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/jbd2.h>
  36 #include <linux/highuid.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/string.h>
  40 #include <linux/slab.h>
  41 #include <linux/falloc.h>
  42 #include <asm/uaccess.h>
  43 #include <linux/fiemap.h>
  44 #include "ext4_jbd2.h"
  45 #include "ext4_extents.h"
  46
  47
  48 /*
  49  * ext_pblock:
  50  * combine low and high parts of physical block number into ext4_fsblk_t
  51  */
  52 ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
  53 {
  54         ext4_fsblk_t block;
  55
  56         block = le32_to_cpu(ex->ee_start_lo);
  57         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
  58         return block;
  59 }
  60
  61 /*
  62  * idx_pblock:
  63  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
  64  */
  65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
  66 {
  67         ext4_fsblk_t block;
  68
  69         block = le32_to_cpu(ix->ei_leaf_lo);
  70         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
  71         return block;
  72 }
  73
  74 /*
  75  * ext4_ext_store_pblock:
  76  * stores a large physical block number into an extent struct,
  77  * breaking it into parts
  78  */
  79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
  80 {
  81         ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  82         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  83 }
  84
  85 /*
  86  * ext4_idx_store_pblock:
  87  * stores a large physical block number into an index struct,
  88  * breaking it into parts
  89  */
  90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
  91 {
  92         ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  93         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  94 }
  95
  96 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  97                                             struct inode *inode,
  98                                             int needed)
  99 {
 100         int err;
 101
 102         if (!ext4_handle_valid(handle))
 103                 return 0;
 104         if (handle->h_buffer_credits > needed)
 105                 return 0;
 106         err = ext4_journal_extend(handle, needed);
 107         if (err <= 0)
 108                 return err;
 109         err = ext4_truncate_restart_trans(handle, inode, needed);
 110         /*
 111          * We have dropped i_data_sem so someone might have cached again
 112          * an extent we are going to truncate.
 113          */
 114         ext4_ext_invalidate_cache(inode);
 115
 116         return err;
 117 }
 118
 119 /*
 120  * could return:
 121  *  - EROFS
 122  *  - ENOMEM
 123  */
 124 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 125                                 struct ext4_ext_path *path)
 126 {
 127         if (path->p_bh) {
 128                 /* path points to block */
 129                 return ext4_journal_get_write_access(handle, path->p_bh);
 130         }
 131         /* path points to leaf/index in inode body */
 132         /* we use in-core data, no need to protect them */
 133         return 0;
 134 }
 135
 136 /*
 137  * could return:
 138  *  - EROFS
 139  *  - ENOMEM
 140  *  - EIO
 141  */
 142 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
 143                                 struct ext4_ext_path *path)
 144 {
 145         int err;
 146         if (path->p_bh) {
 147                 /* path points to block */
 148                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
 149         } else {
 150                 /* path points to leaf/index in inode body */
 151                 err = ext4_mark_inode_dirty(handle, inode);
 152         }
 153         return err;
 154 }
 155
 156 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 157                               struct ext4_ext_path *path,
 158                               ext4_lblk_t block)
 159 {
 160         struct ext4_inode_info *ei = EXT4_I(inode);
 161         ext4_fsblk_t bg_start;
 162         ext4_fsblk_t last_block;
 163         ext4_grpblk_t colour;
 164         ext4_group_t block_group;
 165         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 166         int depth;
 167
 168         if (path) {
 169                 struct ext4_extent *ex;
 170                 depth = path->p_depth;
 171
 172                 /* try to predict block placement */
 173                 ex = path[depth].p_ext;
 174                 if (ex)
 175                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
 176
 177                 /* it looks like index is empty;
 178                  * try to find starting block from index itself */
 179                 if (path[depth].p_bh)
 180                         return path[depth].p_bh->b_blocknr;
 181         }
 182
 183         /* OK. use inode's group */
 184         block_group = ei->i_block_group;
 185         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 186                 /*
 187                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 188                  * block groups per flexgroup, reserve the first block
 189                  * group for directories and special files.  Regular
 190                  * files will start at the second block group.  This
 191                  * tends to speed up directory access and improves
 192                  * fsck times.
 193                  */
 194                 block_group &= ~(flex_size-1);
 195                 if (S_ISREG(inode->i_mode))
 196                         block_group++;
 197         }
 198         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 199         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 200
 201         /*
 202          * If we are doing delayed allocation, we don't need take
 203          * colour into account.
 204          */
 205         if (test_opt(inode->i_sb, DELALLOC))
 206                 return bg_start;
 207
 208         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 209                 colour = (current->pid % 16) *
 210                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 211         else
 212                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 213         return bg_start + colour + block;
 214 }
 215
 216 /*
 217  * Allocation for a meta data block
 218  */
 219 static ext4_fsblk_t
 220 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 221                         struct ext4_ext_path *path,
 222                         struct ext4_extent *ex, int *err)
 223 {
 224         ext4_fsblk_t goal, newblock;
 225
 226         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 227         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
 228         return newblock;
 229 }
 230
 231 static inline int ext4_ext_space_block(struct inode *inode, int check)
 232 {
 233         int size;
 234
 235         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 236                         / sizeof(struct ext4_extent);
 237         if (!check) {
 238 #ifdef AGGRESSIVE_TEST
 239                 if (size > 6)
 240                         size = 6;
 241 #endif
 242         }
 243         return size;
 244 }
 245
 246 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 247 {
 248         int size;
 249
 250         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 251                         / sizeof(struct ext4_extent_idx);
 252         if (!check) {
 253 #ifdef AGGRESSIVE_TEST
 254                 if (size > 5)
 255                         size = 5;
 256 #endif
 257         }
 258         return size;
 259 }
 260
 261 static inline int ext4_ext_space_root(struct inode *inode, int check)
 262 {
 263         int size;
 264
 265         size = sizeof(EXT4_I(inode)->i_data);
 266         size -= sizeof(struct ext4_extent_header);
 267         size /= sizeof(struct ext4_extent);
 268         if (!check) {
 269 #ifdef AGGRESSIVE_TEST
 270                 if (size > 3)
 271                         size = 3;
 272 #endif
 273         }
 274         return size;
 275 }
 276
 277 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 278 {
 279         int size;
 280
 281         size = sizeof(EXT4_I(inode)->i_data);
 282         size -= sizeof(struct ext4_extent_header);
 283         size /= sizeof(struct ext4_extent_idx);
 284         if (!check) {
 285 #ifdef AGGRESSIVE_TEST
 286                 if (size > 4)
 287                         size = 4;
 288 #endif
 289         }
 290         return size;
 291 }
 292
 293 /*
 294  * Calculate the number of metadata blocks needed
 295  * to allocate @blocks
 296  * Worse case is one block per extent
 297  */
 298 int ext4_ext_calc_metadata_amount(struct inode *inode, sector_t lblock)
 299 {
 300         struct ext4_inode_info *ei = EXT4_I(inode);
 301         int idxs, num = 0;
 302
 303         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 304                 / sizeof(struct ext4_extent_idx));
 305
 306         /*
 307          * If the new delayed allocation block is contiguous with the
 308          * previous da block, it can share index blocks with the
 309          * previous block, so we only need to allocate a new index
 310          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 311          * an additional index block, and at ldxs**3 blocks, yet
 312          * another index blocks.
 313          */
 314         if (ei->i_da_metadata_calc_len &&
 315             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 316                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 317                         num++;
 318                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 319                         num++;
 320                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 321                         num++;
 322                         ei->i_da_metadata_calc_len = 0;
 323                 } else
 324                         ei->i_da_metadata_calc_len++;
 325                 ei->i_da_metadata_calc_last_lblock++;
 326                 return num;
 327         }
 328
 329         /*
 330          * In the worst case we need a new set of index blocks at
 331          * every level of the inode's extent tree.
 332          */
 333         ei->i_da_metadata_calc_len = 1;
 334         ei->i_da_metadata_calc_last_lblock = lblock;
 335         return ext_depth(inode) + 1;
 336 }
 337
 338 static int
 339 ext4_ext_max_entries(struct inode *inode, int depth)
 340 {
 341         int max;
 342
 343         if (depth == ext_depth(inode)) {
 344                 if (depth == 0)
 345                         max = ext4_ext_space_root(inode, 1);
 346                 else
 347                         max = ext4_ext_space_root_idx(inode, 1);
 348         } else {
 349                 if (depth == 0)
 350                         max = ext4_ext_space_block(inode, 1);
 351                 else
 352                         max = ext4_ext_space_block_idx(inode, 1);
 353         }
 354
 355         return max;
 356 }
 357
 358 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 359 {
 360         ext4_fsblk_t block = ext_pblock(ext);
 361         int len = ext4_ext_get_actual_len(ext);
 362
 363         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 364 }
 365
 366 static int ext4_valid_extent_idx(struct inode *inode,
 367                                 struct ext4_extent_idx *ext_idx)
 368 {
 369         ext4_fsblk_t block = idx_pblock(ext_idx);
 370
 371         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 372 }
 373
 374 static int ext4_valid_extent_entries(struct inode *inode,
 375                                 struct ext4_extent_header *eh,
 376                                 int depth)
 377 {
 378         struct ext4_extent *ext;
 379         struct ext4_extent_idx *ext_idx;
 380         unsigned short entries;
 381         if (eh->eh_entries == 0)
 382                 return 1;
 383
 384         entries = le16_to_cpu(eh->eh_entries);
 385
 386         if (depth == 0) {
 387                 /* leaf entries */
 388                 ext = EXT_FIRST_EXTENT(eh);
 389                 while (entries) {
 390                         if (!ext4_valid_extent(inode, ext))
 391                                 return 0;
 392                         ext++;
 393                         entries--;
 394                 }
 395         } else {
 396                 ext_idx = EXT_FIRST_INDEX(eh);
 397                 while (entries) {
 398                         if (!ext4_valid_extent_idx(inode, ext_idx))
 399                                 return 0;
 400                         ext_idx++;
 401                         entries--;
 402                 }
 403         }
 404         return 1;
 405 }
 406
 407 static int __ext4_ext_check(const char *function, struct inode *inode,
 408                                         struct ext4_extent_header *eh,
 409                                         int depth)
 410 {
 411         const char *error_msg;
 412         int max = 0;
 413
 414         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 415                 error_msg = "invalid magic";
 416                 goto corrupted;
 417         }
 418         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 419                 error_msg = "unexpected eh_depth";
 420                 goto corrupted;
 421         }
 422         if (unlikely(eh->eh_max == 0)) {
 423                 error_msg = "invalid eh_max";
 424                 goto corrupted;
 425         }
 426         max = ext4_ext_max_entries(inode, depth);
 427         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 428                 error_msg = "too large eh_max";
 429                 goto corrupted;
 430         }
 431         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 432                 error_msg = "invalid eh_entries";
 433                 goto corrupted;
 434         }
 435         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 436                 error_msg = "invalid extent entries";
 437                 goto corrupted;
 438         }
 439         return 0;
 440
 441 corrupted:
 442         __ext4_error(inode->i_sb, function,
 443                         "bad header/extent in inode #%lu: %s - magic %x, "
 444                         "entries %u, max %u(%u), depth %u(%u)",
 445                         inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
 446                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 447                         max, le16_to_cpu(eh->eh_depth), depth);
 448
 449         return -EIO;
 450 }
 451
 452 #define ext4_ext_check(inode, eh, depth)        \
 453         __ext4_ext_check(__func__, inode, eh, depth)
 454
 455 int ext4_ext_check_inode(struct inode *inode)
 456 {
 457         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 458 }
 459
 460 #ifdef EXT_DEBUG
 461 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 462 {
 463         int k, l = path->p_depth;
 464
 465         ext_debug("path:");
 466         for (k = 0; k <= l; k++, path++) {
 467                 if (path->p_idx) {
 468                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 469                             idx_pblock(path->p_idx));
 470                 } else if (path->p_ext) {
 471                         ext_debug("  %d:[%d]%d:%llu ",
 472                                   le32_to_cpu(path->p_ext->ee_block),
 473                                   ext4_ext_is_uninitialized(path->p_ext),
 474                                   ext4_ext_get_actual_len(path->p_ext),
 475                                   ext_pblock(path->p_ext));
 476                 } else
 477                         ext_debug("  []");
 478         }
 479         ext_debug("\n");
 480 }
 481
 482 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 483 {
 484         int depth = ext_depth(inode);
 485         struct ext4_extent_header *eh;
 486         struct ext4_extent *ex;
 487         int i;
 488
 489         if (!path)
 490                 return;
 491
 492         eh = path[depth].p_hdr;
 493         ex = EXT_FIRST_EXTENT(eh);
 494
 495         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 496
 497         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 498                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 499                           ext4_ext_is_uninitialized(ex),
 500                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
 501         }
 502         ext_debug("\n");
 503 }
 504 #else
 505 #define ext4_ext_show_path(inode, path)
 506 #define ext4_ext_show_leaf(inode, path)
 507 #endif
 508
 509 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 510 {
 511         int depth = path->p_depth;
 512         int i;
 513
 514         for (i = 0; i <= depth; i++, path++)
 515                 if (path->p_bh) {
 516                         brelse(path->p_bh);
 517                         path->p_bh = NULL;
 518                 }
 519 }
 520
 521 /*
 522  * ext4_ext_binsearch_idx:
 523  * binary search for the closest index of the given block
 524  * the header must be checked before calling this
 525  */
 526 static void
 527 ext4_ext_binsearch_idx(struct inode *inode,
 528                         struct ext4_ext_path *path, ext4_lblk_t block)
 529 {
 530         struct ext4_extent_header *eh = path->p_hdr;
 531         struct ext4_extent_idx *r, *l, *m;
 532
 533
 534         ext_debug("binsearch for %u(idx):  ", block);
 535
 536         l = EXT_FIRST_INDEX(eh) + 1;
 537         r = EXT_LAST_INDEX(eh);
 538         while (l <= r) {
 539                 m = l + (r - l) / 2;
 540                 if (block < le32_to_cpu(m->ei_block))
 541                         r = m - 1;
 542                 else
 543                         l = m + 1;
 544                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 545                                 m, le32_to_cpu(m->ei_block),
 546                                 r, le32_to_cpu(r->ei_block));
 547         }
 548
 549         path->p_idx = l - 1;
 550         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 551                   idx_pblock(path->p_idx));
 552
 553 #ifdef CHECK_BINSEARCH
 554         {
 555                 struct ext4_extent_idx *chix, *ix;
 556                 int k;
 557
 558                 chix = ix = EXT_FIRST_INDEX(eh);
 559                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 560                   if (k != 0 &&
 561                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 562                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 563                                        "first=0x%p\n", k,
 564                                        ix, EXT_FIRST_INDEX(eh));
 565                                 printk(KERN_DEBUG "%u <= %u\n",
 566                                        le32_to_cpu(ix->ei_block),
 567                                        le32_to_cpu(ix[-1].ei_block));
 568                         }
 569                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 570                                            <= le32_to_cpu(ix[-1].ei_block));
 571                         if (block < le32_to_cpu(ix->ei_block))
 572                                 break;
 573                         chix = ix;
 574                 }
 575                 BUG_ON(chix != path->p_idx);
 576         }
 577 #endif
 578
 579 }
 580
 581 /*
 582  * ext4_ext_binsearch:
 583  * binary search for closest extent of the given block
 584  * the header must be checked before calling this
 585  */
 586 static void
 587 ext4_ext_binsearch(struct inode *inode,
 588                 struct ext4_ext_path *path, ext4_lblk_t block)
 589 {
 590         struct ext4_extent_header *eh = path->p_hdr;
 591         struct ext4_extent *r, *l, *m;
 592
 593         if (eh->eh_entries == 0) {
 594                 /*
 595                  * this leaf is empty:
 596                  * we get such a leaf in split/add case
 597                  */
 598                 return;
 599         }
 600
 601         ext_debug("binsearch for %u:  ", block);
 602
 603         l = EXT_FIRST_EXTENT(eh) + 1;
 604         r = EXT_LAST_EXTENT(eh);
 605
 606         while (l <= r) {
 607                 m = l + (r - l) / 2;
 608                 if (block < le32_to_cpu(m->ee_block))
 609                         r = m - 1;
 610                 else
 611                         l = m + 1;
 612                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 613                                 m, le32_to_cpu(m->ee_block),
 614                                 r, le32_to_cpu(r->ee_block));
 615         }
 616
 617         path->p_ext = l - 1;
 618         ext_debug("  -> %d:%llu:[%d]%d ",
 619                         le32_to_cpu(path->p_ext->ee_block),
 620                         ext_pblock(path->p_ext),
 621                         ext4_ext_is_uninitialized(path->p_ext),
 622                         ext4_ext_get_actual_len(path->p_ext));
 623
 624 #ifdef CHECK_BINSEARCH
 625         {
 626                 struct ext4_extent *chex, *ex;
 627                 int k;
 628
 629                 chex = ex = EXT_FIRST_EXTENT(eh);
 630                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 631                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 632                                           <= le32_to_cpu(ex[-1].ee_block));
 633                         if (block < le32_to_cpu(ex->ee_block))
