lustre/kernel_patches/patches/ext-2.4-patch-1-suse-2.4.19.patch

   1  fs/ext3/Makefile           |    2
   2  fs/ext3/dir.c              |  299 +++++++++
   3  fs/ext3/file.c             |    3
   4  fs/ext3/hash.c             |  215 ++++++
   5  fs/ext3/namei.c            | 1388 ++++++++++++++++++++++++++++++++++++++++-----
   6  fs/ext3/super.c            |    7
   7  include/linux/ext3_fs.h    |   85 ++
   8  include/linux/ext3_fs_sb.h |    2
   9  include/linux/ext3_jbd.h   |    2
  10  include/linux/rbtree.h     |    2
  11  lib/rbtree.c               |   42 +
  12  11 files changed, 1887 insertions(+), 160 deletions(-)
  13
  14 Index: linux-2.4.19/fs/ext3/Makefile
  15 ===================================================================
  16 --- linux-2.4.19.orig/fs/ext3/Makefile  2004-04-23 18:25:03.000000000 -0400
  17 +++ linux-2.4.19/fs/ext3/Makefile       2004-04-23 18:26:27.000000000 -0400
  18 @@ -12,7 +12,7 @@
  19  export-objs := super.o inode.o
  20
  21  obj-y    := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
  22 -               ioctl.o namei.o super.o symlink.o
  23 +               ioctl.o namei.o super.o symlink.o hash.o
  24  obj-m    := $(O_TARGET)
  25
  26  obj-$(CONFIG_EXT3_FS_XATTR) += xattr.o
  27 Index: linux-2.4.19/fs/ext3/dir.c
  28 ===================================================================
  29 --- linux-2.4.19.orig/fs/ext3/dir.c     2001-11-09 17:25:04.000000000 -0500
  30 +++ linux-2.4.19/fs/ext3/dir.c  2004-04-23 18:26:27.000000000 -0400
  31 @@ -21,12 +21,16 @@
  32  #include <linux/fs.h>
  33  #include <linux/jbd.h>
  34  #include <linux/ext3_fs.h>
  35 +#include <linux/slab.h>
  36 +#include <linux/rbtree.h>
  37
  38  static unsigned char ext3_filetype_table[] = {
  39         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  40  };
  41
  42  static int ext3_readdir(struct file *, void *, filldir_t);
  43 +static int ext3_dx_readdir(struct file * filp,
  44 +                          void * dirent, filldir_t filldir);
  45
  46  struct file_operations ext3_dir_operations = {
  47         read:           generic_read_dir,
  48 @@ -35,6 +39,17 @@
  49         fsync:          ext3_sync_file,         /* BKL held */
  50  };
  51
  52 +
  53 +static unsigned char get_dtype(struct super_block *sb, int filetype)
  54 +{
  55 +       if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
  56 +           (filetype >= EXT3_FT_MAX))
  57 +               return DT_UNKNOWN;
  58 +
  59 +       return (ext3_filetype_table[filetype]);
  60 +}
  61 +
  62 +
  63  int ext3_check_dir_entry (const char * function, struct inode * dir,
  64                           struct ext3_dir_entry_2 * de,
  65                           struct buffer_head * bh,
  66 @@ -79,6 +94,16 @@
  67
  68         sb = inode->i_sb;
  69
  70 +       if (is_dx(inode)) {
  71 +               err = ext3_dx_readdir(filp, dirent, filldir);
  72 +               if (err != ERR_BAD_DX_DIR)
  73 +                       return err;
  74 +               /*
  75 +                * We don't set the inode dirty flag since it's not
  76 +                * critical that it get flushed back to the disk.
  77 +                */
  78 +               EXT3_I(filp->f_dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
  79 +       }
  80         stored = 0;
  81         bh = NULL;
  82         offset = filp->f_pos & (sb->s_blocksize - 1);
  83 @@ -162,18 +187,12 @@
  84                                  * during the copy operation.
  85                                  */
  86                                 unsigned long version = filp->f_version;
  87 -                               unsigned char d_type = DT_UNKNOWN;
  88
  89 -                               if (EXT3_HAS_INCOMPAT_FEATURE(sb,
  90 -                                               EXT3_FEATURE_INCOMPAT_FILETYPE)
  91 -                                               && de->file_type < EXT3_FT_MAX)
  92 -                                       d_type =
  93 -                                         ext3_filetype_table[de->file_type];
  94                                 error = filldir(dirent, de->name,
  95                                                 de->name_len,
  96                                                 filp->f_pos,
  97                                                 le32_to_cpu(de->inode),
  98 -                                               d_type);
  99 +                                               get_dtype(sb, de->file_type));
 100                                 if (error)
 101                                         break;
 102                                 if (version != filp->f_version)
 103 @@ -188,3 +207,269 @@
 104         UPDATE_ATIME(inode);
 105         return 0;
 106  }
 107 +
 108 +#ifdef CONFIG_EXT3_INDEX
 109 +/*
 110 + * These functions convert from the major/minor hash to an f_pos
 111 + * value.
 112 + *
 113 + * Currently we only use major hash numer.  This is unfortunate, but
 114 + * on 32-bit machines, the same VFS interface is used for lseek and
 115 + * llseek, so if we use the 64 bit offset, then the 32-bit versions of
 116 + * lseek/telldir/seekdir will blow out spectacularly, and from within
 117 + * the ext2 low-level routine, we don't know if we're being called by
 118 + * a 64-bit version of the system call or the 32-bit version of the
 119 + * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
 120 + * cookie.  Sigh.
 121 + */
 122 +#define hash2pos(major, minor) (major >> 1)
 123 +#define pos2maj_hash(pos)      ((pos << 1) & 0xffffffff)
 124 +#define pos2min_hash(pos)      (0)
 125 +
 126 +/*
 127 + * This structure holds the nodes of the red-black tree used to store
 128 + * the directory entry in hash order.
 129 + */
 130 +struct fname {
 131 +       __u32           hash;
 132 +       __u32           minor_hash;
 133 +       rb_node_t       rb_hash;
 134 +       struct fname    *next;
 135 +       __u32           inode;
 136 +       __u8            name_len;
 137 +       __u8            file_type;
 138 +       char            name[0];
 139 +};
 140 +
 141 +/*
 142 + * This functoin implements a non-recursive way of freeing all of the
 143 + * nodes in the red-black tree.
 144 + */
 145 +static void free_rb_tree_fname(rb_root_t *root)
 146 +{
 147 +       rb_node_t       *n = root->rb_node;
 148 +       rb_node_t       *parent;
 149 +       struct fname    *fname;
 150 +
 151 +       while (n) {
 152 +               /* Do the node's children first */
 153 +               if ((n)->rb_left) {
 154 +                       n = n->rb_left;
 155 +                       continue;
 156 +               }
 157 +               if (n->rb_right) {
 158 +                       n = n->rb_right;
 159 +                       continue;
 160 +               }
 161 +               /*
 162 +                * The node has no children; free it, and then zero
 163 +                * out parent's link to it.  Finally go to the
 164 +                * beginning of the loop and try to free the parent
 165 +                * node.
 166 +                */
 167 +               parent = n->rb_parent;
 168 +               fname = rb_entry(n, struct fname, rb_hash);
 169 +               kfree(fname);
 170 +               if (!parent)
 171 +                       root->rb_node = 0;
 172 +               else if (parent->rb_left == n)
 173 +                       parent->rb_left = 0;
 174 +               else if (parent->rb_right == n)
 175 +                       parent->rb_right = 0;
 176 +               n = parent;
 177 +       }
 178 +       root->rb_node = 0;
 179 +}
 180 +
 181 +
 182 +struct dir_private_info *create_dir_info(loff_t pos)
 183 +{
 184 +       struct dir_private_info *p;
 185 +
 186 +       p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
 187 +       if (!p)
 188 +               return NULL;
 189 +       p->root.rb_node = 0;
 190 +       p->curr_node = 0;
 191 +       p->extra_fname = 0;
 192 +       p->last_pos = 0;
 193 +       p->curr_hash = pos2maj_hash(pos);
 194 +       p->curr_minor_hash = pos2min_hash(pos);
 195 +       p->next_hash = 0;
 196 +       return p;
 197 +}
 198 +
 199 +void ext3_htree_free_dir_info(struct dir_private_info *p)
 200 +{
 201 +       free_rb_tree_fname(&p->root);
 202 +       kfree(p);
 203 +}
 204 +
 205 +/*
 206 + * Given a directory entry, enter it into the fname rb tree.
 207 + */
 208 +void ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
 209 +                            __u32 minor_hash,
 210 +                            struct ext3_dir_entry_2 *dirent)
 211 +{
 212 +       rb_node_t **p, *parent = NULL;
 213 +       struct fname * fname, *new_fn;
 214 +       struct dir_private_info *info;
 215 +       int len;
 216 +
 217 +       info = (struct dir_private_info *) dir_file->private_data;
 218 +       p = &info->root.rb_node;
 219 +
 220 +       /* Create and allocate the fname structure */
 221 +       len = sizeof(struct fname) + dirent->name_len + 1;
 222 +       new_fn = kmalloc(len, GFP_KERNEL);
 223 +       memset(new_fn, 0, len);
 224 +       new_fn->hash = hash;
 225 +       new_fn->minor_hash = minor_hash;
 226 +       new_fn->inode = le32_to_cpu(dirent->inode);
 227 +       new_fn->name_len = dirent->name_len;
 228 +       new_fn->file_type = dirent->file_type;
 229 +       memcpy(new_fn->name, dirent->name, dirent->name_len);
 230 +       new_fn->name[dirent->name_len] = 0;
 231 +
 232 +       while (*p) {
 233 +               parent = *p;
 234 +               fname = rb_entry(parent, struct fname, rb_hash);
 235 +
 236 +               /*
 237 +                * If the hash and minor hash match up, then we put
 238 +                * them on a linked list.  This rarely happens...
 239 +                */
 240 +               if ((new_fn->hash == fname->hash) &&
 241 +                   (new_fn->minor_hash == fname->minor_hash)) {
 242 +                       new_fn->next = fname->next;
 243 +                       fname->next = new_fn;
 244 +                       return;
 245 +               }
 246 +
 247 +               if (new_fn->hash < fname->hash)
 248 +                       p = &(*p)->rb_left;
 249 +               else if (new_fn->hash > fname->hash)
 250 +                       p = &(*p)->rb_right;
 251 +               else if (new_fn->minor_hash < fname->minor_hash)
 252 +                       p = &(*p)->rb_left;
 253 +               else /* if (new_fn->minor_hash > fname->minor_hash) */
 254 +                       p = &(*p)->rb_right;
 255 +       }
 256 +
 257 +       rb_link_node(&new_fn->rb_hash, parent, p);
 258 +       rb_insert_color(&new_fn->rb_hash, &info->root);
 259 +}
 260 +
 261 +
 262 +
 263 +/*
 264 + * This is a helper function for ext3_dx_readdir.  It calls filldir
 265 + * for all entres on the fname linked list.  (Normally there is only
 266 + * one entry on the linked list, unless there are 62 bit hash collisions.)
 267 + */
 268 +static int call_filldir(struct file * filp, void * dirent,
 269 +                       filldir_t filldir, struct fname *fname)
 270 +{
 271 +       struct dir_private_info *info = filp->private_data;
 272 +       loff_t  curr_pos;
 273 +       struct inode *inode = filp->f_dentry->d_inode;
 274 +       struct super_block * sb;
 275 +       int error;
 276 +
 277 +       sb = inode->i_sb;
 278 +
 279 +       if (!fname) {
 280 +               printk("call_filldir: called with null fname?!?\n");
 281 +               return 0;
 282 +       }
 283 +       curr_pos = hash2pos(fname->hash, fname->minor_hash);
 284 +       while (fname) {
 285 +               error = filldir(dirent, fname->name,
 286 +                               fname->name_len, curr_pos,
 287 +                               fname->inode,
 288 +                               get_dtype(sb, fname->file_type));
 289 +               if (error) {
 290 +                       filp->f_pos = curr_pos;
 291 +                       info->extra_fname = fname->next;
 292 +                       return error;
 293 +               }
 294 +               fname = fname->next;
 295 +       }
 296 +       return 0;
 297 +}
 298 +
 299 +static int ext3_dx_readdir(struct file * filp,
 300 +                        void * dirent, filldir_t filldir)
 301 +{
 302 +       struct dir_private_info *info = filp->private_data;
 303 +       struct inode *inode = filp->f_dentry->d_inode;
 304 +       struct fname *fname;
 305 +       int     ret;
 306 +
 307 +       if (!info) {
 308 +               info = create_dir_info(filp->f_pos);
 309 +               if (!info)
 310 +                       return -ENOMEM;
 311 +               filp->private_data = info;
 312 +       }
 313 +
 314 +       /* Some one has messed with f_pos; reset the world */
 315 +       if (info->last_pos != filp->f_pos) {
 316 +               free_rb_tree_fname(&info->root);
 317 +               info->curr_node = 0;
 318 +               info->extra_fname = 0;
 319 +               info->curr_hash = pos2maj_hash(filp->f_pos);
 320 +               info->curr_minor_hash = pos2min_hash(filp->f_pos);
 321 +       }
 322 +
 323 +       /*
 324 +        * If there are any leftover names on the hash collision
 325 +        * chain, return them first.
