lustre/extN/htree-ext3-2.4.18.diff

   1 --- ./fs/ext3/super.c   2002/03/05 06:18:59     2.1
   2 +++ ./fs/ext3/super.c   2002/03/05 06:26:56
   3 @@ -529,6 +529,12 @@
   4                                        "EXT3 Check option not supported\n");
   5  #endif
   6                 }
   7 +               else if (!strcmp (this_char, "index"))
   8 +#ifdef CONFIG_EXT3_INDEX
   9 +                       set_opt (*mount_options, INDEX);
  10 +#else
  11 +                       printk("EXT3 index option not supported\n");
  12 +#endif
  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);
  20 +
  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);
  25 +
  26         ext3_commit_super (sb, es, 1);
  27         if (test_opt (sb, DEBUG))
  28                 printk (KERN_INFO
  29 --- ./fs/ext3/namei.c   2002/03/05 06:18:59     2.1
  30 +++ ./fs/ext3/namei.c   2002/03/06 00:13:18
  31 @@ -16,6 +16,10 @@
  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
  39   */
  40
  41  #include <linux/fs.h>
  42 @@ -33,7 +33,7 @@
  43  #include <linux/string.h>
  44  #include <linux/locks.h>
  45  #include <linux/quotaops.h>
  46 -
  47 +#include <linux/slab.h>
  48
  49  /*
  50   * define how far ahead to read directories while searching them.
  51 @@ -38,6 +42,433 @@
  52  #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  53  #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  54
  55 +static struct buffer_head *ext3_append(handle_t *handle,
  56 +                                       struct inode *inode,
  57 +                                       u32 *block, int *err)
  58 +{
  59 +       struct buffer_head *bh;
  60 +
  61 +       *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  62 +
  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);
  67 +       }
  68 +       return bh;
  69 +}
  70 +
  71 +#ifndef assert
  72 +#define assert(test) J_ASSERT(test)
  73 +#endif
  74 +
  75 +#ifndef swap
  76 +#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
  77 +#endif
  78 +
  79 +typedef struct { u32 v; } le_u32;
  80 +typedef struct { u16 v; } le_u16;
  81 +
  82 +#define dxtrace_on(command) command
  83 +#define dxtrace_off(command)
  84 +#define dxtrace dxtrace_off
  85 +
  86 +struct fake_dirent
  87 +{
  88 +       /*le*/u32 inode;
  89 +       /*le*/u16 rec_len;
  90 +       u8 name_len;
  91 +       u8 file_type;
  92 +};
  93 +
  94 +struct dx_countlimit
  95 +{
  96 +       le_u16 limit;
  97 +       le_u16 count;
  98 +};
  99 +
 100 +struct dx_entry
 101 +{
 102 +       le_u32 hash;
 103 +       le_u32 block;
 104 +};
 105 +
 106 +/*
 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.
 110 + */
 111 +
 112 +struct dx_root
 113 +{
 114 +       struct fake_dirent dot;
 115 +       char dot_name[4];
 116 +       struct fake_dirent dotdot;
 117 +       char dotdot_name[4];
 118 +       struct dx_root_info
 119 +       {
 120 +               le_u32 reserved_zero;
 121 +               u8 hash_version; /* 0 now, 1 at release */
 122 +               u8 info_length; /* 8 */
 123 +               u8 indirect_levels;
 124 +               u8 unused_flags;
 125 +       }
 126 +       info;
 127 +       struct dx_entry entries[0];
 128 +};
 129 +
 130 +struct dx_node
 131 +{
 132 +       struct fake_dirent fake;
 133 +       struct dx_entry entries[0];
 134 +};
 135 +
 136 +
 137 +struct dx_frame
 138 +{
 139 +       struct buffer_head *bh;
 140 +       struct dx_entry *entries;
 141 +       struct dx_entry *at;
 142 +};
 143 +
 144 +struct dx_map_entry
 145 +{
 146 +       u32 hash;
 147 +       u32 offs;
 148 +};
 149 +
 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);
 169 +
 170 +
 171 +#ifdef CONFIG_EXT3_INDEX
 172 +/*
 173 + * Future: use high four bits of block for coalesce-on-delete flags
 174 + * Mask them off for now.
