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 #include <trace/events/ext4.h>
  48
  49 static int ext4_split_extent(handle_t *handle,
  50                                 struct inode *inode,
  51                                 struct ext4_ext_path *path,
  52                                 struct ext4_map_blocks *map,
  53                                 int split_flag,
  54                                 int flags);
  55
  56 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  57                                             struct inode *inode,
  58                                             int needed)
  59 {
  60         int err;
  61
  62         if (!ext4_handle_valid(handle))
  63                 return 0;
  64         if (handle->h_buffer_credits > needed)
  65                 return 0;
  66         err = ext4_journal_extend(handle, needed);
  67         if (err <= 0)
  68                 return err;
  69         err = ext4_truncate_restart_trans(handle, inode, needed);
  70         if (err == 0)
  71                 err = -EAGAIN;
  72
  73         return err;
  74 }
  75
  76 /*
  77  * could return:
  78  *  - EROFS
  79  *  - ENOMEM
  80  */
  81 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  82                                 struct ext4_ext_path *path)
  83 {
  84         if (path->p_bh) {
  85                 /* path points to block */
  86                 return ext4_journal_get_write_access(handle, path->p_bh);
  87         }
  88         /* path points to leaf/index in inode body */
  89         /* we use in-core data, no need to protect them */
  90         return 0;
  91 }
  92
  93 /*
  94  * could return:
  95  *  - EROFS
  96  *  - ENOMEM
  97  *  - EIO
  98  */
  99 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
 100                                 struct ext4_ext_path *path)
 101 {
 102         int err;
 103         if (path->p_bh) {
 104                 /* path points to block */
 105                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
 106         } else {
 107                 /* path points to leaf/index in inode body */
 108                 err = ext4_mark_inode_dirty(handle, inode);
 109         }
 110         return err;
 111 }
 112
 113 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 114                               struct ext4_ext_path *path,
 115                               ext4_lblk_t block)
 116 {
 117         struct ext4_inode_info *ei = EXT4_I(inode);
 118         ext4_fsblk_t bg_start;
 119         ext4_fsblk_t last_block;
 120         ext4_grpblk_t colour;
 121         ext4_group_t block_group;
 122         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 123         int depth;
 124
 125         if (path) {
 126                 struct ext4_extent *ex;
 127                 depth = path->p_depth;
 128
 129                 /*
 130                  * Try to predict block placement assuming that we are
 131                  * filling in a file which will eventually be
 132                  * non-sparse --- i.e., in the case of libbfd writing
 133                  * an ELF object sections out-of-order but in a way
 134                  * the eventually results in a contiguous object or
 135                  * executable file, or some database extending a table
 136                  * space file.  However, this is actually somewhat
 137                  * non-ideal if we are writing a sparse file such as
 138                  * qemu or KVM writing a raw image file that is going
 139                  * to stay fairly sparse, since it will end up
 140                  * fragmenting the file system's free space.  Maybe we
 141                  * should have some hueristics or some way to allow
 142                  * userspace to pass a hint to file system,
 143                  * especially if the latter case turns out to be
 144                  * common.
 145                  */
 146                 ex = path[depth].p_ext;
 147                 if (ex) {
 148                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 149                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 150
 151                         if (block > ext_block)
 152                                 return ext_pblk + (block - ext_block);
 153                         else
 154                                 return ext_pblk - (ext_block - block);
 155                 }
 156
 157                 /* it looks like index is empty;
 158                  * try to find starting block from index itself */
 159                 if (path[depth].p_bh)
 160                         return path[depth].p_bh->b_blocknr;
 161         }
 162
 163         /* OK. use inode's group */
 164         block_group = ei->i_block_group;
 165         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 166                 /*
 167                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 168                  * block groups per flexgroup, reserve the first block
 169                  * group for directories and special files.  Regular
 170                  * files will start at the second block group.  This
 171                  * tends to speed up directory access and improves
 172                  * fsck times.
 173                  */
 174                 block_group &= ~(flex_size-1);
 175                 if (S_ISREG(inode->i_mode))
 176                         block_group++;
 177         }
 178         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 179         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 180
 181         /*
 182          * If we are doing delayed allocation, we don't need take
 183          * colour into account.
 184          */
 185         if (test_opt(inode->i_sb, DELALLOC))
 186                 return bg_start;
 187
 188         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 189                 colour = (current->pid % 16) *
 190                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 191         else
 192                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 193         return bg_start + colour + block;
 194 }
 195
 196 /*
 197  * Allocation for a meta data block
 198  */
 199 static ext4_fsblk_t
 200 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 201                         struct ext4_ext_path *path,
 202                         struct ext4_extent *ex, int *err, unsigned int flags)
 203 {
 204         ext4_fsblk_t goal, newblock;
 205
 206         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 207         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 208                                         NULL, err);
 209         return newblock;
 210 }
 211
 212 static inline int ext4_ext_space_block(struct inode *inode, int check)
 213 {
 214         int size;
 215
 216         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 217                         / sizeof(struct ext4_extent);
 218         if (!check) {
 219 #ifdef AGGRESSIVE_TEST
 220                 if (size > 6)
 221                         size = 6;
 222 #endif
 223         }
 224         return size;
 225 }
 226
 227 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 228 {
 229         int size;
 230
 231         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 232                         / sizeof(struct ext4_extent_idx);
 233         if (!check) {
 234 #ifdef AGGRESSIVE_TEST
 235                 if (size > 5)
 236                         size = 5;
 237 #endif
 238         }
 239         return size;
 240 }
 241
 242 static inline int ext4_ext_space_root(struct inode *inode, int check)
 243 {
 244         int size;
 245
 246         size = sizeof(EXT4_I(inode)->i_data);
 247         size -= sizeof(struct ext4_extent_header);
 248         size /= sizeof(struct ext4_extent);
 249         if (!check) {
 250 #ifdef AGGRESSIVE_TEST
 251                 if (size > 3)
 252                         size = 3;
 253 #endif
 254         }
 255         return size;
 256 }
 257
 258 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 259 {
 260         int size;
 261
 262         size = sizeof(EXT4_I(inode)->i_data);
 263         size -= sizeof(struct ext4_extent_header);
 264         size /= sizeof(struct ext4_extent_idx);
 265         if (!check) {
 266 #ifdef AGGRESSIVE_TEST
 267                 if (size > 4)
 268                         size = 4;
 269 #endif
 270         }
 271         return size;
 272 }
 273
 274 /*
 275  * Calculate the number of metadata blocks needed
 276  * to allocate @blocks
 277  * Worse case is one block per extent
 278  */
 279 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 280 {
 281         struct ext4_inode_info *ei = EXT4_I(inode);
 282         int idxs, num = 0;
 283
 284         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 285                 / sizeof(struct ext4_extent_idx));
 286
 287         /*
 288          * If the new delayed allocation block is contiguous with the
 289          * previous da block, it can share index blocks with the
 290          * previous block, so we only need to allocate a new index
 291          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 292          * an additional index block, and at ldxs**3 blocks, yet
 293          * another index blocks.
 294          */
 295         if (ei->i_da_metadata_calc_len &&
 296             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 297                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 298                         num++;
 299                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 300                         num++;
 301                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 302                         num++;
 303                         ei->i_da_metadata_calc_len = 0;
 304                 } else
 305                         ei->i_da_metadata_calc_len++;
 306                 ei->i_da_metadata_calc_last_lblock++;
 307                 return num;
 308         }
 309
 310         /*
 311          * In the worst case we need a new set of index blocks at
 312          * every level of the inode's extent tree.
