ldiskfs/kernel_patches/patches/ext3-extents-2.6.18-vanilla.patch

   1 Index: linux-2.6.18.8/fs/ext3/dir.c
   2 ===================================================================
   3 --- linux-2.6.18.8.orig/fs/ext3/dir.c   2007-02-24 00:52:30.000000000 +0100
   4 +++ linux-2.6.18.8/fs/ext3/dir.c        2007-07-17 09:18:14.000000000 +0200
   5 @@ -131,8 +131,7 @@ static int ext3_readdir(struct file * fi
   6                 struct buffer_head *bh = NULL;
   7
   8                 map_bh.b_state = 0;
   9 -               err = ext3_get_blocks_handle(NULL, inode, blk, 1,
  10 -                                               &map_bh, 0, 0);
  11 +               err = ext3_get_blocks_wrap(NULL, inode, blk, 1, &map_bh, 0, 0);
  12                 if (err > 0) {
  13                         page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
  14                                 &filp->f_ra,
  15 Index: linux-2.6.18.8/fs/ext3/extents.c
  16 ===================================================================
  17 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  18 +++ linux-2.6.18.8/fs/ext3/extents.c    2007-07-17 11:08:59.000000000 +0200
  19 @@ -0,0 +1,2272 @@
  20 +/*
  21 + * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
  22 + * Written by Alex Tomas <alex@clusterfs.com>
  23 + *
  24 + * Architecture independence:
  25 + *   Copyright (c) 2005, Bull S.A.
  26 + *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
  27 + *
  28 + * This program is free software; you can redistribute it and/or modify
  29 + * it under the terms of the GNU General Public License version 2 as
  30 + * published by the Free Software Foundation.
  31 + *
  32 + * This program is distributed in the hope that it will be useful,
  33 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
  34 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  35 + * GNU General Public License for more details.
  36 + *
  37 + * You should have received a copy of the GNU General Public Licens
  38 + * along with this program; if not, write to the Free Software
  39 + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  40 + */
  41 +
  42 +/*
  43 + * Extents support for EXT3
  44 + *
  45 + * TODO:
  46 + *   - ext3*_error() should be used in some situations
  47 + *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  48 + *   - smart tree reduction
  49 + */
  50 +
  51 +#include <linux/module.h>
  52 +#include <linux/fs.h>
  53 +#include <linux/time.h>
  54 +#include <linux/ext3_jbd.h>
  55 +#include <linux/jbd.h>
  56 +#include <linux/smp_lock.h>
  57 +#include <linux/highuid.h>
  58 +#include <linux/pagemap.h>
  59 +#include <linux/quotaops.h>
  60 +#include <linux/string.h>
  61 +#include <linux/slab.h>
  62 +#include <linux/ext3_extents.h>
  63 +#include <asm/uaccess.h>
  64 +
  65 +
  66 +static handle_t *ext3_ext_journal_restart(handle_t *handle, int needed)
  67 +{
  68 +       int err;
  69 +
  70 +       if (handle->h_buffer_credits > needed)
  71 +               return handle;
  72 +       if (!ext3_journal_extend(handle, needed))
  73 +               return handle;
  74 +       err = ext3_journal_restart(handle, needed);
  75 +
  76 +       return handle;
  77 +}
  78 +
  79 +/*
  80 + * could return:
  81 + *  - EROFS
  82 + *  - ENOMEM
  83 + */
  84 +static int ext3_ext_get_access(handle_t *handle, struct inode *inode,
  85 +                               struct ext3_ext_path *path)
  86 +{
  87 +       if (path->p_bh) {
  88 +               /* path points to block */
  89 +               return ext3_journal_get_write_access(handle, path->p_bh);
  90 +       }
  91 +       /* path points to leaf/index in inode body */
  92 +       /* we use in-core data, no need to protect them */
  93 +       return 0;
  94 +}
  95 +
  96 +/*
  97 + * could return:
  98 + *  - EROFS
  99 + *  - ENOMEM
 100 + *  - EIO
 101 + */
 102 +static int ext3_ext_dirty(handle_t *handle, struct inode *inode,
 103 +                               struct ext3_ext_path *path)
 104 +{
 105 +       int err;
 106 +       if (path->p_bh) {
 107 +               /* path points to block */
 108 +               err = ext3_journal_dirty_metadata(handle, path->p_bh);
 109 +       } else {
 110 +               /* path points to leaf/index in inode body */
 111 +               err = ext3_mark_inode_dirty(handle, inode);
 112 +       }
 113 +       return err;
 114 +}
 115 +
 116 +static int ext3_ext_find_goal(struct inode *inode,
 117 +                             struct ext3_ext_path *path,
 118 +                             unsigned long block)
 119 +{
 120 +       struct ext3_inode_info *ei = EXT3_I(inode);
 121 +       unsigned long bg_start;
 122 +       unsigned long colour;
 123 +       int depth;
 124 +
 125 +       if (path) {
 126 +               struct ext3_extent *ex;
 127 +               depth = path->p_depth;
 128 +
 129 +               /* try to predict block placement */
 130 +               if ((ex = path[depth].p_ext))
 131 +                       return le32_to_cpu(ex->ee_start)
 132 +                                       + (block - le32_to_cpu(ex->ee_block));
 133 +
 134 +               /* it looks index is empty
 135 +                * try to find starting from index itself */
 136 +               if (path[depth].p_bh)
 137 +                       return path[depth].p_bh->b_blocknr;
 138 +       }
 139 +
 140 +       /* OK. use inode's group */
 141 +       bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
 142 +               le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
 143 +       colour = (current->pid % 16) *
 144 +                       (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 145 +       return bg_start + colour + block;
 146 +}
 147 +
 148 +static int
 149 +ext3_ext_new_block(handle_t *handle, struct inode *inode,
 150 +                       struct ext3_ext_path *path,
 151 +                       struct ext3_extent *ex, int *err)
 152 +{
 153 +       int goal, newblock;
 154 +
 155 +       goal = ext3_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 156 +       newblock = ext3_new_block(handle, inode, goal, err);
 157 +       return newblock;
 158 +}
 159 +
 160 +static inline int ext3_ext_space_block(struct inode *inode)
 161 +{
 162 +       int size;
 163 +
 164 +       size = (inode->i_sb->s_blocksize - sizeof(struct ext3_extent_header))
 165 +                       / sizeof(struct ext3_extent);
 166 +#ifdef AGRESSIVE_TEST
 167 +       if (size > 6)
 168 +               size = 6;
 169 +#endif
 170 +       return size;
 171 +}
 172 +
 173 +static inline int ext3_ext_space_block_idx(struct inode *inode)
 174 +{
 175 +       int size;
 176 +
 177 +       size = (inode->i_sb->s_blocksize - sizeof(struct ext3_extent_header))
 178 +                       / sizeof(struct ext3_extent_idx);
 179 +#ifdef AGRESSIVE_TEST
 180 +       if (size > 5)
 181 +               size = 5;
 182 +#endif
 183 +       return size;
 184 +}
 185 +
 186 +static inline int ext3_ext_space_root(struct inode *inode)
 187 +{
 188 +       int size;
 189 +
 190 +       size = sizeof(EXT3_I(inode)->i_data);
 191 +       size -= sizeof(struct ext3_extent_header);
 192 +       size /= sizeof(struct ext3_extent);
 193 +#ifdef AGRESSIVE_TEST
 194 +       if (size > 3)
 195 +               size = 3;
 196 +#endif
 197 +       return size;
 198 +}
 199 +
 200 +static inline int ext3_ext_space_root_idx(struct inode *inode)
 201 +{
 202 +       int size;
 203 +
 204 +       size = sizeof(EXT3_I(inode)->i_data);
 205 +       size -= sizeof(struct ext3_extent_header);
 206 +       size /= sizeof(struct ext3_extent_idx);
 207 +#ifdef AGRESSIVE_TEST
 208 +       if (size > 4)
 209 +               size = 4;
 210 +#endif
 211 +       return size;
 212 +}
 213 +
 214 +static inline int
 215 +ext3_ext_max_entries(struct inode *inode, int depth)
 216 +{
 217 +       int max;
 218 +
 219 +       if (depth == ext_depth(inode)) {
 220 +               if (depth == 0)
 221 +                       max = ext3_ext_space_root(inode);
 222 +               else
 223 +                       max = ext3_ext_space_root_idx(inode);
 224 +       } else {
 225 +               if (depth == 0)
 226 +                       max = ext3_ext_space_block(inode);
 227 +               else
 228 +                       max = ext3_ext_space_block_idx(inode);
 229 +       }
 230 +
 231 +       return max;
 232 +}
 233 +
 234 +static int __ext3_ext_check_header(const char *function, struct inode *inode,
 235 +                                       struct ext3_extent_header *eh,
 236 +                                       int depth)
 237 +{
 238 +       const char *error_msg = NULL;
 239 +       int max = 0;
 240 +
 241 +       if (unlikely(eh->eh_magic != cpu_to_le16(EXT3_EXT_MAGIC))) {
 242 +               error_msg = "invalid magic";
 243 +               goto corrupted;
 244 +       }
 245 +       if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 246 +               error_msg = "unexpected eh_depth";
 247 +               goto corrupted;
 248 +       }
 249 +       if (unlikely(eh->eh_max == 0)) {
 250 +               error_msg = "invalid eh_max";
 251 +               goto corrupted;
 252 +       }
 253 +       max = ext3_ext_max_entries(inode, depth);
 254 +#ifdef AGRESSIVE_TEST
 255 +       if (eh->eh_max > 3) {
 256 +               /* inode probably got extent without defining AGRESSIVE_TEST */
 257 +               max = eh->eh_max;
 258 +       }
 259 +#endif
 260 +       if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 261 +               error_msg = "too large eh_max";
 262 +               goto corrupted;
 263 +       }
 264 +       if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 265 +               error_msg = "invalid eh_entries";
 266 +               goto corrupted;
 267 +       }
 268 +       return 0;
 269 +
 270 +corrupted:
 271 +       ext3_error(inode->i_sb, function,
 272 +                       "bad header in inode #%lu: %s - magic %x, "
 273 +                       "entries %u, max %u(%u), depth %u(%u)",
 274 +                       inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
 275 +                       le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 276 +                       max, le16_to_cpu(eh->eh_depth), depth);
 277 +
 278 +       return -EIO;
 279 +}
 280 +
 281 +#define ext3_ext_check_header(inode,eh,depth)  \
 282 +       __ext3_ext_check_header(__FUNCTION__,inode,eh,depth)
 283 +
 284 +#ifdef EXT_DEBUG
 285 +static void ext3_ext_show_path(struct inode *inode, struct ext3_ext_path *path)
 286 +{
 287 +       int k, l = path->p_depth;
 288 +
 289 +       ext_debug(inode, "path:");
 290 +       for (k = 0; k <= l; k++, path++) {
 291 +               if (path->p_idx) {
 292 +                 ext_debug(inode, "  %d->%d", le32_to_cpu(path->p_idx->ei_block),
 293 +                           le32_to_cpu(path->p_idx->ei_leaf));
 294 +               } else if (path->p_ext) {
 295 +                       ext_debug(inode, "  %d:%d:%d",
 296 +                                 le32_to_cpu(path->p_ext->ee_block),
 297 +                                 le16_to_cpu(path->p_ext->ee_len),
 298 +                                 le32_to_cpu(path->p_ext->ee_start));
 299 +               } else
 300 +                       ext_debug(inode, "  []");
 301 +       }
 302 +       ext_debug(inode, "\n");
 303 +}
 304 +
 305 +static void ext3_ext_show_leaf(struct inode *inode, struct ext3_ext_path *path)
 306 +{
 307 +       int depth = ext_depth(inode);
 308 +       struct ext3_extent_header *eh;
 309 +       struct ext3_extent *ex;
 310 +       int i;
 311 +
 312 +       if (!path)
 313 +               return;
 314 +
 315 +       eh = path[depth].p_hdr;
 316 +       ex = EXT_FIRST_EXTENT(eh);
 317 +
 318 +       for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 319 +               ext_debug(inode, "%d:%d:%d ", le32_to_cpu(ex->ee_block),
 320 +                         le16_to_cpu(ex->ee_len),
 321 +                         le32_to_cpu(ex->ee_start));
 322 +       }
 323 +       ext_debug(inode, "\n");
 324 +}
 325 +#else
 326 +#define ext3_ext_show_path(inode,path)
 327 +#define ext3_ext_show_leaf(inode,path)
 328 +#endif
 329 +
 330 +static void ext3_ext_drop_refs(struct ext3_ext_path *path)
 331 +{
 332 +       int depth = path->p_depth;
 333 +       int i;
 334 +
 335 +       for (i = 0; i <= depth; i++, path++)
 336 +               if (path->p_bh) {
 337 +                       brelse(path->p_bh);
 338 +                       path->p_bh = NULL;
 339 +               }
 340 +}
 341 +
 342 +/*
 343 + * binary search for closest index by given block
 344 + * the header must be checked before calling this
 345 + */
 346 +static void
 347 +ext3_ext_binsearch_idx(struct inode *inode, struct ext3_ext_path *path, int block)
 348 +{
 349 +       struct ext3_extent_header *eh = path->p_hdr;
 350 +       struct ext3_extent_idx *r, *l, *m;
 351 +
 352 +       ext_debug(inode, "binsearch for %d(idx):  ", block);
 353 +
 354 +       l = EXT_FIRST_INDEX(eh) + 1;
 355 +       r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
 356 +       while (l <= r) {
 357 +               m = l + (r - l) / 2;
 358 +               if (block < le32_to_cpu(m->ei_block))
 359 +                       r = m - 1;
 360 +               else
 361 +                       l = m + 1;
 362 +               ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
 363 +                               m, m->ei_block, r, r->ei_block);
 364 +       }
 365 +
 366 +       path->p_idx = l - 1;
 367 +       ext_debug(inode, "  -> %d->%d ", le32_to_cpu(path->p_idx->ei_block),
 368 +                 le32_to_cpu(path->p_idx->ei_leaf));
 369 +
 370 +#ifdef CHECK_BINSEARCH
 371 +       {
 372 +               struct ext3_extent_idx *chix, *ix;
 373 +               int k;
 374 +
 375 +               chix = ix = EXT_FIRST_INDEX(eh);
 376 +               for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 377 +                 if (k != 0 &&
 378 +                     le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 379 +                               printk("k=%d, ix=0x%p, first=0x%p\n", k,
 380 +                                       ix, EXT_FIRST_INDEX(eh));
 381 +                               printk("%u <= %u\n",
 382 +                                      le32_to_cpu(ix->ei_block),
 383 +                                      le32_to_cpu(ix[-1].ei_block));
 384 +                       }
 385 +                       BUG_ON(k && le32_to_cpu(ix->ei_block)
 386 +                                          <= le32_to_cpu(ix[-1].ei_block));
 387 +                       if (block < le32_to_cpu(ix->ei_block))
 388 +                               break;
 389 +                       chix = ix;
 390 +               }
 391 +               BUG_ON(chix != path->p_idx);
 392 +       }
 393 +#endif
 394 +
 395 +}
 396 +
 397 +/*
 398 + * binary search for closest extent by given block
 399 + * the header must be checked before calling this
 400 + */
 401 +static void
 402 +ext3_ext_binsearch(struct inode *inode, struct ext3_ext_path *path, int block)
 403 +{
 404 +       struct ext3_extent_header *eh = path->p_hdr;
 405 +       struct ext3_extent *r, *l, *m;
 406 +
 407 +       if (eh->eh_entries == 0) {
 408 +               /*
 409 +                * this leaf is empty yet:
 410 +                *  we get such a leaf in split/add case
 411 +                */
 412 +               return;
 413 +       }
 414 +
 415 +       ext_debug(inode, "binsearch for %d:  ", block);
 416 +
 417 +       l = EXT_FIRST_EXTENT(eh) + 1;
