fs/ext4/extents.c

   1 /*
   2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   3  * Written by Alex Tomas <alex@clusterfs.com>
   4  *
   5  * Architecture independence:
   6  *   Copyright (c) 2005, Bull S.A.
   7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public Licens
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  21  */
  22
  23 /*
  24  * Extents support for EXT4
  25  *
  26  * TODO:
  27  *   - ext4*_error() should be used in some situations
  28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  29  *   - smart tree reduction
  30  */
  31
  32 #include <linux/module.h>
  33 #include <linux/fs.h>
  34 #include <linux/time.h>
  35 #include <linux/ext4_jbd2.h>
  36 #include <linux/jbd.h>
  37 #include <linux/smp_lock.h>
  38 #include <linux/highuid.h>
  39 #include <linux/pagemap.h>
  40 #include <linux/quotaops.h>
  41 #include <linux/string.h>
  42 #include <linux/slab.h>
  43 #include <linux/ext4_fs_extents.h>
  44 #include <asm/uaccess.h>
  45
  46
  47 /*
  48  * ext_pblock:
  49  * combine low and high parts of physical block number into ext4_fsblk_t
  50  */
  51 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
  52 {
  53         ext4_fsblk_t block;
  54
  55         block = le32_to_cpu(ex->ee_start);
  56         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
  57         return block;
  58 }
  59
  60 /*
  61  * idx_pblock:
  62  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
  63  */
  64 static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
  65 {
  66         ext4_fsblk_t block;
  67
  68         block = le32_to_cpu(ix->ei_leaf);
  69         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
  70         return block;
  71 }
  72
  73 /*
  74  * ext4_ext_store_pblock:
  75  * stores a large physical block number into an extent struct,
  76  * breaking it into parts
  77  */
  78 static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
  79 {
  80         ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  81         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  82 }
  83
  84 /*
  85  * ext4_idx_store_pblock:
  86  * stores a large physical block number into an index struct,
  87  * breaking it into parts
  88  */
  89 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
  90 {
  91         ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  92         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  93 }
  94
  95 static int ext4_ext_check_header(const char *function, struct inode *inode,
  96                                 struct ext4_extent_header *eh)
  97 {
  98         const char *error_msg = NULL;
  99
 100         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 101                 error_msg = "invalid magic";
 102                 goto corrupted;
 103         }
 104         if (unlikely(eh->eh_max == 0)) {
 105                 error_msg = "invalid eh_max";
 106                 goto corrupted;
 107         }
 108         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 109                 error_msg = "invalid eh_entries";
 110                 goto corrupted;
 111         }
 112         return 0;
 113
 114 corrupted:
 115         ext4_error(inode->i_sb, function,
 116                         "bad header in inode #%lu: %s - magic %x, "
 117                         "entries %u, max %u, depth %u",
 118                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
 119                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 120                         le16_to_cpu(eh->eh_depth));
 121
 122         return -EIO;
 123 }
 124
 125 static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
 126 {
 127         int err;
 128
 129         if (handle->h_buffer_credits > needed)
 130                 return handle;
 131         if (!ext4_journal_extend(handle, needed))
 132                 return handle;
 133         err = ext4_journal_restart(handle, needed);
 134
 135         return handle;
 136 }
 137
 138 /*
 139  * could return:
 140  *  - EROFS
 141  *  - ENOMEM
 142  */
 143 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 144                                 struct ext4_ext_path *path)
 145 {
 146         if (path->p_bh) {
 147                 /* path points to block */
 148                 return ext4_journal_get_write_access(handle, path->p_bh);
 149         }
 150         /* path points to leaf/index in inode body */
 151         /* we use in-core data, no need to protect them */
 152         return 0;
 153 }
 154
 155 /*
 156  * could return:
 157  *  - EROFS
 158  *  - ENOMEM
 159  *  - EIO
 160  */
 161 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
 162                                 struct ext4_ext_path *path)
 163 {
 164         int err;
 165         if (path->p_bh) {
 166                 /* path points to block */
 167                 err = ext4_journal_dirty_metadata(handle, path->p_bh);
 168         } else {
 169                 /* path points to leaf/index in inode body */
 170                 err = ext4_mark_inode_dirty(handle, inode);
 171         }
 172         return err;
 173 }
 174
 175 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 176                               struct ext4_ext_path *path,
 177                               ext4_fsblk_t block)
 178 {
 179         struct ext4_inode_info *ei = EXT4_I(inode);
 180         ext4_fsblk_t bg_start;
 181         ext4_grpblk_t colour;
 182         int depth;
 183
 184         if (path) {
 185                 struct ext4_extent *ex;
 186                 depth = path->p_depth;
 187
 188                 /* try to predict block placement */
 189                 ex = path[depth].p_ext;
 190                 if (ex)
 191                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
 192
 193                 /* it looks like index is empty;
 194                  * try to find starting block from index itself */
 195                 if (path[depth].p_bh)
 196                         return path[depth].p_bh->b_blocknr;
 197         }
 198
 199         /* OK. use inode's group */
 200         bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
 201                 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
 202         colour = (current->pid % 16) *
 203                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 204         return bg_start + colour + block;
 205 }
 206
 207 static ext4_fsblk_t
 208 ext4_ext_new_block(handle_t *handle, struct inode *inode,
 209                         struct ext4_ext_path *path,
 210                         struct ext4_extent *ex, int *err)
 211 {
 212         ext4_fsblk_t goal, newblock;
 213
 214         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 215         newblock = ext4_new_block(handle, inode, goal, err);
 216         return newblock;
 217 }
 218
 219 static int ext4_ext_space_block(struct inode *inode)
 220 {
 221         int size;
 222
 223         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 224                         / sizeof(struct ext4_extent);
 225 #ifdef AGRESSIVE_TEST
 226         if (size > 6)
 227                 size = 6;
 228 #endif
 229         return size;
 230 }
 231
 232 static int ext4_ext_space_block_idx(struct inode *inode)
 233 {
 234         int size;
 235
 236         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 237                         / sizeof(struct ext4_extent_idx);
 238 #ifdef AGRESSIVE_TEST
 239         if (size > 5)
 240                 size = 5;
 241 #endif
 242         return size;
 243 }
 244
 245 static int ext4_ext_space_root(struct inode *inode)
 246 {
 247         int size;
 248
 249         size = sizeof(EXT4_I(inode)->i_data);
 250         size -= sizeof(struct ext4_extent_header);
 251         size /= sizeof(struct ext4_extent);
 252 #ifdef AGRESSIVE_TEST
 253         if (size > 3)
 254                 size = 3;
 255 #endif
 256         return size;
 257 }
 258
 259 static int ext4_ext_space_root_idx(struct inode *inode)
 260 {
 261         int size;
 262
 263         size = sizeof(EXT4_I(inode)->i_data);
 264         size -= sizeof(struct ext4_extent_header);
 265         size /= sizeof(struct ext4_extent_idx);
 266 #ifdef AGRESSIVE_TEST
 267         if (size > 4)
 268                 size = 4;
 269 #endif
 270         return size;
 271 }
 272
 273 #ifdef EXT_DEBUG
 274 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 275 {
 276         int k, l = path->p_depth;
 277
 278         ext_debug("path:");
 279         for (k = 0; k <= l; k++, path++) {
 280                 if (path->p_idx) {
 281                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 282                             idx_pblock(path->p_idx));
 283                 } else if (path->p_ext) {
 284                         ext_debug("  %d:%d:%llu ",
 285                                   le32_to_cpu(path->p_ext->ee_block),
 286                                   le16_to_cpu(path->p_ext->ee_len),
 287                                   ext_pblock(path->p_ext));
 288                 } else
 289                         ext_debug("  []");
 290         }
 291         ext_debug("\n");
 292 }
 293
 294 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 295 {
 296         int depth = ext_depth(inode);
 297         struct ext4_extent_header *eh;
 298         struct ext4_extent *ex;
 299         int i;
 300
 301         if (!path)
 302                 return;
 303
 304         eh = path[depth].p_hdr;
 305         ex = EXT_FIRST_EXTENT(eh);
 306
 307         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 308                 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
 309                           le16_to_cpu(ex->ee_len), ext_pblock(ex));
 310         }
 311         ext_debug("\n");
 312 }
 313 #else
 314 #define ext4_ext_show_path(inode,path)
 315 #define ext4_ext_show_leaf(inode,path)
 316 #endif
 317
 318 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
 319 {
 320         int depth = path->p_depth;
 321         int i;
 322
 323         for (i = 0; i <= depth; i++, path++)