 313          */
 314         ei->i_da_metadata_calc_len = 1;
 315         ei->i_da_metadata_calc_last_lblock = lblock;
 316         return ext_depth(inode) + 1;
 317 }
 318
 319 static int
 320 ext4_ext_max_entries(struct inode *inode, int depth)
 321 {
 322         int max;
 323
 324         if (depth == ext_depth(inode)) {
 325                 if (depth == 0)
 326                         max = ext4_ext_space_root(inode, 1);
 327                 else
 328                         max = ext4_ext_space_root_idx(inode, 1);
 329         } else {
 330                 if (depth == 0)
 331                         max = ext4_ext_space_block(inode, 1);
 332                 else
 333                         max = ext4_ext_space_block_idx(inode, 1);
 334         }
 335
 336         return max;
 337 }
 338
 339 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 340 {
 341         ext4_fsblk_t block = ext4_ext_pblock(ext);
 342         int len = ext4_ext_get_actual_len(ext);
 343
 344         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 345 }
 346
 347 static int ext4_valid_extent_idx(struct inode *inode,
 348                                 struct ext4_extent_idx *ext_idx)
 349 {
 350         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 351
 352         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 353 }
 354
 355 static int ext4_valid_extent_entries(struct inode *inode,
 356                                 struct ext4_extent_header *eh,
 357                                 int depth)
 358 {
 359         struct ext4_extent *ext;
 360         struct ext4_extent_idx *ext_idx;
 361         unsigned short entries;
 362         if (eh->eh_entries == 0)
 363                 return 1;
 364
 365         entries = le16_to_cpu(eh->eh_entries);
 366
 367         if (depth == 0) {
 368                 /* leaf entries */
 369                 ext = EXT_FIRST_EXTENT(eh);
 370                 while (entries) {
 371                         if (!ext4_valid_extent(inode, ext))
 372                                 return 0;
 373                         ext++;
 374                         entries--;
 375                 }
 376         } else {
 377                 ext_idx = EXT_FIRST_INDEX(eh);
 378                 while (entries) {
 379                         if (!ext4_valid_extent_idx(inode, ext_idx))
 380                                 return 0;
 381                         ext_idx++;
 382                         entries--;
 383                 }
 384         }
 385         return 1;
 386 }
 387
 388 static int __ext4_ext_check(const char *function, unsigned int line,
 389                             struct inode *inode, struct ext4_extent_header *eh,
 390                             int depth)
 391 {
 392         const char *error_msg;
 393         int max = 0;
 394
 395         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 396                 error_msg = "invalid magic";
 397                 goto corrupted;
 398         }
 399         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 400                 error_msg = "unexpected eh_depth";
 401                 goto corrupted;
 402         }
 403         if (unlikely(eh->eh_max == 0)) {
 404                 error_msg = "invalid eh_max";
 405                 goto corrupted;
 406         }
 407         max = ext4_ext_max_entries(inode, depth);
 408         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 409                 error_msg = "too large eh_max";
 410                 goto corrupted;
 411         }
 412         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 413                 error_msg = "invalid eh_entries";
 414                 goto corrupted;
 415         }
 416         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 417                 error_msg = "invalid extent entries";
 418                 goto corrupted;
 419         }
 420         return 0;
 421
 422 corrupted:
 423         ext4_error_inode(inode, function, line, 0,
 424                         "bad header/extent: %s - magic %x, "
 425                         "entries %u, max %u(%u), depth %u(%u)",
 426                         error_msg, le16_to_cpu(eh->eh_magic),
 427                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 428                         max, le16_to_cpu(eh->eh_depth), depth);
 429
 430         return -EIO;
 431 }
 432
 433 #define ext4_ext_check(inode, eh, depth)        \
 434         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 435
 436 int ext4_ext_check_inode(struct inode *inode)
 437 {
 438         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 439 }
 440
 441 #ifdef EXT_DEBUG
 442 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 443 {
 444         int k, l = path->p_depth;
 445
 446         ext_debug("path:");
 447         for (k = 0; k <= l; k++, path++) {
 448                 if (path->p_idx) {
 449                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 450                             ext4_idx_pblock(path->p_idx));
 451                 } else if (path->p_ext) {
 452                         ext_debug("  %d:[%d]%d:%llu ",
 453                                   le32_to_cpu(path->p_ext->ee_block),
 454                                   ext4_ext_is_uninitialized(path->p_ext),
 455                                   ext4_ext_get_actual_len(path->p_ext),
 456                                   ext4_ext_pblock(path->p_ext));
 457                 } else
 458                         ext_debug("  []");
 459         }
 460         ext_debug("\n");
 461 }
 462
 463 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 464 {
 465         int depth = ext_depth(inode);
 466         struct ext4_extent_header *eh;
 467         struct ext4_extent *ex;
 468         int i;
 469
 470         if (!path)
 471                 return;
 472
 473         eh = path[depth].p_hdr;
 474         ex = EXT_FIRST_EXTENT(eh);
 475
 476         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 477
 478         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 479                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 480                           ext4_ext_is_uninitialized(ex),
 481                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 482         }
 483         ext_debug("\n");
 484 }
 485 #else
 486 #define ext4_ext_show_path(inode, path)
 487 #define ext4_ext_show_leaf(inode, path)
 488 #endif
 489
 490 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 491 {
 492         int depth = path->p_depth;
 493         int i;
 494
 495         for (i = 0; i <= depth; i++, path++)
 496                 if (path->p_bh) {
 497                         brelse(path->p_bh);
 498                         path->p_bh = NULL;
 499                 }
 500 }
 501
 502 /*
 503  * ext4_ext_binsearch_idx:
 504  * binary search for the closest index of the given block
 505  * the header must be checked before calling this
 506  */
 507 static void
 508 ext4_ext_binsearch_idx(struct inode *inode,
 509                         struct ext4_ext_path *path, ext4_lblk_t block)
 510 {
 511         struct ext4_extent_header *eh = path->p_hdr;
 512         struct ext4_extent_idx *r, *l, *m;
 513
 514
 515         ext_debug("binsearch for %u(idx):  ", block);
 516
 517         l = EXT_FIRST_INDEX(eh) + 1;
 518         r = EXT_LAST_INDEX(eh);
 519         while (l <= r) {
 520                 m = l + (r - l) / 2;
 521                 if (block < le32_to_cpu(m->ei_block))
 522                         r = m - 1;
 523                 else
 524                         l = m + 1;
 525                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 526                                 m, le32_to_cpu(m->ei_block),
 527                                 r, le32_to_cpu(r->ei_block));
 528         }
 529
 530         path->p_idx = l - 1;
 531         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 532                   ext4_idx_pblock(path->p_idx));
 533
 534 #ifdef CHECK_BINSEARCH
 535         {
 536                 struct ext4_extent_idx *chix, *ix;
 537                 int k;
 538
 539                 chix = ix = EXT_FIRST_INDEX(eh);
 540                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 541                   if (k != 0 &&
 542                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 543                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 544                                        "first=0x%p\n", k,
 545                                        ix, EXT_FIRST_INDEX(eh));
 546                                 printk(KERN_DEBUG "%u <= %u\n",
 547                                        le32_to_cpu(ix->ei_block),
 548                                        le32_to_cpu(ix[-1].ei_block));
 549                         }
 550                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 551                                            <= le32_to_cpu(ix[-1].ei_block));
 552                         if (block < le32_to_cpu(ix->ei_block))
 553                                 break;
 554                         chix = ix;
 555                 }
 556                 BUG_ON(chix != path->p_idx);
 557         }
 558 #endif
 559
 560 }
 561
 562 /*
 563  * ext4_ext_binsearch:
 564  * binary search for closest extent of the given block
 565  * the header must be checked before calling this
 566  */
 567 static void
 568 ext4_ext_binsearch(struct inode *inode,
 569                 struct ext4_ext_path *path, ext4_lblk_t block)
 570 {
 571         struct ext4_extent_header *eh = path->p_hdr;
 572         struct ext4_extent *r, *l, *m;
 573
 574         if (eh->eh_entries == 0) {
 575                 /*
 576                  * this leaf is empty:
 577                  * we get such a leaf in split/add case
 578                  */
 579                 return;
 580         }
 581
 582         ext_debug("binsearch for %u:  ", block);
 583
 584         l = EXT_FIRST_EXTENT(eh) + 1;
 585         r = EXT_LAST_EXTENT(eh);
 586
 587         while (l <= r) {
 588                 m = l + (r - l) / 2;
 589                 if (block < le32_to_cpu(m->ee_block))
 590                         r = m - 1;
 591                 else
 592                         l = m + 1;
 593                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 594                                 m, le32_to_cpu(m->ee_block),