 634                                 break;
 635                         chex = ex;
 636                 }
 637                 BUG_ON(chex != path->p_ext);
 638         }
 639 #endif
 640
 641 }
 642
 643 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 644 {
 645         struct ext4_extent_header *eh;
 646
 647         eh = ext_inode_hdr(inode);
 648         eh->eh_depth = 0;
 649         eh->eh_entries = 0;
 650         eh->eh_magic = EXT4_EXT_MAGIC;
 651         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 652         ext4_mark_inode_dirty(handle, inode);
 653         ext4_ext_invalidate_cache(inode);
 654         return 0;
 655 }
 656
 657 struct ext4_ext_path *
 658 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 659                                         struct ext4_ext_path *path)
 660 {
 661         struct ext4_extent_header *eh;
 662         struct buffer_head *bh;
 663         short int depth, i, ppos = 0, alloc = 0;
 664
 665         eh = ext_inode_hdr(inode);
 666         depth = ext_depth(inode);
 667
 668         /* account possible depth increase */
 669         if (!path) {
 670                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 671                                 GFP_NOFS);
 672                 if (!path)
 673                         return ERR_PTR(-ENOMEM);
 674                 alloc = 1;
 675         }
 676         path[0].p_hdr = eh;
 677         path[0].p_bh = NULL;
 678
 679         i = depth;
 680         /* walk through the tree */
 681         while (i) {
 682                 int need_to_validate = 0;
 683
 684                 ext_debug("depth %d: num %d, max %d\n",
 685                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 686
 687                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 688                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
 689                 path[ppos].p_depth = i;
 690                 path[ppos].p_ext = NULL;
 691
 692                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 693                 if (unlikely(!bh))
 694                         goto err;
 695                 if (!bh_uptodate_or_lock(bh)) {
 696                         if (bh_submit_read(bh) < 0) {
 697                                 put_bh(bh);
 698                                 goto err;
 699                         }
 700                         /* validate the extent entries */
 701                         need_to_validate = 1;
 702                 }
 703                 eh = ext_block_hdr(bh);
 704                 ppos++;
 705                 if (unlikely(ppos > depth)) {
 706                         put_bh(bh);
 707                         EXT4_ERROR_INODE(inode,
 708                                          "ppos %d > depth %d", ppos, depth);
 709                         goto err;
 710                 }
 711                 path[ppos].p_bh = bh;
 712                 path[ppos].p_hdr = eh;
 713                 i--;
 714
 715                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 716                         goto err;
 717         }
 718
 719         path[ppos].p_depth = i;
 720         path[ppos].p_ext = NULL;
 721         path[ppos].p_idx = NULL;
 722
 723         /* find extent */
 724         ext4_ext_binsearch(inode, path + ppos, block);
 725         /* if not an empty leaf */
 726         if (path[ppos].p_ext)
 727                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
 728
 729         ext4_ext_show_path(inode, path);
 730
 731         return path;
 732
 733 err:
 734         ext4_ext_drop_refs(path);
 735         if (alloc)
 736                 kfree(path);
 737         return ERR_PTR(-EIO);
 738 }
 739
 740 /*
 741  * ext4_ext_insert_index:
 742  * insert new index [@logical;@ptr] into the block at @curp;
 743  * check where to insert: before @curp or after @curp
 744  */
 745 int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 746                                 struct ext4_ext_path *curp,
 747                                 int logical, ext4_fsblk_t ptr)
 748 {
 749         struct ext4_extent_idx *ix;
 750         int len, err;
 751
 752         err = ext4_ext_get_access(handle, inode, curp);
 753         if (err)
 754                 return err;
 755
 756         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 757                 EXT4_ERROR_INODE(inode,
 758                                  "logical %d == ei_block %d!",
 759                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 760                 return -EIO;
 761         }
 762         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 763         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 764                 /* insert after */
 765                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 766                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 767                         len = len < 0 ? 0 : len;
 768                         ext_debug("insert new index %d after: %llu. "
 769                                         "move %d from 0x%p to 0x%p\n",
 770                                         logical, ptr, len,
 771                                         (curp->p_idx + 1), (curp->p_idx + 2));
 772                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 773                 }
 774                 ix = curp->p_idx + 1;
 775         } else {
 776                 /* insert before */
 777                 len = len * sizeof(struct ext4_extent_idx);
 778                 len = len < 0 ? 0 : len;
 779                 ext_debug("insert new index %d before: %llu. "
 780                                 "move %d from 0x%p to 0x%p\n",
 781                                 logical, ptr, len,
 782                                 curp->p_idx, (curp->p_idx + 1));
 783                 memmove(curp->p_idx + 1, curp->p_idx, len);
 784                 ix = curp->p_idx;
 785         }
 786
 787         ix->ei_block = cpu_to_le32(logical);
 788         ext4_idx_store_pblock(ix, ptr);
 789         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 790
 791         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 792                              > le16_to_cpu(curp->p_hdr->eh_max))) {
 793                 EXT4_ERROR_INODE(inode,
 794                                  "logical %d == ei_block %d!",
 795                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 796                 return -EIO;
 797         }
 798         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 799                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 800                 return -EIO;
 801         }
 802
 803         err = ext4_ext_dirty(handle, inode, curp);
 804         ext4_std_error(inode->i_sb, err);
 805
 806         return err;
 807 }
 808
 809 /*
 810  * ext4_ext_split:
 811  * inserts new subtree into the path, using free index entry
 812  * at depth @at:
 813  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 814  * - makes decision where to split
 815  * - moves remaining extents and index entries (right to the split point)
 816  *   into the newly allocated blocks
 817  * - initializes subtree
 818  */
 819 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 820                                 struct ext4_ext_path *path,
 821                                 struct ext4_extent *newext, int at)
 822 {
 823         struct buffer_head *bh = NULL;
 824         int depth = ext_depth(inode);
 825         struct ext4_extent_header *neh;
 826         struct ext4_extent_idx *fidx;
 827         struct ext4_extent *ex;
 828         int i = at, k, m, a;
 829         ext4_fsblk_t newblock, oldblock;
 830         __le32 border;
 831         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 832         int err = 0;
 833
 834         /* make decision: where to split? */
 835         /* FIXME: now decision is simplest: at current extent */
 836
 837         /* if current leaf will be split, then we should use
 838          * border from split point */
 839         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 840                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 841                 return -EIO;
 842         }
 843         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 844                 border = path[depth].p_ext[1].ee_block;
 845                 ext_debug("leaf will be split."
 846                                 " next leaf starts at %d\n",
 847                                   le32_to_cpu(border));
 848         } else {
 849                 border = newext->ee_block;
 850                 ext_debug("leaf will be added."
 851                                 " next leaf starts at %d\n",
 852                                 le32_to_cpu(border));
 853         }
 854
 855         /*
 856          * If error occurs, then we break processing
 857          * and mark filesystem read-only. index won't
 858          * be inserted and tree will be in consistent
 859          * state. Next mount will repair buffers too.
 860          */
 861
 862         /*
 863          * Get array to track all allocated blocks.
 864          * We need this to handle errors and free blocks
 865          * upon them.
 866          */
 867         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 868         if (!ablocks)
 869                 return -ENOMEM;
 870
 871         /* allocate all needed blocks */
 872         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 873         for (a = 0; a < depth - at; a++) {
 874                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 875                                                    newext, &err);
 876                 if (newblock == 0)
 877                         goto cleanup;
 878                 ablocks[a] = newblock;
 879         }
 880
 881         /* initialize new leaf */
 882         newblock = ablocks[--a];
 883         if (unlikely(newblock == 0)) {
 884                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 885                 err = -EIO;
 886                 goto cleanup;
 887         }
 888         bh = sb_getblk(inode->i_sb, newblock);
 889         if (!bh) {
 890                 err = -EIO;
 891                 goto cleanup;
 892         }
 893         lock_buffer(bh);
 894
 895         err = ext4_journal_get_create_access(handle, bh);
 896         if (err)
 897                 goto cleanup;
 898
 899         neh = ext_block_hdr(bh);
 900         neh->eh_entries = 0;
 901         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 902         neh->eh_magic = EXT4_EXT_MAGIC;
 903         neh->eh_depth = 0;
 904         ex = EXT_FIRST_EXTENT(neh);
 905
 906         /* move remainder of path[depth] to the new leaf */
 907         if (unlikely(path[depth].p_hdr->eh_entries !=
 908                      path[depth].p_hdr->eh_max)) {
 909                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 910                                  path[depth].p_hdr->eh_entries,
 911                                  path[depth].p_hdr->eh_max);
 912                 err = -EIO;
 913                 goto cleanup;
 914         }
 915         /* start copy from next extent */
 916         /* TODO: we could do it by single memmove */
 917         m = 0;
 918         path[depth].p_ext++;
 919         while (path[depth].p_ext <=
 920                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
 921                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 922                                 le32_to_cpu(path[depth].p_ext->ee_block),
 923                                 ext_pblock(path[depth].p_ext),
 924                                 ext4_ext_is_uninitialized(path[depth].p_ext),
 925                                 ext4_ext_get_actual_len(path[depth].p_ext),
 926                                 newblock);
 927                 /*memmove(ex++, path[depth].p_ext++,
 928                                 sizeof(struct ext4_extent));
 929                 neh->eh_entries++;*/
 930                 path[depth].p_ext++;
 931                 m++;
 932         }
 933         if (m) {
 934                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
 935                 le16_add_cpu(&neh->eh_entries, m);
 936         }
 937
 938         set_buffer_uptodate(bh);
 939         unlock_buffer(bh);
 940
 941         err = ext4_handle_dirty_metadata(handle, inode, bh);
 942         if (err)
 943                 goto cleanup;
 944         brelse(bh);
 945         bh = NULL;
 946
 947         /* correct old leaf */
 948         if (m) {
 949                 err = ext4_ext_get_access(handle, inode, path + depth);
 950                 if (err)
 951                         goto cleanup;
 952                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 953                 err = ext4_ext_dirty(handle, inode, path + depth);
 954                 if (err)
 955                         goto cleanup;
 956
 957         }
 958
 959         /* create intermediate indexes */
 960         k = depth - at - 1;
 961         if (unlikely(k < 0)) {
 962                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 963                 err = -EIO;
 964                 goto cleanup;
 965         }
 966         if (k)
 967                 ext_debug("create %d intermediate indices\n", k);
 968         /* insert new index into current index block */
 969         /* current depth stored in i var */
 970         i = depth - 1;
 971         while (k--) {
 972                 oldblock = newblock;
 973                 newblock = ablocks[--a];
 974                 bh = sb_getblk(inode->i_sb, newblock);
 975                 if (!bh) {
 976                         err = -EIO;
 977                         goto cleanup;
 978                 }
 979                 lock_buffer(bh);
 980
 981                 err = ext4_journal_get_create_access(handle, bh);
 982                 if (err)
 983                         goto cleanup;
 984
 985                 neh = ext_block_hdr(bh);
 986                 neh->eh_entries = cpu_to_le16(1);
 987                 neh->eh_magic = EXT4_EXT_MAGIC;
 988                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 989                 neh->eh_depth = cpu_to_le16(depth - i);
 990                 fidx = EXT_FIRST_INDEX(neh);
 991                 fidx->ei_block = border;
 992                 ext4_idx_store_pblock(fidx, oldblock);
 993
 994                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 995                                 i, newblock, le32_to_cpu(border), oldblock);
 996                 /* copy indexes */
 997                 m = 0;
 998                 path[i].p_idx++;
 999
1000                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1001                                 EXT_MAX_INDEX(path[i].p_hdr));
1002                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1003                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1004                         EXT4_ERROR_INODE(inode,
1005                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1006                                          le32_to_cpu(path[i].p_ext->ee_block));
1007                         err = -EIO;
1008                         goto cleanup;
1009                 }
1010                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
1011                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
1012                                         le32_to_cpu(path[i].p_idx->ei_block),
1013                                         idx_pblock(path[i].p_idx),
1014                                         newblock);
1015                         /*memmove(++fidx, path[i].p_idx++,
1016                                         sizeof(struct ext4_extent_idx));
1017                         neh->eh_entries++;
1018                         BUG_ON(neh->eh_entries > neh->eh_max);*/
1019                         path[i].p_idx++;
1020                         m++;
1021                 }
1022                 if (m) {
1023                         memmove(++fidx, path[i].p_idx - m,
1024                                 sizeof(struct ext4_extent_idx) * m);
1025                         le16_add_cpu(&neh->eh_entries, m);
1026                 }
1027                 set_buffer_uptodate(bh);
1028                 unlock_buffer(bh);
1029
1030                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1031                 if (err)
1032                         goto cleanup;
1033                 brelse(bh);
1034                 bh = NULL;
1035
1036                 /* correct old index */
1037                 if (m) {
1038                         err = ext4_ext_get_access(handle, inode, path + i);
1039                         if (err)
1040                                 goto cleanup;
1041                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1042                         err = ext4_ext_dirty(handle, inode, path + i);
1043                         if (err)
1044                                 goto cleanup;
1045                 }
1046
1047                 i--;
1048         }
1049
1050         /* insert new index */
1051         err = ext4_ext_insert_index(handle, inode, path + at,
1052                                     le32_to_cpu(border), newblock);
1053
1054 cleanup:
1055         if (bh) {
1056                 if (buffer_locked(bh))
1057                         unlock_buffer(bh);
1058                 brelse(bh);
1059         }
1060
1061         if (err) {
1062                 /* free all allocated blocks in error case */
1063                 for (i = 0; i < depth; i++) {
1064                         if (!ablocks[i])
1065                                 continue;
1066                         ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1067                                          EXT4_FREE_BLOCKS_METADATA);
1068                 }
1069         }
1070         kfree(ablocks);
1071
1072         return err;
1073 }
1074
1075 /*
1076  * ext4_ext_grow_indepth:
1077  * implements tree growing procedure:
1078  * - allocates new block
1079  * - moves top-level data (index block or leaf) into the new block
1080  * - initializes new top-level, creating index that points to the
1081  *   just created block
1082  */
1083 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1084                                         struct ext4_ext_path *path,
1085                                         struct ext4_extent *newext)
1086 {
1087         struct ext4_ext_path *curp = path;
1088         struct ext4_extent_header *neh;
1089         struct ext4_extent_idx *fidx;
1090         struct buffer_head *bh;
1091         ext4_fsblk_t newblock;
1092         int err = 0;
1093
1094         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1095         if (newblock == 0)
1096                 return err;
1097
1098         bh = sb_getblk(inode->i_sb, newblock);
1099         if (!bh) {
1100                 err = -EIO;
1101                 ext4_std_error(inode->i_sb, err);
1102                 return err;
1103         }
1104         lock_buffer(bh);
1105
1106         err = ext4_journal_get_create_access(handle, bh);
1107         if (err) {
1108                 unlock_buffer(bh);
1109                 goto out;
1110         }
1111
1112         /* move top-level index/leaf into new block */
1113         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1114
1115         /* set size of new block */
1116         neh = ext_block_hdr(bh);
1117         /* old root could have indexes or leaves
1118          * so calculate e_max right way */
1119         if (ext_depth(inode))
1120                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1121         else
1122                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1123         neh->eh_magic = EXT4_EXT_MAGIC;
1124         set_buffer_uptodate(bh);
1125         unlock_buffer(bh);
1126
1127         err = ext4_handle_dirty_metadata(handle, inode, bh);
1128         if (err)
1129                 goto out;
1130
1131         /* create index in new top-level index: num,max,pointer */
1132         err = ext4_ext_get_access(handle, inode, curp);
1133         if (err)
1134                 goto out;
1135
1136         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1137         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1138         curp->p_hdr->eh_entries = cpu_to_le16(1);
1139         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1140
1141         if (path[0].p_hdr->eh_depth)
1142                 curp->p_idx->ei_block =
1143                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1144         else
1145                 curp->p_idx->ei_block =
1146                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1147         ext4_idx_store_pblock(curp->p_idx, newblock);
1148
1149         neh = ext_inode_hdr(inode);
1150         fidx = EXT_FIRST_INDEX(neh);
1151         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1152                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1153                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1154
1155         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1156         err = ext4_ext_dirty(handle, inode, curp);
1157 out:
1158         brelse(bh);
1159
1160         return err;
1161 }
1162
1163 /*
1164  * ext4_ext_create_new_leaf:
1165  * finds empty index and adds new leaf.