 175 + */
 176 +
 177 +static inline unsigned dx_get_block (struct dx_entry *entry)
 178 +{
 179 +       return le32_to_cpu(entry->block.v) & 0x00ffffff;
 180 +}
 181 +
 182 +static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 183 +{
 184 +       entry->block.v = cpu_to_le32(value);
 185 +}
 186 +
 187 +static inline unsigned dx_get_hash (struct dx_entry *entry)
 188 +{
 189 +       return le32_to_cpu(entry->hash.v);
 190 +}
 191 +
 192 +static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 193 +{
 194 +       entry->hash.v = cpu_to_le32(value);
 195 +}
 196 +
 197 +static inline unsigned dx_get_count (struct dx_entry *entries)
 198 +{
 199 +       return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
 200 +}
 201 +
 202 +static inline unsigned dx_get_limit (struct dx_entry *entries)
 203 +{
 204 +       return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
 205 +}
 206 +
 207 +static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 208 +{
 209 +       ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
 210 +}
 211 +
 212 +static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 213 +{
 214 +       ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
 215 +}
 216 +
 217 +static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 218 +{
 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);
 222 +}
 223 +
 224 +static inline unsigned dx_node_limit (struct inode *dir)
 225 +{
 226 +       unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 227 +       return 0? 22: entry_space / sizeof(struct dx_entry);
 228 +}
 229 +
 230 +/* Hash function - not bad, but still looking for an ideal default */
 231 +
 232 +static unsigned dx_hack_hash (const u8 *name, int len)
 233 +{
 234 +       u32 hash0 = 0x12a3fe2d, hash1 = 0x37abe8f9;
 235 +       while (len--)
 236 +       {
 237 +               u32 hash = hash1 + (hash0 ^ (*name++ * 7152373));
 238 +               if (hash & 0x80000000) hash -= 0x7fffffff;
 239 +               hash1 = hash0;
 240 +               hash0 = hash;
 241 +       }
 242 +       return hash0;
 243 +}
 244 +
 245 +#define dx_hash(s,n) (dx_hack_hash(s,n) << 1)
 246 +
 247 +/*
 248 + * Debug
 249 + */
 250 +static void dx_show_index (char * label, struct dx_entry *entries)
 251 +{
 252 +       int i, n = dx_get_count (entries);
 253 +       printk("%s index ", label);
 254 +       for (i = 0; i < n; i++)
 255 +       {
 256 +               printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
 257 +       }
 258 +       printk("\n");
 259 +}
 260 +
 261 +struct stats
 262 +{
 263 +       unsigned names;
 264 +       unsigned space;
 265 +       unsigned bcount;
 266 +};
 267 +
 268 +static struct stats dx_show_leaf (ext3_dirent *de, int size, int show_names)
 269 +{
 270 +       unsigned names = 0, space = 0;
 271 +       char *base = (char *) de;
 272 +       printk("names: ");
 273 +       while ((char *) de < base + size)
 274 +       {
 275 +               if (de->inode)
 276 +               {
 277 +                       if (show_names)
 278 +                       {
 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));
 283 +                       }
 284 +                       space += EXT3_DIR_REC_LEN(de->name_len);
 285 +                       names++;
 286 +               }
 287 +               de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
 288 +       }
 289 +       printk("(%i)\n", names);
 290 +       return (struct stats) { names, space, 1 };
 291 +}
 292 +
 293 +struct stats dx_show_entries (struct inode *dir, struct dx_entry *entries, int levels)
 294 +{
 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;
 299 +       int err;
 300 +       printk("%i indexed blocks...\n", count);
 301 +       for (i = 0; i < count; i++, entries++)
 302 +       {
 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;
 308 +               stats = levels?
 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;
 314 +               brelse (bh);
 315 +       }
 316 +       if (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};
 320 +}
 321 +
 322 +/*
 323 + * Probe for a directory leaf block to search
 324 + */
 325 +
 326 +static struct dx_frame *
 327 +dx_probe(struct inode *dir, u32 hash, struct dx_frame *frame_in)
 328 +{
 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;
 334 +       int err;
 335 +
 336 +       frame->bh = NULL;
 337 +       if (!(bh = ext3_bread(NULL, dir, 0, 0, &err)))
 338 +               goto fail;
 339 +       root = (struct dx_root *) bh->b_data;
 340 +       if (root->info.hash_version > 0 || root->info.unused_flags & 1) {
 341 +               brelse(bh);
 342 +               goto fail;
 343 +       }
 344 +       if ((indirect = root->info.indirect_levels) > 1) {
 345 +               brelse(bh);
 346 +               goto fail;
 347 +       }
 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));
 351 +       while (1)
 352 +       {
 353 +               count = dx_get_count(entries);
 354 +               assert (count && count <= dx_get_limit(entries));
 355 +               p = entries + 1;
 356 +               q = entries + count - 1;
 357 +               while (p <= q)
 358 +               {
 359 +                       m = p + (q - p)/2;
 360 +                       dxtrace(printk("."));
 361 +                       if (dx_get_hash(m) > hash)
 362 +                               q = m - 1;
 363 +                       else
 364 +                               p = m + 1;
 365 +               }
 366 +
 367 +               if (0) // linear search cross check
 368 +               {
 369 +                       unsigned n = count - 1;
 370 +                       at = entries;
 371 +                       while (n--)
 372 +                       {
 373 +                               dxtrace(printk(","));
 374 +                               if (dx_get_hash(++at) > hash)
 375 +                               {
 376 +                                       at--;
 377 +                                       break;
 378 +                               }
 379 +                       }
 380 +                       assert (at == p - 1);
 381 +               }
 382 +
 383 +               at = p - 1;