 324                 if (path->p_bh) {
 325                         brelse(path->p_bh);
 326                         path->p_bh = NULL;
 327                 }
 328 }
 329
 330 /*
 331  * ext4_ext_binsearch_idx:
 332  * binary search for the closest index of the given block
 333  */
 334 static void
 335 ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
 336 {
 337         struct ext4_extent_header *eh = path->p_hdr;
 338         struct ext4_extent_idx *r, *l, *m;
 339
 340         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
 341         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
 342         BUG_ON(le16_to_cpu(eh->eh_entries) <= 0);
 343
 344         ext_debug("binsearch for %d(idx):  ", block);
 345
 346         l = EXT_FIRST_INDEX(eh) + 1;
 347         r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
 348         while (l <= r) {
 349                 m = l + (r - l) / 2;
 350                 if (block < le32_to_cpu(m->ei_block))
 351                         r = m - 1;
 352                 else
 353                         l = m + 1;
 354                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
 355                                 m, m->ei_block, r, r->ei_block);
 356         }
 357
 358         path->p_idx = l - 1;
 359         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 360                   idx_block(path->p_idx));
 361
 362 #ifdef CHECK_BINSEARCH
 363         {
 364                 struct ext4_extent_idx *chix, *ix;
 365                 int k;
 366
 367                 chix = ix = EXT_FIRST_INDEX(eh);
 368                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 369                   if (k != 0 &&
 370                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 371                                 printk("k=%d, ix=0x%p, first=0x%p\n", k,
 372                                         ix, EXT_FIRST_INDEX(eh));
 373                                 printk("%u <= %u\n",
 374                                        le32_to_cpu(ix->ei_block),
 375                                        le32_to_cpu(ix[-1].ei_block));
 376                         }
 377                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 378                                            <= le32_to_cpu(ix[-1].ei_block));
 379                         if (block < le32_to_cpu(ix->ei_block))
 380                                 break;
 381                         chix = ix;
 382                 }
 383                 BUG_ON(chix != path->p_idx);
 384         }
 385 #endif
 386
 387 }
 388
 389 /*
 390  * ext4_ext_binsearch:
 391  * binary search for closest extent of the given block
 392  */
 393 static void
 394 ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
 395 {
 396         struct ext4_extent_header *eh = path->p_hdr;
 397         struct ext4_extent *r, *l, *m;
 398
 399         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
 400         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
 401
 402         if (eh->eh_entries == 0) {
 403                 /*
 404                  * this leaf is empty:
 405                  * we get such a leaf in split/add case
 406                  */
 407                 return;
 408         }
 409
 410         ext_debug("binsearch for %d:  ", block);
 411
 412         l = EXT_FIRST_EXTENT(eh) + 1;
 413         r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
 414
 415         while (l <= r) {
 416                 m = l + (r - l) / 2;
 417                 if (block < le32_to_cpu(m->ee_block))
 418                         r = m - 1;
 419                 else
 420                         l = m + 1;
 421                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
 422                                 m, m->ee_block, r, r->ee_block);
 423         }
 424
 425         path->p_ext = l - 1;
 426         ext_debug("  -> %d:%llu:%d ",
 427                         le32_to_cpu(path->p_ext->ee_block),
 428                         ext_pblock(path->p_ext),
 429                         le16_to_cpu(path->p_ext->ee_len));
 430
 431 #ifdef CHECK_BINSEARCH
 432         {
 433                 struct ext4_extent *chex, *ex;
 434                 int k;
 435
 436                 chex = ex = EXT_FIRST_EXTENT(eh);
 437                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 438                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 439                                           <= le32_to_cpu(ex[-1].ee_block));
 440                         if (block < le32_to_cpu(ex->ee_block))
 441                                 break;
 442                         chex = ex;
 443                 }
 444                 BUG_ON(chex != path->p_ext);
 445         }
 446 #endif
 447
 448 }
 449
 450 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 451 {
 452         struct ext4_extent_header *eh;
 453
 454         eh = ext_inode_hdr(inode);
 455         eh->eh_depth = 0;
 456         eh->eh_entries = 0;
 457         eh->eh_magic = EXT4_EXT_MAGIC;
 458         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
 459         ext4_mark_inode_dirty(handle, inode);
 460         ext4_ext_invalidate_cache(inode);
 461         return 0;
 462 }
 463
 464 struct ext4_ext_path *
 465 ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
 466 {
 467         struct ext4_extent_header *eh;
 468         struct buffer_head *bh;
 469         short int depth, i, ppos = 0, alloc = 0;
 470
 471         eh = ext_inode_hdr(inode);
 472         BUG_ON(eh == NULL);
 473         if (ext4_ext_check_header(__FUNCTION__, inode, eh))
 474                 return ERR_PTR(-EIO);
 475
 476         i = depth = ext_depth(inode);
 477
 478         /* account possible depth increase */
 479         if (!path) {
 480                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 481                                 GFP_NOFS);
 482                 if (!path)
 483                         return ERR_PTR(-ENOMEM);
 484                 alloc = 1;
 485         }
 486         path[0].p_hdr = eh;
 487
 488         /* walk through the tree */
 489         while (i) {
 490                 ext_debug("depth %d: num %d, max %d\n",
 491                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 492                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 493                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
 494                 path[ppos].p_depth = i;
 495                 path[ppos].p_ext = NULL;
 496
 497                 bh = sb_bread(inode->i_sb, path[ppos].p_block);
 498                 if (!bh)
 499                         goto err;
 500
 501                 eh = ext_block_hdr(bh);
 502                 ppos++;
 503                 BUG_ON(ppos > depth);
 504                 path[ppos].p_bh = bh;
 505                 path[ppos].p_hdr = eh;
 506                 i--;
 507
 508                 if (ext4_ext_check_header(__FUNCTION__, inode, eh))
 509                         goto err;
 510         }
 511
 512         path[ppos].p_depth = i;
 513         path[ppos].p_hdr = eh;
 514         path[ppos].p_ext = NULL;
 515         path[ppos].p_idx = NULL;
 516
 517         if (ext4_ext_check_header(__FUNCTION__, inode, eh))
 518                 goto err;
 519
 520         /* find extent */
 521         ext4_ext_binsearch(inode, path + ppos, block);
 522
 523         ext4_ext_show_path(inode, path);
 524
 525         return path;
 526
 527 err:
 528         ext4_ext_drop_refs(path);
 529         if (alloc)
 530                 kfree(path);
 531         return ERR_PTR(-EIO);
 532 }
 533
 534 /*
 535  * ext4_ext_insert_index:
 536  * insert new index [@logical;@ptr] into the block at @curp;
 537  * check where to insert: before @curp or after @curp
 538  */
 539 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 540                                 struct ext4_ext_path *curp,
 541                                 int logical, ext4_fsblk_t ptr)
 542 {
 543         struct ext4_extent_idx *ix;
 544         int len, err;
 545
 546         err = ext4_ext_get_access(handle, inode, curp);
 547         if (err)
 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 ext4_extent_idx);
 556                         len = len < 0 ? 0 : len;
 557                         ext_debug("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 ext4_extent_idx);
 567                 len = len < 0 ? 0 : len;
 568                 ext_debug("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         ext4_idx_store_pblock(ix, ptr);
 578         curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
 579
 580         BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
 581                              > le16_to_cpu(curp->p_hdr->eh_max));
 582         BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
 583
 584         err = ext4_ext_dirty(handle, inode, curp);
 585         ext4_std_error(inode->i_sb, err);
 586
 587         return err;
 588 }
 589
 590 /*
 591  * ext4_ext_split:
 592  * 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 extents and index entries (right to the split point)
 597  *   into the newly allocated blocks
 598  * - initializes subtree
 599  */
 600 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 601                                 struct ext4_ext_path *path,
 602                                 struct ext4_extent *newext, int at)
 603 {
 604         struct buffer_head *bh = NULL;
 605         int depth = ext_depth(inode);
 606         struct ext4_extent_header *neh;
 607         struct ext4_extent_idx *fidx;
 608         struct ext4_extent *ex;
 609         int i = at, k, m, a;
 610         ext4_fsblk_t newblock, oldblock;
 611         __le32 border;
 612         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 613         int err = 0;
 614
 615         /* make decision: where to split? */
 616         /* FIXME: now decision is simplest: at current extent */
 617
 618         /* if current leaf will be split, 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("leaf will be split."