1166  * if no free index is found, then it requests in-depth growing.
1167  */
1168 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1169                                         struct ext4_ext_path *path,
1170                                         struct ext4_extent *newext)
1171 {
1172         struct ext4_ext_path *curp;
1173         int depth, i, err = 0;
1174
1175 repeat:
1176         i = depth = ext_depth(inode);
1177
1178         /* walk up to the tree and look for free index entry */
1179         curp = path + depth;
1180         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1181                 i--;
1182                 curp--;
1183         }
1184
1185         /* we use already allocated block for index block,
1186          * so subsequent data blocks should be contiguous */
1187         if (EXT_HAS_FREE_INDEX(curp)) {
1188                 /* if we found index with free entry, then use that
1189                  * entry: create all needed subtree and add new leaf */
1190                 err = ext4_ext_split(handle, inode, path, newext, i);
1191                 if (err)
1192                         goto out;
1193
1194                 /* refill path */
1195                 ext4_ext_drop_refs(path);
1196                 path = ext4_ext_find_extent(inode,
1197                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1198                                     path);
1199                 if (IS_ERR(path))
1200                         err = PTR_ERR(path);
1201         } else {
1202                 /* tree is full, time to grow in depth */
1203                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1204                 if (err)
1205                         goto out;
1206
1207                 /* refill path */
1208                 ext4_ext_drop_refs(path);
1209                 path = ext4_ext_find_extent(inode,
1210                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1211                                     path);
1212                 if (IS_ERR(path)) {
1213                         err = PTR_ERR(path);
1214                         goto out;
1215                 }
1216
1217                 /*
1218                  * only first (depth 0 -> 1) produces free space;
1219                  * in all other cases we have to split the grown tree
1220                  */
1221                 depth = ext_depth(inode);
1222                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1223                         /* now we need to split */
1224                         goto repeat;
1225                 }
1226         }
1227
1228 out:
1229         return err;
1230 }
1231
1232 /*
1233  * search the closest allocated block to the left for *logical
1234  * and returns it at @logical + it's physical address at @phys
1235  * if *logical is the smallest allocated block, the function
1236  * returns 0 at @phys
1237  * return value contains 0 (success) or error code
1238  */
1239 int
1240 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1241                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1242 {
1243         struct ext4_extent_idx *ix;
1244         struct ext4_extent *ex;
1245         int depth, ee_len;
1246
1247         if (unlikely(path == NULL)) {
1248                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1249                 return -EIO;
1250         }
1251         depth = path->p_depth;
1252         *phys = 0;
1253
1254         if (depth == 0 && path->p_ext == NULL)
1255                 return 0;
1256
1257         /* usually extent in the path covers blocks smaller
1258          * then *logical, but it can be that extent is the
1259          * first one in the file */
1260
1261         ex = path[depth].p_ext;
1262         ee_len = ext4_ext_get_actual_len(ex);
1263         if (*logical < le32_to_cpu(ex->ee_block)) {
1264                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1265                         EXT4_ERROR_INODE(inode,
1266                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1267                                          *logical, le32_to_cpu(ex->ee_block));
1268                         return -EIO;
1269                 }
1270                 while (--depth >= 0) {
1271                         ix = path[depth].p_idx;
1272                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1273                                 EXT4_ERROR_INODE(inode,
1274                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1275                                   ix != NULL ? ix->ei_block : 0,
1276                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1277                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1278                                   depth);
1279                                 return -EIO;
1280                         }
1281                 }
1282                 return 0;
1283         }
1284
1285         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1286                 EXT4_ERROR_INODE(inode,
1287                                  "logical %d < ee_block %d + ee_len %d!",
1288                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1289                 return -EIO;
1290         }
1291
1292         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1293         *phys = ext_pblock(ex) + ee_len - 1;
1294         return 0;
1295 }
1296
1297 /*
1298  * search the closest allocated block to the right for *logical
1299  * and returns it at @logical + it's physical address at @phys
1300  * if *logical is the smallest allocated block, the function
1301  * returns 0 at @phys
1302  * return value contains 0 (success) or error code
1303  */
1304 int
1305 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1306                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1307 {
1308         struct buffer_head *bh = NULL;
1309         struct ext4_extent_header *eh;
1310         struct ext4_extent_idx *ix;
1311         struct ext4_extent *ex;
1312         ext4_fsblk_t block;
1313         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1314         int ee_len;
1315
1316         if (unlikely(path == NULL)) {
1317                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1318                 return -EIO;
1319         }
1320         depth = path->p_depth;
1321         *phys = 0;
1322
1323         if (depth == 0 && path->p_ext == NULL)
1324                 return 0;
1325
1326         /* usually extent in the path covers blocks smaller
1327          * then *logical, but it can be that extent is the
1328          * first one in the file */
1329
1330         ex = path[depth].p_ext;
1331         ee_len = ext4_ext_get_actual_len(ex);
1332         if (*logical < le32_to_cpu(ex->ee_block)) {
1333                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1334                         EXT4_ERROR_INODE(inode,
1335                                          "first_extent(path[%d].p_hdr) != ex",
1336                                          depth);
1337                         return -EIO;
1338                 }
1339                 while (--depth >= 0) {
1340                         ix = path[depth].p_idx;
1341                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1342                                 EXT4_ERROR_INODE(inode,
1343                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1344                                                  *logical);
1345                                 return -EIO;
1346                         }
1347                 }
1348                 *logical = le32_to_cpu(ex->ee_block);
1349                 *phys = ext_pblock(ex);
1350                 return 0;
1351         }
1352
1353         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1354                 EXT4_ERROR_INODE(inode,
1355                                  "logical %d < ee_block %d + ee_len %d!",
1356                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1357                 return -EIO;
1358         }
1359
1360         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1361                 /* next allocated block in this leaf */
1362                 ex++;
1363                 *logical = le32_to_cpu(ex->ee_block);
1364                 *phys = ext_pblock(ex);
1365                 return 0;
1366         }
1367
1368         /* go up and search for index to the right */
1369         while (--depth >= 0) {
1370                 ix = path[depth].p_idx;
1371                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1372                         goto got_index;
1373         }
1374
1375         /* we've gone up to the root and found no index to the right */
1376         return 0;
1377
1378 got_index:
1379         /* we've found index to the right, let's
1380          * follow it and find the closest allocated
1381          * block to the right */
1382         ix++;
1383         block = idx_pblock(ix);
1384         while (++depth < path->p_depth) {
1385                 bh = sb_bread(inode->i_sb, block);
1386                 if (bh == NULL)
1387                         return -EIO;
1388                 eh = ext_block_hdr(bh);
1389                 /* subtract from p_depth to get proper eh_depth */
1390                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1391                         put_bh(bh);
1392                         return -EIO;
1393                 }
1394                 ix = EXT_FIRST_INDEX(eh);
1395                 block = idx_pblock(ix);
1396                 put_bh(bh);
1397         }
1398
1399         bh = sb_bread(inode->i_sb, block);
1400         if (bh == NULL)
1401                 return -EIO;
1402         eh = ext_block_hdr(bh);
1403         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1404                 put_bh(bh);
1405                 return -EIO;
1406         }
1407         ex = EXT_FIRST_EXTENT(eh);
1408         *logical = le32_to_cpu(ex->ee_block);
1409         *phys = ext_pblock(ex);
1410         put_bh(bh);
1411         return 0;
1412 }
1413
1414 /*
1415  * ext4_ext_next_allocated_block:
1416  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1417  * NOTE: it considers block number from index entry as
1418  * allocated block. Thus, index entries have to be consistent
1419  * with leaves.
1420  */
1421 static ext4_lblk_t
1422 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1423 {
1424         int depth;
1425
1426         BUG_ON(path == NULL);
1427         depth = path->p_depth;
1428
1429         if (depth == 0 && path->p_ext == NULL)
1430                 return EXT_MAX_BLOCK;
1431
1432         while (depth >= 0) {
1433                 if (depth == path->p_depth) {
1434                         /* leaf */
1435                         if (path[depth].p_ext !=
1436                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1437                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1438                 } else {
1439                         /* index */
1440                         if (path[depth].p_idx !=
1441                                         EXT_LAST_INDEX(path[depth].p_hdr))
1442                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1443                 }
1444                 depth--;
1445         }
1446
1447         return EXT_MAX_BLOCK;
1448 }
1449
1450 /*
1451  * ext4_ext_next_leaf_block:
1452  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1453  */
1454 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1455                                         struct ext4_ext_path *path)
1456 {
1457         int depth;
1458
1459         BUG_ON(path == NULL);
1460         depth = path->p_depth;
1461
1462         /* zero-tree has no leaf blocks at all */
1463         if (depth == 0)
1464                 return EXT_MAX_BLOCK;
1465
1466         /* go to index block */
1467         depth--;
1468
1469         while (depth >= 0) {
1470                 if (path[depth].p_idx !=
1471                                 EXT_LAST_INDEX(path[depth].p_hdr))
1472                         return (ext4_lblk_t)
1473                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1474                 depth--;
1475         }
1476
1477         return EXT_MAX_BLOCK;
1478 }
1479
1480 /*
1481  * ext4_ext_correct_indexes:
1482  * if leaf gets modified and modified extent is first in the leaf,
1483  * then we have to correct all indexes above.
1484  * TODO: do we need to correct tree in all cases?
1485  */
1486 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1487                                 struct ext4_ext_path *path)
1488 {
1489         struct ext4_extent_header *eh;
1490         int depth = ext_depth(inode);
1491         struct ext4_extent *ex;
1492         __le32 border;
1493         int k, err = 0;
1494
1495         eh = path[depth].p_hdr;
1496         ex = path[depth].p_ext;
1497
1498         if (unlikely(ex == NULL || eh == NULL)) {
1499                 EXT4_ERROR_INODE(inode,
1500                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1501                 return -EIO;
1502         }
1503
1504         if (depth == 0) {
1505                 /* there is no tree at all */
1506                 return 0;
1507         }
1508
1509         if (ex != EXT_FIRST_EXTENT(eh)) {
1510                 /* we correct tree if first leaf got modified only */
1511                 return 0;
1512         }
1513
1514         /*
1515          * TODO: we need correction if border is smaller than current one
1516          */
1517         k = depth - 1;
1518         border = path[depth].p_ext->ee_block;
1519         err = ext4_ext_get_access(handle, inode, path + k);
1520         if (err)
1521                 return err;
1522         path[k].p_idx->ei_block = border;
1523         err = ext4_ext_dirty(handle, inode, path + k);
1524         if (err)
1525                 return err;
1526
1527         while (k--) {
1528                 /* change all left-side indexes */
1529                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1530                         break;
1531                 err = ext4_ext_get_access(handle, inode, path + k);
1532                 if (err)
1533                         break;
1534                 path[k].p_idx->ei_block = border;
1535                 err = ext4_ext_dirty(handle, inode, path + k);
1536                 if (err)
1537                         break;
1538         }
1539
1540         return err;
1541 }
1542
1543 int
1544 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1545                                 struct ext4_extent *ex2)
1546 {
1547         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1548
1549         /*
1550          * Make sure that either both extents are uninitialized, or
1551          * both are _not_.
1552          */
1553         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1554                 return 0;
1555
1556         if (ext4_ext_is_uninitialized(ex1))
1557                 max_len = EXT_UNINIT_MAX_LEN;
1558         else
1559                 max_len = EXT_INIT_MAX_LEN;
1560
1561         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1562         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1563
1564         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1565                         le32_to_cpu(ex2->ee_block))
1566                 return 0;
1567
1568         /*
1569          * To allow future support for preallocated extents to be added
1570          * as an RO_COMPAT feature, refuse to merge to extents if
1571          * this can result in the top bit of ee_len being set.
1572          */
1573         if (ext1_ee_len + ext2_ee_len > max_len)
1574                 return 0;
1575 #ifdef AGGRESSIVE_TEST
1576         if (ext1_ee_len >= 4)
1577                 return 0;
1578 #endif
1579
1580         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1581                 return 1;
1582         return 0;
1583 }
1584
1585 /*
1586  * This function tries to merge the "ex" extent to the next extent in the tree.
1587  * It always tries to merge towards right. If you want to merge towards
1588  * left, pass "ex - 1" as argument instead of "ex".
1589  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1590  * 1 if they got merged.
1591  */
1592 int ext4_ext_try_to_merge(struct inode *inode,
1593                           struct ext4_ext_path *path,
1594                           struct ext4_extent *ex)
1595 {
1596         struct ext4_extent_header *eh;
1597         unsigned int depth, len;
1598         int merge_done = 0;
1599         int uninitialized = 0;
1600
1601         depth = ext_depth(inode);
1602         BUG_ON(path[depth].p_hdr == NULL);
1603         eh = path[depth].p_hdr;
1604
1605         while (ex < EXT_LAST_EXTENT(eh)) {
1606                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1607                         break;
1608                 /* merge with next extent! */
1609                 if (ext4_ext_is_uninitialized(ex))
1610                         uninitialized = 1;
1611                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1612                                 + ext4_ext_get_actual_len(ex + 1));
1613                 if (uninitialized)
1614                         ext4_ext_mark_uninitialized(ex);
1615
1616                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1617                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1618                                 * sizeof(struct ext4_extent);
1619                         memmove(ex + 1, ex + 2, len);
1620                 }
1621                 le16_add_cpu(&eh->eh_entries, -1);
1622                 merge_done = 1;
1623                 WARN_ON(eh->eh_entries == 0);
1624                 if (!eh->eh_entries)
1625                         ext4_error(inode->i_sb,
1626                                    "inode#%lu, eh->eh_entries = 0!",
1627                                    inode->i_ino);
1628         }
1629
1630         return merge_done;
1631 }
1632
1633 /*
1634  * check if a portion of the "newext" extent overlaps with an
1635  * existing extent.