 384 +               dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 385 +               frame->bh = bh;
 386 +               frame->entries = entries;
 387 +               frame->at = at;
 388 +               if (!indirect--) return frame;
 389 +               if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0,&err)))
 390 +                       goto fail2;
 391 +               at = entries = ((struct dx_node *) bh->b_data)->entries;
 392 +               assert (dx_get_limit(entries) == dx_node_limit (dir));
 393 +               frame++;
 394 +       }
 395 +fail2:
 396 +       while (frame >= frame_in) {
 397 +               brelse(frame->bh);
 398 +               frame--;
 399 +       }
 400 +fail:
 401 +       return NULL;
 402 +}
 403 +
 404 +static void dx_release (struct dx_frame *frames)
 405 +{
 406 +       if (frames[0].bh == NULL)
 407 +               return;
 408 +
 409 +       if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
 410 +               brelse (frames[1].bh);
 411 +       brelse (frames[0].bh);
 412 +}
 413 +
 414 +/*
 415 + * Directory block splitting, compacting
 416 + */
 417 +
 418 +static int dx_make_map (ext3_dirent *de, int size, struct dx_map_entry map[])
 419 +{
 420 +       int count = 0;
 421 +       char *base = (char *) de;
 422 +       while ((char *) de < base + size) {
 423 +               if (de->name_len && de->inode) {
 424 +                       map[count].hash = dx_hash (de->name, de->name_len);
 425 +                       map[count].offs = (u32) ((char *) de - base);
 426 +                       count++;
 427 +               }
 428 +               de = (ext3_dirent *) ((char *) de + le16_to_cpu(de->rec_len));
 429 +       }
 430 +       return count;
 431 +}
 432 +
 433 +static void dx_sort_map (struct dx_map_entry *map, unsigned count)
 434 +{
 435 +        struct dx_map_entry *p, *q, *top = map + count - 1;
 436 +        int more;
 437 +        /* Combsort until bubble sort doesn't suck */
 438 +        while (count > 2)
 439 +       {
 440 +                count = count*10/13;
 441 +                if (count - 9 < 2) /* 9, 10 -> 11 */
 442 +                        count = 11;
 443 +                for (p = top, q = p - count; q >= map; p--, q--)
 444 +                        if (p->hash < q->hash)
 445 +                                swap(*p, *q);
 446 +        }
 447 +        /* Garden variety bubble sort */
 448 +        do {
 449 +                more = 0;
 450 +                q = top;
 451 +                while (q-- > map)
 452 +               {
 453 +                        if (q[1].hash >= q[0].hash)
 454 +                               continue;
 455 +                        swap(*(q+1), *q);
 456 +                        more = 1;
 457 +               }
 458 +       } while(more);
 459 +}
 460 +
 461 +static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
 462 +{
 463 +       struct dx_entry *entries = frame->entries;
 464 +       struct dx_entry *old = frame->at, *new = old + 1;
 465 +       int count = dx_get_count(entries);
 466 +
 467 +       assert(count < dx_get_limit(entries));
 468 +       assert(old < entries + count);
 469 +       memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
 470 +       dx_set_hash(new, hash);
 471 +       dx_set_block(new, block);
 472 +       dx_set_count(entries, count + 1);
 473 +}
 474 +#endif
 475 +
 476 +static void ext3_update_dx_flag(struct inode *inode)
 477 +{
 478 +       if (!test_opt(inode->i_sb, INDEX))
 479 +               EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
 480 +}
 481 +
 482  /*
 483   * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
 484   *
 485 @@ -95,6 +529,15 @@
 486  }
 487
 488  /*
 489 + * p is at least 6 bytes before the end of page
 490 + */
 491 +static inline ext3_dirent *ext3_next_entry(ext3_dirent *p)
 492 +{
 493 +       return (ext3_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 494 +}
 495 +
 496 +
 497 +/*
 498   *     ext3_find_entry()
 499   *
 500   * finds an entry in the specified directory with the wanted name. It
 501 @@ -105,6 +548,8 @@
 502   * The returned buffer_head has ->b_count elevated.  The caller is expected
 503   * to brelse() it when appropriate.
 504   */
 505 +
 506 +
 507  static struct buffer_head * ext3_find_entry (struct dentry *dentry,
 508                                         struct ext3_dir_entry_2 ** res_dir)
 509  {
 510 @@ -119,10 +564,76 @@
 511         int num = 0;
 512         int nblocks, i, err;
 513         struct inode *dir = dentry->d_parent->d_inode;
 514 +       int namelen;
 515 +       const u8 *name;
 516 +       unsigned blocksize;
 517 +       ext3_dirent *de, *top;
 518
 519         *res_dir = NULL;
 520         sb = dir->i_sb;
 521 +       blocksize = sb->s_blocksize;
 522 +       namelen = dentry->d_name.len;
 523 +       name = dentry->d_name.name;
 524 +       if (namelen > EXT3_NAME_LEN)
 525 +               return NULL;
 526 +       if (ext3_dx && is_dx(dir)) {
 527 +               u32 hash = dx_hash (name, namelen);
 528 +               struct dx_frame frames[2], *frame;
 529 +               if (!(frame = dx_probe (dir, hash, frames)))
 530 +                       return NULL;
 531 +dxnext:
 532 +               block = dx_get_block(frame->at);
 533 +               if (!(bh = ext3_bread (NULL,dir, block, 0, &err)))
 534 +                       goto dxfail;
 535 +               de = (ext3_dirent *) bh->b_data;
 536 +               top = (ext3_dirent *) ((char *) de + blocksize -
 537 +                               EXT3_DIR_REC_LEN(0));
 538 +               for (; de < top; de = ext3_next_entry(de))
 539 +                       if (ext3_match (namelen, name, de)) {
 540 +                               if (!ext3_check_dir_entry("ext3_find_entry",
 541 +                                         dir, de, bh,
 542 +                                         (block<<EXT3_BLOCK_SIZE_BITS(sb))
 543 +                                          +((char *)de - bh->b_data))) {
 544 +                                       brelse (bh);
 545 +                                       goto dxfail;
 546 +                               }
 547 +                               *res_dir = de;