 624                                 " next leaf starts at %d\n",
 625                                   le32_to_cpu(border));
 626         } else {
 627                 border = newext->ee_block;
 628                 ext_debug("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 mark filesystem read-only. 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 = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 646         if (!ablocks)
 647                 return -ENOMEM;
 648
 649         /* allocate all needed blocks */
 650         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 651         for (a = 0; a < depth - at; a++) {
 652                 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
 653                 if (newblock == 0)
 654                         goto cleanup;
 655                 ablocks[a] = newblock;
 656         }
 657
 658         /* initialize new leaf */
 659         newblock = ablocks[--a];
 660         BUG_ON(newblock == 0);
 661         bh = sb_getblk(inode->i_sb, newblock);
 662         if (!bh) {
 663                 err = -EIO;
 664                 goto cleanup;
 665         }
 666         lock_buffer(bh);
 667
 668         err = ext4_journal_get_create_access(handle, bh);
 669         if (err)
 670                 goto cleanup;
 671
 672         neh = ext_block_hdr(bh);
 673         neh->eh_entries = 0;
 674         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
 675         neh->eh_magic = EXT4_EXT_MAGIC;
 676         neh->eh_depth = 0;
 677         ex = EXT_FIRST_EXTENT(neh);
 678
 679         /* move remainder 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("move %d:%llu:%d in new leaf %llu\n",
 688                                 le32_to_cpu(path[depth].p_ext->ee_block),
 689                                 ext_pblock(path[depth].p_ext),
 690                                 le16_to_cpu(path[depth].p_ext->ee_len),
 691                                 newblock);
 692                 /*memmove(ex++, path[depth].p_ext++,
 693                                 sizeof(struct ext4_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 ext4_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         err = ext4_journal_dirty_metadata(handle, bh);
 707         if (err)
 708                 goto cleanup;
 709         brelse(bh);
 710         bh = NULL;
 711
 712         /* correct old leaf */
 713         if (m) {
 714                 err = ext4_ext_get_access(handle, inode, path + depth);
 715                 if (err)
 716                         goto cleanup;
 717                 path[depth].p_hdr->eh_entries =
 718                      cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
 719                 err = ext4_ext_dirty(handle, inode, path + depth);
 720                 if (err)
 721                         goto cleanup;
 722
 723         }
 724
 725         /* create intermediate indexes */
 726         k = depth - at - 1;
 727         BUG_ON(k < 0);
 728         if (k)
 729                 ext_debug("create %d intermediate indices\n", k);
 730         /* insert new index into current index block */
 731         /* current depth stored in i var */
 732         i = depth - 1;
 733         while (k--) {
 734                 oldblock = newblock;
 735                 newblock = ablocks[--a];
 736                 bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock);
 737                 if (!bh) {
 738                         err = -EIO;
 739                         goto cleanup;
 740                 }
 741                 lock_buffer(bh);
 742
 743                 err = ext4_journal_get_create_access(handle, bh);
 744                 if (err)
 745                         goto cleanup;
 746
 747                 neh = ext_block_hdr(bh);
 748                 neh->eh_entries = cpu_to_le16(1);
 749                 neh->eh_magic = EXT4_EXT_MAGIC;
 750                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
 751                 neh->eh_depth = cpu_to_le16(depth - i);
 752                 fidx = EXT_FIRST_INDEX(neh);
 753                 fidx->ei_block = border;
 754                 ext4_idx_store_pblock(fidx, oldblock);
 755
 756                 ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i,
 757                                 newblock, (unsigned long) le32_to_cpu(border),
 758                                 oldblock);
 759                 /* copy indexes */
 760                 m = 0;
 761                 path[i].p_idx++;
 762
 763                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 764                                 EXT_MAX_INDEX(path[i].p_hdr));
 765                 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
 766                                 EXT_LAST_INDEX(path[i].p_hdr));
 767                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
 768                         ext_debug("%d: move %d:%d in new index %llu\n", i,
 769                                         le32_to_cpu(path[i].p_idx->ei_block),
 770                                         idx_pblock(path[i].p_idx),
 771                                         newblock);
 772                         /*memmove(++fidx, path[i].p_idx++,
 773                                         sizeof(struct ext4_extent_idx));
 774                         neh->eh_entries++;
 775                         BUG_ON(neh->eh_entries > neh->eh_max);*/
 776                         path[i].p_idx++;
 777                         m++;
 778                 }
 779                 if (m) {
 780                         memmove(++fidx, path[i].p_idx - m,
 781                                 sizeof(struct ext4_extent_idx) * m);
 782                         neh->eh_entries =
 783                                 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
 784                 }
 785                 set_buffer_uptodate(bh);
 786                 unlock_buffer(bh);
 787
 788                 err = ext4_journal_dirty_metadata(handle, bh);
 789                 if (err)
 790                         goto cleanup;
 791                 brelse(bh);
 792                 bh = NULL;
 793
 794                 /* correct old index */
 795                 if (m) {
 796                         err = ext4_ext_get_access(handle, inode, path + i);
 797                         if (err)
 798                                 goto cleanup;
 799                         path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
 800                         err = ext4_ext_dirty(handle, inode, path + i);
 801                         if (err)
 802                                 goto cleanup;
 803                 }
 804
 805                 i--;
 806         }
 807
 808         /* insert new index */
 809         err = ext4_ext_insert_index(handle, inode, path + at,
 810                                     le32_to_cpu(border), newblock);
 811
 812 cleanup:
 813         if (bh) {
 814                 if (buffer_locked(bh))
 815                         unlock_buffer(bh);
 816                 brelse(bh);
 817         }
 818
 819         if (err) {
 820                 /* free all allocated blocks in error case */
 821                 for (i = 0; i < depth; i++) {
 822                         if (!ablocks[i])
 823                                 continue;
 824                         ext4_free_blocks(handle, inode, ablocks[i], 1);
 825                 }
 826         }
 827         kfree(ablocks);
 828
 829         return err;
 830 }
 831
 832 /*
 833  * ext4_ext_grow_indepth:
 834  * implements tree growing procedure:
 835  * - allocates new block
 836  * - moves top-level data (index block or leaf) into the new block
 837  * - initializes new top-level, creating index that points to the
 838  *   just created block
 839  */
 840 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
 841                                         struct ext4_ext_path *path,
 842                                         struct ext4_extent *newext)
 843 {
 844         struct ext4_ext_path *curp = path;
 845         struct ext4_extent_header *neh;
 846         struct ext4_extent_idx *fidx;
 847         struct buffer_head *bh;
 848         ext4_fsblk_t newblock;
 849         int err = 0;
 850
 851         newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
 852         if (newblock == 0)
 853                 return err;
 854
 855         bh = sb_getblk(inode->i_sb, newblock);
 856         if (!bh) {
 857                 err = -EIO;
 858                 ext4_std_error(inode->i_sb, err);
 859                 return err;
 860         }
 861         lock_buffer(bh);
 862
 863         err = ext4_journal_get_create_access(handle, bh);
 864         if (err) {
 865                 unlock_buffer(bh);
 866                 goto out;
 867         }
 868
 869         /* move top-level index/leaf into new block */
 870         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
 871
 872         /* set size of new block */
 873         neh = ext_block_hdr(bh);
 874         /* old root could have indexes or leaves
 875          * so calculate e_max right way */
 876         if (ext_depth(inode))
 877           neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
 878         else
 879           neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
 880         neh->eh_magic = EXT4_EXT_MAGIC;
 881         set_buffer_uptodate(bh);
 882         unlock_buffer(bh);
 883
 884         err = ext4_journal_dirty_metadata(handle, bh);
 885         if (err)
 886                 goto out;
 887
 888         /* create index in new top-level index: num,max,pointer */
 889         err = ext4_ext_get_access(handle, inode, curp);
 890         if (err)
 891                 goto out;
 892
 893         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
 894         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
 895         curp->p_hdr->eh_entries = cpu_to_le16(1);
 896         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
 897         /* FIXME: it works, but actually path[0] can be index */
 898         curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
 899         ext4_idx_store_pblock(curp->p_idx, newblock);
 900
 901         neh = ext_inode_hdr(inode);
 902         fidx = EXT_FIRST_INDEX(neh);
 903         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
 904                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
 905                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
 906
 907         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
 908         err = ext4_ext_dirty(handle, inode, curp);
 909 out:
 910         brelse(bh);
 911
 912         return err;
 913 }
 914
 915 /*
 916  * ext4_ext_create_new_leaf:
 917  * finds empty index and adds new leaf.
 918  * if no free index is found, then it requests in-depth growing.