1636  *
1637  * If there is an overlap discovered, it updates the length of the newext
1638  * such that there will be no overlap, and then returns 1.
1639  * If there is no overlap found, it returns 0.
1640  */
1641 unsigned int ext4_ext_check_overlap(struct inode *inode,
1642                                     struct ext4_extent *newext,
1643                                     struct ext4_ext_path *path)
1644 {
1645         ext4_lblk_t b1, b2;
1646         unsigned int depth, len1;
1647         unsigned int ret = 0;
1648
1649         b1 = le32_to_cpu(newext->ee_block);
1650         len1 = ext4_ext_get_actual_len(newext);
1651         depth = ext_depth(inode);
1652         if (!path[depth].p_ext)
1653                 goto out;
1654         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1655
1656         /*
1657          * get the next allocated block if the extent in the path
1658          * is before the requested block(s)
1659          */
1660         if (b2 < b1) {
1661                 b2 = ext4_ext_next_allocated_block(path);
1662                 if (b2 == EXT_MAX_BLOCK)
1663                         goto out;
1664         }
1665
1666         /* check for wrap through zero on extent logical start block*/
1667         if (b1 + len1 < b1) {
1668                 len1 = EXT_MAX_BLOCK - b1;
1669                 newext->ee_len = cpu_to_le16(len1);
1670                 ret = 1;
1671         }
1672
1673         /* check for overlap */
1674         if (b1 + len1 > b2) {
1675                 newext->ee_len = cpu_to_le16(b2 - b1);
1676                 ret = 1;
1677         }
1678 out:
1679         return ret;
1680 }
1681
1682 /*
1683  * ext4_ext_insert_extent:
1684  * tries to merge requsted extent into the existing extent or
1685  * inserts requested extent as new one into the tree,
1686  * creating new leaf in the no-space case.
1687  */
1688 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1689                                 struct ext4_ext_path *path,
1690                                 struct ext4_extent *newext, int flag)
1691 {
1692         struct ext4_extent_header *eh;
1693         struct ext4_extent *ex, *fex;
1694         struct ext4_extent *nearex; /* nearest extent */
1695         struct ext4_ext_path *npath = NULL;
1696         int depth, len, err;
1697         ext4_lblk_t next;
1698         unsigned uninitialized = 0;
1699
1700         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1701                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1702                 return -EIO;
1703         }
1704         depth = ext_depth(inode);
1705         ex = path[depth].p_ext;
1706         if (unlikely(path[depth].p_hdr == NULL)) {
1707                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1708                 return -EIO;
1709         }
1710
1711         /* try to insert block into found extent and return */
1712         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1713                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1714                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1715                                 ext4_ext_is_uninitialized(newext),
1716                                 ext4_ext_get_actual_len(newext),
1717                                 le32_to_cpu(ex->ee_block),
1718                                 ext4_ext_is_uninitialized(ex),
1719                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1720                 err = ext4_ext_get_access(handle, inode, path + depth);
1721                 if (err)
1722                         return err;
1723
1724                 /*
1725                  * ext4_can_extents_be_merged should have checked that either
1726                  * both extents are uninitialized, or both aren't. Thus we
1727                  * need to check only one of them here.
1728                  */
1729                 if (ext4_ext_is_uninitialized(ex))
1730                         uninitialized = 1;
1731                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1732                                         + ext4_ext_get_actual_len(newext));
1733                 if (uninitialized)
1734                         ext4_ext_mark_uninitialized(ex);
1735                 eh = path[depth].p_hdr;
1736                 nearex = ex;
1737                 goto merge;
1738         }
1739
1740 repeat:
1741         depth = ext_depth(inode);
1742         eh = path[depth].p_hdr;
1743         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1744                 goto has_space;
1745
1746         /* probably next leaf has space for us? */
1747         fex = EXT_LAST_EXTENT(eh);
1748         next = ext4_ext_next_leaf_block(inode, path);
1749         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1750             && next != EXT_MAX_BLOCK) {
1751                 ext_debug("next leaf block - %d\n", next);
1752                 BUG_ON(npath != NULL);
1753                 npath = ext4_ext_find_extent(inode, next, NULL);
1754                 if (IS_ERR(npath))
1755                         return PTR_ERR(npath);
1756                 BUG_ON(npath->p_depth != path->p_depth);
1757                 eh = npath[depth].p_hdr;
1758                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1759                         ext_debug("next leaf isnt full(%d)\n",
1760                                   le16_to_cpu(eh->eh_entries));
1761                         path = npath;
1762                         goto repeat;
1763                 }
1764                 ext_debug("next leaf has no free space(%d,%d)\n",
1765                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1766         }
1767
1768         /*
1769          * There is no free space in the found leaf.
1770          * We're gonna add a new leaf in the tree.
1771          */
1772         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1773         if (err)
1774                 goto cleanup;
1775         depth = ext_depth(inode);
1776         eh = path[depth].p_hdr;
1777
1778 has_space:
1779         nearex = path[depth].p_ext;
1780
1781         err = ext4_ext_get_access(handle, inode, path + depth);
1782         if (err)
1783                 goto cleanup;
1784
1785         if (!nearex) {
1786                 /* there is no extent in this leaf, create first one */
1787                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1788                                 le32_to_cpu(newext->ee_block),
1789                                 ext_pblock(newext),
1790                                 ext4_ext_is_uninitialized(newext),
1791                                 ext4_ext_get_actual_len(newext));
1792                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1793         } else if (le32_to_cpu(newext->ee_block)
1794                            > le32_to_cpu(nearex->ee_block)) {
1795 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1796                 if (nearex != EXT_LAST_EXTENT(eh)) {
1797                         len = EXT_MAX_EXTENT(eh) - nearex;
1798                         len = (len - 1) * sizeof(struct ext4_extent);
1799                         len = len < 0 ? 0 : len;
1800                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1801                                         "move %d from 0x%p to 0x%p\n",
1802                                         le32_to_cpu(newext->ee_block),
1803                                         ext_pblock(newext),
1804                                         ext4_ext_is_uninitialized(newext),
1805                                         ext4_ext_get_actual_len(newext),
1806                                         nearex, len, nearex + 1, nearex + 2);
1807                         memmove(nearex + 2, nearex + 1, len);
1808                 }
1809                 path[depth].p_ext = nearex + 1;
1810         } else {
1811                 BUG_ON(newext->ee_block == nearex->ee_block);
1812                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1813                 len = len < 0 ? 0 : len;
1814                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1815                                 "move %d from 0x%p to 0x%p\n",
1816                                 le32_to_cpu(newext->ee_block),
1817                                 ext_pblock(newext),
1818                                 ext4_ext_is_uninitialized(newext),
1819                                 ext4_ext_get_actual_len(newext),
1820                                 nearex, len, nearex + 1, nearex + 2);
1821                 memmove(nearex + 1, nearex, len);
1822                 path[depth].p_ext = nearex;
1823         }
1824
1825         le16_add_cpu(&eh->eh_entries, 1);
1826         nearex = path[depth].p_ext;
1827         nearex->ee_block = newext->ee_block;
1828         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1829         nearex->ee_len = newext->ee_len;
1830
1831 merge:
1832         /* try to merge extents to the right */
1833         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1834                 ext4_ext_try_to_merge(inode, path, nearex);
1835
1836         /* try to merge extents to the left */
1837
1838         /* time to correct all indexes above */
1839         err = ext4_ext_correct_indexes(handle, inode, path);
1840         if (err)
1841                 goto cleanup;
1842
1843         err = ext4_ext_dirty(handle, inode, path + depth);
1844
1845 cleanup:
1846         if (npath) {
1847                 ext4_ext_drop_refs(npath);
1848                 kfree(npath);
1849         }
1850         ext4_ext_invalidate_cache(inode);
1851         return err;
1852 }
1853
1854 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1855                         ext4_lblk_t num, ext_prepare_callback func,
1856                         void *cbdata)
1857 {
1858         struct ext4_ext_path *path = NULL;
1859         struct ext4_ext_cache cbex;
1860         struct ext4_extent *ex;
1861         ext4_lblk_t next, start = 0, end = 0;
1862         ext4_lblk_t last = block + num;
1863         int depth, exists, err = 0;
1864
1865         BUG_ON(func == NULL);
1866         BUG_ON(inode == NULL);
1867
1868         while (block < last && block != EXT_MAX_BLOCK) {
1869                 num = last - block;
1870                 /* find extent for this block */
1871                 down_read(&EXT4_I(inode)->i_data_sem);
1872                 path = ext4_ext_find_extent(inode, block, path);
1873                 up_read(&EXT4_I(inode)->i_data_sem);
1874                 if (IS_ERR(path)) {
1875                         err = PTR_ERR(path);
1876                         path = NULL;
1877                         break;
1878                 }
1879
1880                 depth = ext_depth(inode);
1881                 if (unlikely(path[depth].p_hdr == NULL)) {
1882                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1883                         err = -EIO;
1884                         break;
1885                 }
1886                 ex = path[depth].p_ext;
1887                 next = ext4_ext_next_allocated_block(path);
1888
1889                 exists = 0;
1890                 if (!ex) {
1891                         /* there is no extent yet, so try to allocate
1892                          * all requested space */
1893                         start = block;
1894                         end = block + num;
1895                 } else if (le32_to_cpu(ex->ee_block) > block) {
1896                         /* need to allocate space before found extent */
1897                         start = block;
1898                         end = le32_to_cpu(ex->ee_block);
1899                         if (block + num < end)
1900                                 end = block + num;
1901                 } else if (block >= le32_to_cpu(ex->ee_block)
1902                                         + ext4_ext_get_actual_len(ex)) {
1903                         /* need to allocate space after found extent */
1904                         start = block;
1905                         end = block + num;
1906                         if (end >= next)
1907                                 end = next;
1908                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1909                         /*
1910                          * some part of requested space is covered
1911                          * by found extent
1912                          */
1913                         start = block;
1914                         end = le32_to_cpu(ex->ee_block)
1915                                 + ext4_ext_get_actual_len(ex);
1916                         if (block + num < end)
1917                                 end = block + num;
1918                         exists = 1;
1919                 } else {
1920                         BUG();
1921                 }
1922                 BUG_ON(end <= start);
1923
1924                 if (!exists) {
1925                         cbex.ec_block = start;
1926                         cbex.ec_len = end - start;
1927                         cbex.ec_start = 0;
1928                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1929                 } else {
1930                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1931                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1932                         cbex.ec_start = ext_pblock(ex);
1933                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1934                 }
1935
1936                 if (unlikely(cbex.ec_len == 0)) {
1937                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1938                         err = -EIO;
1939                         break;
1940                 }
1941                 err = func(inode, path, &cbex, ex, cbdata);
1942                 ext4_ext_drop_refs(path);
1943
1944                 if (err < 0)
1945                         break;
1946
1947                 if (err == EXT_REPEAT)
1948                         continue;
1949                 else if (err == EXT_BREAK) {
1950                         err = 0;
1951                         break;
1952                 }
1953
1954                 if (ext_depth(inode) != depth) {
1955                         /* depth was changed. we have to realloc path */
1956                         kfree(path);
1957                         path = NULL;
1958                 }
1959
1960                 block = cbex.ec_block + cbex.ec_len;
1961         }
1962
1963         if (path) {
1964                 ext4_ext_drop_refs(path);
1965                 kfree(path);
1966         }
1967
1968         return err;
1969 }
1970
1971 static void
1972 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1973                         __u32 len, ext4_fsblk_t start, int type)
1974 {
1975         struct ext4_ext_cache *cex;
1976         BUG_ON(len == 0);
1977         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1978         cex = &EXT4_I(inode)->i_cached_extent;
1979         cex->ec_type = type;
1980         cex->ec_block = block;
1981         cex->ec_len = len;
1982         cex->ec_start = start;
1983         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1984 }
1985
1986 /*
1987  * ext4_ext_put_gap_in_cache:
1988  * calculate boundaries of the gap that the requested block fits into
1989  * and cache this gap
1990  */
1991 static void
1992 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1993                                 ext4_lblk_t block)
1994 {
1995         int depth = ext_depth(inode);
1996         unsigned long len;
1997         ext4_lblk_t lblock;
1998         struct ext4_extent *ex;
1999
2000         ex = path[depth].p_ext;
2001         if (ex == NULL) {
2002                 /* there is no extent yet, so gap is [0;-] */
2003                 lblock = 0;
2004                 len = EXT_MAX_BLOCK;
2005                 ext_debug("cache gap(whole file):");
2006         } else if (block < le32_to_cpu(ex->ee_block)) {
2007                 lblock = block;
2008                 len = le32_to_cpu(ex->ee_block) - block;
2009                 ext_debug("cache gap(before): %u [%u:%u]",
2010                                 block,
2011                                 le32_to_cpu(ex->ee_block),
2012                                  ext4_ext_get_actual_len(ex));
2013         } else if (block >= le32_to_cpu(ex->ee_block)
2014                         + ext4_ext_get_actual_len(ex)) {
2015                 ext4_lblk_t next;
2016                 lblock = le32_to_cpu(ex->ee_block)
2017                         + ext4_ext_get_actual_len(ex);
2018
2019                 next = ext4_ext_next_allocated_block(path);
2020                 ext_debug("cache gap(after): [%u:%u] %u",
2021                                 le32_to_cpu(ex->ee_block),
2022                                 ext4_ext_get_actual_len(ex),
2023                                 block);
2024                 BUG_ON(next == lblock);
2025                 len = next - lblock;
2026         } else {
2027                 lblock = len = 0;
2028                 BUG();
2029         }
2030
2031         ext_debug(" -> %u:%lu\n", lblock, len);
2032         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
2033 }
2034
2035 static int
2036 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2037                         struct ext4_extent *ex)
2038 {
2039         struct ext4_ext_cache *cex;
2040         int ret = EXT4_EXT_CACHE_NO;
2041
2042         /*
2043          * We borrow i_block_reservation_lock to protect i_cached_extent
2044          */
2045         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2046         cex = &EXT4_I(inode)->i_cached_extent;
2047
2048         /* has cache valid data? */
2049         if (cex->ec_type == EXT4_EXT_CACHE_NO)
2050                 goto errout;
2051
2052         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2053                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2054         if (in_range(block, cex->ec_block, cex->ec_len)) {
2055                 ex->ee_block = cpu_to_le32(cex->ec_block);
2056                 ext4_ext_store_pblock(ex, cex->ec_start);
2057                 ex->ee_len = cpu_to_le16(cex->ec_len);
2058                 ext_debug("%u cached by %u:%u:%llu\n",
2059                                 block,
2060                                 cex->ec_block, cex->ec_len, cex->ec_start);
2061                 ret = cex->ec_type;
2062         }
2063 errout:
2064         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2065         return ret;
2066 }
2067
2068 /*
2069  * ext4_ext_rm_idx:
2070  * removes index from the index block.
2071  * It's used in truncate case only, thus all requests are for
2072  * last index in the block only.