 595                                 r, le32_to_cpu(r->ee_block));
 596         }
 597
 598         path->p_ext = l - 1;
 599         ext_debug("  -> %d:%llu:[%d]%d ",
 600                         le32_to_cpu(path->p_ext->ee_block),
 601                         ext4_ext_pblock(path->p_ext),
 602                         ext4_ext_is_uninitialized(path->p_ext),
 603                         ext4_ext_get_actual_len(path->p_ext));
 604
 605 #ifdef CHECK_BINSEARCH
 606         {
 607                 struct ext4_extent *chex, *ex;
 608                 int k;
 609
 610                 chex = ex = EXT_FIRST_EXTENT(eh);
 611                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 612                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 613                                           <= le32_to_cpu(ex[-1].ee_block));
 614                         if (block < le32_to_cpu(ex->ee_block))
 615                                 break;
 616                         chex = ex;
 617                 }
 618                 BUG_ON(chex != path->p_ext);
 619         }
 620 #endif
 621
 622 }
 623
 624 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 625 {
 626         struct ext4_extent_header *eh;
 627
 628         eh = ext_inode_hdr(inode);
 629         eh->eh_depth = 0;
 630         eh->eh_entries = 0;
 631         eh->eh_magic = EXT4_EXT_MAGIC;
 632         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 633         ext4_mark_inode_dirty(handle, inode);
 634         ext4_ext_invalidate_cache(inode);
 635         return 0;
 636 }
 637
 638 struct ext4_ext_path *
 639 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 640                                         struct ext4_ext_path *path)
 641 {
 642         struct ext4_extent_header *eh;
 643         struct buffer_head *bh;
 644         short int depth, i, ppos = 0, alloc = 0;
 645
 646         eh = ext_inode_hdr(inode);
 647         depth = ext_depth(inode);
 648
 649         /* account possible depth increase */
 650         if (!path) {
 651                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 652                                 GFP_NOFS);
 653                 if (!path)
 654                         return ERR_PTR(-ENOMEM);
 655                 alloc = 1;
 656         }
 657         path[0].p_hdr = eh;
 658         path[0].p_bh = NULL;
 659
 660         i = depth;
 661         /* walk through the tree */
 662         while (i) {
 663                 int need_to_validate = 0;
 664
 665                 ext_debug("depth %d: num %d, max %d\n",
 666                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 667
 668                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 669                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 670                 path[ppos].p_depth = i;
 671                 path[ppos].p_ext = NULL;
 672
 673                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 674                 if (unlikely(!bh))
 675                         goto err;
 676                 if (!bh_uptodate_or_lock(bh)) {
 677                         trace_ext4_ext_load_extent(inode, block,
 678                                                 path[ppos].p_block);
 679                         if (bh_submit_read(bh) < 0) {
 680                                 put_bh(bh);
 681                                 goto err;
 682                         }
 683                         /* validate the extent entries */
 684                         need_to_validate = 1;
 685                 }
 686                 eh = ext_block_hdr(bh);
 687                 ppos++;
 688                 if (unlikely(ppos > depth)) {
 689                         put_bh(bh);
 690                         EXT4_ERROR_INODE(inode,
 691                                          "ppos %d > depth %d", ppos, depth);
 692                         goto err;
 693                 }
 694                 path[ppos].p_bh = bh;
 695                 path[ppos].p_hdr = eh;
 696                 i--;
 697
 698                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 699                         goto err;
 700         }
 701
 702         path[ppos].p_depth = i;
 703         path[ppos].p_ext = NULL;
 704         path[ppos].p_idx = NULL;
 705
 706         /* find extent */
 707         ext4_ext_binsearch(inode, path + ppos, block);
 708         /* if not an empty leaf */
 709         if (path[ppos].p_ext)
 710                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 711
 712         ext4_ext_show_path(inode, path);
 713
 714         return path;
 715
 716 err:
 717         ext4_ext_drop_refs(path);
 718         if (alloc)
 719                 kfree(path);
 720         return ERR_PTR(-EIO);
 721 }
 722
 723 /*
 724  * ext4_ext_insert_index:
 725  * insert new index [@logical;@ptr] into the block at @curp;
 726  * check where to insert: before @curp or after @curp
 727  */
 728 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 729                                  struct ext4_ext_path *curp,
 730                                  int logical, ext4_fsblk_t ptr)
 731 {
 732         struct ext4_extent_idx *ix;
 733         int len, err;
 734
 735         err = ext4_ext_get_access(handle, inode, curp);
 736         if (err)
 737                 return err;
 738
 739         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 740                 EXT4_ERROR_INODE(inode,
 741                                  "logical %d == ei_block %d!",
 742                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 743                 return -EIO;
 744         }
 745         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 746         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 747                 /* insert after */
 748                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 749                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 750                         len = len < 0 ? 0 : len;
 751                         ext_debug("insert new index %d after: %llu. "
 752                                         "move %d from 0x%p to 0x%p\n",
 753                                         logical, ptr, len,
 754                                         (curp->p_idx + 1), (curp->p_idx + 2));
 755                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 756                 }
 757                 ix = curp->p_idx + 1;
 758         } else {
 759                 /* insert before */
 760                 len = len * sizeof(struct ext4_extent_idx);
 761                 len = len < 0 ? 0 : len;
 762                 ext_debug("insert new index %d before: %llu. "
 763                                 "move %d from 0x%p to 0x%p\n",
 764                                 logical, ptr, len,
 765                                 curp->p_idx, (curp->p_idx + 1));
 766                 memmove(curp->p_idx + 1, curp->p_idx, len);
 767                 ix = curp->p_idx;
 768         }
 769
 770         ix->ei_block = cpu_to_le32(logical);
 771         ext4_idx_store_pblock(ix, ptr);
 772         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 773
 774         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 775                              > le16_to_cpu(curp->p_hdr->eh_max))) {
 776                 EXT4_ERROR_INODE(inode,
 777                                  "logical %d == ei_block %d!",
 778                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 779                 return -EIO;
 780         }
 781         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 782                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 783                 return -EIO;
 784         }
 785
 786         err = ext4_ext_dirty(handle, inode, curp);
 787         ext4_std_error(inode->i_sb, err);
 788
 789         return err;
 790 }
 791
 792 /*
 793  * ext4_ext_split:
 794  * inserts new subtree into the path, using free index entry
 795  * at depth @at:
 796  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 797  * - makes decision where to split
 798  * - moves remaining extents and index entries (right to the split point)
 799  *   into the newly allocated blocks
 800  * - initializes subtree
 801  */
 802 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 803                           unsigned int flags,
 804                           struct ext4_ext_path *path,
 805                           struct ext4_extent *newext, int at)
 806 {
 807         struct buffer_head *bh = NULL;
 808         int depth = ext_depth(inode);
 809         struct ext4_extent_header *neh;
 810         struct ext4_extent_idx *fidx;
 811         struct ext4_extent *ex;
 812         int i = at, k, m, a;
 813         ext4_fsblk_t newblock, oldblock;
 814         __le32 border;
 815         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 816         int err = 0;
 817
 818         /* make decision: where to split? */
 819         /* FIXME: now decision is simplest: at current extent */
 820
 821         /* if current leaf will be split, then we should use
 822          * border from split point */
 823         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 824                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 825                 return -EIO;
 826         }
 827         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 828                 border = path[depth].p_ext[1].ee_block;
 829                 ext_debug("leaf will be split."
 830                                 " next leaf starts at %d\n",
 831                                   le32_to_cpu(border));
 832         } else {
 833                 border = newext->ee_block;
 834                 ext_debug("leaf will be added."
 835                                 " next leaf starts at %d\n",
 836                                 le32_to_cpu(border));
 837         }
 838
 839         /*
 840          * If error occurs, then we break processing
 841          * and mark filesystem read-only. index won't
 842          * be inserted and tree will be in consistent
 843          * state. Next mount will repair buffers too.
 844          */
 845
 846         /*
 847          * Get array to track all allocated blocks.
 848          * We need this to handle errors and free blocks
 849          * upon them.