 919  */
 920 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
 921                                         struct ext4_ext_path *path,
 922                                         struct ext4_extent *newext)
 923 {
 924         struct ext4_ext_path *curp;
 925         int depth, i, err = 0;
 926
 927 repeat:
 928         i = depth = ext_depth(inode);
 929
 930         /* walk up to the tree and look for free index entry */
 931         curp = path + depth;
 932         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
 933                 i--;
 934                 curp--;
 935         }
 936
 937         /* we use already allocated block for index block,
 938          * so subsequent data blocks should be contiguous */
 939         if (EXT_HAS_FREE_INDEX(curp)) {
 940                 /* if we found index with free entry, then use that
 941                  * entry: create all needed subtree and add new leaf */
 942                 err = ext4_ext_split(handle, inode, path, newext, i);
 943
 944                 /* refill path */
 945                 ext4_ext_drop_refs(path);
 946                 path = ext4_ext_find_extent(inode,
 947                                             le32_to_cpu(newext->ee_block),
 948                                             path);
 949                 if (IS_ERR(path))
 950                         err = PTR_ERR(path);
 951         } else {
 952                 /* tree is full, time to grow in depth */
 953                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
 954                 if (err)
 955                         goto out;
 956
 957                 /* refill path */
 958                 ext4_ext_drop_refs(path);
 959                 path = ext4_ext_find_extent(inode,
 960                                             le32_to_cpu(newext->ee_block),
 961                                             path);
 962                 if (IS_ERR(path)) {
 963                         err = PTR_ERR(path);
 964                         goto out;
 965                 }
 966
 967                 /*
 968                  * only first (depth 0 -> 1) produces free space;
 969                  * in all other cases we have to split the grown tree
 970                  */
 971                 depth = ext_depth(inode);
 972                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
 973                         /* now we need to split */
 974                         goto repeat;
 975                 }
 976         }
 977
 978 out:
 979         return err;
 980 }
 981
 982 /*
 983  * ext4_ext_next_allocated_block:
 984  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
 985  * NOTE: it considers block number from index entry as
 986  * allocated block. Thus, index entries have to be consistent
 987  * with leaves.
 988  */
 989 static unsigned long
 990 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
 991 {
 992         int depth;
 993
 994         BUG_ON(path == NULL);
 995         depth = path->p_depth;
 996
 997         if (depth == 0 && path->p_ext == NULL)
 998                 return EXT_MAX_BLOCK;
 999
1000         while (depth >= 0) {
1001                 if (depth == path->p_depth) {
1002                         /* leaf */
1003                         if (path[depth].p_ext !=
1004                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1005                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1006                 } else {
1007                         /* index */
1008                         if (path[depth].p_idx !=
1009                                         EXT_LAST_INDEX(path[depth].p_hdr))
1010                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1011                 }
1012                 depth--;
1013         }
1014
1015         return EXT_MAX_BLOCK;
1016 }
1017
1018 /*
1019  * ext4_ext_next_leaf_block:
1020  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1021  */
1022 static unsigned ext4_ext_next_leaf_block(struct inode *inode,
1023                                         struct ext4_ext_path *path)
1024 {
1025         int depth;
1026
1027         BUG_ON(path == NULL);
1028         depth = path->p_depth;
1029
1030         /* zero-tree has no leaf blocks at all */
1031         if (depth == 0)
1032                 return EXT_MAX_BLOCK;
1033
1034         /* go to index block */
1035         depth--;
1036
1037         while (depth >= 0) {
1038                 if (path[depth].p_idx !=
1039                                 EXT_LAST_INDEX(path[depth].p_hdr))
1040                   return le32_to_cpu(path[depth].p_idx[1].ei_block);
1041                 depth--;
1042         }
1043
1044         return EXT_MAX_BLOCK;
1045 }
1046
1047 /*
1048  * ext4_ext_correct_indexes:
1049  * if leaf gets modified and modified extent is first in the leaf,
1050  * then we have to correct all indexes above.
1051  * TODO: do we need to correct tree in all cases?
1052  */
1053 int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1054                                 struct ext4_ext_path *path)
1055 {
1056         struct ext4_extent_header *eh;
1057         int depth = ext_depth(inode);
1058         struct ext4_extent *ex;
1059         __le32 border;
1060         int k, err = 0;
1061
1062         eh = path[depth].p_hdr;
1063         ex = path[depth].p_ext;
1064         BUG_ON(ex == NULL);
1065         BUG_ON(eh == NULL);
1066
1067         if (depth == 0) {
1068                 /* there is no tree at all */
1069                 return 0;
1070         }
1071
1072         if (ex != EXT_FIRST_EXTENT(eh)) {
1073                 /* we correct tree if first leaf got modified only */
1074                 return 0;
1075         }
1076
1077         /*
1078          * TODO: we need correction if border is smaller than current one
1079          */
1080         k = depth - 1;
1081         border = path[depth].p_ext->ee_block;
1082         err = ext4_ext_get_access(handle, inode, path + k);
1083         if (err)
1084                 return err;
1085         path[k].p_idx->ei_block = border;
1086         err = ext4_ext_dirty(handle, inode, path + k);
1087         if (err)
1088                 return err;
1089
1090         while (k--) {
1091                 /* change all left-side indexes */
1092                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1093                         break;
1094                 err = ext4_ext_get_access(handle, inode, path + k);
1095                 if (err)
1096                         break;
1097                 path[k].p_idx->ei_block = border;
1098                 err = ext4_ext_dirty(handle, inode, path + k);
1099                 if (err)
1100                         break;
1101         }
1102
1103         return err;
1104 }
1105
1106 static int
1107 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1108                                 struct ext4_extent *ex2)
1109 {
1110         if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len) !=
1111                         le32_to_cpu(ex2->ee_block))
1112                 return 0;
1113
1114         /*
1115          * To allow future support for preallocated extents to be added
1116          * as an RO_COMPAT feature, refuse to merge to extents if
1117          * this can result in the top bit of ee_len being set.
1118          */
1119         if (le16_to_cpu(ex1->ee_len) + le16_to_cpu(ex2->ee_len) > EXT_MAX_LEN)
1120                 return 0;
1121 #ifdef AGRESSIVE_TEST
1122         if (le16_to_cpu(ex1->ee_len) >= 4)
1123                 return 0;
1124 #endif
1125
1126         if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2))
1127                 return 1;
1128         return 0;
1129 }
1130
1131 /*
1132  * ext4_ext_insert_extent:
1133  * tries to merge requsted extent into the existing extent or
1134  * inserts requested extent as new one into the tree,
1135  * creating new leaf in the no-space case.
1136  */
1137 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1138                                 struct ext4_ext_path *path,
1139                                 struct ext4_extent *newext)
1140 {
1141         struct ext4_extent_header * eh;
1142         struct ext4_extent *ex, *fex;
1143         struct ext4_extent *nearex; /* nearest extent */
1144         struct ext4_ext_path *npath = NULL;
1145         int depth, len, err, next;
1146
1147         BUG_ON(newext->ee_len == 0);
1148         depth = ext_depth(inode);
1149         ex = path[depth].p_ext;
1150         BUG_ON(path[depth].p_hdr == NULL);
1151
1152         /* try to insert block into found extent and return */
1153         if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1154                 ext_debug("append %d block to %d:%d (from %llu)\n",
1155                                 le16_to_cpu(newext->ee_len),
1156                                 le32_to_cpu(ex->ee_block),
1157                                 le16_to_cpu(ex->ee_len), ext_pblock(ex));
1158                 err = ext4_ext_get_access(handle, inode, path + depth);
1159                 if (err)
1160                         return err;
1161                 ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len)
1162                                          + le16_to_cpu(newext->ee_len));
1163                 eh = path[depth].p_hdr;
1164                 nearex = ex;
1165                 goto merge;
1166         }
1167
1168 repeat:
1169         depth = ext_depth(inode);
1170         eh = path[depth].p_hdr;
1171         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1172                 goto has_space;
1173
1174         /* probably next leaf has space for us? */
1175         fex = EXT_LAST_EXTENT(eh);
1176         next = ext4_ext_next_leaf_block(inode, path);
1177         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1178             && next != EXT_MAX_BLOCK) {
1179                 ext_debug("next leaf block - %d\n", next);
1180                 BUG_ON(npath != NULL);
1181                 npath = ext4_ext_find_extent(inode, next, NULL);
1182                 if (IS_ERR(npath))
1183                         return PTR_ERR(npath);
1184                 BUG_ON(npath->p_depth != path->p_depth);
1185                 eh = npath[depth].p_hdr;
1186                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1187                         ext_debug("next leaf isnt full(%d)\n",
1188                                   le16_to_cpu(eh->eh_entries));
1189                         path = npath;
1190                         goto repeat;
1191                 }
1192                 ext_debug("next leaf has no free space(%d,%d)\n",
1193                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1194         }
1195
1196         /*
1197          * There is no free space in the found leaf.