 418 +       r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
 419 +
 420 +       while (l <= r) {
 421 +               m = l + (r - l) / 2;
 422 +               if (block < le32_to_cpu(m->ee_block))
 423 +                       r = m - 1;
 424 +               else
 425 +                       l = m + 1;
 426 +               ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
 427 +                               m, m->ee_block, r, r->ee_block);
 428 +       }
 429 +
 430 +       path->p_ext = l - 1;
 431 +       ext_debug(inode, "  -> %d:%d:%d ",
 432 +                       le32_to_cpu(path->p_ext->ee_block),
 433 +                       le32_to_cpu(path->p_ext->ee_start),
 434 +                       le16_to_cpu(path->p_ext->ee_len));
 435 +
 436 +#ifdef CHECK_BINSEARCH
 437 +       {
 438 +               struct ext3_extent *chex, *ex;
 439 +               int k;
 440 +
 441 +               chex = ex = EXT_FIRST_EXTENT(eh);
 442 +               for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 443 +                       BUG_ON(k && le32_to_cpu(ex->ee_block)
 444 +                                         <= le32_to_cpu(ex[-1].ee_block));
 445 +                       if (block < le32_to_cpu(ex->ee_block))
 446 +                               break;
 447 +                       chex = ex;
 448 +               }
 449 +               BUG_ON(chex != path->p_ext);
 450 +       }
 451 +#endif
 452 +
 453 +}
 454 +
 455 +int ext3_ext_tree_init(handle_t *handle, struct inode *inode)
 456 +{
 457 +       struct ext3_extent_header *eh;
 458 +
 459 +       eh = ext_inode_hdr(inode);
 460 +       eh->eh_depth = 0;
 461 +       eh->eh_entries = 0;
 462 +       eh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 463 +       eh->eh_max = cpu_to_le16(ext3_ext_space_root(inode));
 464 +       ext3_mark_inode_dirty(handle, inode);
 465 +       ext3_ext_invalidate_cache(inode);
 466 +       return 0;
 467 +}
 468 +
 469 +struct ext3_ext_path *
 470 +ext3_ext_find_extent(struct inode *inode, int block, struct ext3_ext_path *path)
 471 +{
 472 +       struct ext3_extent_header *eh;
 473 +       struct buffer_head *bh;
 474 +       short int depth, i, ppos = 0, alloc = 0;
 475 +
 476 +       eh = ext_inode_hdr(inode);
 477 +       i = depth = ext_depth(inode);
 478 +       if (ext3_ext_check_header(inode, eh, depth))
 479 +               return ERR_PTR(-EIO);
 480 +
 481 +       /* account possible depth increase */
 482 +       if (!path) {
 483 +               path = kmalloc(sizeof(struct ext3_ext_path) * (depth + 2),
 484 +                               GFP_NOFS);
 485 +               if (!path)
 486 +                       return ERR_PTR(-ENOMEM);
 487 +               alloc = 1;
 488 +       }
 489 +       memset(path, 0, sizeof(struct ext3_ext_path) * (depth + 1));
 490 +       path[0].p_hdr = eh;
 491 +
 492 +       /* walk through the tree */
 493 +       while (i) {
 494 +               ext_debug(inode, "depth %d: num %d, max %d\n",
 495 +                         ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 496 +
 497 +               ext3_ext_binsearch_idx(inode, path + ppos, block);
 498 +               path[ppos].p_block = le32_to_cpu(path[ppos].p_idx->ei_leaf);
 499 +               path[ppos].p_depth = i;
 500 +               path[ppos].p_ext = NULL;
 501 +
 502 +               bh = sb_bread(inode->i_sb, path[ppos].p_block);
 503 +               if (!bh)
 504 +                       goto err;
 505 +
 506 +               eh = ext_block_hdr(bh);
 507 +               ppos++;
 508 +               BUG_ON(ppos > depth);
 509 +               path[ppos].p_bh = bh;
 510 +               path[ppos].p_hdr = eh;
 511 +               i--;
 512 +
 513 +               if (ext3_ext_check_header(inode, eh, i))
 514 +                       goto err;
 515 +       }
 516 +
 517 +       path[ppos].p_depth = i;
 518 +       path[ppos].p_hdr = eh;
 519 +       path[ppos].p_ext = NULL;
 520 +       path[ppos].p_idx = NULL;
 521 +
 522 +       /* find extent */
 523 +       ext3_ext_binsearch(inode, path + ppos, block);
 524 +
 525 +       ext3_ext_show_path(inode, path);
 526 +
 527 +       return path;
 528 +
 529 +err:
 530 +       ext3_ext_drop_refs(path);
 531 +       if (alloc)
 532 +               kfree(path);
 533 +       return ERR_PTR(-EIO);
 534 +}
 535 +
 536 +/*
 537 + * insert new index [logical;ptr] into the block at cupr
 538 + * it check where to insert: before curp or after curp
 539 + */
 540 +static int ext3_ext_insert_index(handle_t *handle, struct inode *inode,
 541 +                               struct ext3_ext_path *curp,
 542 +                               int logical, int ptr)
 543 +{
 544 +       struct ext3_extent_idx *ix;
 545 +       int len, err;
 546 +
 547 +       if ((err = ext3_ext_get_access(handle, inode, curp)))
 548 +               return err;
 549 +
 550 +       BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
 551 +       len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 552 +       if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 553 +               /* insert after */
 554 +               if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 555 +                       len = (len - 1) * sizeof(struct ext3_extent_idx);
 556 +                       len = len < 0 ? 0 : len;
 557 +                       ext_debug(inode, "insert new index %d after: %d. "
 558 +                                       "move %d from 0x%p to 0x%p\n",
 559 +                                       logical, ptr, len,
 560 +                                       (curp->p_idx + 1), (curp->p_idx + 2));
 561 +                       memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 562 +               }
 563 +               ix = curp->p_idx + 1;
 564 +       } else {
 565 +               /* insert before */
 566 +               len = len * sizeof(struct ext3_extent_idx);
 567 +               len = len < 0 ? 0 : len;
 568 +               ext_debug(inode, "insert new index %d before: %d. "
 569 +                               "move %d from 0x%p to 0x%p\n",
 570 +                               logical, ptr, len,
 571 +                               curp->p_idx, (curp->p_idx + 1));
 572 +               memmove(curp->p_idx + 1, curp->p_idx, len);
 573 +               ix = curp->p_idx;
 574 +       }
 575 +
 576 +       ix->ei_block = cpu_to_le32(logical);
 577 +       ix->ei_leaf = cpu_to_le32(ptr);
 578 +       ix->ei_leaf_hi = ix->ei_unused = 0;
 579 +       curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
 580 +
 581 +       BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
 582 +                            > le16_to_cpu(curp->p_hdr->eh_max));
 583 +       BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
 584 +
 585 +       err = ext3_ext_dirty(handle, inode, curp);
 586 +       ext3_std_error(inode->i_sb, err);
 587 +
 588 +       return err;
 589 +}
 590 +
 591 +/*
 592 + * routine inserts new subtree into the path, using free index entry
 593 + * at depth 'at:
 594 + *  - allocates all needed blocks (new leaf and all intermediate index blocks)
 595 + *  - makes decision where to split
 596 + *  - moves remaining extens and index entries (right to the split point)
 597 + *    into the newly allocated blocks
 598 + *  - initialize subtree
 599 + */
 600 +static int ext3_ext_split(handle_t *handle, struct inode *inode,
 601 +                               struct ext3_ext_path *path,
 602 +                               struct ext3_extent *newext, int at)
 603 +{
 604 +       struct buffer_head *bh = NULL;
 605 +       int depth = ext_depth(inode);
 606 +       struct ext3_extent_header *neh;
 607 +       struct ext3_extent_idx *fidx;
 608 +       struct ext3_extent *ex;
 609 +       int i = at, k, m, a;
 610 +       unsigned long newblock, oldblock;
 611 +       __le32 border;
 612 +       int *ablocks = NULL; /* array of allocated blocks */
 613 +       int err = 0;
 614 +
 615 +       /* make decision: where to split? */
 616 +       /* FIXME: now desicion is simplest: at current extent */
 617 +
 618 +       /* if current leaf will be splitted, then we should use
 619 +        * border from split point */
 620 +       BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
 621 +       if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 622 +               border = path[depth].p_ext[1].ee_block;
 623 +               ext_debug(inode, "leaf will be splitted."
 624 +                               " next leaf starts at %d\n",
 625 +                                 le32_to_cpu(border));
 626 +       } else {
 627 +               border = newext->ee_block;
 628 +               ext_debug(inode, "leaf will be added."
 629 +                               " next leaf starts at %d\n",
 630 +                               le32_to_cpu(border));
 631 +       }
 632 +
 633 +       /*
 634 +        * if error occurs, then we break processing
 635 +        * and turn filesystem read-only. so, index won't
 636 +        * be inserted and tree will be in consistent
 637 +        * state. next mount will repair buffers too
 638 +        */
 639 +
 640 +       /*
 641 +        * get array to track all allocated blocks
 642 +        * we need this to handle errors and free blocks
 643 +        * upon them
 644 +        */
 645 +       ablocks = kmalloc(sizeof(unsigned long) * depth, GFP_NOFS);
 646 +       if (!ablocks)
 647 +               return -ENOMEM;
 648 +       memset(ablocks, 0, sizeof(unsigned long) * depth);
 649 +
 650 +       /* allocate all needed blocks */
 651 +       ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
 652 +       for (a = 0; a < depth - at; a++) {
 653 +               newblock = ext3_ext_new_block(handle, inode, path, newext, &err);
 654 +               if (newblock == 0)
 655 +                       goto cleanup;
 656 +               ablocks[a] = newblock;
 657 +       }
 658 +
 659 +       /* initialize new leaf */
 660 +       newblock = ablocks[--a];
 661 +       BUG_ON(newblock == 0);
 662 +       bh = sb_getblk(inode->i_sb, newblock);
 663 +       if (!bh) {
 664 +               err = -EIO;
 665 +               goto cleanup;
 666 +       }
 667 +       lock_buffer(bh);
 668 +
 669 +       if ((err = ext3_journal_get_create_access(handle, bh)))
 670 +               goto cleanup;
 671 +
 672 +       neh = ext_block_hdr(bh);
 673 +       neh->eh_entries = 0;
 674 +       neh->eh_max = cpu_to_le16(ext3_ext_space_block(inode));
 675 +       neh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 676 +       neh->eh_depth = 0;
 677 +       ex = EXT_FIRST_EXTENT(neh);
 678 +
 679 +       /* move remain of path[depth] to the new leaf */
 680 +       BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
 681 +       /* start copy from next extent */
 682 +       /* TODO: we could do it by single memmove */
 683 +       m = 0;
 684 +       path[depth].p_ext++;
 685 +       while (path[depth].p_ext <=
 686 +                       EXT_MAX_EXTENT(path[depth].p_hdr)) {
 687 +               ext_debug(inode, "move %d:%d:%d in new leaf %lu\n",
 688 +                               le32_to_cpu(path[depth].p_ext->ee_block),
 689 +                               le32_to_cpu(path[depth].p_ext->ee_start),
 690 +                               le16_to_cpu(path[depth].p_ext->ee_len),
 691 +                               newblock);
 692 +               /*memmove(ex++, path[depth].p_ext++,
 693 +                               sizeof(struct ext3_extent));
 694 +               neh->eh_entries++;*/
 695 +               path[depth].p_ext++;
 696 +               m++;
 697 +       }
 698 +       if (m) {
 699 +               memmove(ex, path[depth].p_ext-m, sizeof(struct ext3_extent)*m);
 700 +               neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
 701 +       }
 702 +
 703 +       set_buffer_uptodate(bh);
 704 +       unlock_buffer(bh);
 705 +
 706 +       if ((err = ext3_journal_dirty_metadata(handle, bh)))
 707 +               goto cleanup;
 708 +       brelse(bh);
 709 +       bh = NULL;
 710 +
 711 +       /* correct old leaf */
 712 +       if (m) {
 713 +               if ((err = ext3_ext_get_access(handle, inode, path + depth)))
 714 +                       goto cleanup;
 715 +               path[depth].p_hdr->eh_entries =
 716 +                    cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
 717 +               if ((err = ext3_ext_dirty(handle, inode, path + depth)))
 718 +                       goto cleanup;
 719 +
 720 +       }
 721 +
 722 +       /* create intermediate indexes */
 723 +       k = depth - at - 1;
 724 +       BUG_ON(k < 0);
 725 +       if (k)
 726 +               ext_debug(inode, "create %d intermediate indices\n", k);
 727 +       /* insert new index into current index block */
 728 +       /* current depth stored in i var */
 729 +       i = depth - 1;
 730 +       while (k--) {
 731 +               oldblock = newblock;
 732 +               newblock = ablocks[--a];
 733 +               bh = sb_getblk(inode->i_sb, newblock);
 734 +               if (!bh) {
 735 +                       err = -EIO;
 736 +                       goto cleanup;
 737 +               }
 738 +               lock_buffer(bh);
 739 +
 740 +               if ((err = ext3_journal_get_create_access(handle, bh)))
 741 +                       goto cleanup;
 742 +
 743 +               neh = ext_block_hdr(bh);
 744 +               neh->eh_entries = cpu_to_le16(1);
 745 +               neh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 746 +               neh->eh_max = cpu_to_le16(ext3_ext_space_block_idx(inode));
 747 +               neh->eh_depth = cpu_to_le16(depth - i);
 748 +               fidx = EXT_FIRST_INDEX(neh);
 749 +               fidx->ei_block = border;
 750 +               fidx->ei_leaf = cpu_to_le32(oldblock);
 751 +               fidx->ei_leaf_hi = fidx->ei_unused = 0;
 752 +
 753 +               ext_debug(inode, "int.index at %d (block %lu): %lu -> %lu\n", i,
 754 +                               newblock, (unsigned long) le32_to_cpu(border),
 755 +                               oldblock);
 756 +               /* copy indexes */
 757 +               m = 0;
 758 +               path[i].p_idx++;
 759 +
 760 +               ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
 761 +                               EXT_MAX_INDEX(path[i].p_hdr));
 762 +               BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
 763 +                               EXT_LAST_INDEX(path[i].p_hdr));
 764 +               while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
 765 +                       ext_debug(inode, "%d: move %d:%d in new index %lu\n", i,
 766 +                                       le32_to_cpu(path[i].p_idx->ei_block),
 767 +                                       le32_to_cpu(path[i].p_idx->ei_leaf),
 768 +                                       newblock);
 769 +                       /*memmove(++fidx, path[i].p_idx++,
 770 +                                       sizeof(struct ext3_extent_idx));
 771 +                       neh->eh_entries++;
 772 +                       BUG_ON(neh->eh_entries > neh->eh_max);*/
 773 +                       path[i].p_idx++;
 774 +                       m++;
 775 +               }
 776 +               if (m) {
 777 +                       memmove(++fidx, path[i].p_idx - m,
 778 +                               sizeof(struct ext3_extent_idx) * m);
 779 +                       neh->eh_entries =
 780 +                               cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
 781 +               }
 782 +               set_buffer_uptodate(bh);
 783 +               unlock_buffer(bh);