 850          */
 851         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 852         if (!ablocks)
 853                 return -ENOMEM;
 854
 855         /* allocate all needed blocks */
 856         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 857         for (a = 0; a < depth - at; a++) {
 858                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 859                                                    newext, &err, flags);
 860                 if (newblock == 0)
 861                         goto cleanup;
 862                 ablocks[a] = newblock;
 863         }
 864
 865         /* initialize new leaf */
 866         newblock = ablocks[--a];
 867         if (unlikely(newblock == 0)) {
 868                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 869                 err = -EIO;
 870                 goto cleanup;
 871         }
 872         bh = sb_getblk(inode->i_sb, newblock);
 873         if (!bh) {
 874                 err = -EIO;
 875                 goto cleanup;
 876         }
 877         lock_buffer(bh);
 878
 879         err = ext4_journal_get_create_access(handle, bh);
 880         if (err)
 881                 goto cleanup;
 882
 883         neh = ext_block_hdr(bh);
 884         neh->eh_entries = 0;
 885         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 886         neh->eh_magic = EXT4_EXT_MAGIC;
 887         neh->eh_depth = 0;
 888         ex = EXT_FIRST_EXTENT(neh);
 889
 890         /* move remainder of path[depth] to the new leaf */
 891         if (unlikely(path[depth].p_hdr->eh_entries !=
 892                      path[depth].p_hdr->eh_max)) {
 893                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 894                                  path[depth].p_hdr->eh_entries,
 895                                  path[depth].p_hdr->eh_max);
 896                 err = -EIO;
 897                 goto cleanup;
 898         }
 899         /* start copy from next extent */
 900         /* TODO: we could do it by single memmove */
 901         m = 0;
 902         path[depth].p_ext++;
 903         while (path[depth].p_ext <=
 904                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
 905                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 906                                 le32_to_cpu(path[depth].p_ext->ee_block),
 907                                 ext4_ext_pblock(path[depth].p_ext),
 908                                 ext4_ext_is_uninitialized(path[depth].p_ext),
 909                                 ext4_ext_get_actual_len(path[depth].p_ext),
 910                                 newblock);
 911                 /*memmove(ex++, path[depth].p_ext++,
 912                                 sizeof(struct ext4_extent));
 913                 neh->eh_entries++;*/
 914                 path[depth].p_ext++;
 915                 m++;
 916         }
 917         if (m) {
 918                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
 919                 le16_add_cpu(&neh->eh_entries, m);
 920         }
 921
 922         set_buffer_uptodate(bh);
 923         unlock_buffer(bh);
 924
 925         err = ext4_handle_dirty_metadata(handle, inode, bh);
 926         if (err)
 927                 goto cleanup;
 928         brelse(bh);
 929         bh = NULL;
 930
 931         /* correct old leaf */
 932         if (m) {
 933                 err = ext4_ext_get_access(handle, inode, path + depth);
 934                 if (err)
 935                         goto cleanup;
 936                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 937                 err = ext4_ext_dirty(handle, inode, path + depth);
 938                 if (err)
 939                         goto cleanup;
 940
 941         }
 942
 943         /* create intermediate indexes */
 944         k = depth - at - 1;
 945         if (unlikely(k < 0)) {
 946                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 947                 err = -EIO;
 948                 goto cleanup;
 949         }
 950         if (k)
 951                 ext_debug("create %d intermediate indices\n", k);
 952         /* insert new index into current index block */
 953         /* current depth stored in i var */
 954         i = depth - 1;
 955         while (k--) {
 956                 oldblock = newblock;
 957                 newblock = ablocks[--a];
 958                 bh = sb_getblk(inode->i_sb, newblock);
 959                 if (!bh) {
 960                         err = -EIO;
 961                         goto cleanup;
 962                 }
 963                 lock_buffer(bh);
 964
 965                 err = ext4_journal_get_create_access(handle, bh);
 966                 if (err)
 967                         goto cleanup;
 968
 969                 neh = ext_block_hdr(bh);
 970                 neh->eh_entries = cpu_to_le16(1);
 971                 neh->eh_magic = EXT4_EXT_MAGIC;
 972                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 973                 neh->eh_depth = cpu_to_le16(depth - i);
 974                 fidx = EXT_FIRST_INDEX(neh);
 975                 fidx->ei_block = border;
 976                 ext4_idx_store_pblock(fidx, oldblock);
 977
 978                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 979                                 i, newblock, le32_to_cpu(border), oldblock);
 980                 /* copy indexes */
 981                 m = 0;
 982                 path[i].p_idx++;
 983
 984                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 985                                 EXT_MAX_INDEX(path[i].p_hdr));
 986                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 987                                         EXT_LAST_INDEX(path[i].p_hdr))) {
 988                         EXT4_ERROR_INODE(inode,
 989                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 990                                          le32_to_cpu(path[i].p_ext->ee_block));
 991                         err = -EIO;
 992                         goto cleanup;
 993                 }
 994                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
 995                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
 996                                         le32_to_cpu(path[i].p_idx->ei_block),
 997                                         ext4_idx_pblock(path[i].p_idx),
 998                                         newblock);
 999                         /*memmove(++fidx, path[i].p_idx++,
1000                                         sizeof(struct ext4_extent_idx));
1001                         neh->eh_entries++;
1002                         BUG_ON(neh->eh_entries > neh->eh_max);*/
1003                         path[i].p_idx++;
1004                         m++;
1005                 }
1006                 if (m) {
1007                         memmove(++fidx, path[i].p_idx - m,
1008                                 sizeof(struct ext4_extent_idx) * m);
1009                         le16_add_cpu(&neh->eh_entries, m);
1010                 }
1011                 set_buffer_uptodate(bh);
1012                 unlock_buffer(bh);
1013
1014                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1015                 if (err)
1016                         goto cleanup;
1017                 brelse(bh);
1018                 bh = NULL;
1019
1020                 /* correct old index */
1021                 if (m) {
1022                         err = ext4_ext_get_access(handle, inode, path + i);
1023                         if (err)
1024                                 goto cleanup;
1025                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1026                         err = ext4_ext_dirty(handle, inode, path + i);
1027                         if (err)
1028                                 goto cleanup;
1029                 }
1030
1031                 i--;
1032         }
1033
1034         /* insert new index */
1035         err = ext4_ext_insert_index(handle, inode, path + at,
1036                                     le32_to_cpu(border), newblock);
1037
1038 cleanup:
1039         if (bh) {
1040                 if (buffer_locked(bh))
1041                         unlock_buffer(bh);
1042                 brelse(bh);
1043         }
1044
1045         if (err) {
1046                 /* free all allocated blocks in error case */
1047                 for (i = 0; i < depth; i++) {
1048                         if (!ablocks[i])
1049                                 continue;
1050                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1051                                          EXT4_FREE_BLOCKS_METADATA);
1052                 }
1053         }
1054         kfree(ablocks);
1055
1056         return err;
1057 }
1058
1059 /*
1060  * ext4_ext_grow_indepth:
1061  * implements tree growing procedure:
1062  * - allocates new block
1063  * - moves top-level data (index block or leaf) into the new block
1064  * - initializes new top-level, creating index that points to the
1065  *   just created block
1066  */
1067 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1068                                  unsigned int flags,
1069                                  struct ext4_ext_path *path,
1070                                  struct ext4_extent *newext)
1071 {
1072         struct ext4_ext_path *curp = path;
1073         struct ext4_extent_header *neh;
1074         struct buffer_head *bh;
1075         ext4_fsblk_t newblock;
1076         int err = 0;
1077
1078         newblock = ext4_ext_new_meta_block(handle, inode, path,
1079                 newext, &err, flags);
1080         if (newblock == 0)
1081                 return err;
1082
1083         bh = sb_getblk(inode->i_sb, newblock);
1084         if (!bh) {
1085                 err = -EIO;
1086                 ext4_std_error(inode->i_sb, err);
1087                 return err;
1088         }
1089         lock_buffer(bh);
1090
1091         err = ext4_journal_get_create_access(handle, bh);
1092         if (err) {
1093                 unlock_buffer(bh);
1094                 goto out;
1095         }
1096
1097         /* move top-level index/leaf into new block */
1098         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1099
1100         /* set size of new block */
1101         neh = ext_block_hdr(bh);
1102         /* old root could have indexes or leaves
1103          * so calculate e_max right way */
1104         if (ext_depth(inode))
1105                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1106         else
1107                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1108         neh->eh_magic = EXT4_EXT_MAGIC;
1109         set_buffer_uptodate(bh);
1110         unlock_buffer(bh);
1111
1112         err = ext4_handle_dirty_metadata(handle, inode, bh);
1113         if (err)
1114                 goto out;
1115
1116         /* create index in new top-level index: num,max,pointer */
1117         err = ext4_ext_get_access(handle, inode, curp);
1118         if (err)
1119                 goto out;
1120
1121         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1122         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1123         curp->p_hdr->eh_entries = cpu_to_le16(1);
1124         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1125
1126         if (path[0].p_hdr->eh_depth)
1127                 curp->p_idx->ei_block =
1128                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1129         else
1130                 curp->p_idx->ei_block =
1131                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1132         ext4_idx_store_pblock(curp->p_idx, newblock);
1133
1134         neh = ext_inode_hdr(inode);
1135         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1136                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1137                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1138                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1139
1140         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1141         err = ext4_ext_dirty(handle, inode, curp);
1142 out:
1143         brelse(bh);
1144
1145         return err;
1146 }
1147
1148 /*
1149  * ext4_ext_create_new_leaf:
1150  * finds empty index and adds new leaf.
1151  * if no free index is found, then it requests in-depth growing.