1198          * We're gonna add a new leaf in the tree.
1199          */
1200         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1201         if (err)
1202                 goto cleanup;
1203         depth = ext_depth(inode);
1204         eh = path[depth].p_hdr;
1205
1206 has_space:
1207         nearex = path[depth].p_ext;
1208
1209         err = ext4_ext_get_access(handle, inode, path + depth);
1210         if (err)
1211                 goto cleanup;
1212
1213         if (!nearex) {
1214                 /* there is no extent in this leaf, create first one */
1215                 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1216                                 le32_to_cpu(newext->ee_block),
1217                                 ext_pblock(newext),
1218                                 le16_to_cpu(newext->ee_len));
1219                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1220         } else if (le32_to_cpu(newext->ee_block)
1221                            > le32_to_cpu(nearex->ee_block)) {
1222 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1223                 if (nearex != EXT_LAST_EXTENT(eh)) {
1224                         len = EXT_MAX_EXTENT(eh) - nearex;
1225                         len = (len - 1) * sizeof(struct ext4_extent);
1226                         len = len < 0 ? 0 : len;
1227                         ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1228                                         "move %d from 0x%p to 0x%p\n",
1229                                         le32_to_cpu(newext->ee_block),
1230                                         ext_pblock(newext),
1231                                         le16_to_cpu(newext->ee_len),
1232                                         nearex, len, nearex + 1, nearex + 2);
1233                         memmove(nearex + 2, nearex + 1, len);
1234                 }
1235                 path[depth].p_ext = nearex + 1;
1236         } else {
1237                 BUG_ON(newext->ee_block == nearex->ee_block);
1238                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1239                 len = len < 0 ? 0 : len;
1240                 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1241                                 "move %d from 0x%p to 0x%p\n",
1242                                 le32_to_cpu(newext->ee_block),
1243                                 ext_pblock(newext),
1244                                 le16_to_cpu(newext->ee_len),
1245                                 nearex, len, nearex + 1, nearex + 2);
1246                 memmove(nearex + 1, nearex, len);
1247                 path[depth].p_ext = nearex;
1248         }
1249
1250         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1251         nearex = path[depth].p_ext;
1252         nearex->ee_block = newext->ee_block;
1253         nearex->ee_start = newext->ee_start;
1254         nearex->ee_start_hi = newext->ee_start_hi;
1255         nearex->ee_len = newext->ee_len;
1256
1257 merge:
1258         /* try to merge extents to the right */
1259         while (nearex < EXT_LAST_EXTENT(eh)) {
1260                 if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1))
1261                         break;
1262                 /* merge with next extent! */
1263                 nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len)
1264                                              + le16_to_cpu(nearex[1].ee_len));
1265                 if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1266                         len = (EXT_LAST_EXTENT(eh) - nearex - 1)
1267                                         * sizeof(struct ext4_extent);
1268                         memmove(nearex + 1, nearex + 2, len);
1269                 }
1270                 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1271                 BUG_ON(eh->eh_entries == 0);
1272         }
1273
1274         /* try to merge extents to the left */
1275
1276         /* time to correct all indexes above */
1277         err = ext4_ext_correct_indexes(handle, inode, path);
1278         if (err)
1279                 goto cleanup;
1280
1281         err = ext4_ext_dirty(handle, inode, path + depth);
1282
1283 cleanup:
1284         if (npath) {
1285                 ext4_ext_drop_refs(npath);
1286                 kfree(npath);
1287         }
1288         ext4_ext_tree_changed(inode);
1289         ext4_ext_invalidate_cache(inode);
1290         return err;
1291 }
1292
1293 int ext4_ext_walk_space(struct inode *inode, unsigned long block,
1294                         unsigned long num, ext_prepare_callback func,
1295                         void *cbdata)
1296 {
1297         struct ext4_ext_path *path = NULL;
1298         struct ext4_ext_cache cbex;
1299         struct ext4_extent *ex;
1300         unsigned long next, start = 0, end = 0;
1301         unsigned long last = block + num;
1302         int depth, exists, err = 0;
1303
1304         BUG_ON(func == NULL);
1305         BUG_ON(inode == NULL);
1306
1307         while (block < last && block != EXT_MAX_BLOCK) {
1308                 num = last - block;
1309                 /* find extent for this block */
1310                 path = ext4_ext_find_extent(inode, block, path);
1311                 if (IS_ERR(path)) {
1312                         err = PTR_ERR(path);
1313                         path = NULL;
1314                         break;
1315                 }
1316
1317                 depth = ext_depth(inode);
1318                 BUG_ON(path[depth].p_hdr == NULL);
1319                 ex = path[depth].p_ext;
1320                 next = ext4_ext_next_allocated_block(path);
1321
1322                 exists = 0;
1323                 if (!ex) {
1324                         /* there is no extent yet, so try to allocate
1325                          * all requested space */
1326                         start = block;
1327                         end = block + num;
1328                 } else if (le32_to_cpu(ex->ee_block) > block) {
1329                         /* need to allocate space before found extent */
1330                         start = block;
1331                         end = le32_to_cpu(ex->ee_block);
1332                         if (block + num < end)
1333                                 end = block + num;
1334                 } else if (block >=
1335                              le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) {
1336                         /* need to allocate space after found extent */
1337                         start = block;
1338                         end = block + num;
1339                         if (end >= next)
1340                                 end = next;
1341                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1342                         /*
1343                          * some part of requested space is covered
1344                          * by found extent
1345                          */
1346                         start = block;
1347                         end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len);
1348                         if (block + num < end)
1349                                 end = block + num;
1350                         exists = 1;
1351                 } else {
1352                         BUG();
1353                 }
1354                 BUG_ON(end <= start);
1355
1356                 if (!exists) {
1357                         cbex.ec_block = start;
1358                         cbex.ec_len = end - start;
1359                         cbex.ec_start = 0;
1360                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1361                 } else {
1362                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1363                         cbex.ec_len = le16_to_cpu(ex->ee_len);
1364                         cbex.ec_start = ext_pblock(ex);
1365                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1366                 }
1367
1368                 BUG_ON(cbex.ec_len == 0);
1369                 err = func(inode, path, &cbex, cbdata);
1370                 ext4_ext_drop_refs(path);
1371
1372                 if (err < 0)
1373                         break;
1374                 if (err == EXT_REPEAT)
1375                         continue;
1376                 else if (err == EXT_BREAK) {
1377                         err = 0;
1378                         break;
1379                 }
1380
1381                 if (ext_depth(inode) != depth) {
1382                         /* depth was changed. we have to realloc path */
1383                         kfree(path);
1384                         path = NULL;
1385                 }
1386
1387                 block = cbex.ec_block + cbex.ec_len;
1388         }
1389
1390         if (path) {
1391                 ext4_ext_drop_refs(path);
1392                 kfree(path);
1393         }
1394
1395         return err;
1396 }
1397
1398 static void
1399 ext4_ext_put_in_cache(struct inode *inode, __u32 block,
1400                         __u32 len, __u32 start, int type)
1401 {
1402         struct ext4_ext_cache *cex;
1403         BUG_ON(len == 0);
1404         cex = &EXT4_I(inode)->i_cached_extent;
1405         cex->ec_type = type;
1406         cex->ec_block = block;
1407         cex->ec_len = len;
1408         cex->ec_start = start;
1409 }
1410
1411 /*
1412  * ext4_ext_put_gap_in_cache:
1413  * calculate boundaries of the gap that the requested block fits into
1414  * and cache this gap
1415  */
1416 static void
1417 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1418                                 unsigned long block)
1419 {
1420         int depth = ext_depth(inode);
1421         unsigned long lblock, len;
1422         struct ext4_extent *ex;
1423
1424         ex = path[depth].p_ext;
1425         if (ex == NULL) {
1426                 /* there is no extent yet, so gap is [0;-] */
1427                 lblock = 0;
1428                 len = EXT_MAX_BLOCK;
1429                 ext_debug("cache gap(whole file):");
1430         } else if (block < le32_to_cpu(ex->ee_block)) {
1431                 lblock = block;
1432                 len = le32_to_cpu(ex->ee_block) - block;
1433                 ext_debug("cache gap(before): %lu [%lu:%lu]",
1434                                 (unsigned long) block,
1435                                 (unsigned long) le32_to_cpu(ex->ee_block),
1436                                 (unsigned long) le16_to_cpu(ex->ee_len));
1437         } else if (block >= le32_to_cpu(ex->ee_block)
1438                             + le16_to_cpu(ex->ee_len)) {
1439                 lblock = le32_to_cpu(ex->ee_block)
1440                          + le16_to_cpu(ex->ee_len);
1441                 len = ext4_ext_next_allocated_block(path);
1442                 ext_debug("cache gap(after): [%lu:%lu] %lu",
1443                                 (unsigned long) le32_to_cpu(ex->ee_block),
1444                                 (unsigned long) le16_to_cpu(ex->ee_len),
1445                                 (unsigned long) block);
1446                 BUG_ON(len == lblock);
1447                 len = len - lblock;
1448         } else {
1449                 lblock = len = 0;
1450                 BUG();
1451         }
1452
1453         ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len);
1454         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1455 }
1456
1457 static int
1458 ext4_ext_in_cache(struct inode *inode, unsigned long block,
1459                         struct ext4_extent *ex)
1460 {
1461         struct ext4_ext_cache *cex;
1462
1463         cex = &EXT4_I(inode)->i_cached_extent;
1464
1465         /* has cache valid data? */
1466         if (cex->ec_type == EXT4_EXT_CACHE_NO)
1467                 return EXT4_EXT_CACHE_NO;
1468
1469         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1470                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1471         if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1472                 ex->ee_block = cpu_to_le32(cex->ec_block);
1473                 ext4_ext_store_pblock(ex, cex->ec_start);
1474                 ex->ee_len = cpu_to_le16(cex->ec_len);
1475                 ext_debug("%lu cached by %lu:%lu:%llu\n",
1476                                 (unsigned long) block,
1477                                 (unsigned long) cex->ec_block,
1478                                 (unsigned long) cex->ec_len,
1479                                 cex->ec_start);
1480                 return cex->ec_type;
1481         }
1482
1483         /* not in cache */
1484         return EXT4_EXT_CACHE_NO;
1485 }
1486
1487 /*
1488  * ext4_ext_rm_idx:
1489  * removes index from the index block.