 784 +
 785 +               if ((err = ext3_journal_dirty_metadata(handle, bh)))
 786 +                       goto cleanup;
 787 +               brelse(bh);
 788 +               bh = NULL;
 789 +
 790 +               /* correct old index */
 791 +               if (m) {
 792 +                       err = ext3_ext_get_access(handle, inode, path + i);
 793 +                       if (err)
 794 +                               goto cleanup;
 795 +                       path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
 796 +                       err = ext3_ext_dirty(handle, inode, path + i);
 797 +                       if (err)
 798 +                               goto cleanup;
 799 +               }
 800 +
 801 +               i--;
 802 +       }
 803 +
 804 +       /* insert new index */
 805 +       if (err)
 806 +               goto cleanup;
 807 +
 808 +       err = ext3_ext_insert_index(handle, inode, path + at,
 809 +                                   le32_to_cpu(border), newblock);
 810 +
 811 +cleanup:
 812 +       if (bh) {
 813 +               if (buffer_locked(bh))
 814 +                       unlock_buffer(bh);
 815 +               brelse(bh);
 816 +       }
 817 +
 818 +       if (err) {
 819 +               /* free all allocated blocks in error case */
 820 +               for (i = 0; i < depth; i++) {
 821 +                       if (!ablocks[i])
 822 +                               continue;
 823 +                       ext3_free_blocks(handle, inode, ablocks[i], 1);
 824 +               }
 825 +       }
 826 +       kfree(ablocks);
 827 +
 828 +       return err;
 829 +}
 830 +
 831 +/*
 832 + * routine implements tree growing procedure:
 833 + *  - allocates new block
 834 + *  - moves top-level data (index block or leaf) into the new block
 835 + *  - initialize new top-level, creating index that points to the
 836 + *    just created block
 837 + */
 838 +static int ext3_ext_grow_indepth(handle_t *handle, struct inode *inode,
 839 +                                       struct ext3_ext_path *path,
 840 +                                       struct ext3_extent *newext)
 841 +{
 842 +       struct ext3_ext_path *curp = path;
 843 +       struct ext3_extent_header *neh;
 844 +       struct ext3_extent_idx *fidx;
 845 +       struct buffer_head *bh;
 846 +       unsigned long newblock;
 847 +       int err = 0;
 848 +
 849 +       newblock = ext3_ext_new_block(handle, inode, path, newext, &err);
 850 +       if (newblock == 0)
 851 +               return err;
 852 +
 853 +       bh = sb_getblk(inode->i_sb, newblock);
 854 +       if (!bh) {
 855 +               err = -EIO;
 856 +               ext3_std_error(inode->i_sb, err);
 857 +               return err;
 858 +       }
 859 +       lock_buffer(bh);
 860 +
 861 +       if ((err = ext3_journal_get_create_access(handle, bh))) {
 862 +               unlock_buffer(bh);
 863 +               goto out;
 864 +       }
 865 +
 866 +       /* move top-level index/leaf into new block */
 867 +       memmove(bh->b_data, curp->p_hdr, sizeof(EXT3_I(inode)->i_data));
 868 +
 869 +       /* set size of new block */
 870 +       neh = ext_block_hdr(bh);
 871 +       /* old root could have indexes or leaves
 872 +        * so calculate e_max right way */
 873 +       if (ext_depth(inode))
 874 +         neh->eh_max = cpu_to_le16(ext3_ext_space_block_idx(inode));
 875 +       else
 876 +         neh->eh_max = cpu_to_le16(ext3_ext_space_block(inode));
 877 +       neh->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 878 +       set_buffer_uptodate(bh);
 879 +       unlock_buffer(bh);
 880 +
 881 +       if ((err = ext3_journal_dirty_metadata(handle, bh)))
 882 +               goto out;
 883 +
 884 +       /* create index in new top-level index: num,max,pointer */
 885 +       if ((err = ext3_ext_get_access(handle, inode, curp)))
 886 +               goto out;
 887 +
 888 +       curp->p_hdr->eh_magic = cpu_to_le16(EXT3_EXT_MAGIC);
 889 +       curp->p_hdr->eh_max = cpu_to_le16(ext3_ext_space_root_idx(inode));
 890 +       curp->p_hdr->eh_entries = cpu_to_le16(1);
 891 +       curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
 892 +       /* FIXME: it works, but actually path[0] can be index */
 893 +       curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
 894 +       curp->p_idx->ei_leaf = cpu_to_le32(newblock);
 895 +       curp->p_idx->ei_leaf_hi = curp->p_idx->ei_unused = 0;
 896 +
 897 +       neh = ext_inode_hdr(inode);
 898 +       fidx = EXT_FIRST_INDEX(neh);
 899 +       ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %d\n",
 900 +                 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
 901 +                 le32_to_cpu(fidx->ei_block), le32_to_cpu(fidx->ei_leaf));
 902 +
 903 +       neh->eh_depth = cpu_to_le16(path->p_depth + 1);
 904 +       err = ext3_ext_dirty(handle, inode, curp);
 905 +out:
 906 +       brelse(bh);
 907 +
 908 +       return err;
 909 +}
 910 +
 911 +/*
 912 + * routine finds empty index and adds new leaf. if no free index found
 913 + * then it requests in-depth growing
 914 + */
 915 +static int ext3_ext_create_new_leaf(handle_t *handle, struct inode *inode,
 916 +                                       struct ext3_ext_path *path,
 917 +                                       struct ext3_extent *newext)
 918 +{
 919 +       struct ext3_ext_path *curp;
 920 +       int depth, i, err = 0;
 921 +
 922 +repeat:
 923 +       i = depth = ext_depth(inode);
 924 +
 925 +       /* walk up to the tree and look for free index entry */
 926 +       curp = path + depth;
 927 +       while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
 928 +               i--;
 929 +               curp--;
 930 +       }
 931 +
 932 +       /* we use already allocated block for index block
 933 +        * so, subsequent data blocks should be contigoues */
 934 +       if (EXT_HAS_FREE_INDEX(curp)) {
 935 +               /* if we found index with free entry, then use that
 936 +                * entry: create all needed subtree and add new leaf */
 937 +               err = ext3_ext_split(handle, inode, path, newext, i);
 938 +               if (err)
 939 +                       goto out;
 940 +
 941 +               /* refill path */
 942 +               ext3_ext_drop_refs(path);
 943 +               path = ext3_ext_find_extent(inode,
 944 +                                           le32_to_cpu(newext->ee_block),
 945 +                                           path);
 946 +               if (IS_ERR(path))
 947 +                       err = PTR_ERR(path);
 948 +       } else {
 949 +               /* tree is full, time to grow in depth */
 950 +               err = ext3_ext_grow_indepth(handle, inode, path, newext);
 951 +               if (err)
 952 +                       goto out;
 953 +
 954 +               /* refill path */
 955 +               ext3_ext_drop_refs(path);
 956 +               path = ext3_ext_find_extent(inode,
 957 +                                           le32_to_cpu(newext->ee_block),
 958 +                                           path);
 959 +               if (IS_ERR(path)) {
 960 +                       err = PTR_ERR(path);
 961 +                       goto out;
 962 +               }
 963 +
 964 +               /*
 965 +                * only first (depth 0 -> 1) produces free space
 966 +                * in all other cases we have to split growed tree
 967 +                */
 968 +               depth = ext_depth(inode);
 969 +               if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
 970 +                       /* now we need split */
 971 +                       goto repeat;
 972 +               }
 973 +       }
 974 +
 975 +out:
 976 +       return err;
 977 +}
 978 +
 979 +/*
 980 + * search the closest allocated block to the left for *logical
 981 + * and returns it at @logical + it's physical address at @phys
 982 + * if *logical is the smallest allocated block, the function
 983 + * returns 0 at @phys
 984 + * return value contains 0 (success) or error code
 985 + */
 986 +int
 987 +ext3_ext_search_left(struct inode *inode, struct ext3_ext_path *path,
 988 +                       unsigned long *logical, unsigned long *phys)
 989 +{
 990 +       struct ext3_extent_idx *ix;
 991 +       struct ext3_extent *ex;
 992 +       int depth;
 993 +
 994 +       BUG_ON(path == NULL);
 995 +       depth = path->p_depth;
 996 +       *phys = 0;
 997 +
 998 +       if (depth == 0 && path->p_ext == NULL)
 999 +               return 0;
1000 +
1001 +       /* usually extent in the path covers blocks smaller
1002 +        * then *logical, but it can be that extent is the
1003 +        * first one in the file */
1004 +
1005 +       ex = path[depth].p_ext;
1006 +       if (*logical < le32_to_cpu(ex->ee_block)) {
1007 +               BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1008 +               while (--depth >= 0) {
1009 +                       ix = path[depth].p_idx;
1010 +                       BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1011 +               }
1012 +               return 0;
1013 +       }
1014 +
1015 +       BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1016 +
1017 +       *logical = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1;
1018 +       *phys = le32_to_cpu(ex->ee_start) + le16_to_cpu(ex->ee_len) - 1;
1019 +       return 0;
1020 +}
1021 +EXPORT_SYMBOL(ext3_ext_search_left);
1022 +
1023 +/*
1024 + * search the closest allocated block to the right for *logical
1025 + * and returns it at @logical + it's physical address at @phys
1026 + * if *logical is the smallest allocated block, the function
1027 + * returns 0 at @phys
1028 + * return value contains 0 (success) or error code
1029 + */
1030 +int
1031 +ext3_ext_search_right(struct inode *inode, struct ext3_ext_path *path,
1032 +                       unsigned long *logical, unsigned long *phys)
1033 +{
1034 +       struct buffer_head *bh = NULL;
1035 +       struct ext3_extent_header *eh;
1036 +       struct ext3_extent_idx *ix;
1037 +       struct ext3_extent *ex;
1038 +       unsigned long block;
1039 +       int depth;
1040 +
1041 +       BUG_ON(path == NULL);
1042 +       depth = path->p_depth;
1043 +       *phys = 0;
1044 +
1045 +       if (depth == 0 && path->p_ext == NULL)
1046 +               return 0;
1047 +
1048 +       /* usually extent in the path covers blocks smaller
1049 +        * then *logical, but it can be that extent is the
1050 +        * first one in the file */
1051 +
1052 +       ex = path[depth].p_ext;
1053 +       if (*logical < le32_to_cpu(ex->ee_block)) {
1054 +               BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1055 +               while (--depth >= 0) {
1056 +                       ix = path[depth].p_idx;
1057 +                       BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1058 +               }
1059 +               *logical = le32_to_cpu(ex->ee_block);
1060 +               *phys = le32_to_cpu(ex->ee_start);
1061 +               return 0;
1062 +       }
1063 +
1064 +       BUG_ON(*logical < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
1065 +
1066 +       if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1067 +               /* next allocated block in this leaf */
1068 +               ex++;
1069 +               *logical = le32_to_cpu(ex->ee_block);
1070 +               *phys = le32_to_cpu(ex->ee_start);
1071 +               return 0;
1072 +       }
1073 +
1074 +       /* go up and search for index to the right */
1075 +       while (--depth >= 0) {
1076 +               ix = path[depth].p_idx;
1077 +               if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1078 +                       break;
1079 +       }
1080 +
1081 +       if (depth < 0) {
1082 +               /* we've gone up to the root and
1083 +                * found no index to the right */
1084 +               return 0;
1085 +       }
1086 +
1087 +       /* we've found index to the right, let's
1088 +        * follow it and find the closest allocated
1089 +        * block to the right */
1090 +       ix++;
1091 +       block = le32_to_cpu(ix->ei_leaf);
1092 +       while (++depth < path->p_depth) {
1093 +               bh = sb_bread(inode->i_sb, block);
1094 +               if (bh == NULL)
1095 +                       return -EIO;
1096 +               eh = ext_block_hdr(bh);
1097 +               if (ext3_ext_check_header(inode, eh, path->p_depth - depth)) {
1098 +                       brelse(bh);
1099 +                       return -EIO;
1100 +               }
1101 +               ix = EXT_FIRST_INDEX(eh);
1102 +               block = le32_to_cpu(ix->ei_leaf);
1103 +               brelse(bh);
1104 +       }
1105 +
1106 +       bh = sb_bread(inode->i_sb, block);
1107 +       if (bh == NULL)
1108 +               return -EIO;
1109 +       eh = ext_block_hdr(bh);
1110 +       if (ext3_ext_check_header(inode, eh, 0)) {
1111 +               brelse(bh);
1112 +               return -EIO;
1113 +       }
1114 +       ex = EXT_FIRST_EXTENT(eh);
1115 +       *logical = le32_to_cpu(ex->ee_block);
1116 +       *phys = le32_to_cpu(ex->ee_start);
1117 +       brelse(bh);
1118 +       return 0;
1119 +
1120 +}
1121 +EXPORT_SYMBOL(ext3_ext_search_right);
1122 +
1123 +
1124 +
1125 +/*
1126 + * returns allocated block in subsequent extent or EXT_MAX_BLOCK
1127 + * NOTE: it consider block number from index entry as
1128 + * allocated block. thus, index entries have to be consistent
1129 + * with leafs
1130 + */
1131 +static unsigned long
1132 +ext3_ext_next_allocated_block(struct ext3_ext_path *path)
1133 +{
1134 +       int depth;
1135 +
1136 +       BUG_ON(path == NULL);
1137 +       depth = path->p_depth;
1138 +
1139 +       if (depth == 0 && path->p_ext == NULL)
1140 +               return EXT_MAX_BLOCK;
1141 +
1142 +       while (depth >= 0) {
1143 +               if (depth == path->p_depth) {
1144 +                       /* leaf */
1145 +                       if (path[depth].p_ext !=
1146 +                                       EXT_LAST_EXTENT(path[depth].p_hdr))
1147 +                         return le32_to_cpu(path[depth].p_ext[1].ee_block);
1148 +               } else {
1149 +                       /* index */
1150 +                       if (path[depth].p_idx !=
1151 +                                       EXT_LAST_INDEX(path[depth].p_hdr))
1152 +                         return le32_to_cpu(path[depth].p_idx[1].ei_block);
1153 +               }
1154 +               depth--;
1155 +       }
1156 +
1157 +       return EXT_MAX_BLOCK;
1158 +}
1159 +
1160 +/*
1161 + * returns first allocated block from next leaf or EXT_MAX_BLOCK
1162 + */
1163 +static unsigned ext3_ext_next_leaf_block(struct inode *inode,
1164 +                                               struct ext3_ext_path *path)
1165 +{
1166 +       int depth;
1167 +
1168 +       BUG_ON(path == NULL);
1169 +       depth = path->p_depth;
1170 +
1171 +       /* zero-tree has no leaf blocks at all */
1172 +       if (depth == 0)
1173 +               return EXT_MAX_BLOCK;
1174 +
1175 +       /* go to index block */
1176 +       depth--;
1177 +
1178 +       while (depth >= 0) {
1179 +               if (path[depth].p_idx !=
1180 +                               EXT_LAST_INDEX(path[depth].p_hdr))
1181 +                 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1182 +               depth--;
1183 +       }
1184 +
1185 +       return EXT_MAX_BLOCK;
1186 +}
1187 +
1188 +/*
1189 + * if leaf gets modified and modified extent is first in the leaf
1190 + * then we have to correct all indexes above
1191 + * TODO: do we need to correct tree in all cases?