 326 +        */
 327 +       if (info->extra_fname &&
 328 +           call_filldir(filp, dirent, filldir, info->extra_fname))
 329 +               goto finished;
 330 +
 331 +       if (!info->curr_node)
 332 +               info->curr_node = rb_get_first(&info->root);
 333 +
 334 +       while (1) {
 335 +               /*
 336 +                * Fill the rbtree if we have no more entries,
 337 +                * or the inode has changed since we last read in the
 338 +                * cached entries.
 339 +                */
 340 +               if ((!info->curr_node) ||
 341 +                   (filp->f_version != inode->i_version)) {
 342 +                       info->curr_node = 0;
 343 +                       free_rb_tree_fname(&info->root);
 344 +                       filp->f_version = inode->i_version;
 345 +                       ret = ext3_htree_fill_tree(filp, info->curr_hash,
 346 +                                                  info->curr_minor_hash,
 347 +                                                  &info->next_hash);
 348 +                       if (ret < 0)
 349 +                               return ret;
 350 +                       if (ret == 0)
 351 +                               break;
 352 +                       info->curr_node = rb_get_first(&info->root);
 353 +               }
 354 +
 355 +               fname = rb_entry(info->curr_node, struct fname, rb_hash);
 356 +               info->curr_hash = fname->hash;
 357 +               info->curr_minor_hash = fname->minor_hash;
 358 +               if (call_filldir(filp, dirent, filldir, fname))
 359 +                       break;
 360 +
 361 +               info->curr_node = rb_get_next(info->curr_node);
 362 +               if (!info->curr_node) {
 363 +                       info->curr_hash = info->next_hash;
 364 +                       info->curr_minor_hash = 0;
 365 +               }
 366 +       }
 367 +finished:
 368 +       info->last_pos = filp->f_pos;
 369 +       UPDATE_ATIME(inode);
 370 +       return 0;
 371 +}
 372 +#endif
 373 Index: linux-2.4.19/fs/ext3/file.c
 374 ===================================================================
 375 --- linux-2.4.19.orig/fs/ext3/file.c    2004-04-23 17:54:02.000000000 -0400
 376 +++ linux-2.4.19/fs/ext3/file.c 2004-04-23 18:26:27.000000000 -0400
 377 @@ -38,6 +38,9 @@
 378  {
 379         if (filp->f_mode & FMODE_WRITE)
 380                 ext3_discard_prealloc (inode);
 381 +       if (is_dx(inode) && filp->private_data)
 382 +               ext3_htree_free_dir_info(filp->private_data);
 383 +
 384         return 0;
 385  }
 386
 387 Index: linux-2.4.19/fs/ext3/hash.c
 388 ===================================================================
 389 --- linux-2.4.19.orig/fs/ext3/hash.c    2003-01-30 05:24:37.000000000 -0500
 390 +++ linux-2.4.19/fs/ext3/hash.c 2004-04-23 18:26:27.000000000 -0400
 391 @@ -0,0 +1,215 @@
 392 +/*
 393 + *  linux/fs/ext3/hash.c
 394 + *
 395 + * Copyright (C) 2002 by Theodore Ts'o
 396 + *
 397 + * This file is released under the GPL v2.
 398 + *
 399 + * This file may be redistributed under the terms of the GNU Public
 400 + * License.
 401 + */
 402 +
 403 +#include <linux/fs.h>
 404 +#include <linux/jbd.h>
 405 +#include <linux/sched.h>
 406 +#include <linux/ext3_fs.h>
 407 +
 408 +#define DELTA 0x9E3779B9
 409 +
 410 +static void TEA_transform(__u32 buf[4], __u32 const in[])
 411 +{
 412 +       __u32   sum = 0;
 413 +       __u32   b0 = buf[0], b1 = buf[1];
 414 +       __u32   a = in[0], b = in[1], c = in[2], d = in[3];
 415 +       int     n = 16;
 416 +
 417 +       do {
 418 +               sum += DELTA;
 419 +               b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
 420 +               b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
 421 +       } while(--n);
 422 +
 423 +       buf[0] += b0;
 424 +       buf[1] += b1;
 425 +}
 426 +
 427 +/* F, G and H are basic MD4 functions: selection, majority, parity */
 428 +#define F(x, y, z) ((z) ^ ((x) & ((y) ^ (z))))
 429 +#define G(x, y, z) (((x) & (y)) + (((x) ^ (y)) & (z)))
 430 +#define H(x, y, z) ((x) ^ (y) ^ (z))
 431 +
 432 +/*
 433 + * The generic round function.  The application is so specific that
 434 + * we don't bother protecting all the arguments with parens, as is generally
 435 + * good macro practice, in favor of extra legibility.
 436 + * Rotation is separate from addition to prevent recomputation
 437 + */
 438 +#define ROUND(f, a, b, c, d, x, s)     \
 439 +       (a += f(b, c, d) + x, a = (a << s) | (a >> (32-s)))
 440 +#define K1 0
 441 +#define K2 013240474631UL
 442 +#define K3 015666365641UL
 443 +
 444 +/*
 445 + * Basic cut-down MD4 transform.  Returns only 32 bits of result.
 446 + */
 447 +static void halfMD4Transform (__u32 buf[4], __u32 const in[])
 448 +{
 449 +       __u32   a = buf[0], b = buf[1], c = buf[2], d = buf[3];
 450 +
 451 +       /* Round 1 */
 452 +       ROUND(F, a, b, c, d, in[0] + K1,  3);
 453 +       ROUND(F, d, a, b, c, in[1] + K1,  7);
 454 +       ROUND(F, c, d, a, b, in[2] + K1, 11);
 455 +       ROUND(F, b, c, d, a, in[3] + K1, 19);
 456 +       ROUND(F, a, b, c, d, in[4] + K1,  3);
 457 +       ROUND(F, d, a, b, c, in[5] + K1,  7);
 458 +       ROUND(F, c, d, a, b, in[6] + K1, 11);
 459 +       ROUND(F, b, c, d, a, in[7] + K1, 19);
 460 +
 461 +       /* Round 2 */
 462 +       ROUND(G, a, b, c, d, in[1] + K2,  3);
 463 +       ROUND(G, d, a, b, c, in[3] + K2,  5);
 464 +       ROUND(G, c, d, a, b, in[5] + K2,  9);
 465 +       ROUND(G, b, c, d, a, in[7] + K2, 13);
 466 +       ROUND(G, a, b, c, d, in[0] + K2,  3);
 467 +       ROUND(G, d, a, b, c, in[2] + K2,  5);
 468 +       ROUND(G, c, d, a, b, in[4] + K2,  9);
 469 +       ROUND(G, b, c, d, a, in[6] + K2, 13);
 470 +
 471 +       /* Round 3 */
 472 +       ROUND(H, a, b, c, d, in[3] + K3,  3);
 473 +       ROUND(H, d, a, b, c, in[7] + K3,  9);
 474 +       ROUND(H, c, d, a, b, in[2] + K3, 11);
 475 +       ROUND(H, b, c, d, a, in[6] + K3, 15);
 476 +       ROUND(H, a, b, c, d, in[1] + K3,  3);
 477 +       ROUND(H, d, a, b, c, in[5] + K3,  9);
 478 +       ROUND(H, c, d, a, b, in[0] + K3, 11);
 479 +       ROUND(H, b, c, d, a, in[4] + K3, 15);
 480 +
 481 +       buf[0] += a;
 482 +       buf[1] += b;
 483 +       buf[2] += c;
 484 +       buf[3] += d;
 485 +}
 486 +
 487 +#undef ROUND
 488 +#undef F
 489 +#undef G
 490 +#undef H
 491 +#undef K1
 492 +#undef K2
 493 +#undef K3
 494 +
 495 +/* The old legacy hash */
 496 +static __u32 dx_hack_hash (const char *name, int len)
 497 +{
 498 +       __u32 hash0 = 0x12a3fe2d, hash1 = 0x37abe8f9;
 499 +       while (len--) {
 500 +               __u32 hash = hash1 + (hash0 ^ (*name++ * 7152373));
 501 +
 502 +               if (hash & 0x80000000) hash -= 0x7fffffff;
 503 +               hash1 = hash0;
 504 +               hash0 = hash;
 505 +       }
 506 +       return (hash0 << 1);
 507 +}
 508 +
 509 +static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
 510 +{
 511 +       __u32   pad, val;
 512 +       int     i;
 513 +
 514 +       pad = (__u32)len | ((__u32)len << 8);
 515 +       pad |= pad << 16;
 516 +
 517 +       val = pad;
 518 +       if (len > num*4)
 519 +               len = num * 4;
 520 +       for (i=0; i < len; i++) {
 521 +               if ((i % 4) == 0)
 522 +                       val = pad;
 523 +               val = msg[i] + (val << 8);
 524 +               if ((i % 4) == 3) {
 525 +                       *buf++ = val;
 526 +                       val = pad;
 527 +                       num--;
 528 +               }
 529 +       }
 530 +       if (--num >= 0)
 531 +               *buf++ = val;
 532 +       while (--num >= 0)
 533 +               *buf++ = pad;
 534 +}
 535 +
 536 +/*
 537 + * Returns the hash of a filename.  If len is 0 and name is NULL, then
 538 + * this function can be used to test whether or not a hash version is
 539 + * supported.
 540 + *
 541 + * The seed is an 4 longword (32 bits) "secret" which can be used to
 542 + * uniquify a hash.  If the seed is all zero's, then some default seed
 543 + * may be used.
 544 + *
 545 + * A particular hash version specifies whether or not the seed is
 546 + * represented, and whether or not the returned hash is 32 bits or 64
 547 + * bits.  32 bit hashes will return 0 for the minor hash.