 548 +                               goto dxfound;
 549 +                       }
 550 +               brelse (bh);
 551 +               /* Same hash continues in next block?  Search on. */
 552 +               if (++(frame->at) == frame->entries + dx_get_count(frame->entries))
 553 +               {
 554 +                       struct buffer_head *bh2;
 555 +                       if (frame == frames)
 556 +                               goto dxfail;
 557 +                       if (++(frames->at) == frames->entries + dx_get_count(frames->entries))
 558 +                               goto dxfail;
 559 +                       /* should omit read if not continued */
 560 +                       if (!(bh2 = ext3_bread (NULL, dir,
 561 +                                               dx_get_block(frames->at),
 562 +                                               0, &err)))
 563 +                               goto dxfail;
 564 +                       brelse (frame->bh);
 565 +                       frame->bh = bh2;
 566 +                       frame->at = frame->entries = ((struct dx_node *) bh2->b_data)->entries;
 567 +                       /* Subtle: the 0th entry has the count, find the hash in frame above */
 568 +                       if ((dx_get_hash(frames->at) & -2) == hash)
 569 +                               goto dxnext;
 570 +                       goto dxfail;
 571 +               }
 572 +               if ((dx_get_hash(frame->at) & -2) == hash)
 573 +                       goto dxnext;
 574 +dxfail:
 575 +               dxtrace(printk("%s not found\n", name));
 576 +               dx_release (frames);
 577 +               return NULL;
 578 +dxfound:
 579 +               dx_release (frames);
 580 +               return bh;
 581
 582 +       }
 583 +
 584         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
 585         start = dir->u.ext3_i.i_dir_start_lookup;
 586         if (start >= nblocks)
 587 @@ -237,6 +748,92 @@
 588                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
 589  }
 590
 591 +static ext3_dirent *
 592 +dx_copy_dirents (char *from, char *to, struct dx_map_entry *map, int count)
 593 +{
 594 +       unsigned rec_len = 0;
 595 +
 596 +       while (count--) {
 597 +               ext3_dirent *de = (ext3_dirent *) (from + map->offs);
 598 +               rec_len = EXT3_DIR_REC_LEN(de->name_len);
 599 +               memcpy (to, de, rec_len);
 600 +               ((ext3_dirent *) to)->rec_len = rec_len;
 601 +               to += rec_len;
 602 +               map++;
 603 +       }
 604 +       return (ext3_dirent *) (to - rec_len);
 605 +}
 606 +
 607 +#ifdef CONFIG_EXT3_INDEX
 608 +static ext3_dirent *do_split(handle_t *handle, struct inode *dir,
 609 +                       struct buffer_head **bh,struct dx_frame *frame,
 610 +                       u32 hash, int *error)
 611 +{
 612 +       unsigned blocksize = dir->i_sb->s_blocksize;
 613 +       unsigned count, continued;
 614 +       struct buffer_head *bh2;
 615 +       u32 newblock;
 616 +       unsigned MAX_DX_MAP = PAGE_CACHE_SIZE/EXT3_DIR_REC_LEN(1) + 1;
 617 +       u32 hash2;
 618 +       struct dx_map_entry *map;
 619 +       char *data1 = (*bh)->b_data, *data2, *data3;
 620 +       unsigned split;
 621 +       ext3_dirent *de, *de2;
 622 +
 623 +       bh2 = ext3_append (handle, dir, &newblock, error);
 624 +       if (!(bh2))
 625 +       {
 626 +               brelse(*bh);
 627 +               *bh = NULL;
 628 +               return (ext3_dirent *)bh2;
 629 +       }
 630 +
 631 +       BUFFER_TRACE(*bh, "get_write_access");
 632 +       ext3_journal_get_write_access(handle, *bh);
 633 +       BUFFER_TRACE(frame->bh, "get_write_access");
 634 +       ext3_journal_get_write_access(handle, frame->bh);
 635 +
 636 +       data2 = bh2->b_data;
 637 +
 638 +       map = kmalloc(sizeof(*map) * MAX_DX_MAP, GFP_KERNEL);
 639 +       if (!map)
 640 +               panic("no memory for do_split\n");
 641 +       count = dx_make_map ((ext3_dirent *) data1, blocksize, map);
 642 +       split = count/2; // need to adjust to actual middle
 643 +       dx_sort_map (map, count);
 644 +       hash2 = map[split].hash;
 645 +       continued = hash2 == map[split - 1].hash;
 646 +       dxtrace(printk("Split block %i at %x, %i/%i\n",
 647 +               dx_get_block(frame->at), hash2, split, count-split));
 648 +
 649 +       /* Fancy dance to stay within two buffers */
 650 +       de2 = dx_copy_dirents (data1, data2, map + split, count - split);
 651 +       data3 = (char *) de2 + de2->rec_len;
 652 +       de = dx_copy_dirents (data1, data3, map, split);
 653 +       memcpy(data1, data3, (char *) de + de->rec_len - data3);
 654 +       de = (ext3_dirent *) ((char *) de - data3 + data1); // relocate de
 655 +       de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
 656 +       de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
 657 +       dxtrace(dx_show_leaf ((ext3_dirent *) data1, blocksize, 1));
 658 +       dxtrace(dx_show_leaf ((ext3_dirent *) data2, blocksize, 1));
 659 +
 660 +       /* Which block gets the new entry? */
 661 +       if (hash >= hash2)
 662 +       {
 663 +               swap(*bh, bh2);
 664 +               de = de2;
 665 +       }
 666 +       dx_insert_block (frame, hash2 + continued, newblock);
 667 +       ext3_journal_dirty_metadata (handle, bh2);
 668 +       brelse (bh2);
 669 +       ext3_journal_dirty_metadata (handle, frame->bh);
 670 +       dxtrace(dx_show_index ("frame", frame->entries));
 671 +       kfree(map);
 672 +       return de;
 673 +}
 674 +#endif
 675 +
 676 +
 677  /*
 678   *     ext3_add_entry()
 679   *
 680 @@ -251,6 +844,7 @@
 681  /*
 682   * AKPM: the journalling code here looks wrong on the error paths
 683   */
 684 +
 685  static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
 686         struct inode *inode)
 687  {
 688 @@ -258,117 +852,281 @@
 689         const char *name = dentry->d_name.name;
 690         int namelen = dentry->d_name.len;
 691         unsigned long offset;
 692 -       unsigned short rec_len;
 693         struct buffer_head * bh;
 694 -       struct ext3_dir_entry_2 * de, * de1;