1490  * It's used in truncate case only, thus all requests are for
1491  * last index in the block only.
1492  */
1493 int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1494                         struct ext4_ext_path *path)
1495 {
1496         struct buffer_head *bh;
1497         int err;
1498         ext4_fsblk_t leaf;
1499
1500         /* free index block */
1501         path--;
1502         leaf = idx_pblock(path->p_idx);
1503         BUG_ON(path->p_hdr->eh_entries == 0);
1504         err = ext4_ext_get_access(handle, inode, path);
1505         if (err)
1506                 return err;
1507         path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1508         err = ext4_ext_dirty(handle, inode, path);
1509         if (err)
1510                 return err;
1511         ext_debug("index is empty, remove it, free block %llu\n", leaf);
1512         bh = sb_find_get_block(inode->i_sb, leaf);
1513         ext4_forget(handle, 1, inode, bh, leaf);
1514         ext4_free_blocks(handle, inode, leaf, 1);
1515         return err;
1516 }
1517
1518 /*
1519  * ext4_ext_calc_credits_for_insert:
1520  * This routine returns max. credits that the extent tree can consume.
1521  * It should be OK for low-performance paths like ->writepage()
1522  * To allow many writing processes to fit into a single transaction,
1523  * the caller should calculate credits under truncate_mutex and
1524  * pass the actual path.
1525  */
1526 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1527                                                 struct ext4_ext_path *path)
1528 {
1529         int depth, needed;
1530
1531         if (path) {
1532                 /* probably there is space in leaf? */
1533                 depth = ext_depth(inode);
1534                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1535                                 < le16_to_cpu(path[depth].p_hdr->eh_max))
1536                         return 1;
1537         }
1538
1539         /*
1540          * given 32-bit logical block (4294967296 blocks), max. tree
1541          * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1542          * Let's also add one more level for imbalance.
1543          */
1544         depth = 5;
1545
1546         /* allocation of new data block(s) */
1547         needed = 2;
1548
1549         /*
1550          * tree can be full, so it would need to grow in depth:
1551          * we need one credit to modify old root, credits for
1552          * new root will be added in split accounting
1553          */
1554         needed += 1;
1555
1556         /*
1557          * Index split can happen, we would need:
1558          *    allocate intermediate indexes (bitmap + group)
1559          *  + change two blocks at each level, but root (already included)
1560          */
1561         needed += (depth * 2) + (depth * 2);
1562
1563         /* any allocation modifies superblock */
1564         needed += 1;
1565
1566         return needed;
1567 }
1568
1569 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1570                                 struct ext4_extent *ex,
1571                                 unsigned long from, unsigned long to)
1572 {
1573         struct buffer_head *bh;
1574         int i;
1575
1576 #ifdef EXTENTS_STATS
1577         {
1578                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1579                 unsigned short ee_len =  le16_to_cpu(ex->ee_len);
1580                 spin_lock(&sbi->s_ext_stats_lock);
1581                 sbi->s_ext_blocks += ee_len;
1582                 sbi->s_ext_extents++;
1583                 if (ee_len < sbi->s_ext_min)
1584                         sbi->s_ext_min = ee_len;
1585                 if (ee_len > sbi->s_ext_max)
1586                         sbi->s_ext_max = ee_len;
1587                 if (ext_depth(inode) > sbi->s_depth_max)
1588                         sbi->s_depth_max = ext_depth(inode);
1589                 spin_unlock(&sbi->s_ext_stats_lock);
1590         }
1591 #endif
1592         if (from >= le32_to_cpu(ex->ee_block)
1593             && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1594                 /* tail removal */
1595                 unsigned long num;
1596                 ext4_fsblk_t start;
1597                 num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from;
1598                 start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num;
1599                 ext_debug("free last %lu blocks starting %llu\n", num, start);
1600                 for (i = 0; i < num; i++) {
1601                         bh = sb_find_get_block(inode->i_sb, start + i);
1602                         ext4_forget(handle, 0, inode, bh, start + i);
1603                 }
1604                 ext4_free_blocks(handle, inode, start, num);
1605         } else if (from == le32_to_cpu(ex->ee_block)
1606                    && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1607                 printk("strange request: removal %lu-%lu from %u:%u\n",
1608                        from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1609         } else {
1610                 printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1611                        from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1612         }
1613         return 0;
1614 }
1615
1616 static int
1617 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1618                 struct ext4_ext_path *path, unsigned long start)
1619 {
1620         int err = 0, correct_index = 0;
1621         int depth = ext_depth(inode), credits;
1622         struct ext4_extent_header *eh;
1623         unsigned a, b, block, num;
1624         unsigned long ex_ee_block;
1625         unsigned short ex_ee_len;
1626         struct ext4_extent *ex;
1627
1628         ext_debug("truncate since %lu in leaf\n", start);
1629         if (!path[depth].p_hdr)
1630                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1631         eh = path[depth].p_hdr;
1632         BUG_ON(eh == NULL);
1633         BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
1634         BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
1635
1636         /* find where to start removing */
1637         ex = EXT_LAST_EXTENT(eh);
1638
1639         ex_ee_block = le32_to_cpu(ex->ee_block);
1640         ex_ee_len = le16_to_cpu(ex->ee_len);
1641
1642         while (ex >= EXT_FIRST_EXTENT(eh) &&
1643                         ex_ee_block + ex_ee_len > start) {
1644                 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1645                 path[depth].p_ext = ex;
1646
1647                 a = ex_ee_block > start ? ex_ee_block : start;
1648                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1649                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1650
1651                 ext_debug("  border %u:%u\n", a, b);
1652
1653                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1654                         block = 0;
1655                         num = 0;
1656                         BUG();
1657                 } else if (a != ex_ee_block) {
1658                         /* remove tail of the extent */
1659                         block = ex_ee_block;
1660                         num = a - block;
1661                 } else if (b != ex_ee_block + ex_ee_len - 1) {
1662                         /* remove head of the extent */
1663                         block = a;
1664                         num = b - a;
1665                         /* there is no "make a hole" API yet */
1666                         BUG();
1667                 } else {
1668                         /* remove whole extent: excellent! */
1669                         block = ex_ee_block;
1670                         num = 0;
1671                         BUG_ON(a != ex_ee_block);
1672                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1673                 }
1674
1675                 /* at present, extent can't cross block group: */
1676                 /* leaf + bitmap + group desc + sb + inode */
1677                 credits = 5;
1678                 if (ex == EXT_FIRST_EXTENT(eh)) {
1679                         correct_index = 1;
1680                         credits += (ext_depth(inode)) + 1;
1681                 }
1682 #ifdef CONFIG_QUOTA
1683                 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1684 #endif
1685
1686                 handle = ext4_ext_journal_restart(handle, credits);
1687                 if (IS_ERR(handle)) {
1688                         err = PTR_ERR(handle);
1689                         goto out;
1690                 }
1691
1692                 err = ext4_ext_get_access(handle, inode, path + depth);
1693                 if (err)
1694                         goto out;
1695
1696                 err = ext4_remove_blocks(handle, inode, ex, a, b);
1697                 if (err)
1698                         goto out;
1699
1700                 if (num == 0) {
1701                         /* this extent is removed; mark slot entirely unused */
1702                         ext4_ext_store_pblock(ex, 0);
1703                         eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1704                 }
1705
1706                 ex->ee_block = cpu_to_le32(block);
1707                 ex->ee_len = cpu_to_le16(num);
1708
1709                 err = ext4_ext_dirty(handle, inode, path + depth);
1710                 if (err)