 548 + */
 549 +int ext3fs_dirhash(const char *name, int len, struct dx_hash_info *hinfo)
 550 +{
 551 +       __u32   hash;
 552 +       __u32   minor_hash = 0;
 553 +       const char      *p;
 554 +       int             i;
 555 +       __u32           in[8], buf[4];
 556 +
 557 +       /* Initialize the default seed for the hash checksum functions */
 558 +       buf[0] = 0x67452301;
 559 +       buf[1] = 0xefcdab89;
 560 +       buf[2] = 0x98badcfe;
 561 +       buf[3] = 0x10325476;
 562 +
 563 +       /* Check to see if the seed is all zero's */
 564 +       if (hinfo->seed) {
 565 +               for (i=0; i < 4; i++) {
 566 +                       if (hinfo->seed[i])
 567 +                               break;
 568 +               }
 569 +               if (i < 4)
 570 +                       memcpy(buf, hinfo->seed, sizeof(buf));
 571 +       }
 572 +
 573 +       switch (hinfo->hash_version) {
 574 +       case DX_HASH_LEGACY:
 575 +               hash = dx_hack_hash(name, len);
 576 +               break;
 577 +       case DX_HASH_HALF_MD4:
 578 +               p = name;
 579 +               while (len > 0) {
 580 +                       str2hashbuf(p, len, in, 8);
 581 +                       halfMD4Transform(buf, in);
 582 +                       len -= 32;
 583 +                       p += 32;
 584 +               }
 585 +               minor_hash = buf[2];
 586 +               hash = buf[1];
 587 +               break;
 588 +       case DX_HASH_TEA:
 589 +               p = name;
 590 +               while (len > 0) {
 591 +                       str2hashbuf(p, len, in, 4);
 592 +                       TEA_transform(buf, in);
 593 +                       len -= 16;
 594 +                       p += 16;
 595 +               }
 596 +               hash = buf[0];
 597 +               minor_hash = buf[1];
 598 +               break;
 599 +       default:
 600 +               hinfo->hash = 0;
 601 +               return -1;
 602 +       }
 603 +       hinfo->hash = hash & ~1;
 604 +       hinfo->minor_hash = minor_hash;
 605 +       return 0;
 606 +}
 607 Index: linux-2.4.19/fs/ext3/namei.c
 608 ===================================================================
 609 --- linux-2.4.19.orig/fs/ext3/namei.c   2004-04-23 17:53:55.000000000 -0400
 610 +++ linux-2.4.19/fs/ext3/namei.c        2004-04-23 22:24:05.000000000 -0400
 611 @@ -16,6 +16,12 @@
 612   *        David S. Miller (davem@caip.rutgers.edu), 1995
 613   *  Directory entry file type support and forward compatibility hooks
 614   *     for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
 615 + *  Hash Tree Directory indexing (c)
 616 + *     Daniel Phillips, 2001
 617 + *  Hash Tree Directory indexing porting
 618 + *     Christopher Li, 2002
 619 + *  Hash Tree Directory indexing cleanup
 620 + *     Theodore Ts'o, 2002
 621   */
 622
 623  #include <linux/fs.h>
 624 @@ -40,6 +46,630 @@
 625  #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
 626  #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
 627
 628 +static struct buffer_head *ext3_append(handle_t *handle,
 629 +                                       struct inode *inode,
 630 +                                       u32 *block, int *err)
 631 +{
 632 +       struct buffer_head *bh;
 633 +
 634 +       *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
 635 +
 636 +       if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
 637 +               inode->i_size += inode->i_sb->s_blocksize;
 638 +               EXT3_I(inode)->i_disksize = inode->i_size;
 639 +               ext3_journal_get_write_access(handle,bh);
 640 +       }
 641 +       return bh;
 642 +}
 643 +
 644 +#ifndef assert
 645 +#define assert(test) J_ASSERT(test)
 646 +#endif
 647 +
 648 +#ifndef swap
 649 +#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
 650 +#endif
 651 +
 652 +typedef struct { u32 v; } le_u32;
 653 +typedef struct { u16 v; } le_u16;
 654 +
 655 +#ifdef DX_DEBUG
 656 +#define dxtrace(command) command
 657 +#else
 658 +#define dxtrace(command)
 659 +#endif
 660 +
 661 +struct fake_dirent
 662 +{
 663 +       /*le*/u32 inode;
 664 +       /*le*/u16 rec_len;
 665 +       u8 name_len;
 666 +       u8 file_type;
 667 +};
 668 +
 669 +struct dx_countlimit
 670 +{
 671 +       le_u16 limit;
 672 +       le_u16 count;
 673 +};
 674 +
 675 +struct dx_entry
 676 +{
 677 +       le_u32 hash;
 678 +       le_u32 block;
 679 +};
 680 +
 681 +/*
 682 + * dx_root_info is laid out so that if it should somehow get overlaid by a
 683 + * dirent the two low bits of the hash version will be zero.  Therefore, the
 684 + * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 685 + */
 686 +
 687 +struct dx_root
 688 +{
 689 +       struct fake_dirent dot;
 690 +       char dot_name[4];
 691 +       struct fake_dirent dotdot;
 692 +       char dotdot_name[4];
 693 +       struct dx_root_info
 694 +       {
 695 +               le_u32 reserved_zero;
 696 +               u8 hash_version;
 697 +               u8 info_length; /* 8 */
 698 +               u8 indirect_levels;
 699 +               u8 unused_flags;
 700 +       }
 701 +       info;
 702 +       struct dx_entry entries[0];
 703 +};
 704 +
 705 +struct dx_node
 706 +{
 707 +       struct fake_dirent fake;
 708 +       struct dx_entry entries[0];
 709 +};
 710 +
 711 +
 712 +struct dx_frame
 713 +{
 714 +       struct buffer_head *bh;
 715 +       struct dx_entry *entries;
 716 +       struct dx_entry *at;
 717 +};
 718 +
 719 +struct dx_map_entry
 720 +{
 721 +       u32 hash;
 722 +       u32 offs;
 723 +};
 724 +
 725 +#ifdef CONFIG_EXT3_INDEX
 726 +static inline unsigned dx_get_block (struct dx_entry *entry);
 727 +static void dx_set_block (struct dx_entry *entry, unsigned value);
 728 +static inline unsigned dx_get_hash (struct dx_entry *entry);
 729 +static void dx_set_hash (struct dx_entry *entry, unsigned value);
 730 +static unsigned dx_get_count (struct dx_entry *entries);
 731 +static unsigned dx_get_limit (struct dx_entry *entries);
 732 +static void dx_set_count (struct dx_entry *entries, unsigned value);
 733 +static void dx_set_limit (struct dx_entry *entries, unsigned value);
 734 +static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
 735 +static unsigned dx_node_limit (struct inode *dir);
 736 +static struct dx_frame *dx_probe(struct dentry *dentry,
 737 +                                struct inode *dir,
 738 +                                struct dx_hash_info *hinfo,
 739 +                                struct dx_frame *frame,
 740 +                                int *err);
 741 +static void dx_release (struct dx_frame *frames);
 742 +static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
 743 +                       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 744 +static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 745 +static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
 746 +               struct dx_map_entry *offsets, int count);
 747 +static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
 748 +static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
 749 +static int ext3_htree_next_block(struct inode *dir, __u32 hash,
 750 +                                struct dx_frame *frame,
 751 +                                struct dx_frame *frames, int *err,
 752 +                                __u32 *start_hash);
 753 +static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
 754 +                      struct ext3_dir_entry_2 **res_dir, int *err);
 755 +static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
 756 +                            struct inode *inode);
 757 +
 758 +/*
 759 + * Future: use high four bits of block for coalesce-on-delete flags
 760 + * Mask them off for now.
 761 + */
 762 +
 763 +static inline unsigned dx_get_block (struct dx_entry *entry)
 764 +{
 765 +       return le32_to_cpu(entry->block.v) & 0x00ffffff;
 766 +}
 767 +
 768 +static inline void dx_set_block (struct dx_entry *entry, unsigned value)
 769 +{
 770 +       entry->block.v = cpu_to_le32(value);
 771 +}
 772 +
 773 +static inline unsigned dx_get_hash (struct dx_entry *entry)
 774 +{
 775 +       return le32_to_cpu(entry->hash.v);
 776 +}
 777 +
 778 +static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
 779 +{
 780 +       entry->hash.v = cpu_to_le32(value);
 781 +}
 782 +
 783 +static inline unsigned dx_get_count (struct dx_entry *entries)
 784 +{
 785 +       return le16_to_cpu(((struct dx_countlimit *) entries)->count.v);
 786 +}
 787 +
 788 +static inline unsigned dx_get_limit (struct dx_entry *entries)
 789 +{
 790 +       return le16_to_cpu(((struct dx_countlimit *) entries)->limit.v);
 791 +}
 792 +
 793 +static inline void dx_set_count (struct dx_entry *entries, unsigned value)
 794 +{
 795 +       ((struct dx_countlimit *) entries)->count.v = cpu_to_le16(value);
 796 +}
 797 +
 798 +static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
 799 +{
 800 +       ((struct dx_countlimit *) entries)->limit.v = cpu_to_le16(value);
 801 +}
 802 +
 803 +static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
 804 +{
 805 +       unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
 806 +               EXT3_DIR_REC_LEN(2) - infosize;
 807 +       return 0? 20: entry_space / sizeof(struct dx_entry);
 808 +}
 809 +
 810 +static inline unsigned dx_node_limit (struct inode *dir)
 811 +{
 812 +       unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
 813 +       return 0? 22: entry_space / sizeof(struct dx_entry);
 814 +}
 815 +
 816 +/*
 817 + * Debug
 818 + */
 819 +#ifdef DX_DEBUG
 820 +struct stats
 821 +{
 822 +       unsigned names;
 823 +       unsigned space;
 824 +       unsigned bcount;
 825 +};
 826 +
 827 +static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
 828 +                                int size, int show_names)
 829 +{
 830 +       unsigned names = 0, space = 0;
 831 +       char *base = (char *) de;
 832 +       struct dx_hash_info h = *hinfo;
 833 +
 834 +       printk("names: ");
 835 +       while ((char *) de < base + size)
 836 +       {
 837 +               if (de->inode)
 838 +               {
 839 +                       if (show_names)
 840 +                       {
 841 +                               int len = de->name_len;
 842 +                               char *name = de->name;
 843 +                               while (len--) printk("%c", *name++);
 844 +                               ext3fs_dirhash(de->name, de->name_len, &h);
 845 +                               printk(":%x.%u ", h.hash,
 846 +                                      ((char *) de - base));
 847 +                       }
 848 +                       space += EXT3_DIR_REC_LEN(de->name_len);
 849 +                       names++;
 850 +               }
 851 +               de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 852 +       }
 853 +       printk("(%i)\n", names);
 854 +       return (struct stats) { names, space, 1 };
 855 +}
 856 +
 857 +struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 858 +                            struct dx_entry *entries, int levels)
 859 +{
 860 +       unsigned blocksize = dir->i_sb->s_blocksize;
 861 +       unsigned count = dx_get_count (entries), names = 0, space = 0, i;
 862 +       unsigned bcount = 0;
 863 +       struct buffer_head *bh;
 864 +       int err;
 865 +       printk("%i indexed blocks...\n", count);
 866 +       for (i = 0; i < count; i++, entries++)
 867 +       {
 868 +               u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
 869 +               u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 870 +               struct stats stats;
 871 +               printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 872 +               if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
 873 +               stats = levels?
 874 +                  dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 875 +                  dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
 876 +               names += stats.names;
 877 +               space += stats.space;
 878 +               bcount += stats.bcount;
 879 +               brelse (bh);
 880 +       }
 881 +       if (bcount)
 882 +               printk("%snames %u, fullness %u (%u%%)\n", levels?"":"   ",
 883 +                       names, space/bcount,(space/bcount)*100/blocksize);
 884 +       return (struct stats) { names, space, bcount};
 885 +}
 886 +#endif /* DX_DEBUG */
 887 +
 888 +/*
 889 + * Probe for a directory leaf block to search.
 890 + *
 891 + * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 892 + * error in the directory index, and the caller should fall back to
 893 + * searching the directory normally.  The callers of dx_probe **MUST**
 894 + * check for this error code, and make sure it never gets reflected
 895 + * back to userspace.
 896 + */
 897 +static struct dx_frame *
 898 +dx_probe(struct dentry *dentry, struct inode *dir,
 899 +        struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 900 +{
 901 +       unsigned count, indirect;
 902 +       struct dx_entry *at, *entries, *p, *q, *m;
 903 +       struct dx_root *root;
 904 +       struct buffer_head *bh;
 905 +       struct dx_frame *frame = frame_in;
 906 +       u32 hash;
 907 +
 908 +       frame->bh = NULL;
 909 +       if (dentry)
 910 +               dir = dentry->d_parent->d_inode;
 911 +       if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
 912 +               goto fail;
 913 +       root = (struct dx_root *) bh->b_data;
 914 +       if (root->info.hash_version != DX_HASH_TEA &&
 915 +           root->info.hash_version != DX_HASH_HALF_MD4 &&
 916 +           root->info.hash_version != DX_HASH_LEGACY) {
 917 +               ext3_warning(dir->i_sb, __FUNCTION__,
 918 +                            "Unrecognised inode hash code %d",
 919 +                            root->info.hash_version);
 920 +               brelse(bh);
 921 +               *err = ERR_BAD_DX_DIR;
 922 +               goto fail;
 923 +       }
 924 +       hinfo->hash_version = root->info.hash_version;
 925 +       hinfo->seed = dir->i_sb->u.ext3_sb.s_hash_seed;
 926 +       if (dentry)
 927 +               ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
 928 +       hash = hinfo->hash;
 929 +
 930 +       if (root->info.unused_flags & 1) {
 931 +               ext3_warning(dir->i_sb, __FUNCTION__,
 932 +                            "Unimplemented inode hash flags: %#06x",
 933 +                            root->info.unused_flags);
 934 +               brelse(bh);
 935 +               *err = ERR_BAD_DX_DIR;
 936 +               goto fail;
 937 +       }
 938 +
 939 +       if ((indirect = root->info.indirect_levels) > 1) {
 940 +               ext3_warning(dir->i_sb, __FUNCTION__,
 941 +                            "Unimplemented inode hash depth: %#06x",
 942 +                            root->info.indirect_levels);
 943 +               brelse(bh);
 944 +               *err = ERR_BAD_DX_DIR;
 945 +               goto fail;
 946 +       }
 947 +
 948 +       entries = (struct dx_entry *) (((char *)&root->info) +
 949 +                                      root->info.info_length);
 950 +       assert(dx_get_limit(entries) == dx_root_limit(dir,
 951 +                                                     root->info.info_length));
 952 +       dxtrace (printk("Look up %x", hash));
 953 +       while (1)
 954 +       {
 955 +               count = dx_get_count(entries);
 956 +               assert (count && count <= dx_get_limit(entries));
 957 +               p = entries + 1;
 958 +               q = entries + count - 1;
 959 +               while (p <= q)
 960 +               {
 961 +                       m = p + (q - p)/2;
 962 +                       dxtrace(printk("."));
 963 +                       if (dx_get_hash(m) > hash)
 964 +                               q = m - 1;
 965 +                       else
 966 +                               p = m + 1;
 967 +               }
 968 +
 969 +               if (0) // linear search cross check
 970 +               {
 971 +                       unsigned n = count - 1;
 972 +                       at = entries;
 973 +                       while (n--)
 974 +                       {
 975 +                               dxtrace(printk(","));
 976 +                               if (dx_get_hash(++at) > hash)
 977 +                               {
 978 +                                       at--;
 979 +                                       break;
 980 +                               }
 981 +                       }
 982 +                       assert (at == p - 1);
 983 +               }
 984 +
 985 +               at = p - 1;
 986 +               dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 987 +               frame->bh = bh;
 988 +               frame->entries = entries;
 989 +               frame->at = at;
 990 +               if (!indirect--) return frame;
 991 +               if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
 992 +                       goto fail2;
 993 +               at = entries = ((struct dx_node *) bh->b_data)->entries;
 994 +               assert (dx_get_limit(entries) == dx_node_limit (dir));
 995 +               frame++;
 996 +       }
 997 +fail2:
 998 +       while (frame >= frame_in) {
 999 +               brelse(frame->bh);
1000 +               frame--;
1001 +       }
1002 +fail:
1003 +       return NULL;
1004 +}
1005 +
1006 +static void dx_release (struct dx_frame *frames)
1007 +{
1008 +       if (frames[0].bh == NULL)
1009 +               return;
1010 +
1011 +       if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
1012 +               brelse(frames[1].bh);
1013 +       brelse(frames[0].bh);
1014 +}
1015 +
1016 +/*
1017 + * This function increments the frame pointer to search the next leaf
1018 + * block, and reads in the necessary intervening nodes if the search
1019 + * should be necessary.  Whether or not the search is necessary is
1020 + * controlled by the hash parameter.  If the hash value is even, then
1021 + * the search is only continued if the next block starts with that
1022 + * hash value.  This is used if we are searching for a specific file.