2073  */
2074 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2075                         struct ext4_ext_path *path)
2076 {
2077         int err;
2078         ext4_fsblk_t leaf;
2079
2080         /* free index block */
2081         path--;
2082         leaf = idx_pblock(path->p_idx);
2083         if (unlikely(path->p_hdr->eh_entries == 0)) {
2084                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2085                 return -EIO;
2086         }
2087         err = ext4_ext_get_access(handle, inode, path);
2088         if (err)
2089                 return err;
2090         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2091         err = ext4_ext_dirty(handle, inode, path);
2092         if (err)
2093                 return err;
2094         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2095         ext4_free_blocks(handle, inode, 0, leaf, 1,
2096                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2097         return err;
2098 }
2099
2100 /*
2101  * ext4_ext_calc_credits_for_single_extent:
2102  * This routine returns max. credits that needed to insert an extent
2103  * to the extent tree.
2104  * When pass the actual path, the caller should calculate credits
2105  * under i_data_sem.
2106  */
2107 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2108                                                 struct ext4_ext_path *path)
2109 {
2110         if (path) {
2111                 int depth = ext_depth(inode);
2112                 int ret = 0;
2113
2114                 /* probably there is space in leaf? */
2115                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2116                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2117
2118                         /*
2119                          *  There are some space in the leaf tree, no
2120                          *  need to account for leaf block credit
2121                          *
2122                          *  bitmaps and block group descriptor blocks
2123                          *  and other metadat blocks still need to be
2124                          *  accounted.
2125                          */
2126                         /* 1 bitmap, 1 block group descriptor */
2127                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2128                         return ret;
2129                 }
2130         }
2131
2132         return ext4_chunk_trans_blocks(inode, nrblocks);
2133 }
2134
2135 /*
2136  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2137  *
2138  * if nrblocks are fit in a single extent (chunk flag is 1), then
2139  * in the worse case, each tree level index/leaf need to be changed
2140  * if the tree split due to insert a new extent, then the old tree
2141  * index/leaf need to be updated too
2142  *
2143  * If the nrblocks are discontiguous, they could cause
2144  * the whole tree split more than once, but this is really rare.
2145  */
2146 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2147 {
2148         int index;
2149         int depth = ext_depth(inode);
2150
2151         if (chunk)
2152                 index = depth * 2;
2153         else
2154                 index = depth * 3;
2155
2156         return index;
2157 }
2158
2159 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2160                                 struct ext4_extent *ex,
2161                                 ext4_lblk_t from, ext4_lblk_t to)
2162 {
2163         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2164         int flags = EXT4_FREE_BLOCKS_FORGET;
2165
2166         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2167                 flags |= EXT4_FREE_BLOCKS_METADATA;
2168 #ifdef EXTENTS_STATS
2169         {
2170                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2171                 spin_lock(&sbi->s_ext_stats_lock);
2172                 sbi->s_ext_blocks += ee_len;
2173                 sbi->s_ext_extents++;
2174                 if (ee_len < sbi->s_ext_min)
2175                         sbi->s_ext_min = ee_len;
2176                 if (ee_len > sbi->s_ext_max)
2177                         sbi->s_ext_max = ee_len;
2178                 if (ext_depth(inode) > sbi->s_depth_max)
2179                         sbi->s_depth_max = ext_depth(inode);
2180                 spin_unlock(&sbi->s_ext_stats_lock);
2181         }
2182 #endif
2183         if (from >= le32_to_cpu(ex->ee_block)
2184             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2185                 /* tail removal */
2186                 ext4_lblk_t num;
2187                 ext4_fsblk_t start;
2188
2189                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2190                 start = ext_pblock(ex) + ee_len - num;
2191                 ext_debug("free last %u blocks starting %llu\n", num, start);
2192                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2193         } else if (from == le32_to_cpu(ex->ee_block)
2194                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2195                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2196                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2197         } else {
2198                 printk(KERN_INFO "strange request: removal(2) "
2199                                 "%u-%u from %u:%u\n",
2200                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2201         }
2202         return 0;
2203 }
2204
2205 static int
2206 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2207                 struct ext4_ext_path *path, ext4_lblk_t start)
2208 {
2209         int err = 0, correct_index = 0;
2210         int depth = ext_depth(inode), credits;
2211         struct ext4_extent_header *eh;
2212         ext4_lblk_t a, b, block;
2213         unsigned num;
2214         ext4_lblk_t ex_ee_block;
2215         unsigned short ex_ee_len;
2216         unsigned uninitialized = 0;
2217         struct ext4_extent *ex;
2218
2219         /* the header must be checked already in ext4_ext_remove_space() */
2220         ext_debug("truncate since %u in leaf\n", start);
2221         if (!path[depth].p_hdr)
2222                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2223         eh = path[depth].p_hdr;
2224         if (unlikely(path[depth].p_hdr == NULL)) {
2225                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2226                 return -EIO;
2227         }
2228         /* find where to start removing */
2229         ex = EXT_LAST_EXTENT(eh);
2230
2231         ex_ee_block = le32_to_cpu(ex->ee_block);
2232         ex_ee_len = ext4_ext_get_actual_len(ex);
2233
2234         while (ex >= EXT_FIRST_EXTENT(eh) &&
2235                         ex_ee_block + ex_ee_len > start) {
2236
2237                 if (ext4_ext_is_uninitialized(ex))
2238                         uninitialized = 1;
2239                 else
2240                         uninitialized = 0;
2241
2242                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2243                          uninitialized, ex_ee_len);
2244                 path[depth].p_ext = ex;
2245
2246                 a = ex_ee_block > start ? ex_ee_block : start;
2247                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2248                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2249
2250                 ext_debug("  border %u:%u\n", a, b);
2251
2252                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2253                         block = 0;
2254                         num = 0;
2255                         BUG();
2256                 } else if (a != ex_ee_block) {
2257                         /* remove tail of the extent */
2258                         block = ex_ee_block;
2259                         num = a - block;
2260                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2261                         /* remove head of the extent */
2262                         block = a;
2263                         num = b - a;
2264                         /* there is no "make a hole" API yet */
2265                         BUG();
2266                 } else {
2267                         /* remove whole extent: excellent! */
2268                         block = ex_ee_block;
2269                         num = 0;
2270                         BUG_ON(a != ex_ee_block);
2271                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2272                 }
2273
2274                 /*
2275                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2276                  * descriptor) for each block group; assume two block
2277                  * groups plus ex_ee_len/blocks_per_block_group for
2278                  * the worst case
2279                  */
2280                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2281                 if (ex == EXT_FIRST_EXTENT(eh)) {
2282                         correct_index = 1;
2283                         credits += (ext_depth(inode)) + 1;
2284                 }
2285                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2286
2287                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2288                 if (err)
2289                         goto out;
2290
2291                 err = ext4_ext_get_access(handle, inode, path + depth);
2292                 if (err)
2293                         goto out;
2294
2295                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2296                 if (err)
2297                         goto out;
2298
2299                 if (num == 0) {
2300                         /* this extent is removed; mark slot entirely unused */
2301                         ext4_ext_store_pblock(ex, 0);
2302                         le16_add_cpu(&eh->eh_entries, -1);
2303                 }
2304
2305                 ex->ee_block = cpu_to_le32(block);
2306                 ex->ee_len = cpu_to_le16(num);
2307                 /*
2308                  * Do not mark uninitialized if all the blocks in the
2309                  * extent have been removed.
2310                  */
2311                 if (uninitialized && num)
2312                         ext4_ext_mark_uninitialized(ex);
2313
2314                 err = ext4_ext_dirty(handle, inode, path + depth);
2315                 if (err)
2316                         goto out;
2317
2318                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2319                                 ext_pblock(ex));
2320                 ex--;
2321                 ex_ee_block = le32_to_cpu(ex->ee_block);
2322                 ex_ee_len = ext4_ext_get_actual_len(ex);
2323         }
2324
2325         if (correct_index && eh->eh_entries)
2326                 err = ext4_ext_correct_indexes(handle, inode, path);
2327
2328         /* if this leaf is free, then we should
2329          * remove it from index block above */
2330         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2331                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2332
2333 out:
2334         return err;
2335 }
2336
2337 /*
2338  * ext4_ext_more_to_rm:
2339  * returns 1 if current index has to be freed (even partial)
2340  */
2341 static int
2342 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2343 {
2344         BUG_ON(path->p_idx == NULL);
2345
2346         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2347                 return 0;
2348
2349         /*
2350          * if truncate on deeper level happened, it wasn't partial,
2351          * so we have to consider current index for truncation
2352          */
2353         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2354                 return 0;
2355         return 1;
2356 }
2357
2358 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2359 {
2360         struct super_block *sb = inode->i_sb;
2361         int depth = ext_depth(inode);
2362         struct ext4_ext_path *path;
2363         handle_t *handle;
2364         int i = 0, err = 0;
2365
2366         ext_debug("truncate since %u\n", start);
2367
2368         /* probably first extent we're gonna free will be last in block */
2369         handle = ext4_journal_start(inode, depth + 1);
2370         if (IS_ERR(handle))
2371                 return PTR_ERR(handle);
2372
2373         ext4_ext_invalidate_cache(inode);
2374
2375         /*
2376          * We start scanning from right side, freeing all the blocks
2377          * after i_size and walking into the tree depth-wise.
2378          */
2379         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2380         if (path == NULL) {
2381                 ext4_journal_stop(handle);
2382                 return -ENOMEM;
2383         }
2384         path[0].p_hdr = ext_inode_hdr(inode);
2385         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2386                 err = -EIO;
2387                 goto out;
2388         }
2389         path[0].p_depth = depth;
2390
2391         while (i >= 0 && err == 0) {
2392                 if (i == depth) {
2393                         /* this is leaf block */
2394                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2395                         /* root level has p_bh == NULL, brelse() eats this */
2396                         brelse(path[i].p_bh);
2397                         path[i].p_bh = NULL;
2398                         i--;
2399                         continue;
2400                 }
2401
2402                 /* this is index block */
2403                 if (!path[i].p_hdr) {
2404                         ext_debug("initialize header\n");
2405                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2406                 }
2407
2408                 if (!path[i].p_idx) {
2409                         /* this level hasn't been touched yet */
2410                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2411                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2412                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2413                                   path[i].p_hdr,
2414                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2415                 } else {
2416                         /* we were already here, see at next index */
2417                         path[i].p_idx--;
2418                 }
2419
2420                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2421                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2422                                 path[i].p_idx);
2423                 if (ext4_ext_more_to_rm(path + i)) {
2424                         struct buffer_head *bh;
2425                         /* go to the next level */
2426                         ext_debug("move to level %d (block %llu)\n",
2427                                   i + 1, idx_pblock(path[i].p_idx));
2428                         memset(path + i + 1, 0, sizeof(*path));
2429                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2430                         if (!bh) {
2431                                 /* should we reset i_size? */
2432                                 err = -EIO;
2433                                 break;
2434                         }
2435                         if (WARN_ON(i + 1 > depth)) {
2436                                 err = -EIO;
2437                                 break;
2438                         }
2439                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2440                                                         depth - i - 1)) {
2441                                 err = -EIO;
2442                                 break;
2443                         }
2444                         path[i + 1].p_bh = bh;
2445
2446                         /* save actual number of indexes since this
2447                          * number is changed at the next iteration */
2448                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2449                         i++;
2450                 } else {
2451                         /* we finished processing this index, go up */
2452                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2453                                 /* index is empty, remove it;
2454                                  * handle must be already prepared by the
2455                                  * truncatei_leaf() */
2456                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2457                         }
2458                         /* root level has p_bh == NULL, brelse() eats this */
2459                         brelse(path[i].p_bh);
2460                         path[i].p_bh = NULL;
2461                         i--;
2462                         ext_debug("return to level %d\n", i);
2463                 }
2464         }
2465
2466         /* TODO: flexible tree reduction should be here */
2467         if (path->p_hdr->eh_entries == 0) {
2468                 /*
2469                  * truncate to zero freed all the tree,
2470                  * so we need to correct eh_depth
2471                  */
2472                 err = ext4_ext_get_access(handle, inode, path);
2473                 if (err == 0) {
2474                         ext_inode_hdr(inode)->eh_depth = 0;
2475                         ext_inode_hdr(inode)->eh_max =
2476                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2477                         err = ext4_ext_dirty(handle, inode, path);
2478                 }
2479         }
2480 out:
2481         ext4_ext_drop_refs(path);
2482         kfree(path);
2483         ext4_journal_stop(handle);
2484
2485         return err;
2486 }
2487
2488 /*
2489  * called at mount time
2490  */
2491 void ext4_ext_init(struct super_block *sb)
2492 {
2493         /*
2494          * possible initialization would be here
2495          */
2496
2497         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2498 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2499                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2500 #ifdef AGGRESSIVE_TEST
2501                 printk(", aggressive tests");
2502 #endif
2503 #ifdef CHECK_BINSEARCH
2504                 printk(", check binsearch");
2505 #endif
2506 #ifdef EXTENTS_STATS
2507                 printk(", stats");
2508 #endif
2509                 printk("\n");
2510 #endif
2511 #ifdef EXTENTS_STATS
2512                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2513                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2514                 EXT4_SB(sb)->s_ext_max = 0;
2515 #endif
2516         }
2517 }
2518
2519 /*
2520  * called at umount time
2521  */
2522 void ext4_ext_release(struct super_block *sb)
2523 {
2524         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2525                 return;
2526
2527 #ifdef EXTENTS_STATS
2528         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2529                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2530                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2531                         sbi->s_ext_blocks, sbi->s_ext_extents,
2532                         sbi->s_ext_blocks / sbi->s_ext_extents);
2533                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2534                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2535         }
2536 #endif
2537 }
2538
2539 static void bi_complete(struct bio *bio, int error)
2540 {
2541         complete((struct completion *)bio->bi_private);
2542 }
2543
2544 /* FIXME!! we need to try to merge to left or right after zero-out  */
2545 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2546 {
2547         int ret = -EIO;
2548         struct bio *bio;
2549         int blkbits, blocksize;
2550         sector_t ee_pblock;
2551         struct completion event;
2552         unsigned int ee_len, len, done, offset;
2553
2554
2555         blkbits   = inode->i_blkbits;
2556         blocksize = inode->i_sb->s_blocksize;
2557         ee_len    = ext4_ext_get_actual_len(ex);
2558         ee_pblock = ext_pblock(ex);
2559
2560         /* convert ee_pblock to 512 byte sectors */
2561         ee_pblock = ee_pblock << (blkbits - 9);
2562
2563         while (ee_len > 0) {
2564
2565                 if (ee_len > BIO_MAX_PAGES)
2566                         len = BIO_MAX_PAGES;
2567                 else
2568                         len = ee_len;
2569
2570                 bio = bio_alloc(GFP_NOIO, len);
2571                 bio->bi_sector = ee_pblock;
2572                 bio->bi_bdev   = inode->i_sb->s_bdev;
2573
2574                 done = 0;
2575                 offset = 0;
2576                 while (done < len) {
2577                         ret = bio_add_page(bio, ZERO_PAGE(0),
2578                                                         blocksize, offset);
2579                         if (ret != blocksize) {
2580                                 /*
2581                                  * We can't add any more pages because of
2582                                  * hardware limitations.  Start a new bio.