 695 -       struct super_block * sb;
 696 +       ext3_dirent *de;
 697 +       struct super_block * sb = dir->i_sb;
 698         int     retval;
 699 +       unsigned short reclen = EXT3_DIR_REC_LEN(namelen);
 700
 701 -       sb = dir->i_sb;
 702 +       unsigned blocksize = sb->s_blocksize;
 703 +       unsigned nlen, rlen;
 704 +       u32 block, blocks;
 705 +       char *top;
 706
 707         if (!namelen)
 708                 return -EINVAL;
 709 -       bh = ext3_bread (handle, dir, 0, 0, &retval);
 710 -       if (!bh)
 711 -               return retval;
 712 -       rec_len = EXT3_DIR_REC_LEN(namelen);
 713 -       offset = 0;
 714 -       de = (struct ext3_dir_entry_2 *) bh->b_data;
 715 -       while (1) {
 716 -               if ((char *)de >= sb->s_blocksize + bh->b_data) {
 717 -                       brelse (bh);
 718 -                       bh = NULL;
 719 -                       bh = ext3_bread (handle, dir,
 720 -                               offset >> EXT3_BLOCK_SIZE_BITS(sb), 1, &retval);
 721 -                       if (!bh)
 722 -                               return retval;
 723 -                       if (dir->i_size <= offset) {
 724 -                               if (dir->i_size == 0) {
 725 -                                       brelse(bh);
 726 -                                       return -ENOENT;
 727 +       if (ext3_dx && is_dx(dir)) {
 728 +               struct dx_frame frames[2], *frame;
 729 +               struct dx_entry *entries, *at;
 730 +               u32 hash;
 731 +               char *data1;
 732 +
 733 +               hash = dx_hash(name, namelen);
 734 +               /* FIXME: do something if dx_probe() fails here */
 735 +               frame = dx_probe(dir, hash, frames);
 736 +               entries = frame->entries;
 737 +               at = frame->at;
 738 +
 739 +               if (!(bh = ext3_bread(handle,dir, dx_get_block(at), 0,&retval)))
 740 +                       goto dxfail1;
 741 +
 742 +               BUFFER_TRACE(bh, "get_write_access");
 743 +               ext3_journal_get_write_access(handle, bh);
 744 +
 745 +               data1 = bh->b_data;
 746 +               de = (ext3_dirent *) data1;
 747 +               top = data1 + (0? 200: blocksize);
 748 +               while ((char *) de < top)
 749 +               {
 750 +                       /* FIXME: check EEXIST and dir */
 751 +                       nlen = EXT3_DIR_REC_LEN(de->name_len);
 752 +                       rlen = le16_to_cpu(de->rec_len);
 753 +                       if ((de->inode? rlen - nlen: rlen) >= reclen)
 754 +                               goto dx_add;
 755 +                       de = (ext3_dirent *) ((char *) de + rlen);
 756 +               }
 757 +               /* Block full, should compress but for now just split */
 758 +               dxtrace(printk("using %u of %u node entries\n",
 759 +                       dx_get_count(entries), dx_get_limit(entries)));
 760 +               /* Need to split index? */
 761 +               if (dx_get_count(entries) == dx_get_limit(entries))
 762 +               {
 763 +                       u32 newblock;
 764 +                       unsigned icount = dx_get_count(entries);
 765 +                       int levels = frame - frames;
 766 +                       struct dx_entry *entries2;
 767 +                       struct dx_node *node2;
 768 +                       struct buffer_head *bh2;
 769 +                       if (levels && dx_get_count(frames->entries) == dx_get_limit(frames->entries))
 770 +                               goto dxfull;
 771 +                       bh2 = ext3_append (handle, dir, &newblock, &retval);
 772 +                       if (!(bh2))
 773 +                               goto dxfail2;
 774 +                       node2 = (struct dx_node *)(bh2->b_data);
 775 +                       entries2 = node2->entries;
 776 +                       node2->fake.rec_len = cpu_to_le16(blocksize);
 777 +                       node2->fake.inode = 0;
 778 +                       BUFFER_TRACE(frame->bh, "get_write_access");
 779 +                       ext3_journal_get_write_access(handle, frame->bh);
 780 +                       if (levels)
 781 +                       {
 782 +                               unsigned icount1 = icount/2, icount2 = icount - icount1;
 783 +                               unsigned hash2 = dx_get_hash(entries + icount1);
 784 +                               dxtrace(printk("Split index %i/%i\n", icount1, icount2));
 785 +
 786 +                               BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
 787 +                               ext3_journal_get_write_access(handle, frames[0].bh);
 788 +
 789 +                               memcpy ((char *) entries2, (char *) (entries + icount1),
 790 +                                       icount2 * sizeof(struct dx_entry));
 791 +                               dx_set_count (entries, icount1);
 792 +                               dx_set_count (entries2, icount2);
 793 +                               dx_set_limit (entries2, dx_node_limit(dir));
 794 +
 795 +                               /* Which index block gets the new entry? */
 796 +                               if (at - entries >= icount1) {
 797 +                                       frame->at = at = at - entries - icount1 + entries2;
 798 +                                       frame->entries = entries = entries2;
 799 +                                       swap(frame->bh, bh2);
 800                                 }
 801 -
 802 -                               ext3_debug ("creating next block\n");
 803 -
 804 -                               BUFFER_TRACE(bh, "get_write_access");
 805 -                               ext3_journal_get_write_access(handle, bh);
 806 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
 807 -                               de->inode = 0;
 808 -                               de->rec_len = le16_to_cpu(sb->s_blocksize);
 809 -                               dir->u.ext3_i.i_disksize =
 810 -                                       dir->i_size = offset + sb->s_blocksize;
 811 -                               dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