1023 + *
1024 + * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
1025 + *
1026 + * This function returns 1 if the caller should continue to search,
1027 + * or 0 if it should not.  If there is an error reading one of the
1028 + * index blocks, it will return -1.
1029 + *
1030 + * If start_hash is non-null, it will be filled in with the starting
1031 + * hash of the next page.
1032 + */
1033 +static int ext3_htree_next_block(struct inode *dir, __u32 hash,
1034 +                                struct dx_frame *frame,
1035 +                                struct dx_frame *frames, int *err,
1036 +                                __u32 *start_hash)
1037 +{
1038 +       struct dx_frame *p;
1039 +       struct buffer_head *bh;
1040 +       int num_frames = 0;
1041 +       __u32 bhash;
1042 +
1043 +       *err = ENOENT;
1044 +       p = frame;
1045 +       /*
1046 +        * Find the next leaf page by incrementing the frame pointer.
1047 +        * If we run out of entries in the interior node, loop around and
1048 +        * increment pointer in the parent node.  When we break out of
1049 +        * this loop, num_frames indicates the number of interior
1050 +        * nodes need to be read.
1051 +        */
1052 +       while (1) {
1053 +               if (++(p->at) < p->entries + dx_get_count(p->entries))
1054 +                       break;
1055 +               if (p == frames)
1056 +                       return 0;
1057 +               num_frames++;
1058 +               p--;
1059 +       }
1060 +
1061 +       /*
1062 +        * If the hash is 1, then continue only if the next page has a
1063 +        * continuation hash of any value.  This is used for readdir
1064 +        * handling.  Otherwise, check to see if the hash matches the
1065 +        * desired contiuation hash.  If it doesn't, return since
1066 +        * there's no point to read in the successive index pages.
1067 +        */
1068 +       bhash = dx_get_hash(p->at);
1069 +       if (start_hash)
1070 +               *start_hash = bhash;
1071 +       if ((hash & 1) == 0) {
1072 +               if ((bhash & ~1) != hash)
1073 +                       return 0;
1074 +       }
1075 +       /*
1076 +        * If the hash is HASH_NB_ALWAYS, we always go to the next
1077 +        * block so no check is necessary
1078 +        */
1079 +       while (num_frames--) {
1080 +               if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
1081 +                                     0, err)))
1082 +                       return -1; /* Failure */
1083 +               p++;
1084 +               brelse (p->bh);
1085 +               p->bh = bh;
1086 +               p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1087 +       }
1088 +       return 1;
1089 +}
1090 +
1091 +
1092 +/*
1093 + * p is at least 6 bytes before the end of page
1094 + */
1095 +static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
1096 +{
1097 +       return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
1098 +}
1099 +
1100 +/*
1101 + * This function fills a red-black tree with information from a
1102 + * directory.  We start scanning the directory in hash order, starting
1103 + * at start_hash and start_minor_hash.
1104 + *
1105 + * This function returns the number of entries inserted into the tree,
1106 + * or a negative error code.
1107 + */
1108 +int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1109 +                        __u32 start_minor_hash, __u32 *next_hash)
1110 +{
1111 +       struct dx_hash_info hinfo;
1112 +       struct buffer_head *bh;
1113 +       struct ext3_dir_entry_2 *de, *top;
1114 +       static struct dx_frame frames[2], *frame;
1115 +       struct inode *dir;
1116 +       int block, err;
1117 +       int count = 0;
1118 +       int ret;
1119 +       __u32 hashval;
1120 +
1121 +       dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
1122 +                      start_minor_hash));
1123 +       dir = dir_file->f_dentry->d_inode;
1124 +       hinfo.hash = start_hash;
1125 +       hinfo.minor_hash = 0;
1126 +       frame = dx_probe(0, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
1127 +       if (!frame)
1128 +               return err;
1129 +
1130 +       while (1) {
1131 +               block = dx_get_block(frame->at);
1132 +               dxtrace(printk("Reading block %d\n", block));
1133 +               if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
1134 +                       goto errout;
1135 +
1136 +               de = (struct ext3_dir_entry_2 *) bh->b_data;
1137 +               top = (struct ext3_dir_entry_2 *) ((char *) de + dir->i_sb->s_blocksize -
1138 +                                      EXT3_DIR_REC_LEN(0));
1139 +               for (; de < top; de = ext3_next_entry(de)) {
1140 +                       ext3fs_dirhash(de->name, de->name_len, &hinfo);
1141 +                       if ((hinfo.hash < start_hash) ||
1142 +                           ((hinfo.hash == start_hash) &&
1143 +                            (hinfo.minor_hash < start_minor_hash)))
1144 +                               continue;
1145 +                       ext3_htree_store_dirent(dir_file, hinfo.hash,
1146 +                                               hinfo.minor_hash, de);
1147 +                       count++;
1148 +               }
1149 +               brelse (bh);
1150 +               hashval = ~1;
1151 +               ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
1152 +                                           frame, frames, &err, &hashval);
1153 +               if (next_hash)
1154 +                       *next_hash = hashval;
1155 +               if (ret == -1)
1156 +                       goto errout;
1157 +               /*
1158 +                * Stop if:  (a) there are no more entries, or
1159 +                * (b) we have inserted at least one entry and the
1160 +                * next hash value is not a continuation
1161 +                */
1162 +               if ((ret == 0) ||
1163 +                   (count && ((hashval & 1) == 0)))
1164 +                       break;
1165 +       }
1166 +       dx_release(frames);
1167 +       dxtrace(printk("Fill tree: returned %d entries\n", count));
1168 +       return count;
1169 +errout:
1170 +       dx_release(frames);
1171 +       return (err);
1172 +}
1173 +
1174 +
1175 +/*
1176 + * Directory block splitting, compacting
1177 + */
1178 +
1179 +static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
1180 +                       struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
1181 +{
1182 +       int count = 0;
1183 +       char *base = (char *) de;
1184 +       struct dx_hash_info h = *hinfo;
1185 +
1186 +       while ((char *) de < base + size)
1187 +       {
1188 +               if (de->name_len && de->inode) {
1189 +                       ext3fs_dirhash(de->name, de->name_len, &h);
1190 +                       map_tail--;
1191 +                       map_tail->hash = h.hash;
1192 +                       map_tail->offs = (u32) ((char *) de - base);
1193 +                       count++;
1194 +               }
1195 +               /* XXX: do we need to check rec_len == 0 case? -Chris */
1196 +               de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
1197 +       }
1198 +       return count;
1199 +}
1200 +
1201 +static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1202 +{
1203 +       struct dx_map_entry *p, *q, *top = map + count - 1;
1204 +       int more;
1205 +       /* Combsort until bubble sort doesn't suck */
1206 +       while (count > 2)
1207 +       {
1208 +               count = count*10/13;
1209 +               if (count - 9 < 2) /* 9, 10 -> 11 */
1210 +                       count = 11;
1211 +               for (p = top, q = p - count; q >= map; p--, q--)
1212 +                       if (p->hash < q->hash)
1213 +                               swap(*p, *q);
1214 +       }
1215 +       /* Garden variety bubble sort */
1216 +       do {
1217 +               more = 0;
1218 +               q = top;
1219 +               while (q-- > map)
1220 +               {
1221 +                       if (q[1].hash >= q[0].hash)
1222 +                               continue;
1223 +                       swap(*(q+1), *q);
1224 +                       more = 1;
1225 +               }
1226 +       } while(more);
1227 +}
1228 +
1229 +static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
1230 +{
1231 +       struct dx_entry *entries = frame->entries;
1232 +       struct dx_entry *old = frame->at, *new = old + 1;
1233 +       int count = dx_get_count(entries);
1234 +
1235 +       assert(count < dx_get_limit(entries));
1236 +       assert(old < entries + count);
1237 +       memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1238 +       dx_set_hash(new, hash);
1239 +       dx_set_block(new, block);
1240 +       dx_set_count(entries, count + 1);
1241 +}
1242 +#endif
1243 +
1244 +
1245 +static void ext3_update_dx_flag(struct inode *inode)
1246 +{
1247 +       if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
1248 +                                    EXT3_FEATURE_COMPAT_DIR_INDEX))
1249 +               EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
1250 +}
1251 +
1252  /*
1253   * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
1254   *
1255 @@ -96,6 +726,7 @@
1256         return 0;
1257  }
1258
1259 +
1260  /*
1261   *     ext3_find_entry()
1262   *
1263 @@ -107,6 +738,8 @@
1264   * The returned buffer_head has ->b_count elevated.  The caller is expected
1265   * to brelse() it when appropriate.
1266   */
1267 +
1268 +
1269  static struct buffer_head * ext3_find_entry (struct dentry *dentry,
1270                                         struct ext3_dir_entry_2 ** res_dir)
1271  {
1272 @@ -121,12 +754,32 @@
1273         int num = 0;
1274         int nblocks, i, err;
1275         struct inode *dir = dentry->d_parent->d_inode;
1276 +       int namelen;
1277 +       const u8 *name;
1278 +       unsigned blocksize;
1279
1280         *res_dir = NULL;
1281         sb = dir->i_sb;
1282 -
1283 +       blocksize = sb->s_blocksize;
1284 +       namelen = dentry->d_name.len;
1285 +       name = dentry->d_name.name;
1286 +       if (namelen > EXT3_NAME_LEN)
1287 +               return NULL;
1288 +#ifdef CONFIG_EXT3_INDEX
1289 +       if (is_dx(dir)) {
1290 +               bh = ext3_dx_find_entry(dentry, res_dir, &err);
1291 +               /*
1292 +                * On success, or if the error was file not found,
1293 +                * return.  Otherwise, fall back to doing a search the
1294 +                * old fashioned way.
