lustre/extN/htree-ext3-2.4.18.diff

   1 --- ./fs/ext3/dir.c     2002/03/05 06:18:59     2.1
   2 +++ ./fs/ext3/dir.c     2002/03/05 06:26:56
   3 @@ -26,7 +26,7 @@
   4         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
   5  };
   6
   7 -static int ext3_readdir(struct file *, void *, filldir_t);
   8 +int ext3_readdir(struct file *, void *, filldir_t);
   9
  10  struct file_operations ext3_dir_operations = {
  11         read:           generic_read_dir,
  12 @@ -65,7 +65,7 @@
  13         return error_msg == NULL ? 1 : 0;
  14  }
  15
  16 -static int ext3_readdir(struct file * filp,
  17 +int ext3_readdir(struct file * filp,
  18                          void * dirent, filldir_t filldir)
  19  {
  20         int error = 0;
  21 --- ./fs/ext3/super.c   2002/03/05 06:18:59     2.1
  22 +++ ./fs/ext3/super.c   2002/03/05 06:26:56
  23 @@ -529,6 +529,12 @@
  24                                        "EXT3 Check option not supported\n");
  25  #endif
  26                 }
  27 +               else if (!strcmp (this_char, "index"))
  28 +#ifdef CONFIG_EXT3_INDEX
  29 +                       set_opt (*mount_options, INDEX);
  30 +#else
  31 +                       printk("EXT3 index option not supported\n");
  32 +#endif
  33                 else if (!strcmp (this_char, "debug"))
  34                         set_opt (*mount_options, DEBUG);
  35                 else if (!strcmp (this_char, "errors")) {
  36 @@ -712,6 +718,10 @@
  37                         EXT3_BLOCKS_PER_GROUP(sb),
  38                         EXT3_INODES_PER_GROUP(sb),
  39                         sbi->s_mount_opt);
  40 +
  41 +       if (EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
  42 +               set_opt (EXT3_SB(sb)->s_mount_opt, INDEX);
  43 +
  44         printk(KERN_INFO "EXT3 FS " EXT3FS_VERSION ", " EXT3FS_DATE " on %s, ",
  45                                 bdevname(sb->s_dev));
  46         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
  47 --- ./fs/ext3/namei.c   2002/03/05 06:18:59     2.1
  48 +++ ./fs/ext3/namei.c   2002/03/06 00:13:18
  49 @@ -16,6 +16,10 @@
  50   *        David S. Miller (davem@caip.rutgers.edu), 1995
  51   *  Directory entry file type support and forward compatibility hooks
  52   *     for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  53 + *  Hash Tree Directory indexing (c)
  54 + *     Daniel Phillips, 2001
  55 + *  Hash Tree Directory indexing porting
  56 + *     Christopher Li, 2002
  57   */
  58
  59  #include <linux/fs.h>
  60 @@ -38,6 +42,439 @@
  61  #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  62  #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  63
  64 +void ext3_add_compat_feature (struct super_block *sb, unsigned feature)
  65 +{
  66 +       if (!EXT3_HAS_COMPAT_FEATURE(sb, feature))
  67 +       {
  68 +               lock_super(sb);
  69 +               ext3_update_dynamic_rev(sb);
  70 +               EXT3_SET_COMPAT_FEATURE(sb, feature);
  71 +               ext3_write_super(sb);
  72 +               unlock_super(sb);
  73 +       }
  74 +}
  75 +
  76 +static struct buffer_head *ext3_append(handle_t *handle,
  77 +                                       struct inode *inode,
  78 +                                       u32 *block, int *err)
  79 +{
  80 +       struct buffer_head *bh;
  81 +
  82 +       *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  83 +
  84 +       if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
  85 +               inode->i_size += inode->i_sb->s_blocksize;
  86 +               EXT3_I(inode)->i_disksize = inode->i_size;
  87 +               ext3_journal_get_write_access(handle,bh);
  88 +       }
  89 +       return bh;
  90 +}
  91 +
  92 +#ifndef assert
  93 +#define assert(test) J_ASSERT(test)
  94 +#endif
  95 +
  96 +#ifndef swap
  97 +#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  98 +#endif
  99 +
 100 +typedef struct { u32 v; } le_u32;
 101 +typedef struct { u16 v; } le_u16;
 102 +
 103 +#define dxtrace_on(command) command
 104 +#define dxtrace_off(command)
 105 +#define dxtrace dxtrace_off
 106 +
 107 +struct fake_dirent
 108 +{
 109 +       /*le*/u32 inode;
 110 +       /*le*/u16 rec_len;
 111 +       u8 name_len;
 112 +       u8 file_type;
 113 +};
 114 +
 115 +struct dx_countlimit
 116 +{
 117 +       le_u16 limit;
 118 +       le_u16 count;
 119 +};
 120 +
 121 +struct dx_entry
 122 +{
 123 +       le_u32 hash;
 124 +       le_u32 block;
 125 +};
 126 +
 127 +/*
 128 + * dx_root_info is laid out so that if it should somehow get overlaid by a
 129 + * dirent the two low bits of the hash version will be zero.  Therefore, the
 130 + * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 131 + */
 132 +
 133 +struct dx_root
 134 +{
 135 +       struct fake_dirent dot;
 136 +       char dot_name[4];
 137 +       struct fake_dirent dotdot;
 138 +       char dotdot_name[4];
 139 +       struct dx_root_info
 140 +       {
 141 +               le_u32 reserved_zero;
 142 +               u8 hash_version; /* 0 now, 1 at release */
 143 +               u8 info_length; /* 8 */
 144 +               u8 indirect_levels;
 145 +               u8 unused_flags;
 146 +       }
 147 +       info;
 148 +       struct dx_entry entries[0];
 149 +};
 150 +
 151 +struct dx_node
 152 +{
 153 +       struct fake_dirent fake;
 154 +       struct dx_entry entries[0];
 155 +};
 156 +
 157 +
 158 +struct dx_frame
 159 +{
 160 +       struct buffer_head *bh;
 161 +       struct dx_entry *entries;
 162 +       struct dx_entry *at;
 163 +};
 164 +
 165 +struct dx_map_entry
 166 +{
 167 +       u32 hash;
 168 +       u32 offs;
 169 +};
 170 +
 171 +typedef struct ext3_dir_entry_2 ext3_dirent;
 172 +static inline unsigned dx_get_block (struct dx_entry *entry);
 173 +static void dx_set_block (struct dx_entry *entry, unsigned value);
 174 +static inline unsigned dx_get_hash (struct dx_entry *entry);
 175 +static void dx_set_hash (struct dx_entry *entry, unsigned value);
 176 +static unsigned dx_get_count (struct dx_entry *entries);
 177 +static unsigned dx_get_limit (struct dx_entry *entries);
 178 +static void dx_set_count (struct dx_entry *entries, unsigned value);
 179 +static void dx_set_limit (struct dx_entry *entries, unsigned value);
 180 +static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 181 +static unsigned dx_node_limit (struct inode *dir);
 182 +static unsigned dx_hack_hash (const u8 *name, int len);
 183 +static struct dx_frame *dx_probe (struct inode *dir, u32 hash, struct dx_frame *frame);
 184 +static void dx_release (struct dx_frame *frames);
 185 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[]);
 186 +static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 187 +static ext3_dirent *dx_copy_dirents (char *from, char *to,
 188 +     struct dx_map_entry *map, int count);
 189 +static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 190 +
 191 +
 192 +#ifdef CONFIG_EXT3_INDEX
 193 +/*
 194 + * Future: use high four bits of block for coalesce-on-delete flags
 195 + * Mask them off for now.