 812 -                               ext3_mark_inode_dirty(handle, dir);
 813 +                               dx_insert_block (frames + 0, hash2, newblock);
 814 +                               dxtrace(dx_show_index ("node", frames[1].entries));
 815 +                               dxtrace(dx_show_index ("node",
 816 +                                       ((struct dx_node *) bh2->b_data)->entries));
 817 +                               ext3_journal_dirty_metadata(handle, bh2);
 818 +                               brelse (bh2);
 819                         } else {
 820 -
 821 -                               ext3_debug ("skipping to next block\n");
 822 -
 823 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
 824 +                               dxtrace(printk("Creating second level index...\n"));
 825 +                               memcpy((char *) entries2, (char *) entries,
 826 +                                       icount * sizeof(struct dx_entry));
 827 +                               dx_set_limit(entries2, dx_node_limit(dir));
 828 +
 829 +                               /* Set up root */
 830 +                               dx_set_count(entries, 1);
 831 +                               dx_set_block(entries + 0, newblock);
 832 +                               ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
 833 +
 834 +                               /* Add new access path frame */
 835 +                               frame = frames + 1;
 836 +                               frame->at = at = at - entries + entries2;
 837 +                               frame->entries = entries = entries2;
 838 +                               frame->bh = bh2;
 839 +                               ext3_journal_get_write_access(handle, frame->bh);
 840                         }
 841 +                       ext3_journal_dirty_metadata(handle, frames[0].bh);
 842                 }
 843 -               if (!ext3_check_dir_entry ("ext3_add_entry", dir, de, bh,
 844 -                                          offset)) {
 845 -                       brelse (bh);
 846 -                       return -ENOENT;
 847 -               }
 848 -               if (ext3_match (namelen, name, de)) {
 849 +               de = do_split(handle, dir, &bh, frame, hash, &retval);
 850 +               dx_release (frames);
 851 +               if (!(de))
 852 +                       goto fail;
 853 +               nlen = EXT3_DIR_REC_LEN(de->name_len);
 854 +               rlen = le16_to_cpu(de->rec_len);
 855 +               goto add;
 856 +
 857 +dx_add:
 858 +               dx_release (frames);
 859 +               goto add;
 860 +
 861 +dxfull:
 862 +               ext3_warning(sb, __FUNCTION__, "Directory index full!\n");
 863 +               retval = -ENOSPC;
 864 +dxfail2:
 865 +               brelse(bh);
 866 +dxfail1:
 867 +               dx_release (frames);
 868 +               goto fail1;
 869 +       }
 870 +
 871 +       blocks = dir->i_size >> sb->s_blocksize_bits;
 872 +       for (block = 0, offset = 0; block < blocks; block++) {
 873 +               bh = ext3_bread(handle, dir, block, 0, &retval);
 874 +               if(!bh)
 875 +                       return retval;
 876 +               de = (ext3_dirent *)bh->b_data;
 877 +               top = bh->b_data + blocksize - reclen;
 878 +               while ((char *) de <= top) {
 879 +                       if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
 880 +                                                 bh, offset)) {
 881 +                               brelse (bh);
 882 +                               return -EIO;
 883 +                       }
 884 +                       if (ext3_match (namelen, name, de)) {
 885                                 brelse (bh);
 886                                 return -EEXIST;
 887 -               }
 888 -               if ((le32_to_cpu(de->inode) == 0 &&
 889 -                               le16_to_cpu(de->rec_len) >= rec_len) ||
 890 -                   (le16_to_cpu(de->rec_len) >=
 891 -                               EXT3_DIR_REC_LEN(de->name_len) + rec_len)) {
 892 -                       BUFFER_TRACE(bh, "get_write_access");
 893 -                       ext3_journal_get_write_access(handle, bh);
 894 -                       /* By now the buffer is marked for journaling */
 895 -                       offset += le16_to_cpu(de->rec_len);
 896 -                       if (le32_to_cpu(de->inode)) {
 897 -                               de1 = (struct ext3_dir_entry_2 *) ((char *) de +
 898 -                                       EXT3_DIR_REC_LEN(de->name_len));
 899 -                               de1->rec_len =
 900 -                                       cpu_to_le16(le16_to_cpu(de->rec_len) -
 901 -                                       EXT3_DIR_REC_LEN(de->name_len));
 902 -                               de->rec_len = cpu_to_le16(
 903 -                                               EXT3_DIR_REC_LEN(de->name_len));
 904 -                               de = de1;
 905                         }
 906 -                       de->file_type = EXT3_FT_UNKNOWN;
 907 -                       if (inode) {
 908 -                               de->inode = cpu_to_le32(inode->i_ino);
 909 -                               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
 910 -                       } else
 911 -                               de->inode = 0;
 912 -                       de->name_len = namelen;
 913 -                       memcpy (de->name, name, namelen);
 914 -                       /*
 915 -                        * XXX shouldn't update any times until successful
 916 -                        * completion of syscall, but too many callers depend
 917 -                        * on this.
 918 -                        *
 919 -                        * XXX similarly, too many callers depend on
 920 -                        * ext3_new_inode() setting the times, but error
 921 -                        * recovery deletes the inode, so the worst that can
 922 -                        * happen is that the times are slightly out of date
 923 -                        * and/or different from the directory change time.