1295 +                */
1296 +               if (bh || (err != ERR_BAD_DX_DIR))
1297 +                       return bh;
1298 +               dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
1299 +       }
1300 +#endif
1301         nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
1302 -       start = dir->u.ext3_i.i_dir_start_lookup;
1303 +       start = EXT3_I(dir)->i_dir_start_lookup;
1304         if (start >= nblocks)
1305                 start = 0;
1306         block = start;
1307 @@ -167,7 +820,7 @@
1308                 i = search_dirblock(bh, dir, dentry,
1309                             block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
1310                 if (i == 1) {
1311 -                       dir->u.ext3_i.i_dir_start_lookup = block;
1312 +                       EXT3_I(dir)->i_dir_start_lookup = block;
1313                         ret = bh;
1314                         goto cleanup_and_exit;
1315                 } else {
1316 @@ -198,6 +851,66 @@
1317         return ret;
1318  }
1319
1320 +#ifdef CONFIG_EXT3_INDEX
1321 +static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
1322 +                      struct ext3_dir_entry_2 **res_dir, int *err)
1323 +{
1324 +       struct super_block * sb;
1325 +       struct dx_hash_info     hinfo;
1326 +       u32 hash;
1327 +       struct dx_frame frames[2], *frame;
1328 +       struct ext3_dir_entry_2 *de, *top;
1329 +       struct buffer_head *bh;
1330 +       unsigned long block;
1331 +       int retval;
1332 +       int namelen = dentry->d_name.len;
1333 +       const u8 *name = dentry->d_name.name;
1334 +       struct inode *dir = dentry->d_parent->d_inode;
1335 +
1336 +       sb = dir->i_sb;
1337 +       if (!(frame = dx_probe (dentry, 0, &hinfo, frames, err)))
1338 +               return NULL;
1339 +       hash = hinfo.hash;
1340 +       do {
1341 +               block = dx_get_block(frame->at);
1342 +               if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
1343 +                       goto errout;
1344 +               de = (struct ext3_dir_entry_2 *) bh->b_data;
1345 +               top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1346 +                                      EXT3_DIR_REC_LEN(0));
1347 +               for (; de < top; de = ext3_next_entry(de))
1348 +               if (ext3_match (namelen, name, de)) {
1349 +                       if (!ext3_check_dir_entry("ext3_find_entry",
1350 +                                                 dir, de, bh,
1351 +                                 (block<<EXT3_BLOCK_SIZE_BITS(sb))
1352 +                                         +((char *)de - bh->b_data))) {
1353 +                               brelse (bh);
1354 +                               goto errout;
1355 +                       }
1356 +                       *res_dir = de;
1357 +                       dx_release (frames);
1358 +                       return bh;
1359 +               }
1360 +               brelse (bh);
1361 +               /* Check to see if we should continue to search */
1362 +               retval = ext3_htree_next_block(dir, hash, frame,
1363 +                                              frames, err, 0);
1364 +               if (retval == -1) {
1365 +                       ext3_warning(sb, __FUNCTION__,
1366 +                            "error reading index page in directory #%lu",
1367 +                            dir->i_ino);
1368 +                       goto errout;
1369 +               }
1370 +       } while (retval == 1);
1371 +
1372 +       *err = -ENOENT;
1373 +errout:
1374 +       dxtrace(printk("%s not found\n", name));
1375 +       dx_release (frames);
1376 +       return NULL;
1377 +}
1378 +#endif
1379 +
1380  static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry)
1381  {
1382         struct inode * inode;
1383 @@ -214,8 +927,9 @@
1384                 brelse (bh);
1385                 inode = iget(dir->i_sb, ino);
1386
1387 -               if (!inode)
1388 +               if (!inode) {
1389                         return ERR_PTR(-EACCES);
1390 +               }
1391         }
1392         d_add(dentry, inode);
1393         return NULL;
1394 @@ -239,6 +953,301 @@
1395                 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1396  }
1397
1398 +#ifdef CONFIG_EXT3_INDEX
1399 +static struct ext3_dir_entry_2 *
1400 +dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1401 +{
1402 +       unsigned rec_len = 0;
1403 +
1404 +       while (count--) {
1405 +               struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1406 +               rec_len = EXT3_DIR_REC_LEN(de->name_len);
1407 +               memcpy (to, de, rec_len);
1408 +               ((struct ext3_dir_entry_2 *) to)->rec_len = rec_len;
1409 +               de->inode = 0;
1410 +               map++;
1411 +               to += rec_len;
1412 +       }
1413 +       return (struct ext3_dir_entry_2 *) (to - rec_len);
1414 +}
1415 +
1416 +static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1417 +{
1418 +       struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1419 +       unsigned rec_len = 0;
1420 +
1421 +       prev = to = de;
1422 +       while ((char*)de < base + size) {
1423 +               next = (struct ext3_dir_entry_2 *) ((char *) de +
1424 +                                                   le16_to_cpu(de->rec_len));
1425 +               if (de->inode && de->name_len) {
1426 +                       rec_len = EXT3_DIR_REC_LEN(de->name_len);
1427 +                       if (de > to)
1428 +                               memmove(to, de, rec_len);
1429 +                       to->rec_len = rec_len;
1430 +                       prev = to;
1431 +                       to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1432 +               }
1433 +               de = next;
1434 +       }
1435 +       return prev;
1436 +}
1437 +
1438 +static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1439 +                       struct buffer_head **bh,struct dx_frame *frame,
1440 +                       struct dx_hash_info *hinfo, int *error)
1441 +{
1442 +       unsigned blocksize = dir->i_sb->s_blocksize;
1443 +       unsigned count, continued;
1444 +       struct buffer_head *bh2;
1445 +       u32 newblock;
1446 +       u32 hash2;
1447 +       struct dx_map_entry *map;
1448 +       char *data1 = (*bh)->b_data, *data2;
1449 +       unsigned split;
1450 +       struct ext3_dir_entry_2 *de = NULL, *de2;
1451 +       int     err;
1452 +
1453 +       bh2 = ext3_append (handle, dir, &newblock, error);
1454 +       if (!(bh2)) {
1455 +               brelse(*bh);
1456 +               *bh = NULL;
1457 +               goto errout;
1458 +       }
1459 +
1460 +       BUFFER_TRACE(*bh, "get_write_access");
1461 +       err = ext3_journal_get_write_access(handle, *bh);
1462 +       if (err) {
1463 +       journal_error:
1464 +               brelse(*bh);
1465 +               brelse(bh2);
1466 +               *bh = NULL;
1467 +               ext3_std_error(dir->i_sb, err);
1468 +               goto errout;
1469 +       }
1470 +       BUFFER_TRACE(frame->bh, "get_write_access");
1471 +       err = ext3_journal_get_write_access(handle, frame->bh);
1472 +       if (err)
1473 +               goto journal_error;
1474 +
1475 +       data2 = bh2->b_data;
1476 +
1477 +       /* create map in the end of data2 block */
1478 +       map = (struct dx_map_entry *) (data2 + blocksize);
1479 +       count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1480 +                            blocksize, hinfo, map);
1481 +       map -= count;
1482 +       split = count/2; // need to adjust to actual middle
1483 +       dx_sort_map (map, count);
1484 +       hash2 = map[split].hash;
1485 +       continued = hash2 == map[split - 1].hash;
1486 +       dxtrace(printk("Split block %i at %x, %i/%i\n",
1487 +               dx_get_block(frame->at), hash2, split, count-split));
1488 +
1489 +       /* Fancy dance to stay within two buffers */
1490 +       de2 = dx_move_dirents(data1, data2, map + split, count - split);
1491 +       de = dx_pack_dirents(data1,blocksize);
1492 +       de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1493 +       de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1494 +       dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1495 +       dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1496 +
1497 +       /* Which block gets the new entry? */
1498 +       if (hinfo->hash >= hash2)
1499 +       {
1500 +               swap(*bh, bh2);
1501 +               de = de2;
1502 +       }
1503 +       dx_insert_block (frame, hash2 + continued, newblock);
1504 +       err = ext3_journal_dirty_metadata (handle, bh2);
1505 +       if (err)
1506 +               goto journal_error;
1507 +       err = ext3_journal_dirty_metadata (handle, frame->bh);
1508 +       if (err)
1509 +               goto journal_error;
1510 +       brelse (bh2);
1511 +       dxtrace(dx_show_index ("frame", frame->entries));
1512 +errout:
1513 +       return de;
1514 +}
1515 +#endif
1516 +
1517 +
1518 +/*
1519 + * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1520 + * it points to a directory entry which is guaranteed to be large
1521 + * enough for new directory entry.  If de is NULL, then
1522 + * add_dirent_to_buf will attempt search the directory block for
1523 + * space.  It will return -ENOSPC if no space is available, and -EIO
1524 + * and -EEXIST if directory entry already exists.
1525 + *
1526 + * NOTE!  bh is NOT released in the case where ENOSPC is returned.  In
1527 + * all other cases bh is released.
1528 + */
1529 +static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1530 +                            struct inode *inode, struct ext3_dir_entry_2 *de,
1531 +                            struct buffer_head * bh)
1532 +{
1533 +       struct inode    *dir = dentry->d_parent->d_inode;
1534 +       const char      *name = dentry->d_name.name;
1535 +       int             namelen = dentry->d_name.len;
1536 +       unsigned long   offset = 0;
1537 +       unsigned short  reclen;
1538 +       int             nlen, rlen, err;
1539 +       char            *top;
1540 +
1541 +       reclen = EXT3_DIR_REC_LEN(namelen);
1542 +       if (!de) {
1543 +               de = (struct ext3_dir_entry_2 *)bh->b_data;
1544 +               top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1545 +               while ((char *) de <= top) {
1546 +                       if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1547 +                                                 bh, offset)) {
1548 +                               brelse (bh);
1549 +                               return -EIO;
1550 +                       }
1551 +                       if (ext3_match (namelen, name, de)) {
1552 +                               brelse (bh);
1553 +                               return -EEXIST;
1554 +                       }
1555 +                       nlen = EXT3_DIR_REC_LEN(de->name_len);
1556 +                       rlen = le16_to_cpu(de->rec_len);
1557 +                       if ((de->inode? rlen - nlen: rlen) >= reclen)
1558 +                               break;
1559 +                       de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1560 +                       offset += rlen;
1561 +               }
1562 +               if ((char *) de > top)
1563 +                       return -ENOSPC;
1564 +       }
1565 +       BUFFER_TRACE(bh, "get_write_access");
1566 +       err = ext3_journal_get_write_access(handle, bh);
1567 +       if (err) {
1568 +               ext3_std_error(dir->i_sb, err);
1569 +               brelse(bh);
1570 +               return err;
1571 +       }
1572 +
1573 +       /* By now the buffer is marked for journaling */
1574 +       nlen = EXT3_DIR_REC_LEN(de->name_len);
1575 +       rlen = le16_to_cpu(de->rec_len);
1576 +       if (de->inode) {
1577 +               struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1578 +               de1->rec_len = cpu_to_le16(rlen - nlen);
1579 +               de->rec_len = cpu_to_le16(nlen);
1580 +               de = de1;
1581 +       }
1582 +       de->file_type = EXT3_FT_UNKNOWN;
1583 +       if (inode) {
1584 +               de->inode = cpu_to_le32(inode->i_ino);
1585 +               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1586 +       } else
1587 +               de->inode = 0;
1588 +       de->name_len = namelen;
1589 +       memcpy (de->name, name, namelen);
1590 +       /*
1591 +        * XXX shouldn't update any times until successful
1592 +        * completion of syscall, but too many callers depend
1593 +        * on this.
1594 +        *
1595 +        * XXX similarly, too many callers depend on
1596 +        * ext3_new_inode() setting the times, but error
1597 +        * recovery deletes the inode, so the worst that can
1598 +        * happen is that the times are slightly out of date
1599 +        * and/or different from the directory change time.
1600 +        */
1601 +       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1602 +       ext3_update_dx_flag(dir);
1603 +       dir->i_version = ++event;
1604 +       ext3_mark_inode_dirty(handle, dir);
1605 +       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1606 +       err = ext3_journal_dirty_metadata(handle, bh);
1607 +       if (err)
1608 +               ext3_std_error(dir->i_sb, err);
1609 +       brelse(bh);
1610 +       return 0;
1611 +}
1612 +
1613 +#ifdef CONFIG_EXT3_INDEX
1614 +/*
1615 + * This converts a one block unindexed directory to a 3 block indexed
1616 + * directory, and adds the dentry to the indexed directory.