 196 + */
 197 +
 198 +static inline unsigned dx_get_block (struct dx_entry *entry)
 199 +{
 200 +       return le32_to_cpu(entry->block.v) & 0x00ffffff;
 201 +}
 202 +
 203 +static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 204 +{
 205 +       entry->block.v = cpu_to_le32(value);
 206 +}
 207 +
 208 +static inline unsigned dx_get_hash (struct dx_entry *entry)
 209 +{
 210 +       return le32_to_cpu(entry->hash.v);
 211 +}
 212 +
 213 +static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 214 +{
 215 +       entry->hash.v = cpu_to_le32(value);
 216 +}
 217 +
 218 +static inline unsigned dx_get_count (struct dx_entry *entries)
 219 +{
 220 +       return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
 221 +}
 222 +
 223 +static inline unsigned dx_get_limit (struct dx_entry *entries)
 224 +{
 225 +       return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
 226 +}
 227 +
 228 +static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 229 +{
 230 +       ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
 231 +}
 232 +
 233 +static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 234 +{
 235 +       ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
 236 +}
 237 +
 238 +static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 239 +{
 240 +       unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 241 +               EXT3_DIR_REC_LEN(2) - infosize;
 242 +       return 0? 20: entry_space / sizeof(struct dx_entry);
 243 +}
 244 +
 245 +static inline unsigned dx_node_limit (struct inode *dir)
 246 +{
 247 +       unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 248 +       return 0? 22: entry_space / sizeof(struct dx_entry);
 249 +}
 250 +
 251 +/* Hash function - not bad, but still looking for an ideal default */
 252 +
 253 +static unsigned dx_hack_hash (const u8 *name, int len)
 254 +{
 255 +       u32 hash0 = 0x12a3fe2d, hash1 = 0x37abe8f9;
 256 +       while (len--)
 257 +       {
 258 +               u32 hash = hash1 + (hash0 ^ (*name++ * 7152373));
 259 +               if (hash & 0x80000000) hash -= 0x7fffffff;
 260 +               hash1 = hash0;
 261 +               hash0 = hash;
 262 +       }
 263 +       return hash0;
 264 +}
 265 +
 266 +#define dx_hash(s,n) (dx_hack_hash(s,n) << 1)
 267 +
 268 +/*
 269 + * Debug
 270 + */
 271 +static void dx_show_index (char * label, struct dx_entry *entries)
 272 +{
 273 +       int i, n = dx_get_count (entries);
 274 +       printk("%s index ", label);
 275 +       for (i = 0; i < n; i++)
 276 +       {
 277 +               printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 278 +       }
 279 +       printk("\n");
 280 +}
 281 +
 282 +struct stats
 283 +{
 284 +       unsigned names;
 285 +       unsigned space;
 286 +       unsigned bcount;
 287 +};
 288 +
 289 +static struct stats dx_show_leaf (ext3_dirent *de, int size, int show_names)
 290 +{
 291 +       unsigned names = 0, space = 0;
 292 +       char *base = (char *) de;
 293 +       printk("names: ");
 294 +       while ((char *) de < base + size)
 295 +       {
 296 +               if (de->inode)
 297 +               {
 298 +                       if (show_names)
 299 +                       {
 300 +                               int len = de->name_len;
 301 +                               char *name = de->name;
 302 +                               while (len--) printk("%c", *name++);
 303 +                               printk(":%x.%u ", dx_hash (de->name, de->name_len), ((char *) de - base));
 304 +                       }
 305 +                       space += EXT3_DIR_REC_LEN(de->name_len);
 306 +                       names++;
 307 +               }
 308 +               de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
 309 +       }
 310 +       printk("(%i)\n", names);
 311 +       return (struct stats) { names, space, 1 };
 312 +}
 313 +
 314 +struct stats dx_show_entries (struct inode *dir, struct dx_entry *entries, int levels)
 315 +{
 316 +       unsigned blocksize = dir->i_sb->s_blocksize;
 317 +       unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 318 +       unsigned bcount = 0;
 319 +       struct buffer_head *bh;
 320 +       int err;
 321 +       printk("%i indexed blocks...\n", count);
 322 +       for (i = 0; i < count; i++, entries++)
 323 +       {
 324 +               u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 325 +               u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 326 +               struct stats stats;
 327 +               printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 328 +               if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 329 +               stats = levels?