1711                         goto out;
1712
1713                 ext_debug("new extent: %u:%u:%llu\n", block, num,
1714                                 ext_pblock(ex));
1715                 ex--;
1716                 ex_ee_block = le32_to_cpu(ex->ee_block);
1717                 ex_ee_len = le16_to_cpu(ex->ee_len);
1718         }
1719
1720         if (correct_index && eh->eh_entries)
1721                 err = ext4_ext_correct_indexes(handle, inode, path);
1722
1723         /* if this leaf is free, then we should
1724          * remove it from index block above */
1725         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1726                 err = ext4_ext_rm_idx(handle, inode, path + depth);
1727
1728 out:
1729         return err;
1730 }
1731
1732 /*
1733  * ext4_ext_more_to_rm:
1734  * returns 1 if current index has to be freed (even partial)
1735  */
1736 static int
1737 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1738 {
1739         BUG_ON(path->p_idx == NULL);
1740
1741         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1742                 return 0;
1743
1744         /*
1745          * if truncate on deeper level happened, it wasn't partial,
1746          * so we have to consider current index for truncation
1747          */
1748         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1749                 return 0;
1750         return 1;
1751 }
1752
1753 int ext4_ext_remove_space(struct inode *inode, unsigned long start)
1754 {
1755         struct super_block *sb = inode->i_sb;
1756         int depth = ext_depth(inode);
1757         struct ext4_ext_path *path;
1758         handle_t *handle;
1759         int i = 0, err = 0;
1760
1761         ext_debug("truncate since %lu\n", start);
1762
1763         /* probably first extent we're gonna free will be last in block */
1764         handle = ext4_journal_start(inode, depth + 1);
1765         if (IS_ERR(handle))
1766                 return PTR_ERR(handle);
1767
1768         ext4_ext_invalidate_cache(inode);
1769
1770         /*
1771          * We start scanning from right side, freeing all the blocks
1772          * after i_size and walking into the tree depth-wise.
1773          */
1774         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1775         if (path == NULL) {
1776                 ext4_journal_stop(handle);
1777                 return -ENOMEM;
1778         }
1779         path[0].p_hdr = ext_inode_hdr(inode);
1780         if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) {
1781                 err = -EIO;
1782                 goto out;
1783         }
1784         path[0].p_depth = depth;
1785
1786         while (i >= 0 && err == 0) {
1787                 if (i == depth) {
1788                         /* this is leaf block */
1789                         err = ext4_ext_rm_leaf(handle, inode, path, start);
1790                         /* root level has p_bh == NULL, brelse() eats this */
1791                         brelse(path[i].p_bh);
1792                         path[i].p_bh = NULL;
1793                         i--;
1794                         continue;
1795                 }
1796
1797                 /* this is index block */
1798                 if (!path[i].p_hdr) {
1799                         ext_debug("initialize header\n");
1800                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1801                         if (ext4_ext_check_header(__FUNCTION__, inode,
1802                                                         path[i].p_hdr)) {
1803                                 err = -EIO;
1804                                 goto out;
1805                         }
1806                 }
1807
1808                 BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries)
1809                            > le16_to_cpu(path[i].p_hdr->eh_max));
1810                 BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC);
1811
1812                 if (!path[i].p_idx) {
1813                         /* this level hasn't been touched yet */
1814                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1815                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1816                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
1817                                   path[i].p_hdr,
1818                                   le16_to_cpu(path[i].p_hdr->eh_entries));
1819                 } else {
1820                         /* we were already here, see at next index */
1821                         path[i].p_idx--;
1822                 }
1823
1824                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1825                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
1826                                 path[i].p_idx);
1827                 if (ext4_ext_more_to_rm(path + i)) {
1828                         /* go to the next level */
1829                         ext_debug("move to level %d (block %llu)\n",
1830                                   i + 1, idx_pblock(path[i].p_idx));
1831                         memset(path + i + 1, 0, sizeof(*path));
1832                         path[i+1].p_bh =
1833                                 sb_bread(sb, idx_pblock(path[i].p_idx));
1834                         if (!path[i+1].p_bh) {
1835                                 /* should we reset i_size? */
1836                                 err = -EIO;
1837                                 break;
1838                         }
1839
1840                         /* save actual number of indexes since this
1841                          * number is changed at the next iteration */
1842                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1843                         i++;
1844                 } else {
1845                         /* we finished processing this index, go up */
1846                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
1847                                 /* index is empty, remove it;
1848                                  * handle must be already prepared by the
1849                                  * truncatei_leaf() */
1850                                 err = ext4_ext_rm_idx(handle, inode, path + i);
1851                         }
1852                         /* root level has p_bh == NULL, brelse() eats this */
1853                         brelse(path[i].p_bh);
1854                         path[i].p_bh = NULL;
1855                         i--;
1856                         ext_debug("return to level %d\n", i);
1857                 }
1858         }
1859
1860         /* TODO: flexible tree reduction should be here */
1861         if (path->p_hdr->eh_entries == 0) {
1862                 /*
1863                  * truncate to zero freed all the tree,
1864                  * so we need to correct eh_depth
1865                  */
1866                 err = ext4_ext_get_access(handle, inode, path);
1867                 if (err == 0) {
1868                         ext_inode_hdr(inode)->eh_depth = 0;
1869                         ext_inode_hdr(inode)->eh_max =
1870                                 cpu_to_le16(ext4_ext_space_root(inode));
1871                         err = ext4_ext_dirty(handle, inode, path);
1872                 }
1873         }
1874 out:
1875         ext4_ext_tree_changed(inode);
1876         ext4_ext_drop_refs(path);
1877         kfree(path);
1878         ext4_journal_stop(handle);
1879
1880         return err;
1881 }
1882
1883 /*
1884  * called at mount time
1885  */
1886 void ext4_ext_init(struct super_block *sb)
1887 {
1888         /*
1889          * possible initialization would be here
1890          */
1891
1892         if (test_opt(sb, EXTENTS)) {
1893                 printk("EXT4-fs: file extents enabled");
1894 #ifdef AGRESSIVE_TEST
1895                 printk(", agressive tests");
1896 #endif
1897 #ifdef CHECK_BINSEARCH
1898                 printk(", check binsearch");
1899 #endif
1900 #ifdef EXTENTS_STATS
1901                 printk(", stats");
1902 #endif
1903                 printk("\n");
1904 #ifdef EXTENTS_STATS
1905                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
1906                 EXT4_SB(sb)->s_ext_min = 1 << 30;
1907                 EXT4_SB(sb)->s_ext_max = 0;
1908 #endif
1909         }
1910 }
1911
1912 /*
1913  * called at umount time
1914  */
1915 void ext4_ext_release(struct super_block *sb)
1916 {
1917         if (!test_opt(sb, EXTENTS))
1918                 return;
1919
1920 #ifdef EXTENTS_STATS
1921         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
1922                 struct ext4_sb_info *sbi = EXT4_SB(sb);
1923                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
1924                         sbi->s_ext_blocks, sbi->s_ext_extents,
1925                         sbi->s_ext_blocks / sbi->s_ext_extents);
1926                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
1927                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
1928         }
1929 #endif
1930 }
1931
1932 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
1933                         ext4_fsblk_t iblock,
1934                         unsigned long max_blocks, struct buffer_head *bh_result,
1935                         int create, int extend_disksize)
1936 {
1937         struct ext4_ext_path *path = NULL;
1938         struct ext4_extent newex, *ex;
1939         ext4_fsblk_t goal, newblock;
1940         int err = 0, depth;
