame - frames;
765 + struct dx_entry *entries2;
766 + struct dx_node *node2;
767 + struct buffer_head *bh2;
768 + if (levels && dx_get_count(frames->entries) == dx_get_limit(frames->entries))
770 + bh2 = ext3_append (handle, dir, &newblock, &retval);
773 + node2 = (struct dx_node *)(bh2->b_data);
774 + entries2 = node2->entries;
775 + node2->fake.rec_len = cpu_to_le16(blocksize);
776 + node2->fake.inode = 0;
777 + BUFFER_TRACE(frame->bh, "get_write_access");
778 + ext3_journal_get_write_access(handle, frame->bh);
781 + unsigned icount1 = icount/2, icount2 = icount - icount1;
782 + unsigned hash2 = dx_get_hash(entries + icount1);
783 + dxtrace(printk("Split index %i/%i\n", icount1, icount2));
785 + BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
786 + ext3_journal_get_write_access(handle, frames[0].bh);
788 + memcpy ((char *) entries2, (char *) (entries + icount1),
789 + icount2 * sizeof(struct dx_entry));
790 + dx_set_count (entries, icount1);
791 + dx_set_count (entries2, icount2);
792 + dx_set_limit (entries2, dx_node_limit(dir));
794 + /* Which index block gets the new entry? */
795 + if (at - entries >= icount1) {
796 + frame->at = at = at - entries - icount1 + entries2;
797 + frame->entries = entries = entries2;
798 + swap(frame->bh, bh2);
801 - ext3_debug ("creating next block\n");
803 - BUFFER_TRACE(bh, "get_write_access");
804 - ext3_journal_get_write_access(handle, bh);
805 - de = (struct ext3_dir_entry_2 *) bh->b_data;
807 - de->rec_len = le16_to_cpu(sb->s_blocksize);
808 - dir->u.ext3_i.i_disksize =
809 - dir->i_size = offset + sb->s_blocksize;
810 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
811 - ext3_mark_inode_dirty(handle, dir);
812 + dx_insert_block (frames + 0, hash2, newblock);
813 + dxtrace(dx_show_index ("node", frames[1].entries));
814 + dxtrace(dx_show_index ("node",
815 + ((struct dx_node *) bh2->b_data)->entries));
816 + ext3_journal_dirty_metadata(handle, bh2);
820 - ext3_debug ("skipping to next block\n");
822 - de = (struct ext3_dir_entry_2 *) bh->b_data;
823 + dxtrace(printk("Creating second level index...\n"));
824 + memcpy((char *) entries2, (char *) entries,
825 + icount * sizeof(struct dx_entry));
826 + dx_set_limit(entries2, dx_node_limit(dir));
829 + dx_set_count(entries, 1);
830 + dx_set_block(entries + 0, newblock);
831 + ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
833 + /* Add new access path frame */
834 + frame = frames + 1;
835 + frame->at = at = at - entries + entries2;
836 + frame->entries = entries = entries2;
838 + ext3_journal_get_write_access(handle, frame->bh);
840 + ext3_journal_dirty_metadata(handle, frames[0].bh);
842 - if (!ext3_check_dir_entry ("ext3_add_entry", dir, de, bh,
847 - if (ext3_match (namelen, name, de)) {
848 + de = do_split(handle, dir, &bh, frame, hash, &retval);
849 + dx_release (frames);
852 + nlen = EXT3_DIR_REC_LEN(de->name_len);
853 + rlen = le16_to_cpu(de->rec_len);
857 + dx_release (frames);
861 + ext3_warning(sb, __FUNCTION__, "Directory index full!\n");
866 + dx_release (frames);
870 + blocks = dir->i_size >> sb->s_blocksize_bits;
871 + for (block = 0, offset = 0; block < blocks; block++) {
872 + bh = ext3_bread(handle, dir, block, 0, &retval);
875 + de = (ext3_dirent *)bh->b_data;
876 + top = bh->b_data + blocksize - reclen;
877 + while ((char *) de <= top) {
878 + if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
883 + if (ext3_match (namelen, name, de)) {
887 - if ((le32_to_cpu(de->inode) == 0 &&
888 - le16_to_cpu(de->rec_len) >= rec_len) ||
889 - (le16_to_cpu(de->rec_len) >=
890 - EXT3_DIR_REC_LEN(de->name_len) + rec_len)) {
891 - BUFFER_TRACE(bh, "get_write_access");
892 - ext3_journal_get_write_access(handle, bh);
893 - /* By now the buffer is marked for journaling */
894 - offset += le16_to_cpu(de->rec_len);
895 - if (le32_to_cpu(de->inode)) {
896 - de1 = (struct ext3_dir_entry_2 *) ((char *) de +
897 - EXT3_DIR_REC_LEN(de->name_len));
899 - cpu_to_le16(le16_to_cpu(de->rec_len) -
900 - EXT3_DIR_REC_LEN(de->name_len));
901 - de->rec_len = cpu_to_le16(
902 - EXT3_DIR_REC_LEN(de->name_len));
905 - de->file_type = EXT3_FT_UNKNOWN;
907 - de->inode = cpu_to_le32(inode->i_ino);
908 - ext3_set_de_type(dir->i_sb, de, inode->i_mode);
911 - de->name_len = namelen;
912 - memcpy (de->name, name, namelen);
914 - * XXX shouldn't update any times until successful
915 - * completion of syscall, but too many callers depend
918 - * XXX similarly, too many callers depend on
919 - * ext3_new_inode() setting the times, but error
920 - * recovery deletes the inode, so the worst that can
921 - * happen is that the times are slightly out of date
922 - * and/or different from the directory change time.