1192 + */
1193 +int ext3_ext_correct_indexes(handle_t *handle, struct inode *inode,
1194 +                               struct ext3_ext_path *path)
1195 +{
1196 +       struct ext3_extent_header *eh;
1197 +       int depth = ext_depth(inode);
1198 +       struct ext3_extent *ex;
1199 +       __le32 border;
1200 +       int k, err = 0;
1201 +
1202 +       eh = path[depth].p_hdr;
1203 +       ex = path[depth].p_ext;
1204 +       BUG_ON(ex == NULL);
1205 +       BUG_ON(eh == NULL);
1206 +
1207 +       if (depth == 0) {
1208 +               /* there is no tree at all */
1209 +               return 0;
1210 +       }
1211 +
1212 +       if (ex != EXT_FIRST_EXTENT(eh)) {
1213 +               /* we correct tree if first leaf got modified only */
1214 +               return 0;
1215 +       }
1216 +
1217 +       /*
1218 +        * TODO: we need correction if border is smaller then current one
1219 +        */
1220 +       k = depth - 1;
1221 +       border = path[depth].p_ext->ee_block;
1222 +       if ((err = ext3_ext_get_access(handle, inode, path + k)))
1223 +               return err;
1224 +       path[k].p_idx->ei_block = border;
1225 +       if ((err = ext3_ext_dirty(handle, inode, path + k)))
1226 +               return err;
1227 +
1228 +       while (k--) {
1229 +               /* change all left-side indexes */
1230 +               if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1231 +                       break;
1232 +               if ((err = ext3_ext_get_access(handle, inode, path + k)))
1233 +                       break;
1234 +               path[k].p_idx->ei_block = border;
1235 +               if ((err = ext3_ext_dirty(handle, inode, path + k)))
1236 +                       break;
1237 +       }
1238 +
1239 +       return err;
1240 +}
1241 +
1242 +static int inline
1243 +ext3_can_extents_be_merged(struct inode *inode, struct ext3_extent *ex1,
1244 +                               struct ext3_extent *ex2)
1245 +{
1246 +       /* FIXME: 48bit support */
1247 +        if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len)
1248 +           != le32_to_cpu(ex2->ee_block))
1249 +               return 0;
1250 +
1251 +#ifdef AGRESSIVE_TEST
1252 +       if (le16_to_cpu(ex1->ee_len) >= 4)
1253 +               return 0;
1254 +#endif
1255 +
1256 +        if (le32_to_cpu(ex1->ee_start) + le16_to_cpu(ex1->ee_len)
1257 +                       == le32_to_cpu(ex2->ee_start))
1258 +               return 1;
1259 +       return 0;
1260 +}
1261 +
1262 +/*
1263 + * this routine tries to merge requsted extent into the existing
1264 + * extent or inserts requested extent as new one into the tree,
1265 + * creating new leaf in no-space case
1266 + */
1267 +int ext3_ext_insert_extent(handle_t *handle, struct inode *inode,
1268 +                               struct ext3_ext_path *path,
1269 +                               struct ext3_extent *newext)
1270 +{
1271 +       struct ext3_extent_header * eh;
1272 +       struct ext3_extent *ex, *fex;
1273 +       struct ext3_extent *nearex; /* nearest extent */
1274 +       struct ext3_ext_path *npath = NULL;
1275 +       int depth, len, err, next;
1276 +
1277 +       BUG_ON(newext->ee_len == 0);
1278 +       depth = ext_depth(inode);
1279 +       ex = path[depth].p_ext;
1280 +       BUG_ON(path[depth].p_hdr == NULL);
1281 +
1282 +       /* try to insert block into found extent and return */
1283 +       if (ex && ext3_can_extents_be_merged(inode, ex, newext)) {
1284 +               ext_debug(inode, "append %d block to %d:%d (from %d)\n",
1285 +                               le16_to_cpu(newext->ee_len),
1286 +                               le32_to_cpu(ex->ee_block),
1287 +                               le16_to_cpu(ex->ee_len),
1288 +                               le32_to_cpu(ex->ee_start));
1289 +               if ((err = ext3_ext_get_access(handle, inode, path + depth)))
1290 +                       return err;
1291 +               ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len)
1292 +                                        + le16_to_cpu(newext->ee_len));
1293 +               eh = path[depth].p_hdr;
1294 +               nearex = ex;
1295 +               goto merge;
1296 +       }
1297 +
1298 +repeat:
1299 +       depth = ext_depth(inode);
1300 +       eh = path[depth].p_hdr;
1301 +       if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1302 +               goto has_space;
1303 +
1304 +       /* probably next leaf has space for us? */
1305 +       fex = EXT_LAST_EXTENT(eh);
1306 +       next = ext3_ext_next_leaf_block(inode, path);
1307 +       if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1308 +           && next != EXT_MAX_BLOCK) {
1309 +               ext_debug(inode, "next leaf block - %d\n", next);
1310 +               BUG_ON(npath != NULL);
1311 +               npath = ext3_ext_find_extent(inode, next, NULL);
1312 +               if (IS_ERR(npath))
1313 +                       return PTR_ERR(npath);
1314 +               BUG_ON(npath->p_depth != path->p_depth);
1315 +               eh = npath[depth].p_hdr;
1316 +               if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1317 +                       ext_debug(inode, "next leaf isnt full(%d)\n",
1318 +                                 le16_to_cpu(eh->eh_entries));
1319 +                       path = npath;
1320 +                       goto repeat;
1321 +               }
1322 +               ext_debug(inode, "next leaf has no free space(%d,%d)\n",
1323 +                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1324 +       }
1325 +
1326 +       /*
1327 +        * there is no free space in found leaf
1328 +        * we're gonna add new leaf in the tree
1329 +        */
1330 +       err = ext3_ext_create_new_leaf(handle, inode, path, newext);
1331 +       if (err)
1332 +               goto cleanup;
1333 +       depth = ext_depth(inode);
1334 +       eh = path[depth].p_hdr;
1335 +
1336 +has_space:
1337 +       nearex = path[depth].p_ext;
1338 +
1339 +       if ((err = ext3_ext_get_access(handle, inode, path + depth)))
1340 +               goto cleanup;
1341 +
1342 +       if (!nearex) {
1343 +               /* there is no extent in this leaf, create first one */
1344 +               ext_debug(inode, "first extent in the leaf: %d:%d:%d\n",
1345 +                               le32_to_cpu(newext->ee_block),
1346 +                               le32_to_cpu(newext->ee_start),
1347 +                               le16_to_cpu(newext->ee_len));
1348 +               path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1349 +       } else if (le32_to_cpu(newext->ee_block)
1350 +                          > le32_to_cpu(nearex->ee_block)) {
1351 +               /* BUG_ON(newext->ee_block == nearex->ee_block); */
1352 +               if (nearex != EXT_LAST_EXTENT(eh)) {
1353 +                       len = EXT_MAX_EXTENT(eh) - nearex;
1354 +                       len = (len - 1) * sizeof(struct ext3_extent);
1355 +                       len = len < 0 ? 0 : len;
1356 +                       ext_debug(inode, "insert %d:%d:%d after: nearest 0x%p, "
1357 +                                       "move %d from 0x%p to 0x%p\n",
1358 +                                       le32_to_cpu(newext->ee_block),
1359 +                                       le32_to_cpu(newext->ee_start),
1360 +                                       le16_to_cpu(newext->ee_len),
1361 +                                       nearex, len, nearex + 1, nearex + 2);
1362 +                       memmove(nearex + 2, nearex + 1, len);
1363 +               }
1364 +               path[depth].p_ext = nearex + 1;
1365 +       } else {
1366 +               BUG_ON(newext->ee_block == nearex->ee_block);
1367 +               len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext3_extent);
1368 +               len = len < 0 ? 0 : len;
1369 +               ext_debug(inode, "insert %d:%d:%d before: nearest 0x%p, "
1370 +                               "move %d from 0x%p to 0x%p\n",
1371 +                               le32_to_cpu(newext->ee_block),
1372 +                               le32_to_cpu(newext->ee_start),
1373 +                               le16_to_cpu(newext->ee_len),
1374 +                               nearex, len, nearex + 1, nearex + 2);
1375 +               memmove(nearex + 1, nearex, len);
1376 +               path[depth].p_ext = nearex;
1377 +       }
1378 +
1379 +       eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1380 +       nearex = path[depth].p_ext;
1381 +       nearex->ee_block = newext->ee_block;
1382 +       nearex->ee_start = newext->ee_start;
1383 +       nearex->ee_len = newext->ee_len;
1384 +       /* FIXME: support for large fs */
1385 +       nearex->ee_start_hi = 0;
1386 +
1387 +merge:
1388 +       /* try to merge extents to the right */
1389 +       while (nearex < EXT_LAST_EXTENT(eh)) {
1390 +               if (!ext3_can_extents_be_merged(inode, nearex, nearex + 1))
1391 +                       break;
1392 +               /* merge with next extent! */
1393 +               nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len)
1394 +                                            + le16_to_cpu(nearex[1].ee_len));
1395 +               if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1396 +                       len = (EXT_LAST_EXTENT(eh) - nearex - 1)
1397 +                                       * sizeof(struct ext3_extent);
1398 +                       memmove(nearex + 1, nearex + 2, len);
1399 +               }
1400 +               eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1401 +               BUG_ON(eh->eh_entries == 0);
1402 +       }
1403 +
1404 +       /* try to merge extents to the left */
1405 +
1406 +       /* time to correct all indexes above */
1407 +       err = ext3_ext_correct_indexes(handle, inode, path);
1408 +       if (err)
1409 +               goto cleanup;
1410 +
1411 +       err = ext3_ext_dirty(handle, inode, path + depth);
1412 +
1413 +cleanup:
1414 +       if (npath) {
1415 +               ext3_ext_drop_refs(npath);
1416 +               kfree(npath);
1417 +       }
1418 +       ext3_ext_tree_changed(inode);
1419 +       ext3_ext_invalidate_cache(inode);
1420 +       return err;
1421 +}
1422 +
1423 +int ext3_ext_walk_space(struct inode *inode, unsigned long block,
1424 +                       unsigned long num, ext_prepare_callback func,
1425 +                       void *cbdata)
1426 +{
1427 +       struct ext3_ext_path *path = NULL;
1428 +       struct ext3_ext_cache cbex;
1429 +       struct ext3_extent *ex;
1430 +       unsigned long next, start = 0, end = 0;
1431 +       unsigned long last = block + num;
1432 +       int depth, exists, err = 0;
1433 +
1434 +       BUG_ON(func == NULL);
1435 +       BUG_ON(inode == NULL);
1436 +
1437 +       while (block < last && block != EXT_MAX_BLOCK) {
1438 +               num = last - block;
1439 +               /* find extent for this block */
1440 +               path = ext3_ext_find_extent(inode, block, path);
1441 +               if (IS_ERR(path)) {
1442 +                       err = PTR_ERR(path);
1443 +                       path = NULL;
1444 +                       break;
1445 +               }
1446 +
1447 +               depth = ext_depth(inode);
1448 +               BUG_ON(path[depth].p_hdr == NULL);
1449 +               ex = path[depth].p_ext;
1450 +               next = ext3_ext_next_allocated_block(path);
1451 +
1452 +               exists = 0;
1453 +               if (!ex) {
1454 +                       /* there is no extent yet, so try to allocate
1455 +                        * all requested space */
1456 +                       start = block;
1457 +                       end = block + num;
1458 +               } else if (le32_to_cpu(ex->ee_block) > block) {
1459 +                       /* need to allocate space before found extent */
1460 +                       start = block;
1461 +                       end = le32_to_cpu(ex->ee_block);
1462 +                       if (block + num < end)
1463 +                               end = block + num;
1464 +               } else if (block >=
1465 +                            le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) {
1466 +                       /* need to allocate space after found extent */
1467 +                       start = block;
1468 +                       end = block + num;
1469 +                       if (end >= next)
1470 +                               end = next;
1471 +               } else if (block >= le32_to_cpu(ex->ee_block)) {
1472 +                       /*
1473 +                        * some part of requested space is covered
1474 +                        * by found extent
1475 +                        */
1476 +                       start = block;
1477 +                       end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len);
1478 +                       if (block + num < end)
1479 +                               end = block + num;
1480 +                       exists = 1;
1481 +               } else {
1482 +                       BUG();
1483 +               }
1484 +               BUG_ON(end <= start);
1485 +
1486 +               if (!exists) {
1487 +                       cbex.ec_block = start;
1488 +                       cbex.ec_len = end - start;
1489 +                       cbex.ec_start = 0;
1490 +                       cbex.ec_type = EXT3_EXT_CACHE_GAP;
1491 +               } else {
1492 +                       cbex.ec_block = le32_to_cpu(ex->ee_block);
1493 +                       cbex.ec_len = le16_to_cpu(ex->ee_len);
1494 +                       cbex.ec_start = le32_to_cpu(ex->ee_start);
1495 +                       cbex.ec_type = EXT3_EXT_CACHE_EXTENT;
1496 +               }
1497 +
1498 +               BUG_ON(cbex.ec_len == 0);
1499 +               err = func(inode, path, &cbex, cbdata);
1500 +               ext3_ext_drop_refs(path);
1501 +
1502 +               if (err < 0)
1503 +                       break;
1504 +               if (err == EXT_REPEAT)
1505 +                       continue;
1506 +               else if (err == EXT_BREAK) {
1507 +                       err = 0;
1508 +                       break;
1509 +               }
1510 +
1511 +               if (ext_depth(inode) != depth) {
1512 +                       /* depth was changed. we have to realloc path */
1513 +                       kfree(path);
1514 +                       path = NULL;
1515 +               }
1516 +
1517 +               block = cbex.ec_block + cbex.ec_len;
1518 +       }
1519 +
1520 +       if (path) {
1521 +               ext3_ext_drop_refs(path);
1522 +               kfree(path);
1523 +       }
1524 +
1525 +       return err;
1526 +}
1527 +
1528 +static inline void
1529 +ext3_ext_put_in_cache(struct inode *inode, __u32 block,
1530 +                       __u32 len, __u32 start, int type)
1531 +{
1532 +       struct ext3_ext_cache *cex;
1533 +       BUG_ON(len == 0);
1534 +       cex = &EXT3_I(inode)->i_cached_extent;
1535 +       cex->ec_type = type;
1536 +       cex->ec_block = block;
1537 +       cex->ec_len = len;
1538 +       cex->ec_start = start;
1539 +}
1540 +
1541 +/*
1542 + * this routine calculate boundaries of the gap requested block fits into
1543 + * and cache this gap
1544 + */
1545 +static inline void
1546 +ext3_ext_put_gap_in_cache(struct inode *inode, struct ext3_ext_path *path,
1547 +                               unsigned long block)
1548 +{
1549 +       int depth = ext_depth(inode);
1550 +       unsigned long lblock, len;
1551 +       struct ext3_extent *ex;
1552 +
1553 +       ex = path[depth].p_ext;
1554 +       if (ex == NULL) {
1555 +               /* there is no extent yet, so gap is [0;-] */
1556 +               lblock = 0;
1557 +               len = EXT_MAX_BLOCK;
1558 +               ext_debug(inode, "cache gap(whole file):");
1559 +       } else if (block < le32_to_cpu(ex->ee_block)) {
1560 +               lblock = block;
1561 +               len = le32_to_cpu(ex->ee_block) - block;
1562 +               ext_debug(inode, "cache gap(before): %lu [%lu:%lu]",
1563 +                               (unsigned long) block,
1564 +                               (unsigned long) le32_to_cpu(ex->ee_block),
1565 +                               (unsigned long) le16_to_cpu(ex->ee_len));
1566 +       } else if (block >= le32_to_cpu(ex->ee_block)
1567 +                           + le16_to_cpu(ex->ee_len)) {
1568 +               lblock = le32_to_cpu(ex->ee_block)
1569 +                        + le16_to_cpu(ex->ee_len);
1570 +               len = ext3_ext_next_allocated_block(path);
1571 +               ext_debug(inode, "cache gap(after): [%lu:%lu] %lu",
1572 +                               (unsigned long) le32_to_cpu(ex->ee_block),
1573 +                               (unsigned long) le16_to_cpu(ex->ee_len),
1574 +                               (unsigned long) block);
1575 +               BUG_ON(len == lblock);
1576 +               len = len - lblock;
1577 +       } else {
1578 +               lblock = len = 0;
1579 +               BUG();
1580 +       }
1581 +
1582 +       ext_debug(inode, " -> %lu:%lu\n", (unsigned long) lblock, len);
1583 +       ext3_ext_put_in_cache(inode, lblock, len, 0, EXT3_EXT_CACHE_GAP);
1584 +}
1585 +
1586 +static inline int
1587 +ext3_ext_in_cache(struct inode *inode, unsigned long block,
1588 +                       struct ext3_extent *ex)
1589 +{
1590 +       struct ext3_ext_cache *cex;
1591 +
1592 +       cex = &EXT3_I(inode)->i_cached_extent;
1593 +
1594 +       /* has cache valid data? */
1595 +       if (cex->ec_type == EXT3_EXT_CACHE_NO)
1596 +               return EXT3_EXT_CACHE_NO;
1597 +
1598 +       BUG_ON(cex->ec_type != EXT3_EXT_CACHE_GAP &&
1599 +                       cex->ec_type != EXT3_EXT_CACHE_EXTENT);
1600 +       if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1601 +               ex->ee_block = cpu_to_le32(cex->ec_block);
1602 +               ex->ee_start = cpu_to_le32(cex->ec_start);
1603 +               ex->ee_start_hi = 0;
1604 +               ex->ee_len = cpu_to_le16(cex->ec_len);
1605 +               ext_debug(inode, "%lu cached by %lu:%lu:%lu\n",
1606 +                               (unsigned long) block,
1607 +                               (unsigned long) cex->ec_block,
1608 +                               (unsigned long) cex->ec_len,
1609 +                               (unsigned long) cex->ec_start);
1610 +               return cex->ec_type;
1611 +       }
1612 +
1613 +       /* not in cache */
1614 +       return EXT3_EXT_CACHE_NO;
1615 +}
1616 +
1617 +/*
1618 + * routine removes index from the index block
1619 + * it's used in truncate case only. thus all requests are for
1620 + * last index in the block only
1621 + */
1622 +int ext3_ext_rm_idx(handle_t *handle, struct inode *inode,
1623 +                       struct ext3_ext_path *path)
1624 +{
1625 +       struct buffer_head *bh;
1626 +       int err;
1627 +       unsigned long leaf;
1628 +
1629 +       /* free index block */
1630 +       path--;
1631 +       leaf = le32_to_cpu(path->p_idx->ei_leaf);
1632 +       BUG_ON(path->p_hdr->eh_entries == 0);
1633 +       if ((err = ext3_ext_get_access(handle, inode, path)))
1634 +               return err;
1635 +       path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1636 +       if ((err = ext3_ext_dirty(handle, inode, path)))
1637 +               return err;
1638 +       ext_debug(inode, "index is empty, remove it, free block %lu\n", leaf);
1639 +       bh = sb_find_get_block(inode->i_sb, leaf);
1640 +       ext3_forget(handle, 1, inode, bh, leaf);
1641 +       ext3_free_blocks(handle, inode, leaf, 1);
1642 +       return err;
1643 +}
1644 +
1645 +/*
1646 + * This routine returns max. credits extent tree can consume.