2583                                  */
2584                                 break;
2585                         }
2586                         done++;
2587                         offset += blocksize;
2588                         if (offset >= PAGE_CACHE_SIZE)
2589                                 offset = 0;
2590                 }
2591
2592                 init_completion(&event);
2593                 bio->bi_private = &event;
2594                 bio->bi_end_io = bi_complete;
2595                 submit_bio(WRITE, bio);
2596                 wait_for_completion(&event);
2597
2598                 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2599                         ret = 0;
2600                 else {
2601                         ret = -EIO;
2602                         break;
2603                 }
2604                 bio_put(bio);
2605                 ee_len    -= done;
2606                 ee_pblock += done  << (blkbits - 9);
2607         }
2608         return ret;
2609 }
2610
2611 #define EXT4_EXT_ZERO_LEN 7
2612 /*
2613  * This function is called by ext4_ext_get_blocks() if someone tries to write
2614  * to an uninitialized extent. It may result in splitting the uninitialized
2615  * extent into multiple extents (upto three - one initialized and two
2616  * uninitialized).
2617  * There are three possibilities:
2618  *   a> There is no split required: Entire extent should be initialized
2619  *   b> Splits in two extents: Write is happening at either end of the extent
2620  *   c> Splits in three extents: Somone is writing in middle of the extent
2621  */
2622 static int ext4_ext_convert_to_initialized(handle_t *handle,
2623                                                 struct inode *inode,
2624                                                 struct ext4_ext_path *path,
2625                                                 ext4_lblk_t iblock,
2626                                                 unsigned int max_blocks)
2627 {
2628         struct ext4_extent *ex, newex, orig_ex;
2629         struct ext4_extent *ex1 = NULL;
2630         struct ext4_extent *ex2 = NULL;
2631         struct ext4_extent *ex3 = NULL;
2632         struct ext4_extent_header *eh;
2633         ext4_lblk_t ee_block;
2634         unsigned int allocated, ee_len, depth;
2635         ext4_fsblk_t newblock;
2636         int err = 0;
2637         int ret = 0;
2638
2639         depth = ext_depth(inode);
2640         eh = path[depth].p_hdr;
2641         ex = path[depth].p_ext;
2642         ee_block = le32_to_cpu(ex->ee_block);
2643         ee_len = ext4_ext_get_actual_len(ex);
2644         allocated = ee_len - (iblock - ee_block);
2645         newblock = iblock - ee_block + ext_pblock(ex);
2646         ex2 = ex;
2647         orig_ex.ee_block = ex->ee_block;
2648         orig_ex.ee_len   = cpu_to_le16(ee_len);
2649         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2650
2651         err = ext4_ext_get_access(handle, inode, path + depth);
2652         if (err)
2653                 goto out;
2654         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2655         if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2656                 err =  ext4_ext_zeroout(inode, &orig_ex);
2657                 if (err)
2658                         goto fix_extent_len;
2659                 /* update the extent length and mark as initialized */
2660                 ex->ee_block = orig_ex.ee_block;
2661                 ex->ee_len   = orig_ex.ee_len;
2662                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2663                 ext4_ext_dirty(handle, inode, path + depth);
2664                 /* zeroed the full extent */
2665                 return allocated;
2666         }
2667
2668         /* ex1: ee_block to iblock - 1 : uninitialized */
2669         if (iblock > ee_block) {
2670                 ex1 = ex;
2671                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2672                 ext4_ext_mark_uninitialized(ex1);
2673                 ex2 = &newex;
2674         }
2675         /*
2676          * for sanity, update the length of the ex2 extent before
2677          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2678          * overlap of blocks.
2679          */
2680         if (!ex1 && allocated > max_blocks)
2681                 ex2->ee_len = cpu_to_le16(max_blocks);
2682         /* ex3: to ee_block + ee_len : uninitialised */
2683         if (allocated > max_blocks) {
2684                 unsigned int newdepth;
2685                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2686                 if (allocated <= EXT4_EXT_ZERO_LEN) {
2687                         /*
2688                          * iblock == ee_block is handled by the zerouout
2689                          * at the beginning.
2690                          * Mark first half uninitialized.
2691                          * Mark second half initialized and zero out the
2692                          * initialized extent
2693                          */
2694                         ex->ee_block = orig_ex.ee_block;
2695                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2696                         ext4_ext_mark_uninitialized(ex);
2697                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2698                         ext4_ext_dirty(handle, inode, path + depth);
2699
2700                         ex3 = &newex;
2701                         ex3->ee_block = cpu_to_le32(iblock);
2702                         ext4_ext_store_pblock(ex3, newblock);
2703                         ex3->ee_len = cpu_to_le16(allocated);
2704                         err = ext4_ext_insert_extent(handle, inode, path,
2705                                                         ex3, 0);
2706                         if (err == -ENOSPC) {
2707                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2708                                 if (err)
2709                                         goto fix_extent_len;
2710                                 ex->ee_block = orig_ex.ee_block;
2711                                 ex->ee_len   = orig_ex.ee_len;
2712                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2713                                 ext4_ext_dirty(handle, inode, path + depth);
2714                                 /* blocks available from iblock */
2715                                 return allocated;
2716
2717                         } else if (err)
2718                                 goto fix_extent_len;
2719
2720                         /*
2721                          * We need to zero out the second half because
2722                          * an fallocate request can update file size and
2723                          * converting the second half to initialized extent
2724                          * implies that we can leak some junk data to user
2725                          * space.
2726                          */
2727                         err =  ext4_ext_zeroout(inode, ex3);
2728                         if (err) {
2729                                 /*
2730                                  * We should actually mark the
2731                                  * second half as uninit and return error
2732                                  * Insert would have changed the extent
2733                                  */
2734                                 depth = ext_depth(inode);
2735                                 ext4_ext_drop_refs(path);
2736                                 path = ext4_ext_find_extent(inode,
2737                                                                 iblock, path);
2738                                 if (IS_ERR(path)) {
2739                                         err = PTR_ERR(path);
2740                                         return err;
2741                                 }
2742                                 /* get the second half extent details */
2743                                 ex = path[depth].p_ext;
2744                                 err = ext4_ext_get_access(handle, inode,
2745                                                                 path + depth);
2746                                 if (err)
2747                                         return err;
2748                                 ext4_ext_mark_uninitialized(ex);
2749                                 ext4_ext_dirty(handle, inode, path + depth);
2750                                 return err;
2751                         }
2752
2753                         /* zeroed the second half */
2754                         return allocated;
2755                 }
2756                 ex3 = &newex;
2757                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2758                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2759                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2760                 ext4_ext_mark_uninitialized(ex3);
2761                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2762                 if (err == -ENOSPC) {
2763                         err =  ext4_ext_zeroout(inode, &orig_ex);
2764                         if (err)
2765                                 goto fix_extent_len;
2766                         /* update the extent length and mark as initialized */
2767                         ex->ee_block = orig_ex.ee_block;
2768                         ex->ee_len   = orig_ex.ee_len;
2769                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2770                         ext4_ext_dirty(handle, inode, path + depth);
2771                         /* zeroed the full extent */
2772                         /* blocks available from iblock */
2773                         return allocated;
2774
2775                 } else if (err)
2776                         goto fix_extent_len;
2777                 /*
2778                  * The depth, and hence eh & ex might change
2779                  * as part of the insert above.
2780                  */
2781                 newdepth = ext_depth(inode);
2782                 /*
2783                  * update the extent length after successful insert of the
2784                  * split extent
2785                  */
2786                 orig_ex.ee_len = cpu_to_le16(ee_len -
2787                                                 ext4_ext_get_actual_len(ex3));
2788                 depth = newdepth;
2789                 ext4_ext_drop_refs(path);
2790                 path = ext4_ext_find_extent(inode, iblock, path);
2791                 if (IS_ERR(path)) {
2792                         err = PTR_ERR(path);
2793                         goto out;
2794                 }
2795                 eh = path[depth].p_hdr;
2796                 ex = path[depth].p_ext;
2797                 if (ex2 != &newex)
2798                         ex2 = ex;
2799
2800                 err = ext4_ext_get_access(handle, inode, path + depth);
2801                 if (err)
2802                         goto out;
2803
2804                 allocated = max_blocks;
2805
2806                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2807                  * to insert a extent in the middle zerout directly
2808                  * otherwise give the extent a chance to merge to left
2809                  */
2810                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2811                                                         iblock != ee_block) {
2812                         err =  ext4_ext_zeroout(inode, &orig_ex);
2813                         if (err)
2814                                 goto fix_extent_len;
2815                         /* update the extent length and mark as initialized */
2816                         ex->ee_block = orig_ex.ee_block;
2817                         ex->ee_len   = orig_ex.ee_len;
2818                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2819                         ext4_ext_dirty(handle, inode, path + depth);
2820                         /* zero out the first half */
2821                         /* blocks available from iblock */
2822                         return allocated;
2823                 }
2824         }
2825         /*
2826          * If there was a change of depth as part of the
2827          * insertion of ex3 above, we need to update the length
2828          * of the ex1 extent again here
2829          */
2830         if (ex1 && ex1 != ex) {
2831                 ex1 = ex;
2832                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2833                 ext4_ext_mark_uninitialized(ex1);
2834                 ex2 = &newex;
2835         }
2836         /* ex2: iblock to iblock + maxblocks-1 : initialised */
2837         ex2->ee_block = cpu_to_le32(iblock);
2838         ext4_ext_store_pblock(ex2, newblock);
2839         ex2->ee_len = cpu_to_le16(allocated);
2840         if (ex2 != ex)
2841                 goto insert;
2842         /*
2843          * New (initialized) extent starts from the first block
2844          * in the current extent. i.e., ex2 == ex
2845          * We have to see if it can be merged with the extent
2846          * on the left.
2847          */
2848         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2849                 /*
2850                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2851                  * since it merges towards right _only_.
2852                  */
2853                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2854                 if (ret) {
2855                         err = ext4_ext_correct_indexes(handle, inode, path);
2856                         if (err)
2857                                 goto out;
2858                         depth = ext_depth(inode);
2859                         ex2--;
2860                 }
2861         }
2862         /*
2863          * Try to Merge towards right. This might be required
2864          * only when the whole extent is being written to.
2865          * i.e. ex2 == ex and ex3 == NULL.
2866          */
2867         if (!ex3) {
2868                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2869                 if (ret) {
2870                         err = ext4_ext_correct_indexes(handle, inode, path);
2871                         if (err)
2872                                 goto out;
2873                 }
2874         }
2875         /* Mark modified extent as dirty */
2876         err = ext4_ext_dirty(handle, inode, path + depth);
2877         goto out;
2878 insert:
2879         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2880         if (err == -ENOSPC) {
2881                 err =  ext4_ext_zeroout(inode, &orig_ex);
2882                 if (err)
2883                         goto fix_extent_len;
2884                 /* update the extent length and mark as initialized */
2885                 ex->ee_block = orig_ex.ee_block;
2886                 ex->ee_len   = orig_ex.ee_len;
2887                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2888                 ext4_ext_dirty(handle, inode, path + depth);
2889                 /* zero out the first half */
2890                 return allocated;
2891         } else if (err)
2892                 goto fix_extent_len;
2893 out:
2894         ext4_ext_show_leaf(inode, path);
2895         return err ? err : allocated;
2896
2897 fix_extent_len:
2898         ex->ee_block = orig_ex.ee_block;
2899         ex->ee_len   = orig_ex.ee_len;
2900         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2901         ext4_ext_mark_uninitialized(ex);
2902         ext4_ext_dirty(handle, inode, path + depth);
2903         return err;
2904 }
2905
2906 /*
2907  * This function is called by ext4_ext_get_blocks() from
2908  * ext4_get_blocks_dio_write() when DIO to write
2909  * to an uninitialized extent.
2910  *
2911  * Writing to an uninitized extent may result in splitting the uninitialized
2912  * extent into multiple /intialized unintialized extents (up to three)
2913  * There are three possibilities:
2914  *   a> There is no split required: Entire extent should be uninitialized
2915  *   b> Splits in two extents: Write is happening at either end of the extent
2916  *   c> Splits in three extents: Somone is writing in middle of the extent
2917  *
2918  * One of more index blocks maybe needed if the extent tree grow after
2919  * the unintialized extent split. To prevent ENOSPC occur at the IO
2920  * complete, we need to split the uninitialized extent before DIO submit
2921  * the IO. The uninitilized extent called at this time will be split
2922  * into three uninitialized extent(at most). After IO complete, the part
2923  * being filled will be convert to initialized by the end_io callback function
2924  * via ext4_convert_unwritten_extents().
2925  *
2926  * Returns the size of uninitialized extent to be written on success.
2927  */
2928 static int ext4_split_unwritten_extents(handle_t *handle,
2929                                         struct inode *inode,
2930                                         struct ext4_ext_path *path,
2931                                         ext4_lblk_t iblock,
2932                                         unsigned int max_blocks,
2933                                         int flags)
2934 {
2935         struct ext4_extent *ex, newex, orig_ex;
2936         struct ext4_extent *ex1 = NULL;
2937         struct ext4_extent *ex2 = NULL;
2938         struct ext4_extent *ex3 = NULL;
2939         struct ext4_extent_header *eh;
2940         ext4_lblk_t ee_block;
2941         unsigned int allocated, ee_len, depth;
2942         ext4_fsblk_t newblock;
2943         int err = 0;
2944
2945         ext_debug("ext4_split_unwritten_extents: inode %lu,"
2946                   "iblock %llu, max_blocks %u\n", inode->i_ino,
2947                   (unsigned long long)iblock, max_blocks);
2948         depth = ext_depth(inode);
2949         eh = path[depth].p_hdr;
2950         ex = path[depth].p_ext;
2951         ee_block = le32_to_cpu(ex->ee_block);
2952         ee_len = ext4_ext_get_actual_len(ex);
2953         allocated = ee_len - (iblock - ee_block);
2954         newblock = iblock - ee_block + ext_pblock(ex);
2955         ex2 = ex;
2956         orig_ex.ee_block = ex->ee_block;
2957         orig_ex.ee_len   = cpu_to_le16(ee_len);
2958         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2959
2960         /*
2961          * If the uninitialized extent begins at the same logical
2962          * block where the write begins, and the write completely
2963          * covers the extent, then we don't need to split it.
2964          */
2965         if ((iblock == ee_block) && (allocated <= max_blocks))
2966                 return allocated;
2967
2968         err = ext4_ext_get_access(handle, inode, path + depth);
2969         if (err)
2970                 goto out;
2971         /* ex1: ee_block to iblock - 1 : uninitialized */
2972         if (iblock > ee_block) {
2973                 ex1 = ex;
2974                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2975                 ext4_ext_mark_uninitialized(ex1);
2976                 ex2 = &newex;
2977         }
2978         /*
2979          * for sanity, update the length of the ex2 extent before
2980          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2981          * overlap of blocks.
2982          */
2983         if (!ex1 && allocated > max_blocks)
2984                 ex2->ee_len = cpu_to_le16(max_blocks);
2985         /* ex3: to ee_block + ee_len : uninitialised */
2986         if (allocated > max_blocks) {
2987                 unsigned int newdepth;
2988                 ex3 = &newex;
2989                 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2990                 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2991                 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2992                 ext4_ext_mark_uninitialized(ex3);
2993                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2994                 if (err == -ENOSPC) {
2995                         err =  ext4_ext_zeroout(inode, &orig_ex);
2996                         if (err)
2997                                 goto fix_extent_len;
2998                         /* update the extent length and mark as initialized */
2999                         ex->ee_block = orig_ex.ee_block;
3000                         ex->ee_len   = orig_ex.ee_len;
3001                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3002                         ext4_ext_dirty(handle, inode, path + depth);
3003                         /* zeroed the full extent */
3004                         /* blocks available from iblock */
3005                         return allocated;
3006
3007                 } else if (err)
3008                         goto fix_extent_len;
3009                 /*
3010                  * The depth, and hence eh & ex might change
3011                  * as part of the insert above.