 330 +                  dx_show_entries (dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 331 +                  dx_show_leaf ((ext3_dirent *) bh->b_data, blocksize, 0);
 332 +               names += stats.names;
 333 +               space += stats.space;
 334 +               bcount += stats.bcount;
 335 +               brelse (bh);
 336 +       }
 337 +       if (bcount)
 338 +               printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 339 +                       names, space/bcount,(space/bcount)*100/blocksize);
 340 +       return (struct stats) { names, space, bcount};
 341 +}
 342 +
 343 +static void dx_show_buckets (struct inode *dir)
 344 +{
 345 +       struct buffer_head *bh;
 346 +       struct dx_root *root;
 347 +       int err;
 348 +       if (!(bh = ext3_bread (NULL,dir, 0, 0,&err))) return;
 349 +       root = (struct dx_root *) bh->b_data;
 350 +       dx_show_entries (dir, root->entries, root->info.indirect_levels);
 351 +       brelse (bh);
 352 +}
 353 +
 354 +ssize_t hack_show_dir (struct file * filp, void * dirent, filldir_t filldir)
 355 +{
 356 +       if (is_dx (filp->f_dentry->d_inode) && !filp->f_pos)
 357 +               dx_show_buckets (filp->f_dentry->d_inode);
 358 +       return ext3_readdir(filp,dirent,filldir);
 359 +}
 360 +
 361 +/*
 362 + * Probe for a directory leaf block to search
 363 + */
 364 +
 365 +static struct dx_frame *dx_probe (struct inode *dir, u32 hash, struct dx_frame *frame)
 366 +{
 367 +       unsigned count, indirect;
 368 +       struct dx_entry *at, *entries, *p, *q, *m;
 369 +       struct dx_root *root;
 370 +       struct buffer_head *bh;
 371 +       int err;
 372 +       if (!(bh = ext3_bread (NULL,dir, 0, 0,&err)))
 373 +               goto fail;
 374 +       root = (struct dx_root *) bh->b_data;
 375 +       if (root->info.hash_version > 0 || root->info.unused_flags & 1)
 376 +               goto fail;
 377 +       if ((indirect = root->info.indirect_levels) > 1)
 378 +               goto fail;
 379 +       entries = (struct dx_entry *) (((char *) &root->info) + root->info.info_length);
 380 +       assert (dx_get_limit(entries) == dx_root_limit(dir, root->info.info_length));
 381 +       dxtrace (printk("Look up %x", hash));
 382 +       while (1)
 383 +       {
 384 +               count = dx_get_count(entries);
 385 +               assert (count && count <= dx_get_limit(entries));
 386 +               p = entries + 1;
 387 +               q = entries + count - 1;
 388 +               while (p <= q)
 389 +               {
 390 +                       m = p + (q - p)/2;
 391 +                       dxtrace(printk("."));
 392 +                       if (dx_get_hash(m) > hash)
 393 +                               q = m - 1;
 394 +                       else
 395 +                               p = m + 1;
 396 +               }
 397 +
 398 +               if (0) // linear search cross check
 399 +               {
 400 +                       unsigned n = count - 1;
 401 +                       at = entries;
 402 +                       while (n--)
 403 +                       {
 404 +                               dxtrace(printk(","));
 405 +                               if (dx_get_hash(++at) > hash)
 406 +                               {
 407 +                                       at--;
 408 +                                       break;
 409 +                               }
 410 +                       }
 411 +                       assert (at == p - 1);
 412 +               }
 413 +
 414 +               at = p - 1;
 415 +               dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 416 +               frame->bh = bh;
 417 +               frame->entries = entries;
 418 +               frame->at = at;
 419 +               if (!indirect--) return frame;
 420 +               if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0,&err)))
 421 +                       goto fail2;
 422 +               at = entries = ((struct dx_node *) bh->b_data)->entries;
 423 +               assert (dx_get_limit(entries) == dx_node_limit (dir));
 424 +               frame++;
 425 +       }
 426 +fail2:
 427 +       brelse(frame->bh);
 428 +fail:
 429 +       return NULL;
 430 +}
 431 +
 432 +static void dx_release (struct dx_frame *frames)
 433 +{
 434 +       if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 435 +               brelse (frames[1].bh);
 436 +       brelse (frames[0].bh);
 437 +}
 438 +
 439 +/*
 440 + * Directory block splitting, compacting
 441 + */
 442 +
 443 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[])
 444 +{
 445 +       int count = 0;
 446 +       char *base = (char *) de;
 447 +       while ((char *) de < base + size)
 448 +       {
 449 +               map[count].hash = dx_hash (de->name, de->name_len);
 450 +               map[count].offs = (u32) ((char *) de - base);
 451 +               de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
 452 +               count++;
 453 +       }
 454 +       return count;
 455 +}
 456 +
 457 +static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 458 +{
 459 +        struct dx_map_entry *p, *q, *top = map + count - 1;
 460 +        int more;
 461 +        /* Combsort until bubble sort doesn't suck */
 462 +        while (count > 2)
 463 +       {
 464 +                count = count*10/13;
 465 +                if (count - 9 < 2) /* 9, 10 -> 11 */
 466 +                        count = 11;
 467 +                for (p = top, q = p - count; q >= map; p--, q--)
 468 +                        if (p->hash < q->hash)
 469 +                                swap(*p, *q);
 470 +        }
 471 +        /* Garden variety bubble sort */
 472 +        do {
 473 +                more = 0;
 474 +                q = top;
 475 +                while (q-- > map)
 476 +               {
 477 +                        if (q[1].hash >= q[0].hash)
 478 +                               continue;
 479 +                        swap(*(q+1), *q);
 480 +                        more = 1;
 481 +               }
 482 +       } while(more);
 483 +}
 484 +
 485 +static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block)
 486 +{
 487 +       struct dx_entry *entries = frame->entries, *at = frame->at;
 488 +       assert (dx_get_count (entries) < dx_get_limit (entries) &&
 489 +               frame->at < frame->entries+dx_get_count(entries));
 490 +       memmove (at + 2, at+1, (char *) (entries + dx_get_count(entries)) - (char *) (at));
 491 +       dx_set_hash(at + 1, hash);
 492 +       dx_set_block(at + 1, block);
 493 +       dx_set_count(entries, dx_get_count(entries) + 1);
 494 +}
 495 +#endif
 496 +
 497  /*
 498   * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 499   *
 500 @@ -95,6 +529,15 @@
 501  }
 502
 503  /*
 504 + * p is at least 6 bytes before the end of page
 505 + */
 506 +static inline ext3_dirent *ext3_next_entry(ext3_dirent *p)
 507 +{
 508 +       return (ext3_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 509 +}
 510 +
 511 +
 512 +/*
 513   *     ext3_find_entry()
 514   *
 515   * finds an entry in the specified directory with the wanted name. It
 516 @@ -105,6 +548,8 @@
 517   * The returned buffer_head has ->b_count elevated.  The caller is expected
 518   * to brelse() it when appropriate.