1647 + * It should be OK for low-performance paths like ->writepage()
1648 + * To allow many writing process to fit a single transaction,
1649 + * caller should calculate credits under truncate_mutex and
1650 + * pass actual path.
1651 + */
1652 +int inline ext3_ext_calc_credits_for_insert(struct inode *inode,
1653 +                                               struct ext3_ext_path *path)
1654 +{
1655 +       int depth, needed;
1656 +
1657 +       if (path) {
1658 +               /* probably there is space in leaf? */
1659 +               depth = ext_depth(inode);
1660 +               if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1661 +                               < le16_to_cpu(path[depth].p_hdr->eh_max))
1662 +                       return 1;
1663 +       }
1664 +
1665 +       /*
1666 +        * given 32bit logical block (4294967296 blocks), max. tree
1667 +        * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1668 +        * let's also add one more level for imbalance.
1669 +        */
1670 +       depth = 5;
1671 +
1672 +       /* allocation of new data block(s) */
1673 +       needed = 2;
1674 +
1675 +       /*
1676 +        * tree can be full, so it'd need to grow in depth:
1677 +        * we need one credit to modify old root, credits for
1678 +        * new root will be added in split accounting
1679 +        */
1680 +       needed += 1;
1681 +
1682 +       /*
1683 +        * Index split can happen, we'd need:
1684 +        *    allocate intermediate indexes (bitmap + group)
1685 +        *  + change two blocks at each level, but root (already included)
1686 +        */
1687 +       needed += (depth * 2) + (depth * 2);
1688 +
1689 +       /* any allocation modifies superblock */
1690 +       needed += 1;
1691 +
1692 +       return needed;
1693 +}
1694 +
1695 +static int ext3_remove_blocks(handle_t *handle, struct inode *inode,
1696 +                               struct ext3_extent *ex,
1697 +                               unsigned long from, unsigned long to)
1698 +{
1699 +       struct buffer_head *bh;
1700 +       int i;
1701 +
1702 +#ifdef EXTENTS_STATS
1703 +       {
1704 +               struct ext3_sb_info *sbi = EXT3_SB(inode->i_sb);
1705 +               unsigned short ee_len =  le16_to_cpu(ex->ee_len);
1706 +               spin_lock(&sbi->s_ext_stats_lock);
1707 +               sbi->s_ext_blocks += ee_len;
1708 +               sbi->s_ext_extents++;
1709 +               if (ee_len < sbi->s_ext_min)
1710 +                       sbi->s_ext_min = ee_len;
1711 +               if (ee_len > sbi->s_ext_max)
1712 +                       sbi->s_ext_max = ee_len;
1713 +               if (ext_depth(inode) > sbi->s_depth_max)
1714 +                       sbi->s_depth_max = ext_depth(inode);
1715 +               spin_unlock(&sbi->s_ext_stats_lock);
1716 +       }
1717 +#endif
1718 +       if (from >= le32_to_cpu(ex->ee_block)
1719 +           && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1720 +               /* tail removal */
1721 +               unsigned long num, start;
1722 +               num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from;
1723 +               start = le32_to_cpu(ex->ee_start) + le16_to_cpu(ex->ee_len) - num;
1724 +               ext_debug(inode, "free last %lu blocks starting %lu\n", num, start);
1725 +               for (i = 0; i < num; i++) {
1726 +                       bh = sb_find_get_block(inode->i_sb, start + i);
1727 +                       ext3_forget(handle, 0, inode, bh, start + i);
1728 +               }
1729 +               ext3_free_blocks(handle, inode, start, num);
1730 +       } else if (from == le32_to_cpu(ex->ee_block)
1731 +                  && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1732 +               printk("strange request: removal %lu-%lu from %u:%u\n",
1733 +                      from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1734 +       } else {
1735 +               printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1736 +                      from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1737 +       }
1738 +       return 0;
1739 +}
1740 +
1741 +static int
1742 +ext3_ext_rm_leaf(handle_t *handle, struct inode *inode,
1743 +               struct ext3_ext_path *path, unsigned long start)
1744 +{
1745 +       int err = 0, correct_index = 0;
1746 +       int depth = ext_depth(inode), credits;
1747 +       struct ext3_extent_header *eh;
1748 +       unsigned a, b, block, num;
1749 +       unsigned long ex_ee_block;
1750 +       unsigned short ex_ee_len;
1751 +       struct ext3_extent *ex;
1752 +
1753 +       /* the header must be checked already in ext3_ext_remove_space() */
1754 +       ext_debug(inode, "truncate since %lu in leaf\n", start);
1755 +       if (!path[depth].p_hdr)
1756 +               path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1757 +       eh = path[depth].p_hdr;
1758 +       BUG_ON(eh == NULL);
1759 +
1760 +       /* find where to start removing */
1761 +       ex = EXT_LAST_EXTENT(eh);
1762 +
1763 +       ex_ee_block = le32_to_cpu(ex->ee_block);
1764 +       ex_ee_len = le16_to_cpu(ex->ee_len);
1765 +
1766 +       while (ex >= EXT_FIRST_EXTENT(eh) &&
1767 +                       ex_ee_block + ex_ee_len > start) {
1768 +               ext_debug(inode, "remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1769 +               path[depth].p_ext = ex;
1770 +
1771 +               a = ex_ee_block > start ? ex_ee_block : start;
1772 +               b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1773 +                       ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1774 +
1775 +               ext_debug(inode, "  border %u:%u\n", a, b);
1776 +
1777 +               if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1778 +                       block = 0;
1779 +                       num = 0;
1780 +                       BUG();
1781 +               } else if (a != ex_ee_block) {
1782 +                       /* remove tail of the extent */
1783 +                       block = ex_ee_block;
1784 +                       num = a - block;
1785 +               } else if (b != ex_ee_block + ex_ee_len - 1) {
1786 +                       /* remove head of the extent */
1787 +                       block = a;
1788 +                       num = b - a;
1789 +                       /* there is no "make a hole" API yet */
1790 +                       BUG();
1791 +               } else {
1792 +                       /* remove whole extent: excellent! */
1793 +                       block = ex_ee_block;
1794 +                       num = 0;
1795 +                       BUG_ON(a != ex_ee_block);
1796 +                       BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1797 +               }
1798 +
1799 +               /* at present, extent can't cross block group */
1800 +               /* leaf + bitmap + group desc + sb + inode */
1801 +               credits = 5;
1802 +               if (ex == EXT_FIRST_EXTENT(eh)) {
1803 +                       correct_index = 1;
1804 +                       credits += (ext_depth(inode)) + 1;
1805 +               }
1806 +#ifdef CONFIG_QUOTA
1807 +               credits += 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
1808 +#endif
1809 +
1810 +               handle = ext3_ext_journal_restart(handle, credits);
1811 +               if (IS_ERR(handle)) {
1812 +                       err = PTR_ERR(handle);
1813 +                       goto out;
1814 +               }
1815 +
1816 +               err = ext3_ext_get_access(handle, inode, path + depth);
1817 +               if (err)
1818 +                       goto out;
1819 +
1820 +               err = ext3_remove_blocks(handle, inode, ex, a, b);
1821 +               if (err)
1822 +                       goto out;
1823 +
1824 +               if (num == 0) {
1825 +                       /* this extent is removed entirely mark slot unused */
1826 +                       ex->ee_start = ex->ee_start_hi = 0;
1827 +                       eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1828 +               }
1829 +
1830 +               ex->ee_block = cpu_to_le32(block);
1831 +               ex->ee_len = cpu_to_le16(num);
1832 +
1833 +               err = ext3_ext_dirty(handle, inode, path + depth);
1834 +               if (err)
1835 +                       goto out;
1836 +
1837 +               ext_debug(inode, "new extent: %u:%u:%u\n", block, num,
1838 +                               le32_to_cpu(ex->ee_start));
1839 +               ex--;
1840 +               ex_ee_block = le32_to_cpu(ex->ee_block);
1841 +               ex_ee_len = le16_to_cpu(ex->ee_len);
1842 +       }
1843 +
1844 +       if (correct_index && eh->eh_entries)
1845 +               err = ext3_ext_correct_indexes(handle, inode, path);
1846 +
1847 +       /* if this leaf is free, then we should
1848 +        * remove it from index block above */
1849 +       if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1850 +               err = ext3_ext_rm_idx(handle, inode, path + depth);
1851 +
1852 +out:
1853 +       return err;
1854 +}
1855 +
1856 +/*
1857 + * returns 1 if current index have to be freed (even partial)
1858 + */
1859 +static int inline
1860 +ext3_ext_more_to_rm(struct ext3_ext_path *path)
1861 +{
1862 +       BUG_ON(path->p_idx == NULL);
1863 +
1864 +       if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1865 +               return 0;
1866 +
1867 +       /*
1868 +        * if truncate on deeper level happened it it wasn't partial
1869 +        * so we have to consider current index for truncation
1870 +        */
1871 +       if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1872 +               return 0;
1873 +       return 1;
1874 +}
1875 +
1876 +int ext3_ext_remove_space(struct inode *inode, unsigned long start)
1877 +{
1878 +       struct super_block *sb = inode->i_sb;
1879 +       int depth = ext_depth(inode);
1880 +       struct ext3_ext_path *path;
1881 +       handle_t *handle;
1882 +       int i = 0, err = 0;
1883 +
1884 +       ext_debug(inode, "truncate since %lu\n", start);
1885 +
1886 +       /* probably first extent we're gonna free will be last in block */
1887 +       handle = ext3_journal_start(inode, depth + 1);
1888 +       if (IS_ERR(handle))
1889 +               return PTR_ERR(handle);
1890 +
1891 +       ext3_ext_invalidate_cache(inode);
1892 +
1893 +       /*
1894 +        * we start scanning from right side freeing all the blocks
1895 +        * after i_size and walking into the deep
1896 +        */
1897 +       path = kmalloc(sizeof(struct ext3_ext_path) * (depth + 1), GFP_KERNEL);
1898 +       if (path == NULL) {
1899 +               ext3_journal_stop(handle);
1900 +               return -ENOMEM;
1901 +       }
1902 +       memset(path, 0, sizeof(struct ext3_ext_path) * (depth + 1));
1903 +       path[0].p_hdr = ext_inode_hdr(inode);
1904 +       if (ext3_ext_check_header(inode, path[0].p_hdr, depth)) {
1905 +               err = -EIO;
1906 +               goto out;
1907 +       }
1908 +       path[0].p_depth = depth;
1909 +
1910 +       while (i >= 0 && err == 0) {
1911 +               if (i == depth) {
1912 +                       /* this is leaf block */
1913 +                       err = ext3_ext_rm_leaf(handle, inode, path, start);
1914 +                       /* root level have p_bh == NULL, brelse() eats this */
1915 +                       brelse(path[i].p_bh);
1916 +                       path[i].p_bh = NULL;
1917 +                       i--;
1918 +                       continue;
1919 +               }
1920 +
1921 +               /* this is index block */
1922 +               if (!path[i].p_hdr) {
1923 +                       ext_debug(inode, "initialize header\n");
1924 +                       path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1925 +               }
1926 +
1927 +               if (!path[i].p_idx) {
1928 +                       /* this level hasn't touched yet */
1929 +                       path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1930 +                       path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1931 +                       ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
1932 +                                 path[i].p_hdr,
1933 +                                 le16_to_cpu(path[i].p_hdr->eh_entries));
1934 +               } else {
1935 +                       /* we've already was here, see at next index */
1936 +                       path[i].p_idx--;
1937 +               }
1938 +
1939 +               ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
1940 +                               i, EXT_FIRST_INDEX(path[i].p_hdr),
1941 +                               path[i].p_idx);
1942 +               if (ext3_ext_more_to_rm(path + i)) {
1943 +                       struct buffer_head *bh;
1944 +                       /* go to the next level */
1945 +                       ext_debug(inode, "move to level %d (block %d)\n",
1946 +                                 i + 1, le32_to_cpu(path[i].p_idx->ei_leaf));
1947 +                       memset(path + i + 1, 0, sizeof(*path));
1948 +                       bh = sb_bread(sb, le32_to_cpu(path[i].p_idx->ei_leaf));
1949 +                       if (!bh) {
1950 +                               /* should we reset i_size? */