3012                  */
3013                 newdepth = ext_depth(inode);
3014                 /*
3015                  * update the extent length after successful insert of the
3016                  * split extent
3017                  */
3018                 orig_ex.ee_len = cpu_to_le16(ee_len -
3019                                                 ext4_ext_get_actual_len(ex3));
3020                 depth = newdepth;
3021                 ext4_ext_drop_refs(path);
3022                 path = ext4_ext_find_extent(inode, iblock, path);
3023                 if (IS_ERR(path)) {
3024                         err = PTR_ERR(path);
3025                         goto out;
3026                 }
3027                 eh = path[depth].p_hdr;
3028                 ex = path[depth].p_ext;
3029                 if (ex2 != &newex)
3030                         ex2 = ex;
3031
3032                 err = ext4_ext_get_access(handle, inode, path + depth);
3033                 if (err)
3034                         goto out;
3035
3036                 allocated = max_blocks;
3037         }
3038         /*
3039          * If there was a change of depth as part of the
3040          * insertion of ex3 above, we need to update the length
3041          * of the ex1 extent again here
3042          */
3043         if (ex1 && ex1 != ex) {
3044                 ex1 = ex;
3045                 ex1->ee_len = cpu_to_le16(iblock - ee_block);
3046                 ext4_ext_mark_uninitialized(ex1);
3047                 ex2 = &newex;
3048         }
3049         /*
3050          * ex2: iblock to iblock + maxblocks-1 : to be direct IO written,
3051          * uninitialised still.
3052          */
3053         ex2->ee_block = cpu_to_le32(iblock);
3054         ext4_ext_store_pblock(ex2, newblock);
3055         ex2->ee_len = cpu_to_le16(allocated);
3056         ext4_ext_mark_uninitialized(ex2);
3057         if (ex2 != ex)
3058                 goto insert;
3059         /* Mark modified extent as dirty */
3060         err = ext4_ext_dirty(handle, inode, path + depth);
3061         ext_debug("out here\n");
3062         goto out;
3063 insert:
3064         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3065         if (err == -ENOSPC) {
3066                 err =  ext4_ext_zeroout(inode, &orig_ex);
3067                 if (err)
3068                         goto fix_extent_len;
3069                 /* update the extent length and mark as initialized */
3070                 ex->ee_block = orig_ex.ee_block;
3071                 ex->ee_len   = orig_ex.ee_len;
3072                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3073                 ext4_ext_dirty(handle, inode, path + depth);
3074                 /* zero out the first half */
3075                 return allocated;
3076         } else if (err)
3077                 goto fix_extent_len;
3078 out:
3079         ext4_ext_show_leaf(inode, path);
3080         return err ? err : allocated;
3081
3082 fix_extent_len:
3083         ex->ee_block = orig_ex.ee_block;
3084         ex->ee_len   = orig_ex.ee_len;
3085         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3086         ext4_ext_mark_uninitialized(ex);
3087         ext4_ext_dirty(handle, inode, path + depth);
3088         return err;
3089 }
3090 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3091                                               struct inode *inode,
3092                                               struct ext4_ext_path *path)
3093 {
3094         struct ext4_extent *ex;
3095         struct ext4_extent_header *eh;
3096         int depth;
3097         int err = 0;
3098         int ret = 0;
3099
3100         depth = ext_depth(inode);
3101         eh = path[depth].p_hdr;
3102         ex = path[depth].p_ext;
3103
3104         err = ext4_ext_get_access(handle, inode, path + depth);
3105         if (err)
3106                 goto out;
3107         /* first mark the extent as initialized */
3108         ext4_ext_mark_initialized(ex);
3109
3110         /*
3111          * We have to see if it can be merged with the extent
3112          * on the left.
3113          */
3114         if (ex > EXT_FIRST_EXTENT(eh)) {
3115                 /*
3116                  * To merge left, pass "ex - 1" to try_to_merge(),
3117                  * since it merges towards right _only_.
3118                  */
3119                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3120                 if (ret) {
3121                         err = ext4_ext_correct_indexes(handle, inode, path);
3122                         if (err)
3123                                 goto out;
3124                         depth = ext_depth(inode);
3125                         ex--;
3126                 }
3127         }
3128         /*
3129          * Try to Merge towards right.
3130          */
3131         ret = ext4_ext_try_to_merge(inode, path, ex);
3132         if (ret) {
3133                 err = ext4_ext_correct_indexes(handle, inode, path);
3134                 if (err)
3135                         goto out;
3136                 depth = ext_depth(inode);
3137         }
3138         /* Mark modified extent as dirty */
3139         err = ext4_ext_dirty(handle, inode, path + depth);
3140 out:
3141         ext4_ext_show_leaf(inode, path);
3142         return err;
3143 }
3144
3145 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3146                         sector_t block, int count)
3147 {
3148         int i;
3149         for (i = 0; i < count; i++)
3150                 unmap_underlying_metadata(bdev, block + i);
3151 }
3152
3153 static int
3154 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3155                         ext4_lblk_t iblock, unsigned int max_blocks,
3156                         struct ext4_ext_path *path, int flags,
3157                         unsigned int allocated, struct buffer_head *bh_result,
3158                         ext4_fsblk_t newblock)
3159 {
3160         int ret = 0;
3161         int err = 0;
3162         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3163
3164         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3165                   "block %llu, max_blocks %u, flags %d, allocated %u",
3166                   inode->i_ino, (unsigned long long)iblock, max_blocks,
3167                   flags, allocated);
3168         ext4_ext_show_leaf(inode, path);
3169
3170         /* get_block() before submit the IO, split the extent */
3171         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3172                 ret = ext4_split_unwritten_extents(handle,
3173                                                 inode, path, iblock,
3174                                                 max_blocks, flags);
3175                 /*
3176                  * Flag the inode(non aio case) or end_io struct (aio case)
3177                  * that this IO needs to convertion to written when IO is
3178                  * completed
3179                  */
3180                 if (io)
3181                         io->flag = EXT4_IO_UNWRITTEN;
3182                 else
3183                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3184                 if (ext4_should_dioread_nolock(inode))
3185                         set_buffer_uninit(bh_result);
3186                 goto out;
3187         }
3188         /* IO end_io complete, convert the filled extent to written */
3189         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3190                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3191                                                         path);
3192                 if (ret >= 0)
3193                         ext4_update_inode_fsync_trans(handle, inode, 1);
3194                 goto out2;
3195         }
3196         /* buffered IO case */
3197         /*
3198          * repeat fallocate creation request
3199          * we already have an unwritten extent
3200          */
3201         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3202                 goto map_out;
3203
3204         /* buffered READ or buffered write_begin() lookup */
3205         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3206                 /*
3207                  * We have blocks reserved already.  We
3208                  * return allocated blocks so that delalloc
3209                  * won't do block reservation for us.  But
3210                  * the buffer head will be unmapped so that
3211                  * a read from the block returns 0s.
3212                  */
3213                 set_buffer_unwritten(bh_result);
3214                 goto out1;
3215         }
3216
3217         /* buffered write, writepage time, convert*/
3218         ret = ext4_ext_convert_to_initialized(handle, inode,
3219                                                 path, iblock,
3220                                                 max_blocks);
3221         if (ret >= 0)
3222                 ext4_update_inode_fsync_trans(handle, inode, 1);
3223 out:
3224         if (ret <= 0) {
3225                 err = ret;
3226                 goto out2;
3227         } else
3228                 allocated = ret;
3229         set_buffer_new(bh_result);
3230         /*
3231          * if we allocated more blocks than requested
3232          * we need to make sure we unmap the extra block
3233          * allocated. The actual needed block will get
3234          * unmapped later when we find the buffer_head marked
3235          * new.
3236          */
3237         if (allocated > max_blocks) {
3238                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3239                                         newblock + max_blocks,
3240                                         allocated - max_blocks);
3241                 allocated = max_blocks;
3242         }
3243
3244         /*
3245          * If we have done fallocate with the offset that is already
3246          * delayed allocated, we would have block reservation
3247          * and quota reservation done in the delayed write path.
3248          * But fallocate would have already updated quota and block
3249          * count for this offset. So cancel these reservation
3250          */
3251         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3252                 ext4_da_update_reserve_space(inode, allocated, 0);
3253
3254 map_out:
3255         set_buffer_mapped(bh_result);
3256 out1:
3257         if (allocated > max_blocks)
3258                 allocated = max_blocks;
3259         ext4_ext_show_leaf(inode, path);
3260         bh_result->b_bdev = inode->i_sb->s_bdev;
3261         bh_result->b_blocknr = newblock;
3262 out2:
3263         if (path) {
3264                 ext4_ext_drop_refs(path);
3265                 kfree(path);
3266         }
3267         return err ? err : allocated;
3268 }
3269 /*
3270  * Block allocation/map/preallocation routine for extents based files
3271  *
3272  *
3273  * Need to be called with
3274  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3275  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3276  *
3277  * return > 0, number of of blocks already mapped/allocated
3278  *          if create == 0 and these are pre-allocated blocks
3279  *              buffer head is unmapped
3280  *          otherwise blocks are mapped
3281  *
3282  * return = 0, if plain look up failed (blocks have not been allocated)
3283  *          buffer head is unmapped
3284  *
3285  * return < 0, error case.
3286  */
3287 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
3288                         ext4_lblk_t iblock,
3289                         unsigned int max_blocks, struct buffer_head *bh_result,
3290                         int flags)
3291 {
3292         struct ext4_ext_path *path = NULL;
3293         struct ext4_extent_header *eh;
3294         struct ext4_extent newex, *ex, *last_ex;
3295         ext4_fsblk_t newblock;
3296         int err = 0, depth, ret, cache_type;
3297         unsigned int allocated = 0;
3298         struct ext4_allocation_request ar;
3299         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3300
3301         __clear_bit(BH_New, &bh_result->b_state);
3302         ext_debug("blocks %u/%u requested for inode %lu\n",
3303                         iblock, max_blocks, inode->i_ino);
3304
3305         /* check in cache */
3306         cache_type = ext4_ext_in_cache(inode, iblock, &newex);
3307         if (cache_type) {
3308                 if (cache_type == EXT4_EXT_CACHE_GAP) {
3309                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3310                                 /*
3311                                  * block isn't allocated yet and
3312                                  * user doesn't want to allocate it
3313                                  */
3314                                 goto out2;
3315                         }
3316                         /* we should allocate requested block */
3317                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3318                         /* block is already allocated */
3319                         newblock = iblock
3320                                    - le32_to_cpu(newex.ee_block)
3321                                    + ext_pblock(&newex);
3322                         /* number of remaining blocks in the extent */
3323                         allocated = ext4_ext_get_actual_len(&newex) -
3324                                         (iblock - le32_to_cpu(newex.ee_block));
3325                         goto out;
3326                 } else {
3327                         BUG();
3328                 }
3329         }
3330
3331         /* find extent for this block */
3332         path = ext4_ext_find_extent(inode, iblock, NULL);
3333         if (IS_ERR(path)) {
3334                 err = PTR_ERR(path);
3335                 path = NULL;
3336                 goto out2;
3337         }
3338
3339         depth = ext_depth(inode);
3340
3341         /*
3342          * consistent leaf must not be empty;
3343          * this situation is possible, though, _during_ tree modification;
3344          * this is why assert can't be put in ext4_ext_find_extent()
3345          */
3346         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3347                 EXT4_ERROR_INODE(inode, "bad extent address "
3348                                  "iblock: %d, depth: %d pblock %lld",
3349                                  iblock, depth, path[depth].p_block);
3350                 err = -EIO;
3351                 goto out2;
3352         }
3353         eh = path[depth].p_hdr;
3354
3355         ex = path[depth].p_ext;
3356         if (ex) {
3357                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3358                 ext4_fsblk_t ee_start = ext_pblock(ex);
3359                 unsigned short ee_len;
3360
3361                 /*
3362                  * Uninitialized extents are treated as holes, except that
3363                  * we split out initialized portions during a write.
3364                  */
3365                 ee_len = ext4_ext_get_actual_len(ex);
3366                 /* if found extent covers block, simply return it */
3367                 if (in_range(iblock, ee_block, ee_len)) {
3368                         newblock = iblock - ee_block + ee_start;
3369                         /* number of remaining blocks in the extent */
3370                         allocated = ee_len - (iblock - ee_block);
3371                         ext_debug("%u fit into %u:%d -> %llu\n", iblock,
3372                                         ee_block, ee_len, newblock);
3373
3374                         /* Do not put uninitialized extent in the cache */
3375                         if (!ext4_ext_is_uninitialized(ex)) {
3376                                 ext4_ext_put_in_cache(inode, ee_block,
3377                                                         ee_len, ee_start,
3378                                                         EXT4_EXT_CACHE_EXTENT);
3379                                 goto out;
3380                         }
3381                         ret = ext4_ext_handle_uninitialized_extents(handle,
3382                                         inode, iblock, max_blocks, path,
3383                                         flags, allocated, bh_result, newblock);
3384                         return ret;
3385                 }
3386         }
3387
3388         /*
3389          * requested block isn't allocated yet;
3390          * we couldn't try to create block if create flag is zero
3391          */
3392         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3393                 /*
3394                  * put just found gap into cache to speed up
3395                  * subsequent requests
3396                  */
3397                 ext4_ext_put_gap_in_cache(inode, path, iblock);
3398                 goto out2;
3399         }
3400         /*
3401          * Okay, we need to do block allocation.
3402          */
3403
3404         /* find neighbour allocated blocks */
3405         ar.lleft = iblock;
3406         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3407         if (err)
3408                 goto out2;
3409         ar.lright = iblock;
3410         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3411         if (err)
3412                 goto out2;
3413
3414         /*
3415          * See if request is beyond maximum number of blocks we can have in
3416          * a single extent. For an initialized extent this limit is
3417          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3418          * EXT_UNINIT_MAX_LEN.
3419          */
3420         if (max_blocks > EXT_INIT_MAX_LEN &&
3421             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3422                 max_blocks = EXT_INIT_MAX_LEN;
3423         else if (max_blocks > EXT_UNINIT_MAX_LEN &&
3424                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3425                 max_blocks = EXT_UNINIT_MAX_LEN;
3426
3427         /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
3428         newex.ee_block = cpu_to_le32(iblock);
3429         newex.ee_len = cpu_to_le16(max_blocks);
3430         err = ext4_ext_check_overlap(inode, &newex, path);
3431         if (err)
3432                 allocated = ext4_ext_get_actual_len(&newex);
3433         else
3434                 allocated = max_blocks;
3435
3436         /* allocate new block */
3437         ar.inode = inode;
3438         ar.goal = ext4_ext_find_goal(inode, path, iblock);
3439         ar.logical = iblock;
3440         ar.len = allocated;
3441         if (S_ISREG(inode->i_mode))
3442                 ar.flags = EXT4_MB_HINT_DATA;
3443         else
3444                 /* disable in-core preallocation for non-regular files */
3445                 ar.flags = 0;
3446         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3447         if (!newblock)
3448                 goto out2;
3449         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3450                   ar.goal, newblock, allocated);
3451
3452         /* try to insert new extent into found leaf and return */
3453         ext4_ext_store_pblock(&newex, newblock);
3454         newex.ee_len = cpu_to_le16(ar.len);
3455         /* Mark uninitialized */
3456         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3457                 ext4_ext_mark_uninitialized(&newex);
3458                 /*
3459                  * io_end structure was created for every IO write to an
3460                  * uninitialized extent. To avoid unecessary conversion,
3461                  * here we flag the IO that really needs the conversion.