 924 -                        */
 925 -                       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 926 -                       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
 927 -                       dir->i_version = ++event;
 928 -                       ext3_mark_inode_dirty(handle, dir);
 929 -                       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
 930 -                       ext3_journal_dirty_metadata(handle, bh);
 931 +                       nlen = EXT3_DIR_REC_LEN(de->name_len);
 932 +                       rlen = le16_to_cpu(de->rec_len);
 933 +                       if ((de->inode? rlen - nlen: rlen) >= reclen)
 934 +                               goto add;
 935 +                       de = (ext3_dirent *)((char *)de + rlen);
 936 +                       offset += rlen;
 937 +               }
 938 +               if (ext3_dx && blocks == 1 && test_opt(sb, INDEX))
 939 +                       goto dx_make_index;
 940 +               brelse(bh);
 941 +       }
 942 +       bh = ext3_append(handle, dir, &block, &retval);
 943 +       if (!bh)
 944 +               return retval;
 945 +       de = (ext3_dirent *) bh->b_data;
 946 +       de->inode = 0;
 947 +       de->rec_len = cpu_to_le16(rlen = blocksize);
 948 +       nlen = 0;
 949 +       goto add;
 950 +
 951 +add:
 952 +       BUFFER_TRACE(bh, "get_write_access");
 953 +       ext3_journal_get_write_access(handle, bh);
 954 +       /* By now the buffer is marked for journaling */
 955 +       if (de->inode) {
 956 +               ext3_dirent *de1 = (ext3_dirent *)((char *)de + nlen);
 957 +               de1->rec_len = cpu_to_le16(rlen - nlen);
 958 +               de->rec_len = cpu_to_le16(nlen);
 959 +               de = de1;
 960 +       }
 961 +       de->file_type = EXT3_FT_UNKNOWN;
 962 +       if (inode) {
 963 +               de->inode = cpu_to_le32(inode->i_ino);
 964 +               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
 965 +       } else
 966 +               de->inode = 0;
 967 +       de->name_len = namelen;
 968 +       memcpy (de->name, name, namelen);
 969 +       /*
 970 +        * XXX shouldn't update any times until successful
 971 +        * completion of syscall, but too many callers depend
 972 +        * on this.
 973 +        *
 974 +        * XXX similarly, too many callers depend on
 975 +        * ext3_new_inode() setting the times, but error
 976 +        * recovery deletes the inode, so the worst that can
 977 +        * happen is that the times are slightly out of date
 978 +        * and/or different from the directory change time.
 979 +        */
 980 +       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 981 +       ext3_update_dx_flag(dir);
 982 +       dir->i_version = ++event;
 983 +       ext3_mark_inode_dirty(handle, dir);
 984 +       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
 985 +       ext3_journal_dirty_metadata(handle, bh);
 986 +       brelse(bh);
 987 +       return 0;
 988 +
 989 +dx_make_index:
 990 +       {
 991 +               struct buffer_head *bh2;
 992 +               struct dx_root *root;
 993 +               struct dx_frame frames[2], *frame;
 994 +               struct dx_entry *entries;
 995 +               ext3_dirent *de2;
 996 +               char *data1;
 997 +               unsigned len;
 998 +               u32 hash;
 999 +
1000 +               dxtrace(printk("Creating index\n"));
1001 +               ext3_journal_get_write_access(handle, bh);
1002 +               root = (struct dx_root *) bh->b_data;
1003 +
1004 +               EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1005 +               bh2 = ext3_append (handle, dir, &block, &retval);
1006 +               if (!(bh2))
1007 +               {
1008                         brelse(bh);
1009 -                       return 0;
1010 +                       return retval;
1011                 }
1012 -               offset += le16_to_cpu(de->rec_len);
1013 -               de = (struct ext3_dir_entry_2 *)
1014 -                       ((char *) de + le16_to_cpu(de->rec_len));
1015 +               data1 = bh2->b_data;
1016 +
1017 +               /* The 0th block becomes the root, move the dirents out */
1018 +               de = (ext3_dirent *) &root->info;
1019 +               len = ((char *) root) + blocksize - (char *) de;
1020 +               memcpy (data1, de, len);
1021 +               de = (ext3_dirent *) data1;
1022 +               top = data1 + len;
1023 +               while (((char *) de2=(char*)de+le16_to_cpu(de->rec_len)) < top)
1024 +                       de = de2;
1025 +               de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1026 +               /* Initialize the root; the dot dirents already exist */
1027 +               de = (ext3_dirent *) (&root->dotdot);
1028 +               de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1029 +               memset (&root->info, 0, sizeof(root->info));
1030 +               root->info.info_length = sizeof(root->info);
1031 +               entries = root->entries;
1032 +               dx_set_block (entries, 1);
1033 +               dx_set_count (entries, 1);
1034 +               dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1035 +
1036 +               /* Initialize as for dx_probe */
1037 +               hash = dx_hash (name, namelen);
1038 +               frame = frames;
1039 +               frame->entries = entries;
1040 +               frame->at = entries;
1041 +               frame->bh = bh;
1042 +               bh = bh2;
1043 +               de = do_split(handle,dir, &bh, frame, hash, &retval);
1044 +               dx_release (frames);
1045 +               if (!(de))
1046 +                       return retval;
1047 +               nlen = EXT3_DIR_REC_LEN(de->name_len);
1048 +               rlen = le16_to_cpu(de->rec_len);
1049 +               goto add;
1050         }
1051 -       brelse (bh);
1052 -       return -ENOSPC;
1053 +fail1:
1054 +       return retval;
1055 +fail:
1056 +       return -ENOENT;
1057  }
1058
1059 +
1060  /*
1061   * ext3_delete_entry deletes a directory entry by merging it with the
1062   * previous entry
1063 @@ -451,7 +1212,8 @@
1064         struct inode * inode;
1065         int err;
1066
1067 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1068 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1069 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1070         if (IS_ERR(handle))
1071                 return PTR_ERR(handle);
1072
1073 @@ -478,7 +1240,8 @@
1074         struct inode *inode;
1075         int err;
1076
1077 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1078 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1079 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1080         if (IS_ERR(handle))
1081                 return PTR_ERR(handle);
1082
1083 @@ -507,7 +1270,8 @@
1084         if (dir->i_nlink >= EXT3_LINK_MAX)
1085                 return -EMLINK;
1086
1087 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
1088 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1089 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
1090         if (IS_ERR(handle))
1091                 return PTR_ERR(handle);
1092
1093 @@ -550,7 +1320,7 @@
1094         if (err)
1095                 goto out_no_entry;
1096         dir->i_nlink++;
1097 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1098 +       ext3_update_dx_flag(dir);
1099         ext3_mark_inode_dirty(handle, dir);
1100         d_instantiate(dentry, inode);
1101  out_stop:
1102 @@ -832,7 +1596,7 @@
1103         ext3_mark_inode_dirty(handle, inode);
1104         dir->i_nlink--;
1105         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1106 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1107 +       ext3_update_dx_flag(dir);
1108         ext3_mark_inode_dirty(handle, dir);
1109
1110  end_rmdir:
1111 @@ -878,7 +1642,7 @@
1112         if (retval)
1113                 goto end_unlink;
1114         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1115 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1116 +       ext3_update_dx_flag(dir);
1117         ext3_mark_inode_dirty(handle, dir);
1118         inode->i_nlink--;
1119         if (!inode->i_nlink)
1120 @@ -904,7 +1668,8 @@
1121         if (l > dir->i_sb->s_blocksize)
1122                 return -ENAMETOOLONG;
1123
1124 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 5);
1125 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1126 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5);
1127         if (IS_ERR(handle))
1128                 return PTR_ERR(handle);
1129
1130 @@ -959,7 +1724,8 @@
1131         if (inode->i_nlink >= EXT3_LINK_MAX)
1132                 return -EMLINK;
1133
1134 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS);
1135 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
1136 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS);
1137         if (IS_ERR(handle))
1138                 return PTR_ERR(handle);
1139
1140 @@ -995,7 +1761,8 @@
1141
1142         old_bh = new_bh = dir_bh = NULL;
1143
1144 -       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS + 2);
1145 +       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
1146 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
1147         if (IS_ERR(handle))
1148                 return PTR_ERR(handle);
1149
1150 @@ -1077,7 +1844,7 @@
1151                 new_inode->i_ctime = CURRENT_TIME;
1152         }
1153         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1154 -       old_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1155 +       ext3_update_dx_flag(old_dir);
1156         if (dir_bh) {
1157                 BUFFER_TRACE(dir_bh, "get_write_access");
1158                 ext3_journal_get_write_access(handle, dir_bh);
1159 @@ -1089,7 +1856,7 @@
1160                         new_inode->i_nlink--;
1161                 } else {
1162                         new_dir->i_nlink++;
1163 -                       new_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1164 +                       ext3_update_dx_flag(new_dir);
1165                         ext3_mark_inode_dirty(handle, new_dir);
1166                 }
1167         }
1168 --- ./include/linux/ext3_fs.h   2002/03/05 06:18:59     2.1
1169 +++ ./include/linux/ext3_fs.h   2002/03/05 06:26:56
1170 @@ -339,6 +339,7 @@
1171    #define EXT3_MOUNT_WRITEBACK_DATA    0x0C00  /* No data ordering */
1172  #define EXT3_MOUNT_UPDATE_JOURNAL      0x1000  /* Update the journal format */
1173  #define EXT3_MOUNT_NO_UID32            0x2000  /* Disable 32-bit UIDs */
1174 +#define EXT3_MOUNT_INDEX               0x4000  /* Enable directory index */
1175
1176  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
1177  #ifndef _LINUX_EXT2_FS_H
1178 @@ -575,6 +576,24 @@
1179  #define EXT3_DIR_ROUND                 (EXT3_DIR_PAD - 1)
1180  #define EXT3_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT3_DIR_ROUND) & \
1181                                          ~EXT3_DIR_ROUND)
1182 +/*
1183 + * Hash Tree Directory indexing
1184 + * (c) Daniel Phillips, 2001
1185 + */
1186 +
1187 +#define CONFIG_EXT3_INDEX
1188 +
1189 +#ifdef CONFIG_EXT3_INDEX
1190 +  enum {ext3_dx = 1};
1191 +  #define is_dx(dir) (EXT3_I(dir)->i_flags & EXT3_INDEX_FL)
1192 +#define EXT3_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT3_LINK_MAX)
1193 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
1194 +#else
1195 +  enum {ext3_dx = 0};
1196 +  #define is_dx(dir) 0
1197 +#define EXT3_DIR_LINK_MAX(dir) ((dir)->i_nlink >= EXT3_LINK_MAX)
1198 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2)
1199 +#endif
1200
1201  #ifdef __KERNEL__
1202  /*
1203 --- ./include/linux/ext3_jbd.h  2002/03/05 06:18:59     2.1
1204 +++ ./include/linux/ext3_jbd.h  2002/03/05 06:33:54
1205 @@ -63,6 +63,8 @@
1206
1207  #define EXT3_RESERVE_TRANS_BLOCKS      12
1208
1209 +#define EXT3_INDEX_EXTRA_TRANS_BLOCKS  8
1210 +
1211  int
1212  ext3_mark_iloc_dirty(handle_t *handle,
1213                      struct inode *inode,