 519   */
 520 +
 521 +
 522  static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 523                                         struct ext3_dir_entry_2 ** res_dir)
 524  {
 525 @@ -119,10 +564,76 @@
 526         int num = 0;
 527         int nblocks, i, err;
 528         struct inode *dir = dentry->d_parent->d_inode;
 529 +       int namelen;
 530 +       const u8 *name;
 531 +       unsigned blocksize;
 532 +       ext3_dirent *de, *top;
 533
 534         *res_dir = NULL;
 535         sb = dir->i_sb;
 536 +       blocksize = sb->s_blocksize;
 537 +       namelen = dentry->d_name.len;
 538 +       name = dentry->d_name.name;
 539 +       if (namelen > EXT3_NAME_LEN)
 540 +               return NULL;
 541 +       if (ext3_dx && is_dx(dir)) {
 542 +               u32 hash = dx_hash (name, namelen);
 543 +               struct dx_frame frames[2], *frame;
 544 +               if (!(frame = dx_probe (dir, hash, frames)))
 545 +                       return NULL;
 546 +dxnext:
 547 +               block = dx_get_block(frame->at);
 548 +               if (!(bh = ext3_bread (NULL,dir, block, 0, &err)))
 549 +                       goto dxfail;
 550 +               de = (ext3_dirent *) bh->b_data;
 551 +               top = (ext3_dirent *) ((char *) de + blocksize -
 552 +                               EXT3_DIR_REC_LEN(0));
 553 +               for (; de < top; de = ext3_next_entry(de))
 554 +                       if (ext3_match (namelen, name, de)) {
 555 +                               if (!ext3_check_dir_entry("ext3_find_entry",
 556 +                                         dir, de, bh,
 557 +                                         (block<<EXT3_BLOCK_SIZE_BITS(sb))
 558 +                                          +((char *)de - bh->b_data))) {
 559 +                                       brelse (bh);
 560 +                                       goto dxfail;
 561 +                               }
 562 +                               *res_dir = de;
 563 +                               goto dxfound;
 564 +                       }
 565 +               brelse (bh);
 566 +               /* Same hash continues in next block?  Search on. */
 567 +               if (++(frame->at) == frame->entries + dx_get_count(frame->entries))
 568 +               {
 569 +                       struct buffer_head *bh2;
 570 +                       if (frame == frames)
 571 +                               goto dxfail;
 572 +                       if (++(frames->at) == frames->entries + dx_get_count(frames->entries))
 573 +                               goto dxfail;
 574 +                       /* should omit read if not continued */
 575 +                       if (!(bh2 = ext3_bread (NULL, dir,
 576 +                                               dx_get_block(frames->at),
 577 +                                               0, &err)))
 578 +                               goto dxfail;
 579 +                       brelse (frame->bh);
 580 +                       frame->bh = bh2;
 581 +                       frame->at = frame->entries = ((struct dx_node *) bh2->b_data)->entries;
 582 +                       /* Subtle: the 0th entry has the count, find the hash in frame above */
 583 +                       if ((dx_get_hash(frames->at) & -2) == hash)
 584 +                               goto dxnext;
 585 +                       goto dxfail;
 586 +               }
 587 +               if ((dx_get_hash(frame->at) & -2) == hash)
 588 +                       goto dxnext;
 589 +dxfail:
 590 +               dxtrace(printk("%s not found\n", name));
 591 +               dx_release (frames);
 592 +               return NULL;
 593 +dxfound:
 594 +               dx_release (frames);
 595 +               return bh;
 596
 597 +       }
 598 +
 599         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 600         start = dir->u.ext3_i.i_dir_start_lookup;
 601         if (start >= nblocks)
 602 @@ -237,6 +748,88 @@
 603                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
 604  }
 605
 606 +static ext3_dirent *
 607 +dx_copy_dirents (char *from, char *to, struct dx_map_entry *map, int count)
 608 +{
 609 +       unsigned rec_len = 0;
 610 +
 611 +       while (count--) {
 612 +               ext3_dirent *de = (ext3_dirent *) (from + map->offs);
 613 +               rec_len = EXT3_DIR_REC_LEN(de->name_len);
 614 +               memcpy (to, de, rec_len);
 615 +               ((ext3_dirent *) to)->rec_len = rec_len;
 616 +               to += rec_len;
 617 +               map++;
 618 +       }
 619 +       return (ext3_dirent *) (to - rec_len);
 620 +}
 621 +
 622 +#ifdef CONFIG_EXT3_INDEX
 623 +static ext3_dirent *do_split(handle_t *handle, struct inode *dir,
 624 +                       struct buffer_head **bh,struct dx_frame *frame,
 625 +                       u32 hash, int *error)
 626 +{
 627 +       unsigned blocksize = dir->i_sb->s_blocksize;
 628 +       unsigned count, continued;
 629 +       struct buffer_head *bh2;
 630 +       u32 newblock;
 631 +       unsigned MAX_DX_MAP = PAGE_CACHE_SIZE/EXT3_DIR_REC_LEN(1) + 1;
 632 +       u32 hash2;
 633 +       struct dx_map_entry map[MAX_DX_MAP];
 634 +       char *data1 = (*bh)->b_data, *data2, *data3;
 635 +       unsigned split;
 636 +       ext3_dirent *de, *de2;
 637 +
 638 +       bh2 = ext3_append (handle, dir, &newblock, error);
 639 +       if (!(bh2))
 640 +       {
 641 +               brelse(*bh);
 642 +               *bh = NULL;
 643 +               return (ext3_dirent *)bh2;
 644 +       }
 645 +
 646 +       BUFFER_TRACE(*bh, "get_write_access");
 647 +       ext3_journal_get_write_access(handle, *bh);
 648 +       BUFFER_TRACE(frame->bh, "get_write_access");
 649 +       ext3_journal_get_write_access(handle, frame->bh);
 650 +
 651 +       data2 = bh2->b_data;
 652 +
 653 +       count = dx_make_map ((ext3_dirent *) data1, blocksize, map);
 654 +       split = count/2; // need to adjust to actual middle
 655 +       dx_sort_map (map, count);
 656 +       hash2 = map[split].hash;
 657 +       continued = hash2 == map[split - 1].hash;
 658 +       dxtrace(printk("Split block %i at %x, %i/%i\n",
 659 +               dx_get_block(frame->at), hash2, split, count-split));
 660 +
 661 +       /* Fancy dance to stay within two buffers */
 662 +       de2 = dx_copy_dirents (data1, data2, map + split, count - split);
 663 +       data3 = (char *) de2 + de2->rec_len;
 664 +       de = dx_copy_dirents (data1, data3, map, split);
 665 +       memcpy(data1, data3, (char *) de + de->rec_len - data3);
 666 +       de = (ext3_dirent *) ((char *) de - data3 + data1); // relocate de