924 - dir->i_mtime = dir->i_ctime = CURRENT_TIME;
925 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
926 - ext3_mark_inode_dirty(handle, dir);
927 - dir->i_version = ++event;
928 - BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
929 - ext3_journal_dirty_metadata(handle, bh);
930 + nlen = EXT3_DIR_REC_LEN(de->name_len);
931 + rlen = le16_to_cpu(de->rec_len);
932 + if ((de->inode? rlen - nlen: rlen) >= reclen)
934 + de = (ext3_dirent *)((char *)de + rlen);
937 + if (ext3_dx && blocks == 1 && test_opt(sb, INDEX))
938 + goto dx_make_index;
941 + bh = ext3_append(handle, dir, &block, &retval);
944 + de = (ext3_dirent *) bh->b_data;
946 + de->rec_len = cpu_to_le16(rlen = blocksize);
951 + BUFFER_TRACE(bh, "get_write_access");
952 + ext3_journal_get_write_access(handle, bh);
953 + /* By now the buffer is marked for journaling */
955 + ext3_dirent *de1 = (ext3_dirent *)((char *)de + nlen);
956 + de1->rec_len = cpu_to_le16(rlen - nlen);
957 + de->rec_len = cpu_to_le16(nlen);
960 + de->file_type = EXT3_FT_UNKNOWN;
962 + de->inode = cpu_to_le32(inode->i_ino);
963 + ext3_set_de_type(dir->i_sb, de, inode->i_mode);
966 + de->name_len = namelen;
967 + memcpy (de->name, name, namelen);
969 + * XXX shouldn't update any times until successful
970 + * completion of syscall, but too many callers depend
973 + * XXX similarly, too many callers depend on
974 + * ext3_new_inode() setting the times, but error
975 + * recovery deletes the inode, so the worst that can
976 + * happen is that the times are slightly out of date
977 + * and/or different from the directory change time.
979 + dir->i_mtime = dir->i_ctime = CURRENT_TIME;
980 + ext3_update_dx_flag(dir);
981 + dir->i_version = ++event;
982 + ext3_mark_inode_dirty(handle, dir);
983 + BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
984 + ext3_journal_dirty_metadata(handle, bh);
990 + struct buffer_head *bh2;
991 + struct dx_root *root;
992 + struct dx_frame frames[2], *frame;
993 + struct dx_entry *entries;
999 + dxtrace(printk("Creating index\n"));
1000 + ext3_journal_get_write_access(handle, bh);
1001 + root = (struct dx_root *) bh->b_data;
1003 + EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1004 + bh2 = ext3_append (handle, dir, &block, &retval);
1011 - offset += le16_to_cpu(de->rec_len);
1012 - de = (struct ext3_dir_entry_2 *)
1013 - ((char *) de + le16_to_cpu(de->rec_len));
1014 + data1 = bh2->b_data;
1016 + /* The 0th block becomes the root, move the dirents out */
1017 + de = (ext3_dirent *) &root->info;
1018 + len = ((char *) root) + blocksize - (char *) de;
1019 + memcpy (data1, de, len);
1020 + de = (ext3_dirent *) data1;
1021 + top = data1 + len;
1022 + while (((char *) de2=(char*)de+le16_to_cpu(de->rec_len)) < top)
1024 + de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1025 + /* Initialize the root; the dot dirents already exist */
1026 + de = (ext3_dirent *) (&root->dotdot);
1027 + de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1028 + memset (&root->info, 0, sizeof(root->info));
1029 + root->info.info_length = sizeof(root->info);
1030 + entries = root->entries;
1031 + dx_set_block (entries, 1);
1032 + dx_set_count (entries, 1);
1033 + dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1035 + /* Initialize as for dx_probe */
1036 + hash = dx_hash (name, namelen);
1038 + frame->entries = entries;
1039 + frame->at = entries;
1042 + de = do_split(handle,dir, &bh, frame, hash, &retval);
1043 + dx_release (frames);
1046 + nlen = EXT3_DIR_REC_LEN(de->name_len);
1047 + rlen = le16_to_cpu(de->rec_len);
1060 * ext3_delete_entry deletes a directory entry by merging it with the
1062 @@ -451,7 +1212,8 @@
1063 struct inode * inode;
1066 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1067 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1068 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1070 return PTR_ERR(handle);
1072 @@ -478,7 +1240,8 @@
1073 struct inode *inode;