1617 + */
1618 +static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1619 +                           struct inode *inode, struct buffer_head *bh)
1620 +{
1621 +       struct inode    *dir = dentry->d_parent->d_inode;
1622 +       const char      *name = dentry->d_name.name;
1623 +       int             namelen = dentry->d_name.len;
1624 +       struct buffer_head *bh2;
1625 +       struct dx_root  *root;
1626 +       struct dx_frame frames[2], *frame;
1627 +       struct dx_entry *entries;
1628 +       struct ext3_dir_entry_2 *de, *de2;
1629 +       char            *data1, *top;
1630 +       unsigned        len;
1631 +       int             retval;
1632 +       unsigned        blocksize;
1633 +       struct dx_hash_info hinfo;
1634 +       u32             block;
1635 +
1636 +       blocksize =  dir->i_sb->s_blocksize;
1637 +       dxtrace(printk("Creating index\n"));
1638 +       retval = ext3_journal_get_write_access(handle, bh);
1639 +       if (retval) {
1640 +               ext3_std_error(dir->i_sb, retval);
1641 +               brelse(bh);
1642 +               return retval;
1643 +       }
1644 +       root = (struct dx_root *) bh->b_data;
1645 +
1646 +       EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1647 +       bh2 = ext3_append (handle, dir, &block, &retval);
1648 +       if (!(bh2)) {
1649 +               brelse(bh);
1650 +               return retval;
1651 +       }
1652 +       data1 = bh2->b_data;
1653 +
1654 +       /* The 0th block becomes the root, move the dirents out */
1655 +       de = (struct ext3_dir_entry_2 *) &root->dotdot;
1656 +       de = (struct ext3_dir_entry_2 *) ((char *)de + de->rec_len);
1657 +       len = ((char *) root) + blocksize - (char *) de;
1658 +       memcpy (data1, de, len);
1659 +       de = (struct ext3_dir_entry_2 *) data1;
1660 +       top = data1 + len;
1661 +       while (((char *) de2=(char*)de+le16_to_cpu(de->rec_len)) < top)
1662 +               de = de2;
1663 +       de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1664 +       /* Initialize the root; the dot dirents already exist */
1665 +       de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1666 +       de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1667 +       memset (&root->info, 0, sizeof(root->info));
1668 +       root->info.info_length = sizeof(root->info);
1669 +       root->info.hash_version = dir->i_sb->u.ext3_sb.s_def_hash_version;
1670 +       entries = root->entries;
1671 +       dx_set_block (entries, 1);
1672 +       dx_set_count (entries, 1);
1673 +       dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1674 +
1675 +       /* Initialize as for dx_probe */
1676 +       hinfo.hash_version = root->info.hash_version;
1677 +       hinfo.seed = dir->i_sb->u.ext3_sb.s_hash_seed;
1678 +       ext3fs_dirhash(name, namelen, &hinfo);
1679 +       frame = frames;
1680 +       frame->entries = entries;
1681 +       frame->at = entries;
1682 +       frame->bh = bh;
1683 +       bh = bh2;
1684 +       de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1685 +       dx_release (frames);
1686 +       if (!(de))
1687 +               return retval;
1688 +
1689 +       return add_dirent_to_buf(handle, dentry, inode, de, bh);
1690 +}
1691 +#endif
1692 +
1693  /*
1694   *     ext3_add_entry()
1695   *
1696 @@ -249,127 +1258,197 @@
1697   * may not sleep between calling this and putting something into
1698   * the entry, as someone else might have used it while you slept.
1699   */
1700 -
1701 -/*
1702 - * AKPM: the journalling code here looks wrong on the error paths
1703 - */
1704  static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1705         struct inode *inode)
1706  {
1707         struct inode *dir = dentry->d_parent->d_inode;
1708 -       const char *name = dentry->d_name.name;
1709 -       int namelen = dentry->d_name.len;
1710         unsigned long offset;
1711 -       unsigned short rec_len;
1712         struct buffer_head * bh;
1713 -       struct ext3_dir_entry_2 * de, * de1;
1714 +       struct ext3_dir_entry_2 *de;
1715         struct super_block * sb;
1716         int     retval;
1717 +#ifdef CONFIG_EXT3_INDEX
1718 +        int     dx_fallback=0;
1719 +#endif
1720 +        unsigned blocksize;
1721 +        unsigned nlen, rlen;
1722 +        u32 block, blocks;
1723
1724         sb = dir->i_sb;
1725 -
1726 -       if (!namelen)
1727 +        blocksize = sb->s_blocksize;
1728 +       if (!dentry->d_name.len)
1729                 return -EINVAL;
1730 -       bh = ext3_bread (handle, dir, 0, 0, &retval);
1731 +#ifdef CONFIG_EXT3_INDEX
1732 +        if (is_dx(dir)) {
1733 +                retval = ext3_dx_add_entry(handle, dentry, inode);
1734 +                if (!retval || (retval != ERR_BAD_DX_DIR))
1735 +                        return retval;
1736 +                EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1737 +                dx_fallback++;
1738 +                ext3_mark_inode_dirty(handle, dir);
1739 +        }
1740 +#endif
1741 +        blocks = dir->i_size >> sb->s_blocksize_bits;
1742 +        for (block = 0, offset = 0; block < blocks; block++) {
1743 +                bh = ext3_bread(handle, dir, block, 0, &retval);
1744 +                if(!bh)
1745 +                        return retval;
1746 +                retval = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1747 +                if (retval != -ENOSPC)
1748 +                        return retval;
1749 +
1750 +#ifdef CONFIG_EXT3_INDEX
1751 +                if (blocks == 1 && !dx_fallback &&
1752 +                    EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1753 +                        return make_indexed_dir(handle, dentry, inode, bh);
1754 +#endif
1755 +                brelse(bh);
1756 +        }
1757 +        bh = ext3_append(handle, dir, &block, &retval);
1758         if (!bh)
1759                 return retval;
1760 -       rec_len = EXT3_DIR_REC_LEN(namelen);
1761 -       offset = 0;
1762         de = (struct ext3_dir_entry_2 *) bh->b_data;
1763 -       while (1) {
1764 -               if ((char *)de >= sb->s_blocksize + bh->b_data) {
1765 -                       brelse (bh);
1766 -                       bh = NULL;
1767 -                       bh = ext3_bread (handle, dir,
1768 -                               offset >> EXT3_BLOCK_SIZE_BITS(sb), 1, &retval);
1769 -                       if (!bh)
1770 -                               return retval;
1771 -                       if (dir->i_size <= offset) {
1772 -                               if (dir->i_size == 0) {
1773 -                                       brelse(bh);
1774 -                                       return -ENOENT;
1775 -                               }
1776 -
1777 -                               ext3_debug ("creating next block\n");
1778 -
1779 -                               BUFFER_TRACE(bh, "get_write_access");
1780 -                               ext3_journal_get_write_access(handle, bh);
1781 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
1782 -                               de->inode = 0;
1783 -                               de->rec_len = le16_to_cpu(sb->s_blocksize);
1784 -                               dir->u.ext3_i.i_disksize =
1785 -                                       dir->i_size = offset + sb->s_blocksize;
1786 -                               dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1787 -                               ext3_mark_inode_dirty(handle, dir);
1788 -                       } else {
1789 -
1790 -                               ext3_debug ("skipping to next block\n");
1791 +        de->inode = 0;
1792 +        de->rec_len = cpu_to_le16(rlen = blocksize);
1793 +        nlen = 0;
1794 +        return add_dirent_to_buf(handle, dentry, inode, de, bh);
1795 +}
1796
1797 -                               de = (struct ext3_dir_entry_2 *) bh->b_data;
1798 -                       }
1799 -               }
1800 -               if (!ext3_check_dir_entry ("ext3_add_entry", dir, de, bh,
1801 -                                          offset)) {
1802 -                       brelse (bh);
1803 -                       return -ENOENT;
1804 -               }
1805 -               if (ext3_match (namelen, name, de)) {
1806 -                               brelse (bh);
1807 -                               return -EEXIST;
1808 +#ifdef CONFIG_EXT3_INDEX
1809 +/*
1810 + * Returns 0 for success, or a negative error value
1811 + */
1812 +static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1813 +                             struct inode *inode)
1814 +{
1815 +        struct dx_frame frames[2], *frame;
1816 +        struct dx_entry *entries, *at;
1817 +        struct dx_hash_info hinfo;
1818 +        struct buffer_head * bh;
1819 +        struct inode *dir = dentry->d_parent->d_inode;
1820 +        struct super_block * sb = dir->i_sb;
1821 +        struct ext3_dir_entry_2 *de;
1822 +        int err;
1823 +
1824 +        frame = dx_probe(dentry, 0, &hinfo, frames, &err);
1825 +        if (!frame)
1826 +                return err;
1827 +        entries = frame->entries;
1828 +        at = frame->at;
1829 +
1830 +        if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1831 +                goto cleanup;
1832 +
1833 +        BUFFER_TRACE(bh, "get_write_access");
1834 +        err = ext3_journal_get_write_access(handle, bh);
1835 +        if (err)
1836 +                goto journal_error;
1837 +
1838 +        err = add_dirent_to_buf(handle, dentry, inode, 0, bh);
1839 +        if (err != -ENOSPC) {
1840 +                bh = 0;
1841 +                goto cleanup;
1842 +        }
1843 +
1844 +        /* Block full, should compress but for now just split */
1845 +        dxtrace(printk("using %u of %u node entries\n",
1846 +                       dx_get_count(entries), dx_get_limit(entries)));
1847 +        /* Need to split index? */
1848 +        if (dx_get_count(entries) == dx_get_limit(entries)) {
1849 +                u32 newblock;
1850 +                unsigned icount = dx_get_count(entries);
1851 +                int levels = frame - frames;
1852 +                struct dx_entry *entries2;
1853 +                struct dx_node *node2;
1854 +                struct buffer_head *bh2;
1855 +
1856 +                if (levels && (dx_get_count(frames->entries) ==
1857 +                               dx_get_limit(frames->entries))) {
1858 +                        ext3_warning(sb, __FUNCTION__,
1859 +                                     "Directory index full!\n");
1860 +                        err = -ENOSPC;
1861 +                        goto cleanup;
1862                 }
1863 -               if ((le32_to_cpu(de->inode) == 0 &&
1864 -                               le16_to_cpu(de->rec_len) >= rec_len) ||
1865 -                   (le16_to_cpu(de->rec_len) >=
1866 -                               EXT3_DIR_REC_LEN(de->name_len) + rec_len)) {
1867 -                       BUFFER_TRACE(bh, "get_write_access");
1868 -                       ext3_journal_get_write_access(handle, bh);
1869 -                       /* By now the buffer is marked for journaling */
1870 -                       offset += le16_to_cpu(de->rec_len);
1871 -                       if (le32_to_cpu(de->inode)) {
1872 -                               de1 = (struct ext3_dir_entry_2 *) ((char *) de +
1873 -                                       EXT3_DIR_REC_LEN(de->name_len));
1874 -                               de1->rec_len =
1875 -                                       cpu_to_le16(le16_to_cpu(de->rec_len) -
1876 -                                       EXT3_DIR_REC_LEN(de->name_len));
1877 -                               de->rec_len = cpu_to_le16(
1878 -                                               EXT3_DIR_REC_LEN(de->name_len));
1879 -                               de = de1;
1880 +
1881 +                bh2 = ext3_append (handle, dir, &newblock, &err);
1882 +                if (!(bh2))
1883 +                        goto cleanup;
1884 +                node2 = (struct dx_node *)(bh2->b_data);
1885 +                entries2 = node2->entries;
1886 +                node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1887 +                node2->fake.inode = 0;
1888 +                BUFFER_TRACE(frame->bh, "get_write_access");
1889 +                err = ext3_journal_get_write_access(handle, frame->bh);
1890 +                if (err)
1891 +                        goto journal_error;
1892 +                if (levels) {
1893 +                        unsigned icount1 = icount/2, icount2 = icount - icount1;+                       unsigned hash2 = dx_get_hash(entries + icount1);
1894 +                        dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1895 +
1896 +                        BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1897 +                        err = ext3_journal_get_write_access(handle,
1898 +                                                            frames[0].bh);
1899 +                        if (err)
1900 +                                goto journal_error;
1901 +
1902 +                        memcpy ((char *) entries2, (char *) (entries + icount1),+                               icount2 * sizeof(struct dx_entry));
1903 +                        dx_set_count (entries, icount1);
1904 +                        dx_set_count (entries2, icount2);
1905 +                        dx_set_limit (entries2, dx_node_limit(dir));
1906 +
1907 +                        /* Which index block gets the new entry? */
1908 +                        if (at - entries >= icount1) {
1909 +                                frame->at = at = at - entries - icount1 + entries2;
1910 +                                frame->entries = entries = entries2;
1911 +                                swap(frame->bh, bh2);
1912                         }
1913 -                       de->file_type = EXT3_FT_UNKNOWN;
1914 -                       if (inode) {
1915 -                               de->inode = cpu_to_le32(inode->i_ino);
1916 -                               ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1917 -                       } else
1918 -                               de->inode = 0;
1919 -                       de->name_len = namelen;
1920 -                       memcpy (de->name, name, namelen);
1921 -                       /*
1922 -                        * XXX shouldn't update any times until successful
1923 -                        * completion of syscall, but too many callers depend
1924 -                        * on this.
1925 -                        *
1926 -                        * XXX similarly, too many callers depend on
1927 -                        * ext3_new_inode() setting the times, but error
1928 -                        * recovery deletes the inode, so the worst that can
1929 -                        * happen is that the times are slightly out of date
1930 -                        * and/or different from the directory change time.