3462                  * For non asycn direct IO case, flag the inode state
3463                  * that we need to perform convertion when IO is done.
3464                  */
3465                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3466                         if (io)
3467                                 io->flag = EXT4_IO_UNWRITTEN;
3468                         else
3469                                 ext4_set_inode_state(inode,
3470                                                      EXT4_STATE_DIO_UNWRITTEN);
3471                 }
3472                 if (ext4_should_dioread_nolock(inode))
3473                         set_buffer_uninit(bh_result);
3474         }
3475
3476         if (unlikely(EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL)) {
3477                 if (unlikely(!eh->eh_entries)) {
3478                         EXT4_ERROR_INODE(inode,
3479                                          "eh->eh_entries == 0 ee_block %d",
3480                                          ex->ee_block);
3481                         err = -EIO;
3482                         goto out2;
3483                 }
3484                 last_ex = EXT_LAST_EXTENT(eh);
3485                 if (iblock + ar.len > le32_to_cpu(last_ex->ee_block)
3486                     + ext4_ext_get_actual_len(last_ex))
3487                         EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL;
3488         }
3489         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3490         if (err) {
3491                 /* free data blocks we just allocated */
3492                 /* not a good idea to call discard here directly,
3493                  * but otherwise we'd need to call it every free() */
3494                 ext4_discard_preallocations(inode);
3495                 ext4_free_blocks(handle, inode, 0, ext_pblock(&newex),
3496                                  ext4_ext_get_actual_len(&newex), 0);
3497                 goto out2;
3498         }
3499
3500         /* previous routine could use block we allocated */
3501         newblock = ext_pblock(&newex);
3502         allocated = ext4_ext_get_actual_len(&newex);
3503         if (allocated > max_blocks)
3504                 allocated = max_blocks;
3505         set_buffer_new(bh_result);
3506
3507         /*
3508          * Update reserved blocks/metadata blocks after successful
3509          * block allocation which had been deferred till now.
3510          */
3511         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3512                 ext4_da_update_reserve_space(inode, allocated, 1);
3513
3514         /*
3515          * Cache the extent and update transaction to commit on fdatasync only
3516          * when it is _not_ an uninitialized extent.
3517          */
3518         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3519                 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
3520                                                 EXT4_EXT_CACHE_EXTENT);
3521                 ext4_update_inode_fsync_trans(handle, inode, 1);
3522         } else
3523                 ext4_update_inode_fsync_trans(handle, inode, 0);
3524 out:
3525         if (allocated > max_blocks)
3526                 allocated = max_blocks;
3527         ext4_ext_show_leaf(inode, path);
3528         set_buffer_mapped(bh_result);
3529         bh_result->b_bdev = inode->i_sb->s_bdev;
3530         bh_result->b_blocknr = newblock;
3531 out2:
3532         if (path) {
3533                 ext4_ext_drop_refs(path);
3534                 kfree(path);
3535         }
3536         return err ? err : allocated;
3537 }
3538
3539 void ext4_ext_truncate(struct inode *inode)
3540 {
3541         struct address_space *mapping = inode->i_mapping;
3542         struct super_block *sb = inode->i_sb;
3543         ext4_lblk_t last_block;
3544         handle_t *handle;
3545         int err = 0;
3546
3547         /*
3548          * probably first extent we're gonna free will be last in block
3549          */
3550         err = ext4_writepage_trans_blocks(inode);
3551         handle = ext4_journal_start(inode, err);
3552         if (IS_ERR(handle))
3553                 return;
3554
3555         if (inode->i_size & (sb->s_blocksize - 1))
3556                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3557
3558         if (ext4_orphan_add(handle, inode))
3559                 goto out_stop;
3560
3561         down_write(&EXT4_I(inode)->i_data_sem);
3562         ext4_ext_invalidate_cache(inode);
3563
3564         ext4_discard_preallocations(inode);
3565
3566         /*
3567          * TODO: optimization is possible here.
3568          * Probably we need not scan at all,
3569          * because page truncation is enough.
3570          */
3571
3572         /* we have to know where to truncate from in crash case */
3573         EXT4_I(inode)->i_disksize = inode->i_size;
3574         ext4_mark_inode_dirty(handle, inode);
3575
3576         last_block = (inode->i_size + sb->s_blocksize - 1)
3577                         >> EXT4_BLOCK_SIZE_BITS(sb);
3578         err = ext4_ext_remove_space(inode, last_block);
3579
3580         /* In a multi-transaction truncate, we only make the final
3581          * transaction synchronous.
3582          */
3583         if (IS_SYNC(inode))
3584                 ext4_handle_sync(handle);
3585
3586 out_stop:
3587         up_write(&EXT4_I(inode)->i_data_sem);
3588         /*
3589          * If this was a simple ftruncate() and the file will remain alive,
3590          * then we need to clear up the orphan record which we created above.
3591          * However, if this was a real unlink then we were called by
3592          * ext4_delete_inode(), and we allow that function to clean up the
3593          * orphan info for us.
3594          */
3595         if (inode->i_nlink)
3596                 ext4_orphan_del(handle, inode);
3597
3598         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3599         ext4_mark_inode_dirty(handle, inode);
3600         ext4_journal_stop(handle);
3601 }
3602
3603 static void ext4_falloc_update_inode(struct inode *inode,
3604                                 int mode, loff_t new_size, int update_ctime)
3605 {
3606         struct timespec now;
3607
3608         if (update_ctime) {
3609                 now = current_fs_time(inode->i_sb);
3610                 if (!timespec_equal(&inode->i_ctime, &now))
3611                         inode->i_ctime = now;
3612         }
3613         /*
3614          * Update only when preallocation was requested beyond
3615          * the file size.
3616          */
3617         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3618                 if (new_size > i_size_read(inode))
3619                         i_size_write(inode, new_size);
3620                 if (new_size > EXT4_I(inode)->i_disksize)
3621                         ext4_update_i_disksize(inode, new_size);
3622         } else {
3623                 /*
3624                  * Mark that we allocate beyond EOF so the subsequent truncate
3625                  * can proceed even if the new size is the same as i_size.
3626                  */
3627                 if (new_size > i_size_read(inode))
3628                         EXT4_I(inode)->i_flags |= EXT4_EOFBLOCKS_FL;
3629         }
3630
3631 }
3632
3633 /*
3634  * preallocate space for a file. This implements ext4's fallocate inode
3635  * operation, which gets called from sys_fallocate system call.
3636  * For block-mapped files, posix_fallocate should fall back to the method
3637  * of writing zeroes to the required new blocks (the same behavior which is
3638  * expected for file systems which do not support fallocate() system call).
3639  */
3640 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3641 {
3642         handle_t *handle;
3643         ext4_lblk_t block;
3644         loff_t new_size;
3645         unsigned int max_blocks;
3646         int ret = 0;
3647         int ret2 = 0;
3648         int retries = 0;
3649         struct buffer_head map_bh;
3650         unsigned int credits, blkbits = inode->i_blkbits;
3651
3652         /*
3653          * currently supporting (pre)allocate mode for extent-based
3654          * files _only_
3655          */
3656         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3657                 return -EOPNOTSUPP;
3658
3659         /* preallocation to directories is currently not supported */
3660         if (S_ISDIR(inode->i_mode))
3661                 return -ENODEV;
3662
3663         block = offset >> blkbits;
3664         /*
3665          * We can't just convert len to max_blocks because
3666          * If blocksize = 4096 offset = 3072 and len = 2048
3667          */
3668         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3669                                                         - block;
3670         /*
3671          * credits to insert 1 extent into extent tree
3672          */
3673         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3674         mutex_lock(&inode->i_mutex);
3675 retry:
3676         while (ret >= 0 && ret < max_blocks) {
3677                 block = block + ret;
3678                 max_blocks = max_blocks - ret;
3679                 handle = ext4_journal_start(inode, credits);
3680                 if (IS_ERR(handle)) {
3681                         ret = PTR_ERR(handle);
3682                         break;
3683                 }
3684                 map_bh.b_state = 0;
3685                 ret = ext4_get_blocks(handle, inode, block,
3686                                       max_blocks, &map_bh,
3687                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3688                 if (ret <= 0) {
3689 #ifdef EXT4FS_DEBUG
3690                         WARN_ON(ret <= 0);
3691                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3692                                     "returned error inode#%lu, block=%u, "
3693                                     "max_blocks=%u", __func__,
3694                                     inode->i_ino, block, max_blocks);
3695 #endif
3696                         ext4_mark_inode_dirty(handle, inode);
3697                         ret2 = ext4_journal_stop(handle);
3698                         break;
3699                 }
3700                 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3701                                                 blkbits) >> blkbits))
3702                         new_size = offset + len;
3703                 else
3704                         new_size = (block + ret) << blkbits;
3705
3706                 ext4_falloc_update_inode(inode, mode, new_size,
3707                                                 buffer_new(&map_bh));
3708                 ext4_mark_inode_dirty(handle, inode);
3709                 ret2 = ext4_journal_stop(handle);
3710                 if (ret2)
3711                         break;
3712         }
3713         if (ret == -ENOSPC &&
3714                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3715                 ret = 0;
3716                 goto retry;
3717         }
3718         mutex_unlock(&inode->i_mutex);
3719         return ret > 0 ? ret2 : ret;
3720 }
3721
3722 /*
3723  * This function convert a range of blocks to written extents
3724  * The caller of this function will pass the start offset and the size.
3725  * all unwritten extents within this range will be converted to
3726  * written extents.
3727  *
3728  * This function is called from the direct IO end io call back
3729  * function, to convert the fallocated extents after IO is completed.
3730  * Returns 0 on success.
3731  */
3732 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3733                                     ssize_t len)
3734 {
3735         handle_t *handle;
3736         ext4_lblk_t block;
3737         unsigned int max_blocks;
3738         int ret = 0;
3739         int ret2 = 0;
3740         struct buffer_head map_bh;
3741         unsigned int credits, blkbits = inode->i_blkbits;
3742
3743         block = offset >> blkbits;
3744         /*
3745          * We can't just convert len to max_blocks because
3746          * If blocksize = 4096 offset = 3072 and len = 2048
3747          */
3748         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3749                                                         - block;
3750         /*
3751          * credits to insert 1 extent into extent tree
3752          */
3753         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3754         while (ret >= 0 && ret < max_blocks) {
3755                 block = block + ret;
3756                 max_blocks = max_blocks - ret;
3757                 handle = ext4_journal_start(inode, credits);
3758                 if (IS_ERR(handle)) {
3759                         ret = PTR_ERR(handle);
3760                         break;
3761                 }
3762                 map_bh.b_state = 0;
3763                 ret = ext4_get_blocks(handle, inode, block,
3764                                       max_blocks, &map_bh,
3765                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3766                 if (ret <= 0) {
3767                         WARN_ON(ret <= 0);
3768                         printk(KERN_ERR "%s: ext4_ext_get_blocks "
3769                                     "returned error inode#%lu, block=%u, "
3770                                     "max_blocks=%u", __func__,
3771                                     inode->i_ino, block, max_blocks);
3772                 }
3773                 ext4_mark_inode_dirty(handle, inode);
3774                 ret2 = ext4_journal_stop(handle);
3775                 if (ret <= 0 || ret2 )
3776                         break;
3777         }
3778         return ret > 0 ? ret2 : ret;
3779 }
3780 /*
3781  * Callback function called for each extent to gather FIEMAP information.
3782  */
3783 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3784                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3785                        void *data)
3786 {
3787         struct fiemap_extent_info *fieinfo = data;
3788         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3789         __u64   logical;
3790         __u64   physical;
3791         __u64   length;
3792         __u32   flags = 0;
3793         int     error;
3794
3795         logical =  (__u64)newex->ec_block << blksize_bits;
3796
3797         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3798                 pgoff_t offset;
3799                 struct page *page;
3800                 struct buffer_head *bh = NULL;
3801
3802                 offset = logical >> PAGE_SHIFT;
3803                 page = find_get_page(inode->i_mapping, offset);
3804                 if (!page || !page_has_buffers(page))
3805                         return EXT_CONTINUE;
3806
3807                 bh = page_buffers(page);
3808
3809                 if (!bh)
3810                         return EXT_CONTINUE;
3811
3812                 if (buffer_delay(bh)) {
3813                         flags |= FIEMAP_EXTENT_DELALLOC;
3814                         page_cache_release(page);
3815                 } else {
3816                         page_cache_release(page);
3817                         return EXT_CONTINUE;
3818                 }
3819         }
3820
3821         physical = (__u64)newex->ec_start << blksize_bits;
3822         length =   (__u64)newex->ec_len << blksize_bits;
3823
3824         if (ex && ext4_ext_is_uninitialized(ex))
3825                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3826
3827         /*
3828          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3829          *
3830          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3831          * this also indicates no more allocated blocks.
3832          *
3833          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3834          */
3835         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3836             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3837                 loff_t size = i_size_read(inode);
3838                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3839
3840                 flags |= FIEMAP_EXTENT_LAST;
3841                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3842                     logical+length > size)
3843                         length = (size - logical + bs - 1) & ~(bs-1);
3844         }
3845
3846         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3847                                         length, flags);
3848         if (error < 0)
3849                 return error;
3850         if (error == 1)
3851                 return EXT_BREAK;
3852
3853         return EXT_CONTINUE;
3854 }
3855
3856 /* fiemap flags we can handle specified here */
3857 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3858
3859 static int ext4_xattr_fiemap(struct inode *inode,
3860                                 struct fiemap_extent_info *fieinfo)
3861 {
3862         __u64 physical = 0;
3863         __u64 length;
3864         __u32 flags = FIEMAP_EXTENT_LAST;
3865         int blockbits = inode->i_sb->s_blocksize_bits;
3866         int error = 0;
3867
3868         /* in-inode? */
3869         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3870                 struct ext4_iloc iloc;
3871                 int offset;     /* offset of xattr in inode */
3872
3873                 error = ext4_get_inode_loc(inode, &iloc);
3874                 if (error)
3875                         return error;
3876                 physical = iloc.bh->b_blocknr << blockbits;
3877                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3878                                 EXT4_I(inode)->i_extra_isize;
3879                 physical += offset;
3880                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3881                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3882                 brelse(iloc.bh);
3883         } else { /* external block */
3884                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3885                 length = inode->i_sb->s_blocksize;
3886         }
3887
3888         if (physical)
3889                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3890                                                 length, flags);
3891         return (error < 0 ? error : 0);
3892 }
3893
3894 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3895                 __u64 start, __u64 len)
3896 {
3897         ext4_lblk_t start_blk;
3898         int error = 0;
3899
3900         /* fallback to generic here if not in extents fmt */
3901         if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3902                 return generic_block_fiemap(inode, fieinfo, start, len,
3903                         ext4_get_block);
3904
3905         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3906                 return -EBADR;
3907
3908         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3909                 error = ext4_xattr_fiemap(inode, fieinfo);
3910         } else {
3911                 ext4_lblk_t len_blks;
3912                 __u64 last_blk;
3913
3914                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3915                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3916                 if (last_blk >= EXT_MAX_BLOCK)
3917                         last_blk = EXT_MAX_BLOCK-1;
3918                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3919
3920                 /*
3921                  * Walk the extent tree gathering extent information.
3922                  * ext4_ext_fiemap_cb will push extents back to user.
3923                  */
3924                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3925                                           ext4_ext_fiemap_cb, fieinfo);
3926         }
3927
3928         return error;
3929 }
3930