1152  */
1153 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1154                                     unsigned int flags,
1155                                     struct ext4_ext_path *path,
1156                                     struct ext4_extent *newext)
1157 {
1158         struct ext4_ext_path *curp;
1159         int depth, i, err = 0;
1160
1161 repeat:
1162         i = depth = ext_depth(inode);
1163
1164         /* walk up to the tree and look for free index entry */
1165         curp = path + depth;
1166         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1167                 i--;
1168                 curp--;
1169         }
1170
1171         /* we use already allocated block for index block,
1172          * so subsequent data blocks should be contiguous */
1173         if (EXT_HAS_FREE_INDEX(curp)) {
1174                 /* if we found index with free entry, then use that
1175                  * entry: create all needed subtree and add new leaf */
1176                 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1177                 if (err)
1178                         goto out;
1179
1180                 /* refill path */
1181                 ext4_ext_drop_refs(path);
1182                 path = ext4_ext_find_extent(inode,
1183                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1184                                     path);
1185                 if (IS_ERR(path))
1186                         err = PTR_ERR(path);
1187         } else {
1188                 /* tree is full, time to grow in depth */
1189                 err = ext4_ext_grow_indepth(handle, inode, flags,
1190                                             path, newext);
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                         goto out;
1202                 }
1203
1204                 /*
1205                  * only first (depth 0 -> 1) produces free space;
1206                  * in all other cases we have to split the grown tree
1207                  */
1208                 depth = ext_depth(inode);
1209                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1210                         /* now we need to split */
1211                         goto repeat;
1212                 }
1213         }
1214
1215 out:
1216         return err;
1217 }
1218
1219 /*
1220  * search the closest allocated block to the left for *logical
1221  * and returns it at @logical + it's physical address at @phys
1222  * if *logical is the smallest allocated block, the function
1223  * returns 0 at @phys
1224  * return value contains 0 (success) or error code
1225  */
1226 static int ext4_ext_search_left(struct inode *inode,
1227                                 struct ext4_ext_path *path,
1228                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1229 {
1230         struct ext4_extent_idx *ix;
1231         struct ext4_extent *ex;
1232         int depth, ee_len;
1233
1234         if (unlikely(path == NULL)) {
1235                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1236                 return -EIO;
1237         }
1238         depth = path->p_depth;
1239         *phys = 0;
1240
1241         if (depth == 0 && path->p_ext == NULL)
1242                 return 0;
1243
1244         /* usually extent in the path covers blocks smaller
1245          * then *logical, but it can be that extent is the
1246          * first one in the file */
1247
1248         ex = path[depth].p_ext;
1249         ee_len = ext4_ext_get_actual_len(ex);
1250         if (*logical < le32_to_cpu(ex->ee_block)) {
1251                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1252                         EXT4_ERROR_INODE(inode,
1253                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1254                                          *logical, le32_to_cpu(ex->ee_block));
1255                         return -EIO;
1256                 }
1257                 while (--depth >= 0) {
1258                         ix = path[depth].p_idx;
1259                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1260                                 EXT4_ERROR_INODE(inode,
1261                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1262                                   ix != NULL ? ix->ei_block : 0,
1263                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1264                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1265                                   depth);
1266                                 return -EIO;
1267                         }
1268                 }
1269                 return 0;
1270         }
1271
1272         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1273                 EXT4_ERROR_INODE(inode,
1274                                  "logical %d < ee_block %d + ee_len %d!",
1275                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1276                 return -EIO;
1277         }
1278
1279         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1280         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1281         return 0;
1282 }
1283
1284 /*
1285  * search the closest allocated block to the right for *logical
1286  * and returns it at @logical + it's physical address at @phys
1287  * if *logical is the smallest allocated block, the function
1288  * returns 0 at @phys
1289  * return value contains 0 (success) or error code
1290  */
1291 static int ext4_ext_search_right(struct inode *inode,
1292                                  struct ext4_ext_path *path,
1293                                  ext4_lblk_t *logical, ext4_fsblk_t *phys)
1294 {
1295         struct buffer_head *bh = NULL;
1296         struct ext4_extent_header *eh;
1297         struct ext4_extent_idx *ix;
1298         struct ext4_extent *ex;
1299         ext4_fsblk_t block;
1300         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1301         int ee_len;
1302
1303         if (unlikely(path == NULL)) {
1304                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1305                 return -EIO;
1306         }
1307         depth = path->p_depth;
1308         *phys = 0;
1309
1310         if (depth == 0 && path->p_ext == NULL)
1311                 return 0;
1312
1313         /* usually extent in the path covers blocks smaller
1314          * then *logical, but it can be that extent is the
1315          * first one in the file */
1316
1317         ex = path[depth].p_ext;
1318         ee_len = ext4_ext_get_actual_len(ex);
1319         if (*logical < le32_to_cpu(ex->ee_block)) {
1320                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1321                         EXT4_ERROR_INODE(inode,
1322                                          "first_extent(path[%d].p_hdr) != ex",
1323                                          depth);
1324                         return -EIO;
1325                 }
1326                 while (--depth >= 0) {
1327                         ix = path[depth].p_idx;
1328                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1329                                 EXT4_ERROR_INODE(inode,
1330                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1331                                                  *logical);
1332                                 return -EIO;
1333                         }
1334                 }
1335                 *logical = le32_to_cpu(ex->ee_block);
1336                 *phys = ext4_ext_pblock(ex);
1337                 return 0;
1338         }
1339
1340         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1341                 EXT4_ERROR_INODE(inode,
1342                                  "logical %d < ee_block %d + ee_len %d!",
1343                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1344                 return -EIO;
1345         }
1346
1347         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1348                 /* next allocated block in this leaf */
1349                 ex++;
1350                 *logical = le32_to_cpu(ex->ee_block);
1351                 *phys = ext4_ext_pblock(ex);
1352                 return 0;
1353         }
1354
1355         /* go up and search for index to the right */
1356         while (--depth >= 0) {
1357                 ix = path[depth].p_idx;
1358                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1359                         goto got_index;
1360         }
1361
1362         /* we've gone up to the root and found no index to the right */
1363         return 0;
1364
1365 got_index:
1366         /* we've found index to the right, let's
1367          * follow it and find the closest allocated
1368          * block to the right */
1369         ix++;
1370         block = ext4_idx_pblock(ix);
1371         while (++depth < path->p_depth) {
1372                 bh = sb_bread(inode->i_sb, block);
1373                 if (bh == NULL)
1374                         return -EIO;
1375                 eh = ext_block_hdr(bh);
1376                 /* subtract from p_depth to get proper eh_depth */
1377                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1378                         put_bh(bh);
1379                         return -EIO;
1380                 }
1381                 ix = EXT_FIRST_INDEX(eh);
1382                 block = ext4_idx_pblock(ix);
1383                 put_bh(bh);
1384         }
1385
1386         bh = sb_bread(inode->i_sb, block);
1387         if (bh == NULL)
1388                 return -EIO;
1389         eh = ext_block_hdr(bh);
1390         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1391                 put_bh(bh);
1392                 return -EIO;
1393         }
1394         ex = EXT_FIRST_EXTENT(eh);
1395         *logical = le32_to_cpu(ex->ee_block);
1396         *phys = ext4_ext_pblock(ex);
1397         put_bh(bh);
1398         return 0;
1399 }
1400
1401 /*
1402  * ext4_ext_next_allocated_block:
1403  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1404  * NOTE: it considers block number from index entry as
1405  * allocated block. Thus, index entries have to be consistent
1406  * with leaves.
1407  */
1408 static ext4_lblk_t
1409 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1410 {
1411         int depth;
1412
1413         BUG_ON(path == NULL);
1414         depth = path->p_depth;
1415
1416         if (depth == 0 && path->p_ext == NULL)
1417                 return EXT_MAX_BLOCK;
1418
1419         while (depth >= 0) {
1420                 if (depth == path->p_depth) {
1421                         /* leaf */
1422                         if (path[depth].p_ext !=
1423                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1424                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1425                 } else {
1426                         /* index */
1427                         if (path[depth].p_idx !=
1428                                         EXT_LAST_INDEX(path[depth].p_hdr))
1429                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1430                 }
1431                 depth--;
1432         }
1433
1434         return EXT_MAX_BLOCK;
1435 }
1436
1437 /*
1438  * ext4_ext_next_leaf_block:
1439  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1440  */
1441 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1442                                         struct ext4_ext_path *path)
1443 {
1444         int depth;
1445
1446         BUG_ON(path == NULL);
1447         depth = path->p_depth;
1448
1449         /* zero-tree has no leaf blocks at all */
1450         if (depth == 0)
1451                 return EXT_MAX_BLOCK;
1452
1453         /* go to index block */
1454         depth--;
1455
1456         while (depth >= 0) {
1457                 if (path[depth].p_idx !=
1458                                 EXT_LAST_INDEX(path[depth].p_hdr))
1459                         return (ext4_lblk_t)
1460                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1461                 depth--;
1462         }
1463
1464         return EXT_MAX_BLOCK;
1465 }
1466
1467 /*
1468  * ext4_ext_correct_indexes:
1469  * if leaf gets modified and modified extent is first in the leaf,
1470  * then we have to correct all indexes above.
1471  * TODO: do we need to correct tree in all cases?