1951 +                               err = -EIO;
1952 +                               break;
1953 +                       }
1954 +                       BUG_ON(i + 1 > depth);
1955 +                       if (ext3_ext_check_header(inode, ext_block_hdr(bh),
1956 +                                                       depth - i - 1)) {
1957 +                               err = -EIO;
1958 +                               break;
1959 +                       }
1960 +                       path[i+1].p_bh = bh;
1961 +
1962 +                       /* put actual number of indexes to know is this
1963 +                        * number got changed at the next iteration */
1964 +                       path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1965 +                       i++;
1966 +               } else {
1967 +                       /* we finish processing this index, go up */
1968 +                       if (path[i].p_hdr->eh_entries == 0 && i > 0) {
1969 +                               /* index is empty, remove it
1970 +                                * handle must be already prepared by the
1971 +                                * truncatei_leaf() */
1972 +                               err = ext3_ext_rm_idx(handle, inode, path + i);
1973 +                       }
1974 +                       /* root level have p_bh == NULL, brelse() eats this */
1975 +                       brelse(path[i].p_bh);
1976 +                       path[i].p_bh = NULL;
1977 +                       i--;
1978 +                       ext_debug(inode, "return to level %d\n", i);
1979 +               }
1980 +       }
1981 +
1982 +       /* TODO: flexible tree reduction should be here */
1983 +       if (path->p_hdr->eh_entries == 0) {
1984 +               /*
1985 +                * truncate to zero freed all the tree
1986 +                * so, we need to correct eh_depth
1987 +                */
1988 +               err = ext3_ext_get_access(handle, inode, path);
1989 +               if (err == 0) {
1990 +                       ext_inode_hdr(inode)->eh_depth = 0;
1991 +                       ext_inode_hdr(inode)->eh_max =
1992 +                               cpu_to_le16(ext3_ext_space_root(inode));
1993 +                       err = ext3_ext_dirty(handle, inode, path);
1994 +               }
1995 +       }
1996 +out:
1997 +       ext3_ext_tree_changed(inode);
1998 +       ext3_ext_drop_refs(path);
1999 +       kfree(path);
2000 +       ext3_journal_stop(handle);
2001 +
2002 +       return err;
2003 +}
2004 +
2005 +/*
2006 + * called at mount time
2007 + */
2008 +void ext3_ext_init(struct super_block *sb)
2009 +{
2010 +       /*
2011 +        * possible initialization would be here
2012 +        */
2013 +
2014 +       if (test_opt(sb, EXTENTS)) {
2015 +               printk("EXT3-fs: file extents enabled");
2016 +#ifdef AGRESSIVE_TEST
2017 +               printk(", agressive tests");
2018 +#endif
2019 +#ifdef CHECK_BINSEARCH
2020 +               printk(", check binsearch");
2021 +#endif
2022 +#ifdef EXTENTS_STATS
2023 +               printk(", stats");
2024 +#endif
2025 +               printk("\n");
2026 +#ifdef EXTENTS_STATS
2027 +               spin_lock_init(&EXT3_SB(sb)->s_ext_stats_lock);
2028 +               EXT3_SB(sb)->s_ext_min = 1 << 30;
2029 +               EXT3_SB(sb)->s_ext_max = 0;
2030 +#endif
2031 +       }
2032 +}
2033 +
2034 +/*
2035 + * called at umount time
2036 + */
2037 +void ext3_ext_release(struct super_block *sb)
2038 +{
2039 +       if (!test_opt(sb, EXTENTS))
2040 +               return;
2041 +
2042 +#ifdef EXTENTS_STATS
2043 +       if (EXT3_SB(sb)->s_ext_blocks && EXT3_SB(sb)->s_ext_extents) {
2044 +               struct ext3_sb_info *sbi = EXT3_SB(sb);
2045 +               printk(KERN_ERR "EXT3-fs: %lu blocks in %lu extents (%lu ave)\n",
2046 +                       sbi->s_ext_blocks, sbi->s_ext_extents,
2047 +                       sbi->s_ext_blocks / sbi->s_ext_extents);
2048 +               printk(KERN_ERR "EXT3-fs: extents: %lu min, %lu max, max depth %lu\n",
2049 +                       sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2050 +       }
2051 +#endif
2052 +}
2053 +
2054 +int ext3_ext_get_blocks(handle_t *handle, struct inode *inode, sector_t iblock,
2055 +                       unsigned long max_blocks, struct buffer_head *bh_result,
2056 +                       int create, int extend_disksize)
2057 +{
2058 +       struct ext3_ext_path *path = NULL;
2059 +       struct ext3_extent newex, *ex;
2060 +       int goal, newblock, err = 0, depth;
2061 +       unsigned long allocated = 0;
2062 +       unsigned long next;
2063 +
2064 +       __clear_bit(BH_New, &bh_result->b_state);
2065 +       ext_debug(inode, "blocks %d/%lu requested for inode %u\n", (int) iblock,
2066 +                       max_blocks, (unsigned) inode->i_ino);
2067 +       mutex_lock(&EXT3_I(inode)->truncate_mutex);
2068 +
2069 +       /* check in cache */
2070 +       if ((goal = ext3_ext_in_cache(inode, iblock, &newex))) {
2071 +               if (goal == EXT3_EXT_CACHE_GAP) {
2072 +                       if (!create) {
2073 +                               /* block isn't allocated yet and
2074 +                                * user don't want to allocate it */
2075 +                               goto out2;
2076 +                       }
2077 +                       /* we should allocate requested block */
2078 +               } else if (goal == EXT3_EXT_CACHE_EXTENT) {
2079 +                       /* block is already allocated */
2080 +                       newblock = iblock
2081 +                                  - le32_to_cpu(newex.ee_block)
2082 +                                  + le32_to_cpu(newex.ee_start);
2083 +                       /* number of remain blocks in the extent */
2084 +                       BUG_ON(iblock < le32_to_cpu(newex.ee_block));
2085 +                       allocated = le16_to_cpu(newex.ee_len) -
2086 +                                       (iblock - le32_to_cpu(newex.ee_block));
2087 +                       goto out;
2088 +               } else {
2089 +                       BUG();
2090 +               }
2091 +       }
2092 +
2093 +       /* find extent for this block */
2094 +       path = ext3_ext_find_extent(inode, iblock, NULL);
2095 +       if (IS_ERR(path)) {
2096 +               err = PTR_ERR(path);
2097 +               path = NULL;
2098 +               goto out2;
2099 +       }
2100 +
2101 +       depth = ext_depth(inode);
2102 +
2103 +       /*
2104 +        * consistent leaf must not be empty
2105 +        * this situations is possible, though, _during_ tree modification
2106 +        * this is why assert can't be put in ext3_ext_find_extent()
2107 +        */
2108 +       BUG_ON(path[depth].p_ext == NULL && depth != 0);
2109 +
2110 +       if ((ex = path[depth].p_ext)) {
2111 +               unsigned long ee_block = le32_to_cpu(ex->ee_block);
2112 +               unsigned long ee_start = le32_to_cpu(ex->ee_start);
2113 +               unsigned short ee_len  = le16_to_cpu(ex->ee_len);
2114 +               /* if found exent covers block, simple return it */
2115 +               if (iblock >= ee_block && iblock < ee_block + ee_len) {
2116 +                       newblock = iblock - ee_block + ee_start;
2117 +                       /* number of remain blocks in the extent */
2118 +                       allocated = ee_len - (iblock - ee_block);
2119 +                       ext_debug(inode, "%d fit into %lu:%d -> %d\n", (int) iblock,
2120 +                                       ee_block, ee_len, newblock);
2121 +                       ext3_ext_put_in_cache(inode, ee_block, ee_len,
2122 +                                               ee_start, EXT3_EXT_CACHE_EXTENT);
2123 +                       goto out;
2124 +               }
2125 +       }
2126 +
2127 +       /*
2128 +        * requested block isn't allocated yet
2129 +        * we couldn't try to create block if create flag is zero
2130 +        */
2131 +       if (!create) {
2132 +               /* put just found gap into cache to speedup subsequest reqs */
2133 +               ext3_ext_put_gap_in_cache(inode, path, iblock);
2134 +               goto out2;
2135 +       }
2136 +
2137 +       /*
2138 +        * Okay, we need to do block allocation.  Lazily initialize the block
2139 +        * allocation info here if necessary
2140 +        */
2141 +       if (S_ISREG(inode->i_mode) && (!EXT3_I(inode)->i_block_alloc_info))
2142 +               ext3_init_block_alloc_info(inode);
2143 +
2144 +       /* find next allocated block so that we know how many
2145 +        * blocks we can allocate without ovelapping next extent */
2146 +       BUG_ON(iblock < le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len));
2147 +       next = ext3_ext_next_allocated_block(path);
2148 +       BUG_ON(next <= iblock);
2149 +       allocated = next - iblock;
2150 +       if (allocated > max_blocks)
2151 +               allocated = max_blocks;
2152 +
2153 +       /* allocate new block */
2154 +       goal = ext3_ext_find_goal(inode, path, iblock);
2155 +       newblock = ext3_new_blocks(handle, inode, goal, &allocated, &err);
2156 +       if (!newblock)
2157 +               goto out2;
2158 +       ext_debug(inode, "allocate new block: goal %d, found %d/%lu\n",
2159 +                       goal, newblock, allocated);
2160 +
2161 +       /* try to insert new extent into found leaf and return */
2162 +       newex.ee_block = cpu_to_le32(iblock);
2163 +       newex.ee_start = cpu_to_le32(newblock);
2164 +       newex.ee_start_hi = 0;
2165 +       newex.ee_len = cpu_to_le16(allocated);
2166 +       err = ext3_ext_insert_extent(handle, inode, path, &newex);
2167 +       if (err) {
2168 +               /* free data blocks we just allocated */
2169 +               ext3_free_blocks(handle, inode, le32_to_cpu(newex.ee_start),
2170 +                               le16_to_cpu(newex.ee_len));
2171 +               goto out2;
2172 +       }
2173 +
2174 +       if (extend_disksize && inode->i_size > EXT3_I(inode)->i_disksize)
2175 +               EXT3_I(inode)->i_disksize = inode->i_size;
2176 +
2177 +       /* previous routine could use block we allocated */
2178 +       newblock = le32_to_cpu(newex.ee_start);
2179 +       __set_bit(BH_New, &bh_result->b_state);
2180 +
2181 +       ext3_ext_put_in_cache(inode, iblock, allocated, newblock,
2182 +                               EXT3_EXT_CACHE_EXTENT);
2183 +out:
2184 +       if (allocated > max_blocks)
2185 +               allocated = max_blocks;
2186 +       ext3_ext_show_leaf(inode, path);
2187 +       __set_bit(BH_Mapped, &bh_result->b_state);
2188 +       bh_result->b_bdev = inode->i_sb->s_bdev;
2189 +       bh_result->b_blocknr = newblock;
2190 +       bh_result->b_size = (allocated << inode->i_blkbits);
2191 +out2:
2192 +       if (path) {
2193 +               ext3_ext_drop_refs(path);
2194 +               kfree(path);
2195 +       }
2196 +       mutex_unlock(&EXT3_I(inode)->truncate_mutex);
2197 +
2198 +       return err ? err : allocated;
2199 +}
2200 +
2201 +void ext3_ext_truncate(struct inode * inode, struct page *page)
2202 +{
2203 +       struct address_space *mapping = inode->i_mapping;
2204 +       struct super_block *sb = inode->i_sb;
2205 +       unsigned long last_block;
2206 +       handle_t *handle;
2207 +       int err = 0;
2208 +
2209 +       /*
2210 +        * probably first extent we're gonna free will be last in block
2211 +        */
2212 +       err = ext3_writepage_trans_blocks(inode) + 3;
2213 +       handle = ext3_journal_start(inode, err);
2214 +       if (IS_ERR(handle)) {
2215 +               if (page) {
2216 +                       clear_highpage(page);
2217 +                       flush_dcache_page(page);
2218 +                       unlock_page(page);
2219 +                       page_cache_release(page);
2220 +               }
2221 +               return;
2222 +       }
2223 +
2224 +       if (page)
2225 +               ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2226 +
2227 +       mutex_lock(&EXT3_I(inode)->truncate_mutex);
2228 +       ext3_ext_invalidate_cache(inode);
2229 +
2230 +       /*
2231 +        * TODO: optimization is possible here
2232 +        * probably we need not scaning at all,
2233 +        * because page truncation is enough
2234 +        */
2235 +       if (ext3_orphan_add(handle, inode))
2236 +               goto out_stop;
2237 +
2238 +       /* we have to know where to truncate from in crash case */
2239 +       EXT3_I(inode)->i_disksize = inode->i_size;
2240 +       ext3_mark_inode_dirty(handle, inode);
2241 +
2242 +       last_block = (inode->i_size + sb->s_blocksize - 1)
2243 +                       >> EXT3_BLOCK_SIZE_BITS(sb);
2244 +       err = ext3_ext_remove_space(inode, last_block);
2245 +
2246 +       /* In a multi-transaction truncate, we only make the final
2247 +        * transaction synchronous */
2248 +       if (IS_SYNC(inode))
2249 +               handle->h_sync = 1;
2250 +
2251 +out_stop:
2252 +       /*
2253 +        * If this was a simple ftruncate(), and the file will remain alive
2254 +        * then we need to clear up the orphan record which we created above.
2255 +        * However, if this was a real unlink then we were called by
2256 +        * ext3_delete_inode(), and we allow that function to clean up the
2257 +        * orphan info for us.