1931 -                        */
1932 -                       dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1933 -                       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
1934 -                       ext3_mark_inode_dirty(handle, dir);
1935 -                       dir->i_version = ++event;
1936 -                       BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1937 -                       ext3_journal_dirty_metadata(handle, bh);
1938 -                       brelse(bh);
1939 -                       return 0;
1940 +                        dx_insert_block (frames + 0, hash2, newblock);
1941 +                        dxtrace(dx_show_index ("node", frames[1].entries));
1942 +                        dxtrace(dx_show_index ("node",
1943 +                               ((struct dx_node *) bh2->b_data)->entries));
1944 +                        err = ext3_journal_dirty_metadata(handle, bh2);
1945 +                        if (err)
1946 +                                goto journal_error;
1947 +                        brelse (bh2);
1948 +                } else {
1949 +                        dxtrace(printk("Creating second level index...\n"));
1950 +                        memcpy((char *) entries2, (char *) entries,
1951 +                               icount * sizeof(struct dx_entry));
1952 +                        dx_set_limit(entries2, dx_node_limit(dir));
1953 +
1954 +                        /* Set up root */
1955 +                        dx_set_count(entries, 1);
1956 +                        dx_set_block(entries + 0, newblock);
1957 +                        ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1958 +
1959 +                        /* Add new access path frame */
1960 +                        frame = frames + 1;
1961 +                        frame->at = at = at - entries + entries2;
1962 +                        frame->entries = entries = entries2;
1963 +                        frame->bh = bh2;
1964 +                        err = ext3_journal_get_write_access(handle,
1965 +                                                            frame->bh);
1966 +                        if (err)
1967 +                                goto journal_error;
1968                 }
1969 -               offset += le16_to_cpu(de->rec_len);
1970 -               de = (struct ext3_dir_entry_2 *)
1971 -                       ((char *) de + le16_to_cpu(de->rec_len));
1972 +                ext3_journal_dirty_metadata(handle, frames[0].bh);
1973         }
1974 -       brelse (bh);
1975 -       return -ENOSPC;
1976 +        de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1977 +        if (!de)
1978 +                goto cleanup;
1979 +        err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1980 +        bh = 0;
1981 +        goto cleanup;
1982 +
1983 +journal_error:
1984 +        ext3_std_error(dir->i_sb, err);
1985 +cleanup:
1986 +        if (bh)
1987 +                brelse(bh);
1988 +        dx_release(frames);
1989 +        return err;
1990  }
1991 +#endif
1992
1993  /*
1994   * ext3_delete_entry deletes a directory entry by merging it with the
1995 @@ -453,9 +1532,11 @@
1996         struct inode * inode;
1997         int err;
1998
1999 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
2000 -       if (IS_ERR(handle))
2001 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2002 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
2003 +       if (IS_ERR(handle)) {
2004                 return PTR_ERR(handle);
2005 +       }
2006
2007         if (IS_SYNC(dir))
2008                 handle->h_sync = 1;
2009 @@ -480,9 +1561,11 @@
2010         struct inode *inode;
2011         int err;
2012
2013 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
2014 -       if (IS_ERR(handle))
2015 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2016 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
2017 +       if (IS_ERR(handle)) {
2018                 return PTR_ERR(handle);
2019 +       }
2020
2021         if (IS_SYNC(dir))
2022                 handle->h_sync = 1;
2023 @@ -512,9 +1595,11 @@
2024         if (dir->i_nlink >= EXT3_LINK_MAX)
2025                 return -EMLINK;
2026
2027 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
2028 -       if (IS_ERR(handle))
2029 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2030 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3);
2031 +       if (IS_ERR(handle)) {
2032                 return PTR_ERR(handle);
2033 +       }
2034
2035         if (IS_SYNC(dir))
2036                 handle->h_sync = 1;
2037 @@ -526,7 +1611,8 @@
2038
2039         inode->i_op = &ext3_dir_inode_operations;
2040         inode->i_fop = &ext3_dir_operations;
2041 -       inode->i_size = inode->u.ext3_i.i_disksize = inode->i_sb->s_blocksize;
2042 +        inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
2043 +        inode->i_blocks = 0;
2044         dir_block = ext3_bread (handle, inode, 0, 1, &err);
2045         if (!dir_block) {
2046                 inode->i_nlink--; /* is this nlink == 0? */
2047 @@ -555,21 +1641,19 @@
2048         brelse (dir_block);
2049         ext3_mark_inode_dirty(handle, inode);
2050         err = ext3_add_entry (handle, dentry, inode);
2051 -       if (err)
2052 -               goto out_no_entry;
2053 +       if (err) {
2054 +               inode->i_nlink = 0;
2055 +               ext3_mark_inode_dirty(handle, inode);
2056 +               iput (inode);
2057 +               goto out_stop;
2058 +       }
2059         dir->i_nlink++;
2060 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
2061 +       ext3_update_dx_flag(dir);
2062         ext3_mark_inode_dirty(handle, dir);
2063         d_instantiate(dentry, inode);
2064  out_stop:
2065         ext3_journal_stop(handle, dir);
2066         return err;
2067 -
2068 -out_no_entry:
2069 -       inode->i_nlink = 0;
2070 -       ext3_mark_inode_dirty(handle, inode);
2071 -       iput (inode);
2072 -       goto out_stop;
2073  }
2074
2075  /*
2076 @@ -656,7 +1740,7 @@
2077         int err = 0, rc;
2078
2079         lock_super(sb);
2080 -       if (!list_empty(&inode->u.ext3_i.i_orphan))
2081 +       if (!list_empty(&EXT3_I(inode)->i_orphan))
2082                 goto out_unlock;
2083
2084         /* Orphan handling is only valid for files with data blocks
2085 @@ -697,7 +1781,7 @@
2086          * This is safe: on error we're going to ignore the orphan list
2087          * anyway on the next recovery. */
2088         if (!err)
2089 -               list_add(&inode->u.ext3_i.i_orphan, &EXT3_SB(sb)->s_orphan);
2090 +               list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
2091
2092         jbd_debug(4, "superblock will point to %ld\n", inode->i_ino);
2093         jbd_debug(4, "orphan inode %ld will point to %d\n",
2094 @@ -715,25 +1799,26 @@
2095  int ext3_orphan_del(handle_t *handle, struct inode *inode)
2096  {
2097         struct list_head *prev;
2098 +        struct ext3_inode_info *ei = EXT3_I(inode);
2099         struct ext3_sb_info *sbi;
2100         ino_t ino_next;
2101         struct ext3_iloc iloc;
2102         int err = 0;
2103
2104         lock_super(inode->i_sb);
2105 -       if (list_empty(&inode->u.ext3_i.i_orphan)) {
2106 +        if (list_empty(&ei->i_orphan)) {
2107                 unlock_super(inode->i_sb);
2108                 return 0;
2109         }
2110
2111         ino_next = NEXT_ORPHAN(inode);
2112 -       prev = inode->u.ext3_i.i_orphan.prev;
2113 +        prev = ei->i_orphan.prev;
2114         sbi = EXT3_SB(inode->i_sb);
2115
2116         jbd_debug(4, "remove inode %ld from orphan list\n", inode->i_ino);
2117
2118 -       list_del(&inode->u.ext3_i.i_orphan);
2119 -       INIT_LIST_HEAD(&inode->u.ext3_i.i_orphan);
2120 +        list_del(&ei->i_orphan);
2121 +        INIT_LIST_HEAD(&ei->i_orphan);
2122
2123         /* If we're on an error path, we may not have a valid
2124          * transaction handle with which to update the orphan list on
2125 @@ -794,8 +1879,9 @@
2126         handle_t *handle;
2127
2128         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2129 -       if (IS_ERR(handle))
2130 +       if (IS_ERR(handle)) {
2131                 return PTR_ERR(handle);
2132 +       }
2133
2134         retval = -ENOENT;
2135         bh = ext3_find_entry (dentry, &de);
2136 @@ -833,7 +1919,7 @@
2137         ext3_mark_inode_dirty(handle, inode);
2138         dir->i_nlink--;
2139         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2140 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
2141 +        ext3_update_dx_flag(dir);
2142         ext3_mark_inode_dirty(handle, dir);
2143
2144  end_rmdir:
2145 @@ -851,8 +1937,9 @@
2146         handle_t *handle;
2147
2148         handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS);
2149 -       if (IS_ERR(handle))
2150 +       if (IS_ERR(handle)) {
2151                 return PTR_ERR(handle);
2152 +       }
2153
2154         if (IS_SYNC(dir))
2155                 handle->h_sync = 1;
2156 @@ -879,7 +1966,7 @@
2157         if (retval)
2158                 goto end_unlink;
2159         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2160 -       dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
2161 +       ext3_update_dx_flag(dir);
2162         ext3_mark_inode_dirty(handle, dir);
2163         inode->i_nlink--;
2164         if (!inode->i_nlink)
2165 @@ -905,9 +1992,11 @@
2166         if (l > dir->i_sb->s_blocksize)
2167                 return -ENAMETOOLONG;
2168
2169 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 5);
2170 -       if (IS_ERR(handle))
2171 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2172 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5);
2173 +       if (IS_ERR(handle)) {
2174                 return PTR_ERR(handle);
2175 +       }
2176
2177         if (IS_SYNC(dir))
2178                 handle->h_sync = 1;
2179 @@ -917,7 +2006,7 @@
2180         if (IS_ERR(inode))
2181                 goto out_stop;
2182
2183 -       if (l > sizeof (inode->u.ext3_i.i_data)) {
2184 +        if (l > sizeof (EXT3_I(inode)->i_data)) {
2185                 inode->i_op = &ext3_symlink_inode_operations;
2186                 inode->i_mapping->a_ops = &ext3_aops;
2187                 /*
2188 @@ -926,25 +2015,23 @@
2189                  * i_size in generic_commit_write().