1472  */
1473 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1474                                 struct ext4_ext_path *path)
1475 {
1476         struct ext4_extent_header *eh;
1477         int depth = ext_depth(inode);
1478         struct ext4_extent *ex;
1479         __le32 border;
1480         int k, err = 0;
1481
1482         eh = path[depth].p_hdr;
1483         ex = path[depth].p_ext;
1484
1485         if (unlikely(ex == NULL || eh == NULL)) {
1486                 EXT4_ERROR_INODE(inode,
1487                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1488                 return -EIO;
1489         }
1490
1491         if (depth == 0) {
1492                 /* there is no tree at all */
1493                 return 0;
1494         }
1495
1496         if (ex != EXT_FIRST_EXTENT(eh)) {
1497                 /* we correct tree if first leaf got modified only */
1498                 return 0;
1499         }
1500
1501         /*
1502          * TODO: we need correction if border is smaller than current one
1503          */
1504         k = depth - 1;
1505         border = path[depth].p_ext->ee_block;
1506         err = ext4_ext_get_access(handle, inode, path + k);
1507         if (err)
1508                 return err;
1509         path[k].p_idx->ei_block = border;
1510         err = ext4_ext_dirty(handle, inode, path + k);
1511         if (err)
1512                 return err;
1513
1514         while (k--) {
1515                 /* change all left-side indexes */
1516                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1517                         break;
1518                 err = ext4_ext_get_access(handle, inode, path + k);
1519                 if (err)
1520                         break;
1521                 path[k].p_idx->ei_block = border;
1522                 err = ext4_ext_dirty(handle, inode, path + k);
1523                 if (err)
1524                         break;
1525         }
1526
1527         return err;
1528 }
1529
1530 int
1531 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1532                                 struct ext4_extent *ex2)
1533 {
1534         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1535
1536         /*
1537          * Make sure that either both extents are uninitialized, or
1538          * both are _not_.
1539          */
1540         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1541                 return 0;
1542
1543         if (ext4_ext_is_uninitialized(ex1))
1544                 max_len = EXT_UNINIT_MAX_LEN;
1545         else
1546                 max_len = EXT_INIT_MAX_LEN;
1547
1548         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1549         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1550
1551         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1552                         le32_to_cpu(ex2->ee_block))
1553                 return 0;
1554
1555         /*
1556          * To allow future support for preallocated extents to be added
1557          * as an RO_COMPAT feature, refuse to merge to extents if
1558          * this can result in the top bit of ee_len being set.
1559          */
1560         if (ext1_ee_len + ext2_ee_len > max_len)
1561                 return 0;
1562 #ifdef AGGRESSIVE_TEST
1563         if (ext1_ee_len >= 4)
1564                 return 0;
1565 #endif
1566
1567         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1568                 return 1;
1569         return 0;
1570 }
1571
1572 /*
1573  * This function tries to merge the "ex" extent to the next extent in the tree.
1574  * It always tries to merge towards right. If you want to merge towards
1575  * left, pass "ex - 1" as argument instead of "ex".
1576  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1577  * 1 if they got merged.
1578  */
1579 static int ext4_ext_try_to_merge_right(struct inode *inode,
1580                                  struct ext4_ext_path *path,
1581                                  struct ext4_extent *ex)
1582 {
1583         struct ext4_extent_header *eh;
1584         unsigned int depth, len;
1585         int merge_done = 0;
1586         int uninitialized = 0;
1587
1588         depth = ext_depth(inode);
1589         BUG_ON(path[depth].p_hdr == NULL);
1590         eh = path[depth].p_hdr;
1591
1592         while (ex < EXT_LAST_EXTENT(eh)) {
1593                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1594                         break;
1595                 /* merge with next extent! */
1596                 if (ext4_ext_is_uninitialized(ex))
1597                         uninitialized = 1;
1598                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1599                                 + ext4_ext_get_actual_len(ex + 1));
1600                 if (uninitialized)
1601                         ext4_ext_mark_uninitialized(ex);
1602
1603                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1604                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1605                                 * sizeof(struct ext4_extent);
1606                         memmove(ex + 1, ex + 2, len);
1607                 }
1608                 le16_add_cpu(&eh->eh_entries, -1);
1609                 merge_done = 1;
1610                 WARN_ON(eh->eh_entries == 0);
1611                 if (!eh->eh_entries)
1612                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1613         }
1614
1615         return merge_done;
1616 }
1617
1618 /*
1619  * This function tries to merge the @ex extent to neighbours in the tree.
1620  * return 1 if merge left else 0.
1621  */
1622 static int ext4_ext_try_to_merge(struct inode *inode,
1623                                   struct ext4_ext_path *path,
1624                                   struct ext4_extent *ex) {
1625         struct ext4_extent_header *eh;
1626         unsigned int depth;
1627         int merge_done = 0;
1628         int ret = 0;
1629
1630         depth = ext_depth(inode);
1631         BUG_ON(path[depth].p_hdr == NULL);
1632         eh = path[depth].p_hdr;
1633
1634         if (ex > EXT_FIRST_EXTENT(eh))
1635                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1636
1637         if (!merge_done)
1638                 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1639
1640         return ret;
1641 }
1642
1643 /*
1644  * check if a portion of the "newext" extent overlaps with an
1645  * existing extent.
1646  *
1647  * If there is an overlap discovered, it updates the length of the newext
1648  * such that there will be no overlap, and then returns 1.
1649  * If there is no overlap found, it returns 0.
1650  */
1651 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1652                                            struct ext4_extent *newext,
1653                                            struct ext4_ext_path *path)
1654 {
1655         ext4_lblk_t b1, b2;
1656         unsigned int depth, len1;
1657         unsigned int ret = 0;
1658
1659         b1 = le32_to_cpu(newext->ee_block);
1660         len1 = ext4_ext_get_actual_len(newext);
1661         depth = ext_depth(inode);
1662         if (!path[depth].p_ext)
1663                 goto out;
1664         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1665
1666         /*
1667          * get the next allocated block if the extent in the path
1668          * is before the requested block(s)
1669          */
1670         if (b2 < b1) {
1671                 b2 = ext4_ext_next_allocated_block(path);
1672                 if (b2 == EXT_MAX_BLOCK)
1673                         goto out;
1674         }
1675
1676         /* check for wrap through zero on extent logical start block*/
1677         if (b1 + len1 < b1) {
1678                 len1 = EXT_MAX_BLOCK - b1;
1679                 newext->ee_len = cpu_to_le16(len1);
1680                 ret = 1;
1681         }
1682
1683         /* check for overlap */
1684         if (b1 + len1 > b2) {
1685                 newext->ee_len = cpu_to_le16(b2 - b1);
1686                 ret = 1;
1687         }
1688 out:
1689         return ret;
1690 }
1691
1692 /*
1693  * ext4_ext_insert_extent:
1694  * tries to merge requsted extent into the existing extent or
1695  * inserts requested extent as new one into the tree,
1696  * creating new leaf in the no-space case.
1697  */
1698 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1699                                 struct ext4_ext_path *path,
1700                                 struct ext4_extent *newext, int flag)
1701 {
1702         struct ext4_extent_header *eh;
1703         struct ext4_extent *ex, *fex;
1704         struct ext4_extent *nearex; /* nearest extent */
1705         struct ext4_ext_path *npath = NULL;
1706         int depth, len, err;
1707         ext4_lblk_t next;
1708         unsigned uninitialized = 0;
1709         int flags = 0;
1710
1711         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1712                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1713                 return -EIO;
1714         }
1715         depth = ext_depth(inode);
1716         ex = path[depth].p_ext;
1717         if (unlikely(path[depth].p_hdr == NULL)) {
1718                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1719                 return -EIO;
1720         }
1721
1722         /* try to insert block into found extent and return */
1723         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1724                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1725                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1726                           ext4_ext_is_uninitialized(newext),
1727                           ext4_ext_get_actual_len(newext),
1728                           le32_to_cpu(ex->ee_block),
1729                           ext4_ext_is_uninitialized(ex),
1730                           ext4_ext_get_actual_len(ex),
1731                           ext4_ext_pblock(ex));
1732                 err = ext4_ext_get_access(handle, inode, path + depth);
1733                 if (err)
1734                         return err;
1735
1736                 /*
1737                  * ext4_can_extents_be_merged should have checked that either
1738                  * both extents are uninitialized, or both aren't. Thus we
1739                  * need to check only one of them here.
1740                  */
1741                 if (ext4_ext_is_uninitialized(ex))
1742                         uninitialized = 1;
1743                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1744                                         + ext4_ext_get_actual_len(newext));
1745                 if (uninitialized)
1746                         ext4_ext_mark_uninitialized(ex);
1747                 eh = path[depth].p_hdr;
1748                 nearex = ex;
1749                 goto merge;
1750         }
1751
1752 repeat:
1753         depth = ext_depth(inode);
1754         eh = path[depth].p_hdr;
1755         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1756                 goto has_space;
1757
1758         /* probably next leaf has space for us? */
1759         fex = EXT_LAST_EXTENT(eh);
1760         next = ext4_ext_next_leaf_block(inode, path);
1761         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1762             && next != EXT_MAX_BLOCK) {
1763                 ext_debug("next leaf block - %d\n", next);
1764                 BUG_ON(npath != NULL);
1765                 npath = ext4_ext_find_extent(inode, next, NULL);
1766                 if (IS_ERR(npath))
1767                         return PTR_ERR(npath);
1768                 BUG_ON(npath->p_depth != path->p_depth);
1769                 eh = npath[depth].p_hdr;
1770                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1771                         ext_debug("next leaf isn't full(%d)\n",
1772                                   le16_to_cpu(eh->eh_entries));
1773                         path = npath;
1774                         goto repeat;
1775                 }
1776                 ext_debug("next leaf has no free space(%d,%d)\n",
1777                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1778         }
1779
1780         /*
1781          * There is no free space in the found leaf.