2258 +        */
2259 +       if (inode->i_nlink)
2260 +               ext3_orphan_del(handle, inode);
2261 +
2262 +       mutex_unlock(&EXT3_I(inode)->truncate_mutex);
2263 +       ext3_journal_stop(handle);
2264 +}
2265 +
2266 +/*
2267 + * this routine calculate max number of blocks we could modify
2268 + * in order to allocate new block for an inode
2269 + */
2270 +int ext3_ext_writepage_trans_blocks(struct inode *inode, int num)
2271 +{
2272 +       int needed;
2273 +
2274 +       needed = ext3_ext_calc_credits_for_insert(inode, NULL);
2275 +
2276 +       /* caller want to allocate num blocks, but note it includes sb */
2277 +       needed = needed * num - (num - 1);
2278 +
2279 +#ifdef CONFIG_QUOTA
2280 +       needed += 2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
2281 +#endif
2282 +
2283 +       return needed;
2284 +}
2285 +
2286 +EXPORT_SYMBOL(ext3_mark_inode_dirty);
2287 +EXPORT_SYMBOL(ext3_ext_invalidate_cache);
2288 +EXPORT_SYMBOL(ext3_ext_insert_extent);
2289 +EXPORT_SYMBOL(ext3_ext_walk_space);
2290 +EXPORT_SYMBOL(ext3_ext_find_goal);
2291 +EXPORT_SYMBOL(ext3_ext_calc_credits_for_insert);
2292 Index: linux-2.6.18.8/fs/ext3/ialloc.c
2293 ===================================================================
2294 --- linux-2.6.18.8.orig/fs/ext3/ialloc.c        2007-07-17 09:18:09.000000000 +0200
2295 +++ linux-2.6.18.8/fs/ext3/ialloc.c     2007-07-17 11:08:09.000000000 +0200
2296 @@ -652,6 +652,17 @@ got:
2297                 ext3_std_error(sb, err);
2298                 goto fail_free_drop;
2299         }
2300 +       if (test_opt(sb, EXTENTS) && S_ISREG(inode->i_mode)) {
2301 +               EXT3_I(inode)->i_flags |= EXT3_EXTENTS_FL;
2302 +               ext3_ext_tree_init(handle, inode);
2303 +               if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_EXTENTS)) {
2304 +                       err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
2305 +                       if (err) goto fail;
2306 +                       EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_EXTENTS);
2307 +                       BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "call ext3_journal_dirty_metadata");
2308 +                       err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
2309 +               }
2310 +       }
2311
2312         ext3_debug("allocating inode %lu\n", inode->i_ino);
2313         goto really_out;
2314 Index: linux-2.6.18.8/fs/ext3/inode.c
2315 ===================================================================
2316 --- linux-2.6.18.8.orig/fs/ext3/inode.c 2007-07-17 09:18:12.000000000 +0200
2317 +++ linux-2.6.18.8/fs/ext3/inode.c      2007-07-17 11:08:11.000000000 +0200
2318 @@ -40,8 +40,6 @@
2319  #include "iopen.h"
2320  #include "acl.h"
2321
2322 -static int ext3_writepage_trans_blocks(struct inode *inode);
2323 -
2324  /*
2325   * Test whether an inode is a fast symlink.
2326   */
2327 @@ -804,6 +802,7 @@ int ext3_get_blocks_handle(handle_t *han
2328         ext3_fsblk_t first_block = 0;
2329
2330
2331 +       J_ASSERT(!(EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL));
2332         J_ASSERT(handle != NULL || create == 0);
2333         depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
2334
2335 @@ -984,12 +983,10 @@ static int ext3_get_block(struct inode *
2336
2337  get_block:
2338         if (ret == 0) {
2339 -               ret = ext3_get_blocks_handle(handle, inode, iblock,
2340 +               ret = ext3_get_blocks_wrap(handle, inode, iblock,
2341                                         max_blocks, bh_result, create, 0);
2342 -               if (ret > 0) {
2343 -                       bh_result->b_size = (ret << inode->i_blkbits);
2344 +               if (ret > 0)
2345                         ret = 0;
2346 -               }
2347         }
2348         return ret;
2349  }
2350 @@ -1008,7 +1005,7 @@ struct buffer_head *ext3_getblk(handle_t
2351         dummy.b_state = 0;
2352         dummy.b_blocknr = -1000;
2353         buffer_trace_init(&dummy.b_history);
2354 -       err = ext3_get_blocks_handle(handle, inode, block, 1,
2355 +       err = ext3_get_blocks_wrap(handle, inode, block, 1,
2356                                         &dummy, create, 1);
2357         /*
2358          * ext3_get_blocks_handle() returns number of blocks
2359 @@ -1759,7 +1756,7 @@ void ext3_set_aops(struct inode *inode)
2360   * This required during truncate. We need to physically zero the tail end
2361   * of that block so it doesn't yield old data if the file is later grown.
2362   */
2363 -static int ext3_block_truncate_page(handle_t *handle, struct page *page,
2364 +int ext3_block_truncate_page(handle_t *handle, struct page *page,
2365                 struct address_space *mapping, loff_t from)
2366  {
2367         ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2368 @@ -2263,6 +2260,9 @@ void ext3_truncate(struct inode *inode)
2369                         return;
2370         }
2371
2372 +       if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL)
2373 +               return ext3_ext_truncate(inode, page);
2374 +
2375         handle = start_transaction(inode);
2376         if (IS_ERR(handle)) {
2377                 if (page) {
2378 @@ -3008,12 +3008,15 @@ err_out:
2379   * block and work out the exact number of indirects which are touched.  Pah.
2380   */
2381
2382 -static int ext3_writepage_trans_blocks(struct inode *inode)
2383 +int ext3_writepage_trans_blocks(struct inode *inode)
2384  {
2385         int bpp = ext3_journal_blocks_per_page(inode);
2386         int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2387         int ret;
2388
2389 +       if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL)
2390 +               return ext3_ext_writepage_trans_blocks(inode, bpp);
2391 +
2392         if (ext3_should_journal_data(inode))
2393                 ret = 3 * (bpp + indirects) + 2;
2394         else
2395 @@ -3260,7 +3263,7 @@ int ext3_map_inode_page(struct inode *in
2396                 if (blocks[i] != 0)
2397                         continue;
2398
2399 -               rc = ext3_get_blocks_handle(handle, inode, iblock, 1, &dummy, 1, 1);
2400 +               rc = ext3_get_blocks_wrap(handle, inode, iblock, 1, &dummy, 1, 1);
2401                 if (rc < 0) {
2402                         printk(KERN_INFO "ext3_map_inode_page: error reading "
2403                                         "block %ld\n", iblock);
2404 Index: linux-2.6.18.8/fs/ext3/Makefile
2405 ===================================================================
2406 --- linux-2.6.18.8.orig/fs/ext3/Makefile        2007-07-17 09:18:11.000000000 +0200
2407 +++ linux-2.6.18.8/fs/ext3/Makefile     2007-07-17 11:08:11.000000000 +0200
2408 @@ -5,7 +5,7 @@
2409  obj-$(CONFIG_EXT3_FS) += ext3.o
2410
2411  ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
2412 -          ioctl.o namei.o super.o symlink.o hash.o resize.o
2413 +          ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o
2414
2415  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
2416  ext3-$(CONFIG_EXT3_FS_POSIX_ACL) += acl.o
2417 Index: linux-2.6.18.8/fs/ext3/super.c
2418 ===================================================================
2419 --- linux-2.6.18.8.orig/fs/ext3/super.c 2007-07-17 09:18:12.000000000 +0200
2420 +++ linux-2.6.18.8/fs/ext3/super.c      2007-07-17 11:08:12.000000000 +0200
2421 @@ -391,6 +391,7 @@ static void ext3_put_super (struct super
2422         struct ext3_super_block *es = sbi->s_es;
2423         int i;
2424
2425 +       ext3_ext_release(sb);
2426         ext3_xattr_put_super(sb);
2427         journal_destroy(sbi->s_journal);
2428         if (!(sb->s_flags & MS_RDONLY)) {
2429 @@ -455,6 +456,8 @@ static struct inode *ext3_alloc_inode(st
2430  #endif
2431         ei->i_block_alloc_info = NULL;
2432         ei->vfs_inode.i_version = 1;
2433 +
2434 +       memset(&ei->i_cached_extent, 0, sizeof(ei->i_cached_extent));
2435         return &ei->vfs_inode;
2436  }
2437
2438 @@ -680,7 +683,8 @@ enum {
2439         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
2440         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
2441         Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
2442 -       Opt_grpquota
2443 +       Opt_grpquota,
2444 +       Opt_extents, Opt_noextents, Opt_extdebug,
2445  };
2446
2447  static match_table_t tokens = {
2448 @@ -733,6 +737,9 @@ static match_table_t tokens = {
2449         {Opt_noiopen, "noiopen"},
2450         {Opt_iopen_nopriv, "iopen_nopriv"},
2451         {Opt_barrier, "barrier=%u"},
2452 +       {Opt_extents, "extents"},
2453 +       {Opt_noextents, "noextents"},
2454 +       {Opt_extdebug, "extdebug"},
2455         {Opt_err, NULL},
2456         {Opt_resize, "resize"},
2457  };
2458 @@ -1077,6 +1084,15 @@ clear_qf_name:
2459                 case Opt_bh:
2460                         clear_opt(sbi->s_mount_opt, NOBH);
2461                         break;
2462 +               case Opt_extents:
2463 +                       set_opt (sbi->s_mount_opt, EXTENTS);
2464 +                       break;
2465 +               case Opt_noextents:
2466 +                       clear_opt (sbi->s_mount_opt, EXTENTS);
2467 +                       break;
2468 +               case Opt_extdebug:
2469 +                       set_opt (sbi->s_mount_opt, EXTDEBUG);
2470 +                       break;
2471                 default:
2472                         printk (KERN_ERR
2473                                 "EXT3-fs: Unrecognized mount option \"%s\" "
2474 @@ -1806,6 +1822,8 @@ static int ext3_fill_super (struct super
2475                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2476                 "writeback");
2477
2478 +       ext3_ext_init(sb);
2479 +
2480         lock_kernel();
2481         return 0;
2482
2483 Index: linux-2.6.18.8/include/linux/ext3_extents.h
2484 ===================================================================
2485 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
2486 +++ linux-2.6.18.8/include/linux/ext3_extents.h 2007-07-17 09:18:14.000000000 +0200
2487 @@ -0,0 +1,231 @@
2488 +/*
2489 + * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
2490 + * Written by Alex Tomas <alex@clusterfs.com>
2491 + *
2492 + * This program is free software; you can redistribute it and/or modify
2493 + * it under the terms of the GNU General Public License version 2 as
2494 + * published by the Free Software Foundation.
2495 + *
2496 + * This program is distributed in the hope that it will be useful,
2497 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2498 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2499 + * GNU General Public License for more details.
2500 + *
2501 + * You should have received a copy of the GNU General Public Licens
2502 + * along with this program; if not, write to the Free Software
2503 + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
2504 + */
2505 +
2506 +#ifndef _LINUX_EXT3_EXTENTS
2507 +#define _LINUX_EXT3_EXTENTS
2508 +
2509 +#include <linux/ext3_fs.h>
2510 +
2511 +/*
2512 + * with AGRESSIVE_TEST defined capacity of index/leaf blocks
2513 + * become very little, so index split, in-depth growing and
2514 + * other hard changes happens much more often
2515 + * this is for debug purposes only
2516 + */
2517 +#define AGRESSIVE_TEST_
2518 +
2519 +/*
2520 + * with EXTENTS_STATS defined number of blocks and extents
2521 + * are collected in truncate path. they'll be showed at
2522 + * umount time
2523 + */
2524 +#define EXTENTS_STATS__
2525 +
2526 +/*
2527 + * if CHECK_BINSEARCH defined, then results of binary search
2528 + * will be checked by linear search
2529 + */
2530 +#define CHECK_BINSEARCH__
2531 +
2532 +/*
2533 + * if EXT_DEBUG is defined you can use 'extdebug' mount option
2534 + * to get lots of info what's going on
2535 + */
2536 +#define EXT_DEBUG_
2537 +#ifdef EXT_DEBUG
2538 +#define ext_debug(inode,fmt,a...)                       \
2539 +do {                                                    \
2540 +       if (test_opt(inode->i_sb, EXTDEBUG))             \
2541 +               printk(fmt, ##a);                        \
2542 +} while (0);
2543 +#else
2544 +#define ext_debug(inode,fmt,a...)