1941         unsigned long allocated = 0;
1942
1943         __clear_bit(BH_New, &bh_result->b_state);
1944         ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock,
1945                         max_blocks, (unsigned) inode->i_ino);
1946         mutex_lock(&EXT4_I(inode)->truncate_mutex);
1947
1948         /* check in cache */
1949         goal = ext4_ext_in_cache(inode, iblock, &newex);
1950         if (goal) {
1951                 if (goal == EXT4_EXT_CACHE_GAP) {
1952                         if (!create) {
1953                                 /* block isn't allocated yet and
1954                                  * user doesn't want to allocate it */
1955                                 goto out2;
1956                         }
1957                         /* we should allocate requested block */
1958                 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
1959                         /* block is already allocated */
1960                         newblock = iblock
1961                                    - le32_to_cpu(newex.ee_block)
1962                                    + ext_pblock(&newex);
1963                         /* number of remaining blocks in the extent */
1964                         allocated = le16_to_cpu(newex.ee_len) -
1965                                         (iblock - le32_to_cpu(newex.ee_block));
1966                         goto out;
1967                 } else {
1968                         BUG();
1969                 }
1970         }
1971
1972         /* find extent for this block */
1973         path = ext4_ext_find_extent(inode, iblock, NULL);
1974         if (IS_ERR(path)) {
1975                 err = PTR_ERR(path);
1976                 path = NULL;
1977                 goto out2;
1978         }
1979
1980         depth = ext_depth(inode);
1981
1982         /*
1983          * consistent leaf must not be empty;
1984          * this situation is possible, though, _during_ tree modification;
1985          * this is why assert can't be put in ext4_ext_find_extent()
1986          */
1987         BUG_ON(path[depth].p_ext == NULL && depth != 0);
1988
1989         ex = path[depth].p_ext;
1990         if (ex) {
1991                 unsigned long ee_block = le32_to_cpu(ex->ee_block);
1992                 ext4_fsblk_t ee_start = ext_pblock(ex);
1993                 unsigned short ee_len  = le16_to_cpu(ex->ee_len);
1994
1995                 /*
1996                  * Allow future support for preallocated extents to be added
1997                  * as an RO_COMPAT feature:
1998                  * Uninitialized extents are treated as holes, except that
1999                  * we avoid (fail) allocating new blocks during a write.
2000                  */
2001                 if (ee_len > EXT_MAX_LEN)
2002                         goto out2;
2003                 /* if found extent covers block, simply return it */
2004                 if (iblock >= ee_block && iblock < ee_block + ee_len) {
2005                         newblock = iblock - ee_block + ee_start;
2006                         /* number of remaining blocks in the extent */
2007                         allocated = ee_len - (iblock - ee_block);
2008                         ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
2009                                         ee_block, ee_len, newblock);
2010                         ext4_ext_put_in_cache(inode, ee_block, ee_len,
2011                                                 ee_start, EXT4_EXT_CACHE_EXTENT);
2012                         goto out;
2013                 }
2014         }
2015
2016         /*
2017          * requested block isn't allocated yet;
2018          * we couldn't try to create block if create flag is zero
2019          */
2020         if (!create) {
2021                 /* put just found gap into cache to speed up
2022                  * subsequent requests */
2023                 ext4_ext_put_gap_in_cache(inode, path, iblock);
2024                 goto out2;
2025         }
2026         /*
2027          * Okay, we need to do block allocation.  Lazily initialize the block
2028          * allocation info here if necessary.
2029          */
2030         if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2031                 ext4_init_block_alloc_info(inode);
2032
2033         /* allocate new block */
2034         goal = ext4_ext_find_goal(inode, path, iblock);
2035         allocated = max_blocks;
2036         newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2037         if (!newblock)
2038                 goto out2;
2039         ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2040                         goal, newblock, allocated);
2041
2042         /* try to insert new extent into found leaf and return */
2043         newex.ee_block = cpu_to_le32(iblock);
2044         ext4_ext_store_pblock(&newex, newblock);
2045         newex.ee_len = cpu_to_le16(allocated);
2046         err = ext4_ext_insert_extent(handle, inode, path, &newex);
2047         if (err)
2048                 goto out2;
2049
2050         if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2051                 EXT4_I(inode)->i_disksize = inode->i_size;
2052
2053         /* previous routine could use block we allocated */
2054         newblock = ext_pblock(&newex);
2055         __set_bit(BH_New, &bh_result->b_state);
2056
2057         ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2058                                 EXT4_EXT_CACHE_EXTENT);
2059 out:
2060         if (allocated > max_blocks)
2061                 allocated = max_blocks;
2062         ext4_ext_show_leaf(inode, path);
2063         __set_bit(BH_Mapped, &bh_result->b_state);
2064         bh_result->b_bdev = inode->i_sb->s_bdev;
2065         bh_result->b_blocknr = newblock;
2066 out2:
2067         if (path) {
2068                 ext4_ext_drop_refs(path);
2069                 kfree(path);
2070         }
2071         mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2072
2073         return err ? err : allocated;
2074 }
2075
2076 void ext4_ext_truncate(struct inode * inode, struct page *page)
2077 {
2078         struct address_space *mapping = inode->i_mapping;
2079         struct super_block *sb = inode->i_sb;
2080         unsigned long last_block;
2081         handle_t *handle;
2082         int err = 0;
2083
2084         /*
2085          * probably first extent we're gonna free will be last in block
2086          */
2087         err = ext4_writepage_trans_blocks(inode) + 3;
2088         handle = ext4_journal_start(inode, err);
2089         if (IS_ERR(handle)) {
2090                 if (page) {
2091                         clear_highpage(page);
2092                         flush_dcache_page(page);
2093                         unlock_page(page);
2094                         page_cache_release(page);
2095                 }
2096                 return;
2097         }
2098
2099         if (page)
2100                 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2101
2102         mutex_lock(&EXT4_I(inode)->truncate_mutex);
2103         ext4_ext_invalidate_cache(inode);
2104
2105         /*
2106          * TODO: optimization is possible here.
2107          * Probably we need not scan at all,
2108          * because page truncation is enough.
2109          */
2110         if (ext4_orphan_add(handle, inode))
2111                 goto out_stop;
2112
2113         /* we have to know where to truncate from in crash case */
2114         EXT4_I(inode)->i_disksize = inode->i_size;
2115         ext4_mark_inode_dirty(handle, inode);
2116
2117         last_block = (inode->i_size + sb->s_blocksize - 1)
2118                         >> EXT4_BLOCK_SIZE_BITS(sb);
2119         err = ext4_ext_remove_space(inode, last_block);
2120
2121         /* In a multi-transaction truncate, we only make the final
2122          * transaction synchronous. */
2123         if (IS_SYNC(inode))
2124                 handle->h_sync = 1;
2125
2126 out_stop:
2127         /*
2128          * If this was a simple ftruncate() and the file will remain alive,
2129          * then we need to clear up the orphan record which we created above.
2130          * However, if this was a real unlink then we were called by
2131          * ext4_delete_inode(), and we allow that function to clean up the
2132          * orphan info for us.
2133          */
2134         if (inode->i_nlink)
2135                 ext4_orphan_del(handle, inode);
2136
2137         mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2138         ext4_journal_stop(handle);
2139 }
2140
2141 /*
2142  * ext4_ext_writepage_trans_blocks:
2143  * calculate max number of blocks we could modify
2144  * in order to allocate new block for an inode
2145  */
2146 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2147 {
2148         int needed;
2149
2150         needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2151
2152         /* caller wants to allocate num blocks, but note it includes sb */
2153         needed = needed * num - (num - 1);
2154
2155 #ifdef CONFIG_QUOTA
2156         needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2157 #endif
2158
2159         return needed;
2160 }
2161
2162 EXPORT_SYMBOL(ext4_mark_inode_dirty);
2163 EXPORT_SYMBOL(ext4_ext_invalidate_cache);
2164 EXPORT_SYMBOL(ext4_ext_insert_extent);
2165 EXPORT_SYMBOL(ext4_ext_walk_space);
2166 EXPORT_SYMBOL(ext4_ext_find_goal);
2167 EXPORT_SYMBOL(ext4_ext_calc_credits_for_insert);
2168