 667 +       de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
 668 +       de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
 669 +       dxtrace(dx_show_leaf ((ext3_dirent *) data1, blocksize, 1));
 670 +       dxtrace(dx_show_leaf ((ext3_dirent *) data2, blocksize, 1));
 671 +
 672 +       /* Which block gets the new entry? */
 673 +       if (hash >= hash2)
 674 +       {
 675 +               swap(*bh, bh2);
 676 +               de = de2;
 677 +       }
 678 +       dx_insert_block (frame, hash2 + continued, newblock);
 679 +       ext3_journal_dirty_metadata (handle, bh2);
 680 +       brelse (bh2);
 681 +       ext3_journal_dirty_metadata (handle, frame->bh);
 682 +       dxtrace(dx_show_index ("frame", frame->entries));
 683 +       return de;
 684 +}
 685 +#endif
 686 +
 687 +
 688  /*
 689   *     ext3_add_entry()
 690   *
 691 @@ -251,6 +844,7 @@
 692  /*
 693   * AKPM: the journalling code here looks wrong on the error paths
 694   */
 695 +
 696  static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
 697         struct inode *inode)
 698  {
 699 @@ -258,117 +852,282 @@
 700         const char *name = dentry->d_name.name;
 701         int namelen = dentry->d_name.len;
 702         unsigned long offset;
 703 -       unsigned short rec_len;
 704         struct buffer_head * bh;
 705 -       struct ext3_dir_entry_2 * de, * de1;
 706 -       struct super_block * sb;
 707 +       ext3_dirent *de;
 708 +       struct super_block * sb = dir->i_sb;
 709         int     retval;
 710 +       unsigned short reclen = EXT3_DIR_REC_LEN(namelen);
 711
 712 -       sb = dir->i_sb;
 713 +       unsigned blocksize = sb->s_blocksize;
 714 +       unsigned nlen, rlen;
 715 +       u32 block, blocks;
 716 +       char *top;
 717
 718         if (!namelen)
 719                 return -EINVAL;
 720 -       bh = ext3_bread (handle, dir, 0, 0, &retval);
 721 -       if (!bh)
 722 -               return retval;
 723 -       rec_len = EXT3_DIR_REC_LEN(namelen);
 724 -       offset = 0;
 725 -       de = (struct ext3_dir_entry_2 *) bh->b_data;
 726 -       while (1) {
 727 -               if ((char *)de >= sb->s_blocksize + bh->b_data) {
 728 -                       brelse (bh);
 729 -                       bh = NULL;
 730 -                       bh = ext3_bread (handle, dir,
 731 -                               offset >> EXT3_BLOCK_SIZE_BITS(sb), 1, &retval);
 732 -                       if (!bh)
 733 -                               return retval;
 734 -                       if (dir->i_size <= offset) {
 735 -                               if (dir->i_size == 0) {
 736 -                                       brelse(bh);
 737 -                                       return -ENOENT;
 738 +       if (ext3_dx && is_dx(dir))
 739 +       {
 740 +               struct dx_frame frames[2], *frame;
 741 +               struct dx_entry *entries, *at;
 742 +               u32 hash;
 743 +               char *data1;
 744 +
 745 +               hash = dx_hash (name, namelen);
 746 +               frame = dx_probe (dir, hash, frames); // do something if null
 747 +               entries = frame->entries;
 748 +               at = frame->at;
 749 +
 750 +               if (!(bh = ext3_bread (handle,dir, dx_get_block(frame->at), 0,&retval)))
 751 +                       goto dxfail1;
 752 +
 753 +               BUFFER_TRACE(bh, "get_write_access");
 754 +               ext3_journal_get_write_access(handle, bh);
 755 +
 756 +               data1 = bh->b_data;
 757 +               de = (ext3_dirent *) data1;
 758 +               top = data1 + (0? 200: blocksize);
 759 +               while ((char *) de < top)
 760 +               {
 761 +                       /* FIXME: check EEXIST and dir */
 762 +                       nlen = EXT3_DIR_REC_LEN(de->name_len);
 763 +                       rlen = le16_to_cpu(de->rec_len);
 764 +                       if ((de->inode? rlen - nlen: rlen) >= reclen)
 765 +                               goto dx_add;
 766 +                       de = (ext3_dirent *) ((char *) de + rlen);
 767 +               }
 768 +               /* Block full, should compress but for now just split */
 769 +               dxtrace(printk("using %u of %u node entries\n",
 770 +                       dx_get_count(entries), dx_get_limit(entries)));
 771 +               /* Need to split index? */
 772 +               if (dx_get_count(entries) == dx_get_limit(entries))
 773 +               {
 774 +                       u32 newblock;
 775 +                       unsigned icount = dx_get_count(entries);
 776 +                       int levels = frame - frames;
 777 +                       struct dx_entry *entries2;
 778 +                       struct dx_node *node2;
 779 +                       struct buffer_head *bh2;
 780 +                       if (levels && dx_get_count(frames->entries) == dx_get_limit(frames->entries))
 781 +                               goto dxfull;
 782 +                       bh2 = ext3_append (handle, dir, &newblock, &retval);
 783 +                       if (!(bh2))
 784 +                               goto dxfail2;
 785 +                       node2 = (struct dx_node *)(bh2->b_data);
 786 +                       entries2 = node2->entries;
 787 +                       node2->fake.rec_len = cpu_to_le16(blocksize);
 788 +                       node2->fake.inode = 0;
 789 +                       BUFFER_TRACE(frame->bh, "get_write_access");
 790 +                       ext3_journal_get_write_access(handle, frame->bh);
 791 +                       if (levels)
 792 +                       {
 793 +                               unsigned icount1 = icount/2, icount2 = icount - icount1;
 794 +                               unsigned hash2 = dx_get_hash(entries + icount1);
 795 +                               dxtrace(printk("Split index %i/%i\n", icount1, icount2));
 796 +
 797 +                               BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
 798 +                               ext3_journal_get_write_access(handle, frames[0].bh);
 799 +
 800 +                               memcpy ((char *) entries2, (char *) (entries + icount1),
 801 +                                       icount2 * sizeof(struct dx_entry));
 802 +                               dx_set_count (entries, icount1);
 803 +                               dx_set_count (entries2, icount2);
 804 +                               dx_set_limit (entries2, dx_node_limit(dir));
 805 +
 806 +                               /* Which index block gets the new entry? */
 807 +                               if (at - entries >= icount1) {
 808 +                                       frame->at = at = at - entries - icount1 + entries2;