2545 +#endif
2546 +
2547 +
2548 +/*
2549 + * if EXT_STATS is defined then stats numbers are collected
2550 + * these number will be displayed at umount time
2551 + */
2552 +#define EXT_STATS_
2553 +
2554 +/*
2555 + * define EXT3_ALLOC_NEEDED to 0 since block bitmap, group desc. and sb
2556 + * are now accounted in ext3_ext_calc_credits_for_insert()
2557 + */
2558 +#define EXT3_ALLOC_NEEDED 0
2559 +
2560 +/*
2561 + * ext3_inode has i_block array (60 bytes total)
2562 + * first 12 bytes store ext3_extent_header
2563 + * the remain stores array of ext3_extent
2564 + */
2565 +
2566 +/*
2567 + * this is extent on-disk structure
2568 + * it's used at the bottom of the tree
2569 + */
2570 +struct ext3_extent {
2571 +       __le32  ee_block;       /* first logical block extent covers */
2572 +       __le16  ee_len;         /* number of blocks covered by extent */
2573 +       __le16  ee_start_hi;    /* high 16 bits of physical block */
2574 +       __le32  ee_start;       /* low 32 bigs of physical block */
2575 +};
2576 +
2577 +/*
2578 + * this is index on-disk structure
2579 + * it's used at all the levels, but the bottom
2580 + */
2581 +struct ext3_extent_idx {
2582 +       __le32  ei_block;       /* index covers logical blocks from 'block' */
2583 +       __le32  ei_leaf;        /* pointer to the physical block of the next *
2584 +                                * level. leaf or next index could bet here */
2585 +       __le16  ei_leaf_hi;     /* high 16 bits of physical block */
2586 +       __u16   ei_unused;
2587 +};
2588 +
2589 +/*
2590 + * each block (leaves and indexes), even inode-stored has header
2591 + */
2592 +struct ext3_extent_header {
2593 +       __le16  eh_magic;       /* probably will support different formats */
2594 +       __le16  eh_entries;     /* number of valid entries */
2595 +       __le16  eh_max;         /* capacity of store in entries */
2596 +       __le16  eh_depth;       /* has tree real underlaying blocks? */
2597 +       __le32  eh_generation;  /* flags(8 bits) | generation of the tree */
2598 +};
2599 +
2600 +#define EXT3_EXT_MAGIC         0xf30a
2601 +
2602 +/*
2603 + * array of ext3_ext_path contains path to some extent
2604 + * creation/lookup routines use it for traversal/splitting/etc
2605 + * truncate uses it to simulate recursive walking
2606 + */
2607 +struct ext3_ext_path {
2608 +       __u32                           p_block;
2609 +       __u16                           p_depth;
2610 +       struct ext3_extent              *p_ext;
2611 +       struct ext3_extent_idx          *p_idx;
2612 +       struct ext3_extent_header       *p_hdr;
2613 +       struct buffer_head              *p_bh;
2614 +};
2615 +
2616 +/*
2617 + * structure for external API
2618 + */
2619 +
2620 +#define EXT3_EXT_CACHE_NO      0
2621 +#define EXT3_EXT_CACHE_GAP     1
2622 +#define EXT3_EXT_CACHE_EXTENT  2
2623 +#define EXT3_EXT_HAS_NO_TREE  /* ext3_extents_tree struct is not used*/
2624 +
2625 +/*
2626 + * to be called by ext3_ext_walk_space()
2627 + * negative retcode - error
2628 + * positive retcode - signal for ext3_ext_walk_space(), see below
2629 + * callback must return valid extent (passed or newly created)
2630 + */
2631 +typedef int (*ext_prepare_callback)(struct inode *, struct ext3_ext_path *,
2632 +                                       struct ext3_ext_cache *,
2633 +                                       void *);
2634 +
2635 +#define EXT_CONTINUE   0
2636 +#define EXT_BREAK      1
2637 +#define EXT_REPEAT     2
2638 +
2639 +
2640 +#define EXT_MAX_BLOCK  0xffffffff
2641 +
2642 +#define EXT_FLAGS_CLR_UNKNOWN  0x7  /* Flags cleared on modification */
2643 +#define EXT_HDR_GEN_BITS       24
2644 +#define EXT_HDR_GEN_MASK       ((1 << EXT_HDR_GEN_BITS) - 1)
2645 +
2646 +#define EXT_FIRST_EXTENT(__hdr__) \
2647 +       ((struct ext3_extent *) (((char *) (__hdr__)) +         \
2648 +                                sizeof(struct ext3_extent_header)))
2649 +#define EXT_FIRST_INDEX(__hdr__) \
2650 +       ((struct ext3_extent_idx *) (((char *) (__hdr__)) +     \
2651 +                                    sizeof(struct ext3_extent_header)))
2652 +#define EXT_HAS_FREE_INDEX(__path__) \
2653 +        (le16_to_cpu((__path__)->p_hdr->eh_entries) \
2654 +                                    < le16_to_cpu((__path__)->p_hdr->eh_max))
2655 +#define EXT_LAST_EXTENT(__hdr__) \
2656 +       (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
2657 +#define EXT_LAST_INDEX(__hdr__) \
2658 +       (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
2659 +#define EXT_MAX_EXTENT(__hdr__) \
2660 +       (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
2661 +#define EXT_MAX_INDEX(__hdr__) \
2662 +       (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)
2663 +
2664 +
2665 +static inline struct ext3_extent_header *ext_inode_hdr(struct inode *inode)
2666 +{
2667 +       return (struct ext3_extent_header *) EXT3_I(inode)->i_data;
2668 +}
2669 +
2670 +static inline struct ext3_extent_header *ext_block_hdr(struct buffer_head *bh)
2671 +{
2672 +       return (struct ext3_extent_header *) bh->b_data;
2673 +}
2674 +
2675 +static inline unsigned short ext_depth(struct inode *inode)
2676 +{
2677 +       return le16_to_cpu(ext_inode_hdr(inode)->eh_depth);
2678 +}
2679 +
2680 +static inline unsigned short ext_flags(struct ext3_extent_header *neh)
2681 +{
2682 +       return le16_to_cpu(neh->eh_generation) >> EXT_HDR_GEN_BITS;
2683 +}
2684 +
2685 +static inline unsigned short ext_hdr_gen(struct ext3_extent_header *neh)
2686 +{
2687 +       return le16_to_cpu(neh->eh_generation) & EXT_HDR_GEN_MASK;
2688 +}
2689 +
2690 +static inline unsigned short ext_generation(struct inode *inode)
2691 +{
2692 +       return ext_hdr_gen(ext_inode_hdr(inode));
2693 +}
2694 +
2695 +static inline void ext3_ext_tree_changed(struct inode *inode)
2696 +{
2697 +       struct ext3_extent_header *neh = ext_inode_hdr(inode);
2698 +       neh->eh_generation = cpu_to_le32(
2699 +               ((ext_flags(neh) & ~EXT_FLAGS_CLR_UNKNOWN) << EXT_HDR_GEN_BITS)
2700 +                   | ((ext_hdr_gen(neh) + 1) & EXT_HDR_GEN_MASK));
2701 +}
2702 +
2703 +static inline void
2704 +ext3_ext_invalidate_cache(struct inode *inode)
2705 +{
2706 +       EXT3_I(inode)->i_cached_extent.ec_type = EXT3_EXT_CACHE_NO;
2707 +}
2708 +
2709 +extern int ext3_ext_search_left(struct inode *, struct ext3_ext_path *, unsigned long *, unsigned long *);
2710 +extern int ext3_ext_search_right(struct inode *, struct ext3_ext_path *, unsigned long *, unsigned long *);
2711 +extern int ext3_extent_tree_init(handle_t *, struct inode *);
2712 +extern int ext3_ext_calc_credits_for_insert(struct inode *, struct ext3_ext_path *);
2713 +extern int ext3_ext_insert_extent(handle_t *, struct inode *, struct ext3_ext_path *, struct ext3_extent *);
2714 +extern int ext3_ext_walk_space(struct inode *, unsigned long, unsigned long, ext_prepare_callback, void *);
2715 +extern struct ext3_ext_path * ext3_ext_find_extent(struct inode *, int, struct ext3_ext_path *);
2716 +
2717 +#endif /* _LINUX_EXT3_EXTENTS */
2718 +
2719 Index: linux-2.6.18.8/include/linux/ext3_fs.h
2720 ===================================================================
2721 --- linux-2.6.18.8.orig/include/linux/ext3_fs.h 2007-07-17 09:18:13.000000000 +0200
2722 +++ linux-2.6.18.8/include/linux/ext3_fs.h      2007-07-17 11:08:12.000000000 +0200
2723 @@ -182,8 +182,9 @@ struct ext3_group_desc
2724  #define EXT3_DIRSYNC_FL                        0x00010000 /* dirsync behaviour (directories only) */
2725  #define EXT3_TOPDIR_FL                 0x00020000 /* Top of directory hierarchies*/
2726  #define EXT3_RESERVED_FL               0x80000000 /* reserved for ext3 lib */
2727 +#define EXT3_EXTENTS_FL                        0x00080000 /* Inode uses extents */
2728
2729 -#define EXT3_FL_USER_VISIBLE           0x0003DFFF /* User visible flags */
2730 +#define EXT3_FL_USER_VISIBLE           0x000BDFFF /* User visible flags */
2731  #define EXT3_FL_USER_MODIFIABLE                0x000380FF /* User modifiable flags */
2732
2733  /*
2734 @@ -373,6 +374,8 @@ struct ext3_inode {
2735  #define EXT3_MOUNT_GRPQUOTA            0x200000 /* "old" group quota */
2736  #define EXT3_MOUNT_IOPEN               0x400000        /* Allow access via iopen */
2737  #define EXT3_MOUNT_IOPEN_NOPRIV                0x800000/* Make iopen world-readable */
2738 +#define EXT3_MOUNT_EXTENTS             0x2000000/* Extents support */
2739 +#define EXT3_MOUNT_EXTDEBUG             0x4000000/* Extents debug */
2740
2741  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
2742  #ifndef clear_opt
2743 @@ -572,11 +575,13 @@ static inline int ext3_valid_inum(struct
2744  #define EXT3_FEATURE_INCOMPAT_RECOVER          0x0004 /* Needs recovery */
2745  #define EXT3_FEATURE_INCOMPAT_JOURNAL_DEV      0x0008 /* Journal device */
2746  #define EXT3_FEATURE_INCOMPAT_META_BG          0x0010
2747 +#define EXT3_FEATURE_INCOMPAT_EXTENTS          0x0040 /* extents support */
2748
2749  #define EXT3_FEATURE_COMPAT_SUPP       EXT2_FEATURE_COMPAT_EXT_ATTR
2750  #define EXT3_FEATURE_INCOMPAT_SUPP     (EXT3_FEATURE_INCOMPAT_FILETYPE| \
2751                                          EXT3_FEATURE_INCOMPAT_RECOVER| \
2752 -                                        EXT3_FEATURE_INCOMPAT_META_BG)
2753 +                                        EXT3_FEATURE_INCOMPAT_META_BG| \
2754 +                                        EXT3_FEATURE_INCOMPAT_EXTENTS)
2755  #define EXT3_FEATURE_RO_COMPAT_SUPP    (EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER| \
2756                                          EXT3_FEATURE_RO_COMPAT_LARGE_FILE| \
2757                                          EXT3_FEATURE_RO_COMPAT_BTREE_DIR)
2758 @@ -816,6 +821,9 @@ extern int ext3_get_inode_loc(struct ino
2759  extern void ext3_truncate (struct inode *);
2760  extern void ext3_set_inode_flags(struct inode *);
2761  extern void ext3_set_aops(struct inode *inode);
2762 +extern int ext3_writepage_trans_blocks(struct inode *);
2763 +extern int ext3_block_truncate_page(handle_t *handle, struct page *page,
2764 +               struct address_space *mapping, loff_t from);
2765
2766  /* ioctl.c */
2767  extern int ext3_ioctl (struct inode *, struct file *, unsigned int,
2768 @@ -869,6 +877,30 @@ extern struct inode_operations ext3_spec
2769  extern struct inode_operations ext3_symlink_inode_operations;
2770  extern struct inode_operations ext3_fast_symlink_inode_operations;
2771
2772 +/* extents.c */
2773 +extern int ext3_ext_tree_init(handle_t *handle, struct inode *);
2774 +extern int ext3_ext_writepage_trans_blocks(struct inode *, int);
2775 +extern int ext3_ext_get_blocks(handle_t *, struct inode *, sector_t,
2776 +                               unsigned long, struct buffer_head *, int, int);
2777 +extern void ext3_ext_truncate(struct inode *, struct page *);
2778 +extern void ext3_ext_init(struct super_block *);
2779 +extern void ext3_ext_release(struct super_block *);
2780 +static inline int
2781 +ext3_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
2782 +                       unsigned long max_blocks, struct buffer_head *bh,
2783 +                       int create, int extend_disksize)
2784 +{
2785 +       int ret;
2786 +       if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL)
2787 +               return ext3_ext_get_blocks(handle, inode, block, max_blocks,
2788 +                                       bh, create, extend_disksize);
2789 +       ret = ext3_get_blocks_handle(handle, inode, block, max_blocks, bh, create,
2790 +                                    extend_disksize);
2791 +       if (ret > 0)
2792 +               bh->b_size = (ret << inode->i_blkbits);
2793 +       return ret;
2794 +}
2795 +
2796
2797  #endif /* __KERNEL__ */
2798
2799 Index: linux-2.6.18.8/include/linux/ext3_fs_i.h
2800 ===================================================================
2801 --- linux-2.6.18.8.orig/include/linux/ext3_fs_i.h       2007-02-24 00:52:30.000000000 +0100
2802 +++ linux-2.6.18.8/include/linux/ext3_fs_i.h    2007-07-17 11:08:11.000000000 +0200
2803 @@ -65,6 +65,16 @@ struct ext3_block_alloc_info {
2804  #define rsv_end rsv_window._rsv_end
2805
2806  /*
2807 + * storage for cached extent
2808 + */
2809 +struct ext3_ext_cache {
2810 +       __u32   ec_start;
2811 +       __u32   ec_block;
2812 +       __u32   ec_len; /* must be 32bit to return holes */
2813 +       __u32   ec_type;
2814 +};
2815 +
2816 +/*
2817   * third extended file system inode data in memory
2818   */
2819  struct ext3_inode_info {
2820 @@ -142,6 +152,8 @@ struct ext3_inode_info {
2821          */
2822         struct mutex truncate_mutex;
2823         struct inode vfs_inode;
2824 +
2825 +       struct ext3_ext_cache i_cached_extent;
2826  };
2827
2828  #endif /* _LINUX_EXT3_FS_I */
2829 Index: linux-2.6.18.8/include/linux/ext3_fs_sb.h
2830 ===================================================================
2831 --- linux-2.6.18.8.orig/include/linux/ext3_fs_sb.h      2007-02-24 00:52:30.000000000 +0100
2832 +++ linux-2.6.18.8/include/linux/ext3_fs_sb.h   2007-07-17 11:08:12.000000000 +0200
2833 @@ -78,6 +78,16 @@ struct ext3_sb_info {
2834         char *s_qf_names[MAXQUOTAS];            /* Names of quota files with journalled quota */
2835         int s_jquota_fmt;                       /* Format of quota to use */
2836  #endif
2837 +
2838 +#ifdef EXTENTS_STATS
2839 +       /* ext3 extents stats */
2840 +       unsigned long s_ext_min;
2841 +       unsigned long s_ext_max;
2842 +       unsigned long s_depth_max;
2843 +       spinlock_t s_ext_stats_lock;
2844 +       unsigned long s_ext_blocks;
2845 +       unsigned long s_ext_extents;
2846 +#endif
2847  };
2848
2849  #endif /* _LINUX_EXT3_FS_SB */
2850 Index: linux-2.6.18.8/include/linux/ext3_jbd.h
2851 ===================================================================
2852 --- linux-2.6.18.8.orig/include/linux/ext3_jbd.h        2007-02-24 00:52:30.000000000 +0100
2853 +++ linux-2.6.18.8/include/linux/ext3_jbd.h     2007-07-17 09:18:14.000000000 +0200
2854 @@ -23,12 +23,20 @@
2855
2856  /* Define the number of blocks we need to account to a transaction to
2857   * modify one block of data.
2858 - *
2859 + *
2860   * We may have to touch one inode, one bitmap buffer, up to three
2861   * indirection blocks, the group and superblock summaries, and the data
2862 - * block to complete the transaction.  */
2863 + * block to complete the transaction.
2864 + *
2865 + * For extents-enabled fs we may have to allocate and modify upto
2866 + * 5 levels of tree + root which is stored in inode. */
2867 +
2868 +#define EXT3_SINGLEDATA_TRANS_BLOCKS(sb)                               \
2869 +       (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_EXTENTS)   \
2870 +               || test_opt(sb, EXTENTS) ? 27U : 8U)
2871
2872 -#define EXT3_SINGLEDATA_TRANS_BLOCKS   8U
2873 +/* Indicate that EXT3_SINGLEDATA_TRANS_BLOCKS takes the sb as argument */
2874 +#define EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
2875
2876  /* Extended attribute operations touch at most two data buffers,
2877   * two bitmap buffers, and two group summaries, in addition to the inode
2878 @@ -42,7 +50,7 @@
2879   * superblock only gets updated once, of course, so don't bother
2880   * counting that again for the quota updates. */
2881
2882 -#define EXT3_DATA_TRANS_BLOCKS(sb)     (EXT3_SINGLEDATA_TRANS_BLOCKS + \
2883 +#define EXT3_DATA_TRANS_BLOCKS(sb)     (EXT3_SINGLEDATA_TRANS_BLOCKS(sb) + \
2884                                          EXT3_XATTR_TRANS_BLOCKS - 2 + \
2885                                          2*EXT3_QUOTA_TRANS_BLOCKS(sb))
2886
2887 @@ -78,9 +86,9 @@
2888  /* Amount of blocks needed for quota insert/delete - we do some block writes
2889   * but inode, sb and group updates are done only once */
2890  #define EXT3_QUOTA_INIT_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_INIT_ALLOC*\
2891 -               (EXT3_SINGLEDATA_TRANS_BLOCKS-3)+3+DQUOT_INIT_REWRITE) : 0)
2892 +               (EXT3_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_INIT_REWRITE) : 0)
2893  #define EXT3_QUOTA_DEL_BLOCKS(sb) (test_opt(sb, QUOTA) ? (DQUOT_DEL_ALLOC*\
2894 -               (EXT3_SINGLEDATA_TRANS_BLOCKS-3)+3+DQUOT_DEL_REWRITE) : 0)
2895 +               (EXT3_SINGLEDATA_TRANS_BLOCKS(sb)-3)+3+DQUOT_DEL_REWRITE) : 0)
2896  #else
2897  #define EXT3_QUOTA_TRANS_BLOCKS(sb) 0
2898  #define EXT3_QUOTA_INIT_BLOCKS(sb) 0