1782          * We're gonna add a new leaf in the tree.
1783          */
1784         if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1785                 flags = EXT4_MB_USE_ROOT_BLOCKS;
1786         err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1787         if (err)
1788                 goto cleanup;
1789         depth = ext_depth(inode);
1790         eh = path[depth].p_hdr;
1791
1792 has_space:
1793         nearex = path[depth].p_ext;
1794
1795         err = ext4_ext_get_access(handle, inode, path + depth);
1796         if (err)
1797                 goto cleanup;
1798
1799         if (!nearex) {
1800                 /* there is no extent in this leaf, create first one */
1801                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1802                                 le32_to_cpu(newext->ee_block),
1803                                 ext4_ext_pblock(newext),
1804                                 ext4_ext_is_uninitialized(newext),
1805                                 ext4_ext_get_actual_len(newext));
1806                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1807         } else if (le32_to_cpu(newext->ee_block)
1808                            > le32_to_cpu(nearex->ee_block)) {
1809 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1810                 if (nearex != EXT_LAST_EXTENT(eh)) {
1811                         len = EXT_MAX_EXTENT(eh) - nearex;
1812                         len = (len - 1) * sizeof(struct ext4_extent);
1813                         len = len < 0 ? 0 : len;
1814                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1815                                         "move %d from 0x%p to 0x%p\n",
1816                                         le32_to_cpu(newext->ee_block),
1817                                         ext4_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 + 2, nearex + 1, len);
1822                 }
1823                 path[depth].p_ext = nearex + 1;
1824         } else {
1825                 BUG_ON(newext->ee_block == nearex->ee_block);
1826                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1827                 len = len < 0 ? 0 : len;
1828                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1829                                 "move %d from 0x%p to 0x%p\n",
1830                                 le32_to_cpu(newext->ee_block),
1831                                 ext4_ext_pblock(newext),
1832                                 ext4_ext_is_uninitialized(newext),
1833                                 ext4_ext_get_actual_len(newext),
1834                                 nearex, len, nearex + 1, nearex + 2);
1835                 memmove(nearex + 1, nearex, len);
1836                 path[depth].p_ext = nearex;
1837         }
1838
1839         le16_add_cpu(&eh->eh_entries, 1);
1840         nearex = path[depth].p_ext;
1841         nearex->ee_block = newext->ee_block;
1842         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1843         nearex->ee_len = newext->ee_len;
1844
1845 merge:
1846         /* try to merge extents to the right */
1847         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1848                 ext4_ext_try_to_merge(inode, path, nearex);
1849
1850         /* try to merge extents to the left */
1851
1852         /* time to correct all indexes above */
1853         err = ext4_ext_correct_indexes(handle, inode, path);
1854         if (err)
1855                 goto cleanup;
1856
1857         err = ext4_ext_dirty(handle, inode, path + depth);
1858
1859 cleanup:
1860         if (npath) {
1861                 ext4_ext_drop_refs(npath);
1862                 kfree(npath);
1863         }
1864         ext4_ext_invalidate_cache(inode);
1865         return err;
1866 }
1867
1868 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1869                                ext4_lblk_t num, ext_prepare_callback func,
1870                                void *cbdata)
1871 {
1872         struct ext4_ext_path *path = NULL;
1873         struct ext4_ext_cache cbex;
1874         struct ext4_extent *ex;
1875         ext4_lblk_t next, start = 0, end = 0;
1876         ext4_lblk_t last = block + num;
1877         int depth, exists, err = 0;
1878
1879         BUG_ON(func == NULL);
1880         BUG_ON(inode == NULL);
1881
1882         while (block < last && block != EXT_MAX_BLOCK) {
1883                 num = last - block;
1884                 /* find extent for this block */
1885                 down_read(&EXT4_I(inode)->i_data_sem);
1886                 path = ext4_ext_find_extent(inode, block, path);
1887                 up_read(&EXT4_I(inode)->i_data_sem);
1888                 if (IS_ERR(path)) {
1889                         err = PTR_ERR(path);
1890                         path = NULL;
1891                         break;
1892                 }
1893
1894                 depth = ext_depth(inode);
1895                 if (unlikely(path[depth].p_hdr == NULL)) {
1896                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1897                         err = -EIO;
1898                         break;
1899                 }
1900                 ex = path[depth].p_ext;
1901                 next = ext4_ext_next_allocated_block(path);
1902
1903                 exists = 0;
1904                 if (!ex) {
1905                         /* there is no extent yet, so try to allocate
1906                          * all requested space */
1907                         start = block;
1908                         end = block + num;
1909                 } else if (le32_to_cpu(ex->ee_block) > block) {
1910                         /* need to allocate space before found extent */
1911                         start = block;
1912                         end = le32_to_cpu(ex->ee_block);
1913                         if (block + num < end)
1914                                 end = block + num;
1915                 } else if (block >= le32_to_cpu(ex->ee_block)
1916                                         + ext4_ext_get_actual_len(ex)) {
1917                         /* need to allocate space after found extent */
1918                         start = block;
1919                         end = block + num;
1920                         if (end >= next)
1921                                 end = next;
1922                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1923                         /*
1924                          * some part of requested space is covered
1925                          * by found extent
1926                          */
1927                         start = block;
1928                         end = le32_to_cpu(ex->ee_block)
1929                                 + ext4_ext_get_actual_len(ex);
1930                         if (block + num < end)
1931                                 end = block + num;
1932                         exists = 1;
1933                 } else {
1934                         BUG();
1935                 }
1936                 BUG_ON(end <= start);
1937
1938                 if (!exists) {
1939                         cbex.ec_block = start;
1940                         cbex.ec_len = end - start;
1941                         cbex.ec_start = 0;
1942                 } else {
1943                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1944                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1945                         cbex.ec_start = ext4_ext_pblock(ex);
1946                 }
1947
1948                 if (unlikely(cbex.ec_len == 0)) {
1949                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1950                         err = -EIO;
1951                         break;
1952                 }
1953                 err = func(inode, path, &cbex, ex, cbdata);
1954                 ext4_ext_drop_refs(path);
1955
1956                 if (err < 0)
1957                         break;
1958
1959                 if (err == EXT_REPEAT)
1960                         continue;
1961                 else if (err == EXT_BREAK) {
1962                         err = 0;
1963                         break;
1964                 }
1965
1966                 if (ext_depth(inode) != depth) {
1967                         /* depth was changed. we have to realloc path */
1968                         kfree(path);
1969                         path = NULL;
1970                 }
1971
1972                 block = cbex.ec_block + cbex.ec_len;
1973         }
1974
1975         if (path) {
1976                 ext4_ext_drop_refs(path);
1977                 kfree(path);
1978         }
1979
1980         return err;
1981 }
1982
1983 static void
1984 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1985                         __u32 len, ext4_fsblk_t start)
1986 {
1987         struct ext4_ext_cache *cex;
1988         BUG_ON(len == 0);
1989         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1990         cex = &EXT4_I(inode)->i_cached_extent;
1991         cex->ec_block = block;
1992         cex->ec_len = len;
1993         cex->ec_start = start;
1994         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1995 }
1996
1997 /*
1998  * ext4_ext_put_gap_in_cache:
1999  * calculate boundaries of the gap that the requested block fits into
2000  * and cache this gap
2001  */
2002 static void
2003 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2004                                 ext4_lblk_t block)
2005 {
2006         int depth = ext_depth(inode);
2007         unsigned long len;
2008         ext4_lblk_t lblock;
2009         struct ext4_extent *ex;
2010
2011         ex = path[depth].p_ext;
2012         if (ex == NULL) {
2013                 /* there is no extent yet, so gap is [0;-] */
2014                 lblock = 0;
2015                 len = EXT_MAX_BLOCK;
2016                 ext_debug("cache gap(whole file):");
2017         } else if (block < le32_to_cpu(ex->ee_block)) {
2018                 lblock = block;
2019                 len = le32_to_cpu(ex->ee_block) - block;
2020                 ext_debug("cache gap(before): %u [%u:%u]",
2021                                 block,
2022                                 le32_to_cpu(ex->ee_block),
2023                                  ext4_ext_get_actual_len(ex));
2024         } else if (block >= le32_to_cpu(ex->ee_block)
2025                         + ext4_ext_get_actual_len(ex)) {
2026                 ext4_lblk_t next;
2027                 lblock = le32_to_cpu(ex->ee_block)
2028                         + ext4_ext_get_actual_len(ex);
2029
2030                 next = ext4_ext_next_allocated_block(path);
2031                 ext_debug("cache gap(after): [%u:%u] %u",
2032                                 le32_to_cpu(ex->ee_block),
2033                                 ext4_ext_get_actual_len(ex),
2034                                 block);
2035                 BUG_ON(next == lblock);
2036                 len = next - lblock;
2037         } else {
2038                 lblock = len = 0;
2039                 BUG();
2040         }
2041
2042         ext_debug(" -> %u:%lu\n", lblock, len);
2043         ext4_ext_put_in_cache(inode, lblock, len, 0);
2044 }
2045
2046 /*
2047  * ext4_ext_in_cache()
2048  * Checks to see if the given block is in the cache.