 809 +                                       frame->entries = entries = entries2;
 810 +                                       swap(frame->bh, bh2);
 811                                 }
 812 -
 813 -                               ext3_debug ("creating next block\n");
 814 -
 815 -                               BUFFER_TRACE(bh, "get_write_access");
 816 -                               ext3_journal_get_write_access(handle, bh);
 817 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
 818 -                               de->inode = 0;
 819 -                               de->rec_len = le16_to_cpu(sb->s_blocksize);
 820 -                               dir->u.ext3_i.i_disksize =
 821 -                                       dir->i_size = offset + sb->s_blocksize;
 822 -                               dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
 823 -                               ext3_mark_inode_dirty(handle, dir);
 824 +                               dx_insert_block (frames + 0, hash2, newblock);
 825 +                               dxtrace(dx_show_index ("node", frames[1].entries));
 826 +                               dxtrace(dx_show_index ("node",
 827 +                                       ((struct dx_node *) bh2->b_data)->entries));
 828 +                               ext3_journal_dirty_metadata(handle, bh2);
 829 +                               brelse (bh2);
 830                         } else {
 831 -
 832 -                               ext3_debug ("skipping to next block\n");
 833 -
 834 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
 835 +                               dxtrace(printk("Creating second level index...\n"));
 836 +                               memcpy((char *) entries2, (char *) entries,
 837 +                                       icount * sizeof(struct dx_entry));
 838 +                               dx_set_limit(entries2, dx_node_limit(dir));
 839 +
 840 +                               /* Set up root */
 841 +                               dx_set_count(entries, 1);
 842 +                               dx_set_block(entries + 0, newblock);
 843 +                               ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
 844 +
 845 +                               /* Add new access path frame */
 846 +                               frame = frames + 1;
 847 +                               frame->at = at = at - entries + entries2;
 848 +                               frame->entries = entries = entries2;
 849 +                               frame->bh = bh2;
 850 +                               ext3_journal_get_write_access(handle, frame->bh);
 851                         }
 852 +                       ext3_journal_dirty_metadata(handle, frames[0].bh);
 853                 }
 854 -               if (!ext3_check_dir_entry ("ext3_add_entry", dir, de, bh,
 855 -                                          offset)) {
 856 -                       brelse (bh);
 857 -                       return -ENOENT;
 858 -               }
 859 -               if (ext3_match (namelen, name, de)) {
 860 +               de = do_split(handle, dir, &bh, frame, hash, &retval);
 861 +               dx_release (frames);
 862 +               if (!(de))
 863 +                       goto fail;
 864 +               nlen = EXT3_DIR_REC_LEN(de->name_len);
 865 +               rlen = le16_to_cpu(de->rec_len);
 866 +               goto add;
 867 +
 868 +dx_add:
 869 +               dx_release (frames);
 870 +               goto add;
 871 +
 872 +dxfull:
 873 +               ext3_warning(sb, __FUNCTION__, "Directory index full!\n");
 874 +               retval = -ENOSPC;
 875 +dxfail2:
 876 +               brelse(bh);
 877 +dxfail1:
 878 +               dx_release (frames);
 879 +               goto fail1;
 880 +       }
 881 +
 882 +       blocks = dir->i_size >> sb->s_blocksize_bits;
 883 +       for (block = 0, offset = 0; block < blocks; block++) {
 884 +               bh = ext3_bread(handle, dir, block, 0, &retval);
 885 +               if(!bh)
 886 +                       return retval;
 887 +               de = (ext3_dirent *)bh->b_data;
 888 +               top = bh->b_data + blocksize - reclen;
 889 +               while ((char *) de <= top) {
 890 +                       if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
 891 +                                                 bh, offset)) {
 892 +                               brelse (bh);
 893 +                               return -EIO;
 894 +                       }
 895 +                       if (ext3_match (namelen, name, de)) {
 896                                 brelse (bh);
 897                                 return -EEXIST;
 898 -               }
 899 -               if ((le32_to_cpu(de->inode) == 0 &&
 900 -                               le16_to_cpu(de->rec_len) >= rec_len) ||
 901 -                   (le16_to_cpu(de->rec_len) >=
 902 -                               EXT3_DIR_REC_LEN(de->name_len) + rec_len)) {
 903 -                       BUFFER_TRACE(bh, "get_write_access");
 904 -                       ext3_journal_get_write_access(handle, bh);
 905 -                       /* By now the buffer is marked for journaling */
 906 -                       offset += le16_to_cpu(de->rec_len);
 907 -                       if (le32_to_cpu(de->inode)) {
 908 -                               de1 = (struct ext3_dir_entry_2 *) ((char *) de +
 909 -                                       EXT3_DIR_REC_LEN(de->name_len));
 910 -                               de1->rec_len =
 911 -                                       cpu_to_le16(le16_to_cpu(de->rec_len) -
 912 -                                       EXT3_DIR_REC_LEN(de->name_len));
 913 -                               de->rec_len = cpu_to_le16(
 914 -                                               EXT3_DIR_REC_LEN(de->name_len));
 915 -                               de = de1;
 916                         }
 917 -                       de->file_type = EXT3_FT_UNKNOWN;
 918 -                       if (inode) {
 919 -                               de->inode = cpu_to_le32(inode->i_ino);
 920 -                               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
 921 -                       } else
 922 -                               de->inode = 0;
 923 -                       de->name_len = namelen;
 924 -                       memcpy (de->name, name, namelen);
 925 -                       /*
 926 -                        * XXX shouldn't update any times until successful
 927 -                        * completion of syscall, but too many callers depend
 928 -                        * on this.