2190                  */
2191                 err = block_symlink(inode, symname, l);
2192 -               if (err)
2193 -                       goto out_no_entry;
2194 +                if (err) {
2195 +                        ext3_dec_count(handle, inode);
2196 +                        ext3_mark_inode_dirty(handle, inode);
2197 +                        iput (inode);
2198 +                        goto out_stop;
2199 +                }
2200         } else {
2201                 inode->i_op = &ext3_fast_symlink_inode_operations;
2202 -               memcpy((char*)&inode->u.ext3_i.i_data,symname,l);
2203 +                memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2204                 inode->i_size = l-1;
2205         }
2206 -       inode->u.ext3_i.i_disksize = inode->i_size;
2207 +        EXT3_I(inode)->i_disksize = inode->i_size;
2208         ext3_mark_inode_dirty(handle, inode);
2209         err = ext3_add_nondir(handle, dentry, inode);
2210  out_stop:
2211         ext3_journal_stop(handle, dir);
2212         return err;
2213 -
2214 -out_no_entry:
2215 -       ext3_dec_count(handle, inode);
2216 -       ext3_mark_inode_dirty(handle, inode);
2217 -       iput (inode);
2218 -       goto out_stop;
2219  }
2220
2221  static int ext3_link (struct dentry * old_dentry,
2222 @@ -957,12 +2044,15 @@
2223         if (S_ISDIR(inode->i_mode))
2224                 return -EPERM;
2225
2226 -       if (inode->i_nlink >= EXT3_LINK_MAX)
2227 +       if (inode->i_nlink >= EXT3_LINK_MAX) {
2228                 return -EMLINK;
2229 +       }
2230
2231 -       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS);
2232 -       if (IS_ERR(handle))
2233 +       handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS +
2234 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2235 +       if (IS_ERR(handle)) {
2236                 return PTR_ERR(handle);
2237 +       }
2238
2239         if (IS_SYNC(dir))
2240                 handle->h_sync = 1;
2241 @@ -996,9 +2086,11 @@
2242
2243         old_bh = new_bh = dir_bh = NULL;
2244
2245 -       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS + 2);
2246 -       if (IS_ERR(handle))
2247 +       handle = ext3_journal_start(old_dir, 2 * EXT3_DATA_TRANS_BLOCKS +
2248 +                                       EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2249 +       if (IS_ERR(handle)) {
2250                 return PTR_ERR(handle);
2251 +       }
2252
2253         if (IS_SYNC(old_dir) || IS_SYNC(new_dir))
2254                 handle->h_sync = 1;
2255 @@ -1078,7 +2170,7 @@
2256                 new_inode->i_ctime = CURRENT_TIME;
2257         }
2258         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
2259 -       old_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
2260 +       ext3_update_dx_flag(old_dir);
2261         if (dir_bh) {
2262                 BUFFER_TRACE(dir_bh, "get_write_access");
2263                 ext3_journal_get_write_access(handle, dir_bh);
2264 @@ -1090,7 +2182,7 @@
2265                         new_inode->i_nlink--;
2266                 } else {
2267                         new_dir->i_nlink++;
2268 -                       new_dir->u.ext3_i.i_flags &= ~EXT3_INDEX_FL;
2269 +                       ext3_update_dx_flag(new_dir);
2270                         ext3_mark_inode_dirty(handle, new_dir);
2271                 }
2272         }
2273 Index: linux-2.4.19/fs/ext3/super.c
2274 ===================================================================
2275 --- linux-2.4.19.orig/fs/ext3/super.c   2004-04-23 18:25:03.000000000 -0400
2276 +++ linux-2.4.19/fs/ext3/super.c        2004-04-23 18:26:27.000000000 -0400
2277 @@ -741,6 +741,7 @@
2278         es->s_mtime = cpu_to_le32(CURRENT_TIME);
2279         ext3_update_dynamic_rev(sb);
2280         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2281 +
2282         ext3_commit_super (sb, es, 1);
2283         if (test_opt (sb, DEBUG))
2284                 printk (KERN_INFO
2285 @@ -751,6 +752,7 @@
2286                         EXT3_BLOCKS_PER_GROUP(sb),
2287                         EXT3_INODES_PER_GROUP(sb),
2288                         sbi->s_mount_opt);
2289 +
2290         printk(KERN_INFO "EXT3 FS " EXT3FS_VERSION ", " EXT3FS_DATE " on %s, ",
2291                                 bdevname(sb->s_dev));
2292         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
2293 @@ -925,6 +927,7 @@
2294         return res;
2295  }
2296
2297 +
2298  struct super_block * ext3_read_super (struct super_block * sb, void * data,
2299                                       int silent)
2300  {
2301 @@ -1113,6 +1116,9 @@
2302         sbi->s_mount_state = le16_to_cpu(es->s_state);
2303         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
2304         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
2305 +       for (i=0; i < 4; i++)
2306 +               sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2307 +       sbi->s_def_hash_version = es->s_def_hash_version;
2308
2309         if (sbi->s_blocks_per_group > blocksize * 8) {
2310                 printk (KERN_ERR
2311 @@ -1821,6 +1827,7 @@
2312         exit_ext3_xattr();
2313  }
2314
2315 +EXPORT_SYMBOL(ext3_force_commit);
2316  EXPORT_SYMBOL(ext3_bread);
2317
2318  MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2319 Index: linux-2.4.19/include/linux/ext3_fs.h
2320 ===================================================================
2321 --- linux-2.4.19.orig/include/linux/ext3_fs.h   2004-04-23 17:53:55.000000000 -0400
2322 +++ linux-2.4.19/include/linux/ext3_fs.h        2004-04-23 18:26:27.000000000 -0400
2323 @@ -40,6 +40,11 @@
2324  #define EXT3FS_VERSION         "2.4-0.9.18"
2325
2326  /*
2327 + * Always enable hashed directories
2328 + */
2329 +#define CONFIG_EXT3_INDEX
2330 +
2331 +/*
2332   * Debug code
2333   */
2334  #ifdef EXT3FS_DEBUG
2335 @@ -414,8 +419,11 @@
2336  /*E0*/ __u32   s_journal_inum;         /* inode number of journal file */
2337         __u32   s_journal_dev;          /* device number of journal file */
2338         __u32   s_last_orphan;          /* start of list of inodes to delete */
2339 -
2340 -/*EC*/ __u32   s_reserved[197];        /* Padding to the end of the block */
2341 +       __u32   s_hash_seed[4];         /* HTREE hash seed */
2342 +       __u8    s_def_hash_version;     /* Default hash version to use */
2343 +       __u8    s_reserved_char_pad;
2344 +       __u16   s_reserved_word_pad;
2345 +       __u32   s_reserved[192];        /* Padding to the end of the block */
2346  };
2347
2348  #ifdef __KERNEL__
2349 @@ -552,9 +560,46 @@
2350  #define EXT3_DIR_ROUND                 (EXT3_DIR_PAD - 1)
2351  #define EXT3_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT3_DIR_ROUND) & \
2352                                          ~EXT3_DIR_ROUND)
2353 +/*
2354 + * Hash Tree Directory indexing
2355 + * (c) Daniel Phillips, 2001
2356 + */
2357 +
2358 +#ifdef CONFIG_EXT3_INDEX
2359 +  #define is_dx(dir) (EXT3_HAS_COMPAT_FEATURE(dir->i_sb, \
2360 +                                             EXT3_FEATURE_COMPAT_DIR_INDEX) && \
2361 +                     (EXT3_I(dir)->i_flags & EXT3_INDEX_FL))
2362 +#define EXT3_DIR_LINK_MAX(dir) (!is_dx(dir) && (dir)->i_nlink >= EXT3_LINK_MAX)
2363 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2 || (dir)->i_nlink == 1)
2364 +#else
2365 +  #define is_dx(dir) 0
2366 +#define EXT3_DIR_LINK_MAX(dir) ((dir)->i_nlink >= EXT3_LINK_MAX)
2367 +#define EXT3_DIR_LINK_EMPTY(dir) ((dir)->i_nlink == 2)
2368 +#endif
2369 +
2370 +/* Legal values for the dx_root hash_version field: */
2371 +
2372 +#define DX_HASH_LEGACY         0
2373 +#define DX_HASH_HALF_MD4       1
2374 +#define DX_HASH_TEA            2
2375 +
2376 +/* hash info structure used by the directory hash */
2377 +struct dx_hash_info
2378 +{
2379 +       u32             hash;
2380 +       u32             minor_hash;
2381 +       int             hash_version;
2382 +       u32             *seed;
2383 +};
2384
2385  #ifdef __KERNEL__
2386  /*
2387 + * Control parameters used by ext3_htree_next_block
2388 + */
2389 +#define HASH_NB_ALWAYS         1
2390 +
2391 +
2392 +/*
2393   * Describe an inode's exact location on disk and in memory
2394   */
2395  struct ext3_iloc
2396 @@ -564,6 +609,27 @@
2397         unsigned long block_group;
2398  };
2399
2400 +
2401 +/*
2402 + * This structure is stuffed into the struct file's private_data field
2403 + * for directories.  It is where we put information so that we can do
2404 + * readdir operations in hash tree order.
2405 + */
2406 +struct dir_private_info {
2407 +       rb_root_t       root;
2408 +       rb_node_t       *curr_node;
2409 +       struct fname    *extra_fname;
2410 +       loff_t          last_pos;
2411 +       __u32           curr_hash;
2412 +       __u32           curr_minor_hash;
2413 +       __u32           next_hash;
2414 +};
2415 +
2416 +/*
2417 + * Special error return code only used by dx_probe() and its callers.
2418 + */
2419 +#define ERR_BAD_DX_DIR -75000
2420 +
2421  /*
2422   * Function prototypes
2423   */
2424 @@ -591,11 +657,20 @@
2425
2426  /* dir.c */
2427  extern int ext3_check_dir_entry(const char *, struct inode *,
2428 -                               struct ext3_dir_entry_2 *, struct buffer_head *,
2429 -                               unsigned long);
2430 +                               struct ext3_dir_entry_2 *,
2431 +                               struct buffer_head *, unsigned long);
2432 +extern void ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
2433 +                                   __u32 minor_hash,
2434 +                                   struct ext3_dir_entry_2 *dirent);
2435 +extern void ext3_htree_free_dir_info(struct dir_private_info *p);
2436 +
2437  /* fsync.c */
2438  extern int ext3_sync_file (struct file *, struct dentry *, int);
2439
2440 +/* hash.c */
2441 +extern int ext3fs_dirhash(const char *name, int len, struct
2442 +                         dx_hash_info *hinfo);
2443 +
2444  /* ialloc.c */
2445  extern struct inode * ext3_new_inode (handle_t *, struct inode *, int);
2446  extern void ext3_free_inode (handle_t *, struct inode *);
2447 @@ -628,6 +703,8 @@
2448  /* namei.c */
2449  extern int ext3_orphan_add(handle_t *, struct inode *);
2450  extern int ext3_orphan_del(handle_t *, struct inode *);
2451 +extern int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
2452 +                               __u32 start_minor_hash, __u32 *next_hash);
2453
2454  /* super.c */
2455  extern void ext3_error (struct super_block *, const char *, const char *, ...)
2456 Index: linux-2.4.19/include/linux/ext3_fs_sb.h
2457 ===================================================================
2458 --- linux-2.4.19.orig/include/linux/ext3_fs_sb.h        2004-04-23 17:53:54.000000000 -0400
2459 +++ linux-2.4.19/include/linux/ext3_fs_sb.h     2004-04-23 18:26:27.000000000 -0400
2460 @@ -62,6 +62,8 @@
2461         int s_inode_size;
2462         int s_first_ino;
2463         u32 s_next_generation;
2464 +       u32 s_hash_seed[4];
2465 +       int s_def_hash_version;
2466
2467         /* Journaling */
2468         struct inode * s_journal_inode;
2469 Index: linux-2.4.19/include/linux/ext3_jbd.h
2470 ===================================================================
2471 --- linux-2.4.19.orig/include/linux/ext3_jbd.h  2004-04-23 17:53:54.000000000 -0400
2472 +++ linux-2.4.19/include/linux/ext3_jbd.h       2004-04-23 18:26:27.000000000 -0400
2473 @@ -69,6 +69,8 @@
2474
2475  #define EXT3_RESERVE_TRANS_BLOCKS      12
2476
2477 +#define EXT3_INDEX_EXTRA_TRANS_BLOCKS  8
2478 +
2479  int
2480  ext3_mark_iloc_dirty(handle_t *handle,
2481                      struct inode *inode,
2482 Index: linux-2.4.19/include/linux/rbtree.h
2483 ===================================================================
2484 --- linux-2.4.19.orig/include/linux/rbtree.h    2001-11-22 14:46:18.000000000 -0500
2485 +++ linux-2.4.19/include/linux/rbtree.h 2004-04-23 18:26:27.000000000 -0400
2486 @@ -120,6 +120,8 @@
2487
2488  extern void rb_insert_color(rb_node_t *, rb_root_t *);
2489  extern void rb_erase(rb_node_t *, rb_root_t *);
2490 +extern rb_node_t *rb_get_first(rb_root_t *root);
2491 +extern rb_node_t *rb_get_next(rb_node_t *n);
2492
2493  static inline void rb_link_node(rb_node_t * node, rb_node_t * parent, rb_node_t ** rb_link)
2494  {
2495 Index: linux-2.4.19/lib/rbtree.c
2496 ===================================================================
2497 --- linux-2.4.19.orig/lib/rbtree.c      2002-08-02 20:39:46.000000000 -0400
2498 +++ linux-2.4.19/lib/rbtree.c   2004-04-23 18:26:27.000000000 -0400
2499 @@ -17,6 +17,8 @@
2500    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
2501
2502    linux/lib/rbtree.c
2503 +
2504 +  rb_get_first and rb_get_next written by Theodore Ts'o, 9/8/2002
2505  */
2506
2507  #include <linux/rbtree.h>
2508 @@ -294,3 +296,43 @@
2509                 __rb_erase_color(child, parent, root);
2510  }
2511  EXPORT_SYMBOL(rb_erase);
2512 +
2513 +/*
2514 + * This function returns the first node (in sort order) of the tree.
2515 + */
2516 +rb_node_t *rb_get_first(rb_root_t *root)
2517 +{
2518 +       rb_node_t       *n;
2519 +
2520 +       n = root->rb_node;
2521 +       if (!n)
2522 +               return 0;
2523 +       while (n->rb_left)
2524 +               n = n->rb_left;
2525 +       return n;
2526 +}
2527 +EXPORT_SYMBOL(rb_get_first);
2528 +
2529 +/*
2530 + * Given a node, this function will return the next node in the tree.
2531 + */
2532 +rb_node_t *rb_get_next(rb_node_t *n)
2533 +{
2534 +       rb_node_t       *parent;
2535 +
2536 +       if (n->rb_right) {
2537 +               n = n->rb_right;
2538 +               while (n->rb_left)
2539 +                       n = n->rb_left;
2540 +               return n;
2541 +       } else {
2542 +               while ((parent = n->rb_parent)) {
2543 +                       if (n == parent->rb_left)
2544 +                               return parent;
2545 +                       n = parent;
2546 +               }
2547 +               return 0;
2548 +       }
2549 +}
2550 +EXPORT_SYMBOL(rb_get_next);
2551 +