1076 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1077 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1078 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1080 return PTR_ERR(handle);
1082 @@ -507,7 +1270,8 @@
1083 if (dir->i_nlink >= EXT3_LINK_MAX)
1086 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1087 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1088 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1090 return PTR_ERR(handle);
1092 @@ -550,7 +1320,7 @@
1096 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1097 + ext3_update_dx_flag(dir);
1098 ext3_mark_inode_dirty(handle, dir);
1099 d_instantiate(dentry, inode);
1101 @@ -832,7 +1596,7 @@
1102 ext3_mark_inode_dirty(handle, inode);
1104 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1105 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1106 + ext3_update_dx_flag(dir);
1107 ext3_mark_inode_dirty(handle, dir);
1110 @@ -878,7 +1642,7 @@
1113 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1114 - dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1115 + ext3_update_dx_flag(dir);
1116 ext3_mark_inode_dirty(handle, dir);
1118 if (!inode->i_nlink)
1119 @@ -904,7 +1668,8 @@
1120 if (l > dir->i_sb->s_blocksize)
1121 return -ENAMETOOLONG;
1123 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 5);
1124 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1125 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5);
1127 return PTR_ERR(handle);
1129 @@ -959,7 +1724,8 @@
1130 if (inode->i_nlink >= EXT3_LINK_MAX)
1133 - handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS);
1134 + handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1135 + EXT3_INDEX_EXTRA_TRANS_BLOCKS);
1137 return PTR_ERR(handle);
1139 @@ -995,7 +1761,8 @@
1141 old_bh = new_bh = dir_bh = NULL;
1143 - handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS + 2);
1144 + handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
1145 + EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
1147 return PTR_ERR(handle);
1149 @@ -1077,7 +1844,7 @@
1150 new_inode->i_ctime = CURRENT_TIME;
1152 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1153 - old_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1154 + ext3_update_dx_flag(old_dir);
1156 BUFFER_TRACE(dir_bh, "get_write_access");
1157 ext3_journal_get_write_access(handle, dir_bh);
1158 @@ -1089,7 +1856,7 @@
1159 new_inode->i_nlink--;
1162 - new_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1163 + ext3_update_dx_flag(new_dir);
1164 ext3_mark_inode_dirty(handle, new_dir);
1167 --- ./include/linux/ext3_fs.h 2002/03/05 06:18:59 2.1
1168 +++ ./include/linux/ext3_fs.h 2002/03/05 06:26:56
1170 #define EXT3_MOUNT_WRITEBACK_DATA 0x0C00 /* No data ordering */
1171 #define EXT3_MOUNT_UPDATE_JOURNAL 0x1000 /* Update the journal format */
1172 #define EXT3_MOUNT_NO_UID32 0x2000 /* Disable 32-bit UIDs */
1173 +#define EXT3_MOUNT_INDEX 0x4000 /* Enable directory index */
1175 /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
1176 #ifndef _LINUX_EXT2_FS_H
1177 @@ -575,6 +576,24 @@
1178 #define EXT3_DIR_ROUND (EXT3_DIR_PAD - 1)
1179 #define EXT3_DIR_REC_LEN(name_len) (((name_len) + 8 + EXT3_DIR_ROUND) & \
1182 + * Hash Tree Directory indexing
1183 + * (c) Daniel Phillips, 2001
1186 +#define CONFIG_EXT3_INDEX
1188 +#ifdef CONFIG_EXT3_INDEX
1189 + enum {ext3_dx = 1};
1190 + #define is_dx(dir) (EXT3_I(dir)->i_flags & EXT3_INDEX_FL)
1191 +#define EXT3_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT3_LINK_MAX)
1192 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
1194 + enum {ext3_dx = 0};
1195 + #define is_dx(dir) 0
1196 +#define EXT3_DIR_LINK_MAX(dir) ((dir)->i_nlink >= EXT3_LINK_MAX)
1197 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2)
1202 --- ./include/linux/ext3_jbd.h 2002/03/05 06:18:59 2.1
1203 +++ ./include/linux/ext3_jbd.h 2002/03/05 06:33:54
1206 #define EXT3_RESERVE_TRANS_BLOCKS 12
1208 +#define EXT3_INDEX_EXTRA_TRANS_BLOCKS 8
1211 ext3_mark_iloc_dirty(handle_t *handle,
1212 struct inode *inode,