 929 -                        *
 930 -                        * XXX similarly, too many callers depend on
 931 -                        * ext3_new_inode() setting the times, but error
 932 -                        * recovery deletes the inode, so the worst that can
 933 -                        * happen is that the times are slightly out of date
 934 -                        * and/or different from the directory change time.
 935 -                        */
 936 -                       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 937 -                       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
 938 -                       ext3_mark_inode_dirty(handle, dir);
 939 -                       dir->i_version = ++event;
 940 -                       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
 941 -                       ext3_journal_dirty_metadata(handle, bh);
 942 +                       nlen = EXT3_DIR_REC_LEN(de->name_len);
 943 +                       rlen = le16_to_cpu(de->rec_len);
 944 +                       if ((de->inode? rlen - nlen: rlen) >= reclen)
 945 +                               goto add;
 946 +                       de = (ext3_dirent *)((char *)de + rlen);
 947 +                       offset += rlen;
 948 +               }
 949 +               if (ext3_dx && blocks == 1 && test_opt(sb, INDEX))
 950 +                       goto dx_make_index;
 951 +               brelse(bh);
 952 +       }
 953 +       bh = ext3_append(handle, dir, &block, &retval);
 954 +       if (!bh)
 955 +               return retval;
 956 +       de = (ext3_dirent *) bh->b_data;
 957 +       de->inode = 0;
 958 +       de->rec_len = cpu_to_le16(rlen = blocksize);
 959 +       nlen = 0;
 960 +       goto add;
 961 +
 962 +add:
 963 +       BUFFER_TRACE(bh, "get_write_access");
 964 +       ext3_journal_get_write_access(handle, bh);
 965 +       /* By now the buffer is marked for journaling */
 966 +       if (de->inode) {
 967 +               ext3_dirent *de1 = (ext3_dirent *)((char *)de + nlen);
 968 +               de1->rec_len = cpu_to_le16(rlen - nlen);
 969 +               de->rec_len = cpu_to_le16(nlen);
 970 +               de = de1;
 971 +       }
 972 +       de->file_type = EXT3_FT_UNKNOWN;
 973 +       if (inode) {
 974 +               de->inode = cpu_to_le32(inode->i_ino);
 975 +               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
 976 +       } else
 977 +               de->inode = 0;
 978 +       de->name_len = namelen;
 979 +       memcpy (de->name, name, namelen);
 980 +       /*
 981 +        * XXX shouldn't update any times until successful
 982 +        * completion of syscall, but too many callers depend
 983 +        * on this.
 984 +        *
 985 +        * XXX similarly, too many callers depend on
 986 +        * ext3_new_inode() setting the times, but error
 987 +        * recovery deletes the inode, so the worst that can
 988 +        * happen is that the times are slightly out of date
 989 +        * and/or different from the directory change time.
 990 +        */
 991 +       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 992 +       /* EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL; */
 993 +       ext3_mark_inode_dirty(handle, dir);
 994 +       dir->i_version = ++event;
 995 +       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
 996 +       ext3_journal_dirty_metadata(handle, bh);
 997 +       brelse(bh);
 998 +       return 0;
 999 +
1000 +dx_make_index:
1001 +       {
1002 +               struct buffer_head *bh2;
1003 +               struct dx_root *root;
1004 +               struct dx_frame frames[2], *frame;
1005 +               struct dx_entry *entries;
1006 +               ext3_dirent *de2;
1007 +               char *data1;
1008 +               unsigned len;
1009 +               u32 hash;
1010 +
1011 +               dxtrace(printk("Creating index\n"));
1012 +               ext3_journal_get_write_access(handle, bh);
1013 +               root = (struct dx_root *) bh->b_data;
1014 +
1015 +               ext3_add_compat_feature (sb, EXT3_FEATURE_COMPAT_DIR_INDEX);
1016 +               EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1017 +               bh2 = ext3_append (handle, dir, &block, &retval);
1018 +               if (!(bh2))
1019 +               {
1020                         brelse(bh);
1021 -                       return 0;
1022 +                       return retval;
1023                 }
1024 -               offset += le16_to_cpu(de->rec_len);
1025 -               de = (struct ext3_dir_entry_2 *)
1026 -                       ((char *) de + le16_to_cpu(de->rec_len));
1027 +               data1 = bh2->b_data;
1028 +
1029 +               /* The 0th block becomes the root, move the dirents out */
1030 +               de = (ext3_dirent *) &root->info;
1031 +               len = ((char *) root) + blocksize - (char *) de;
1032 +               memcpy (data1, de, len);
1033 +               de = (ext3_dirent *) data1;
1034 +               top = data1 + len;
1035 +               while (((char *) de2=(char*)de+le16_to_cpu(de->rec_len)) < top)
1036 +                       de = de2;
1037 +               de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1038 +               /* Initialize the root; the dot dirents already exist */
1039 +               de = (ext3_dirent *) (&root->dotdot);
1040 +               de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1041 +               memset (&root->info, 0, sizeof(root->info));
1042 +               root->info.info_length = sizeof(root->info);
1043 +               entries = root->entries;
1044 +               dx_set_block (entries, 1);
1045 +               dx_set_count (entries, 1);
1046 +               dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1047 +
1048 +               /* Initialize as for dx_probe */
1049 +               hash = dx_hash (name, namelen);
1050 +               frame = frames;
1051 +               frame->entries = entries;
1052 +               frame->at = entries;
1053 +               frame->bh = bh;
1054 +               bh = bh2;
1055 +               de = do_split(handle,dir, &bh, frame, hash, &retval);
1056 +               dx_release (frames);
1057 +               if (!(de))
1058 +                       return retval;
1059 +               nlen = EXT3_DIR_REC_LEN(de->name_len);
1060 +               rlen = le16_to_cpu(de->rec_len);
1061 +               goto add;
1062         }
1063 -       brelse (bh);
1064 -       return -ENOSPC;
1065 +fail1:
1066 +       return retval;
1067 +fail:
1068 +       return -ENOENT;
1069  }
1070
1071 +
1072  /*
1073   * ext3_delete_entry deletes a directory entry by merging it with the
1074   * previous entry
1075 @@ -451,7 +1212,8 @@
1076         struct inode * inode;
1077         int err;
1078
1079 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1080 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1081 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1082         if (IS_ERR(handle))
1083                 return PTR_ERR(handle);
1084
1085 @@ -478,7 +1240,8 @@
1086         struct inode *inode;
1087         int err;
1088
1089 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1090 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1091 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1092         if (IS_ERR(handle))
1093                 return PTR_ERR(handle);
1094
1095 @@ -507,7 +1270,8 @@
1096         if (dir->i_nlink >= EXT3_LINK_MAX)
1097                 return -EMLINK;
1098
1099 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1100 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1101 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1102         if (IS_ERR(handle))
1103                 return PTR_ERR(handle);
1104
1105 @@ -832,7 +1596,7 @@
1106         ext3_mark_inode_dirty(handle, inode);
1107         dir->i_nlink--;
1108         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1109 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1110 +       //      EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1111         ext3_mark_inode_dirty(handle, dir);
1112
1113  end_rmdir:
1114 @@ -878,7 +1642,7 @@
1115         if (retval)
1116                 goto end_unlink;
1117         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1118 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1119 +       //      EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1120         ext3_mark_inode_dirty(handle, dir);
1121         inode->i_nlink--;
1122         if (!inode->i_nlink)
1123 @@ -904,7 +1668,8 @@
1124         if (l > dir->i_sb->s_blocksize)
1125                 return -ENAMETOOLONG;
1126
1127 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 5);
1128 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1129 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5);
1130         if (IS_ERR(handle))
1131                 return PTR_ERR(handle);
1132
1133 @@ -959,7 +1724,8 @@
1134         if (inode->i_nlink >= EXT3_LINK_MAX)
1135                 return -EMLINK;
1136
1137 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS);
1138 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1139 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS);
1140         if (IS_ERR(handle))
1141                 return PTR_ERR(handle);
1142
1143 @@ -995,7 +1761,8 @@
1144
1145         old_bh = new_bh = dir_bh = NULL;
1146
1147 -       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS + 2);
1148 +       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
1149 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
1150         if (IS_ERR(handle))
1151                 return PTR_ERR(handle);
1152
1153 @@ -1077,7 +1844,7 @@
1154                 new_inode->i_ctime = CURRENT_TIME;
1155         }
1156         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1157 -       old_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1158 +       // EXT3_I(old_dir)->i_flags &= ~EXT3_INDEX_FL;
1159         if (dir_bh) {
1160                 BUFFER_TRACE(dir_bh, "get_write_access");
1161                 ext3_journal_get_write_access(handle, dir_bh);
1162 @@ -1089,7 +1856,7 @@
1163                         new_inode->i_nlink--;
1164                 } else {
1165                         new_dir->i_nlink++;
1166 -                       new_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1167 +       //              EXT3_I(new_dir)->i_flags &= ~EXT3_INDEX_FL;
1168                         ext3_mark_inode_dirty(handle, new_dir);
1169                 }
1170         }
1171 --- ./include/linux/ext3_fs.h   2002/03/05 06:18:59     2.1
1172 +++ ./include/linux/ext3_fs.h   2002/03/05 06:26:56
1173 @@ -339,6 +339,7 @@
1174    #define EXT3_MOUNT_WRITEBACK_DATA    0x0C00  /* No data ordering */
1175  #define EXT3_MOUNT_UPDATE_JOURNAL      0x1000  /* Update the journal format */
1176  #define EXT3_MOUNT_NO_UID32            0x2000  /* Disable 32-bit UIDs */
1177 +#define EXT3_MOUNT_INDEX               0x4000  /* Enable directory index */
1178
1179  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
1180  #ifndef _LINUX_EXT2_FS_H
1181 @@ -575,6 +576,24 @@
1182  #define EXT3_DIR_ROUND                 (EXT3_DIR_PAD - 1)
1183  #define EXT3_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT3_DIR_ROUND) & \
1184                                          ~EXT3_DIR_ROUND)
1185 +/*
1186 + * Hash Tree Directory indexing
1187 + * (c) Daniel Phillips, 2001
1188 + */
1189 +
1190 +#define CONFIG_EXT3_INDEX
1191 +
1192 +#ifdef CONFIG_EXT3_INDEX
1193 +  enum {ext3_dx = 1};
1194 +  #define is_dx(dir) (EXT3_I(dir)->i_flags & EXT3_INDEX_FL)
1195 +#define EXT3_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT3_LINK_MAX)
1196 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
1197 +#else
1198 +  enum {ext3_dx = 0};
1199 +  #define is_dx(dir) 0
1200 +#define EXT3_DIR_LINK_MAX(dir) ((dir)->i_nlink >= EXT3_LINK_MAX)
1201 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2)
1202 +#endif
1203
1204  #ifdef __KERNEL__
1205  /*
1206 --- ./include/linux/ext3_jbd.h  2002/03/05 06:18:59     2.1
1207 +++ ./include/linux/ext3_jbd.h  2002/03/05 06:33:54
1208 @@ -63,6 +63,8 @@
1209
1210  #define EXT3_RESERVE_TRANS_BLOCKS      12
1211
1212 +#define EXT3_INDEX_EXTRA_TRANS_BLOCKS  8
1213 +
1214  int
1215  ext3_mark_iloc_dirty(handle_t *handle,
1216                      struct inode *inode,