2049  * If it is, the cached extent is stored in the given
2050  * cache extent pointer.  If the cached extent is a hole,
2051  * this routine should be used instead of
2052  * ext4_ext_in_cache if the calling function needs to
2053  * know the size of the hole.
2054  *
2055  * @inode: The files inode
2056  * @block: The block to look for in the cache
2057  * @ex:    Pointer where the cached extent will be stored
2058  *         if it contains block
2059  *
2060  * Return 0 if cache is invalid; 1 if the cache is valid
2061  */
2062 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2063         struct ext4_ext_cache *ex){
2064         struct ext4_ext_cache *cex;
2065         struct ext4_sb_info *sbi;
2066         int ret = 0;
2067
2068         /*
2069          * We borrow i_block_reservation_lock to protect i_cached_extent
2070          */
2071         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2072         cex = &EXT4_I(inode)->i_cached_extent;
2073         sbi = EXT4_SB(inode->i_sb);
2074
2075         /* has cache valid data? */
2076         if (cex->ec_len == 0)
2077                 goto errout;
2078
2079         if (in_range(block, cex->ec_block, cex->ec_len)) {
2080                 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2081                 ext_debug("%u cached by %u:%u:%llu\n",
2082                                 block,
2083                                 cex->ec_block, cex->ec_len, cex->ec_start);
2084                 ret = 1;
2085         }
2086 errout:
2087         if (!ret)
2088                 sbi->extent_cache_misses++;
2089         else
2090                 sbi->extent_cache_hits++;
2091         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2092         return ret;
2093 }
2094
2095 /*
2096  * ext4_ext_in_cache()
2097  * Checks to see if the given block is in the cache.
2098  * If it is, the cached extent is stored in the given
2099  * extent pointer.
2100  *
2101  * @inode: The files inode
2102  * @block: The block to look for in the cache
2103  * @ex:    Pointer where the cached extent will be stored
2104  *         if it contains block
2105  *
2106  * Return 0 if cache is invalid; 1 if the cache is valid
2107  */
2108 static int
2109 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2110                         struct ext4_extent *ex)
2111 {
2112         struct ext4_ext_cache cex;
2113         int ret = 0;
2114
2115         if (ext4_ext_check_cache(inode, block, &cex)) {
2116                 ex->ee_block = cpu_to_le32(cex.ec_block);
2117                 ext4_ext_store_pblock(ex, cex.ec_start);
2118                 ex->ee_len = cpu_to_le16(cex.ec_len);
2119                 ret = 1;
2120         }
2121
2122         return ret;
2123 }
2124
2125
2126 /*
2127  * ext4_ext_rm_idx:
2128  * removes index from the index block.
2129  * It's used in truncate case only, thus all requests are for
2130  * last index in the block only.
2131  */
2132 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2133                         struct ext4_ext_path *path)
2134 {
2135         int err;
2136         ext4_fsblk_t leaf;
2137
2138         /* free index block */
2139         path--;
2140         leaf = ext4_idx_pblock(path->p_idx);
2141         if (unlikely(path->p_hdr->eh_entries == 0)) {
2142                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2143                 return -EIO;
2144         }
2145         err = ext4_ext_get_access(handle, inode, path);
2146         if (err)
2147                 return err;
2148         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2149         err = ext4_ext_dirty(handle, inode, path);
2150         if (err)
2151                 return err;
2152         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2153         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2154                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2155         return err;
2156 }
2157
2158 /*
2159  * ext4_ext_calc_credits_for_single_extent:
2160  * This routine returns max. credits that needed to insert an extent
2161  * to the extent tree.
2162  * When pass the actual path, the caller should calculate credits
2163  * under i_data_sem.
2164  */
2165 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2166                                                 struct ext4_ext_path *path)
2167 {
2168         if (path) {
2169                 int depth = ext_depth(inode);
2170                 int ret = 0;
2171
2172                 /* probably there is space in leaf? */
2173                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2174                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2175
2176                         /*
2177                          *  There are some space in the leaf tree, no
2178                          *  need to account for leaf block credit
2179                          *
2180                          *  bitmaps and block group descriptor blocks
2181                          *  and other metadat blocks still need to be
2182                          *  accounted.
2183                          */
2184                         /* 1 bitmap, 1 block group descriptor */
2185                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2186                         return ret;
2187                 }
2188         }
2189
2190         return ext4_chunk_trans_blocks(inode, nrblocks);
2191 }
2192
2193 /*
2194  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2195  *
2196  * if nrblocks are fit in a single extent (chunk flag is 1), then
2197  * in the worse case, each tree level index/leaf need to be changed
2198  * if the tree split due to insert a new extent, then the old tree
2199  * index/leaf need to be updated too
2200  *
2201  * If the nrblocks are discontiguous, they could cause
2202  * the whole tree split more than once, but this is really rare.
2203  */
2204 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2205 {
2206         int index;
2207         int depth = ext_depth(inode);
2208
2209         if (chunk)
2210                 index = depth * 2;
2211         else
2212                 index = depth * 3;
2213
2214         return index;
2215 }
2216
2217 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2218                                 struct ext4_extent *ex,
2219                                 ext4_lblk_t from, ext4_lblk_t to)
2220 {
2221         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2222         int flags = EXT4_FREE_BLOCKS_FORGET;
2223
2224         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2225                 flags |= EXT4_FREE_BLOCKS_METADATA;
2226 #ifdef EXTENTS_STATS
2227         {
2228                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2229                 spin_lock(&sbi->s_ext_stats_lock);
2230                 sbi->s_ext_blocks += ee_len;
2231                 sbi->s_ext_extents++;
2232                 if (ee_len < sbi->s_ext_min)
2233                         sbi->s_ext_min = ee_len;
2234                 if (ee_len > sbi->s_ext_max)
2235                         sbi->s_ext_max = ee_len;
2236                 if (ext_depth(inode) > sbi->s_depth_max)
2237                         sbi->s_depth_max = ext_depth(inode);
2238                 spin_unlock(&sbi->s_ext_stats_lock);
2239         }
2240 #endif
2241         if (from >= le32_to_cpu(ex->ee_block)
2242             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2243                 /* tail removal */
2244                 ext4_lblk_t num;
2245                 ext4_fsblk_t start;
2246
2247                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2248                 start = ext4_ext_pblock(ex) + ee_len - num;
2249                 ext_debug("free last %u blocks starting %llu\n", num, start);
2250                 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2251         } else if (from == le32_to_cpu(ex->ee_block)
2252                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2253                 /* head removal */
2254                 ext4_lblk_t num;
2255                 ext4_fsblk_t start;
2256
2257                 num = to - from;
2258                 start = ext4_ext_pblock(ex);
2259
2260                 ext_debug("free first %u blocks starting %llu\n", num, start);
2261                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2262
2263         } else {
2264                 printk(KERN_INFO "strange request: removal(2) "
2265                                 "%u-%u from %u:%u\n",
2266                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2267         }
2268         return 0;
2269 }
2270
2271
2272 /*
2273  * ext4_ext_rm_leaf() Removes the extents associated with the
2274  * blocks appearing between "start" and "end", and splits the extents
2275  * if "start" and "end" appear in the same extent
2276  *
2277  * @handle: The journal handle
2278  * @inode:  The files inode
2279  * @path:   The path to the leaf
2280  * @start:  The first block to remove
2281  * @end:   The last block to remove
2282  */
2283 static int
2284 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2285                 struct ext4_ext_path *path, ext4_lblk_t start,
2286                 ext4_lblk_t end)
2287 {
2288         int err = 0, correct_index = 0;
2289         int depth = ext_depth(inode), credits;
2290         struct ext4_extent_header *eh;
2291         ext4_lblk_t a, b, block;
2292         unsigned num;
2293         ext4_lblk_t ex_ee_block;
2294         unsigned short ex_ee_len;
2295         unsigned uninitialized = 0;
2296         struct ext4_extent *ex;
2297         struct ext4_map_blocks map;
2298
2299         /* the header must be checked already in ext4_ext_remove_space() */
2300         ext_debug("truncate since %u in leaf\n", start);
2301         if (!path[depth].p_hdr)
2302                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2303         eh = path[depth].p_hdr;
2304         if (unlikely(path[depth].p_hdr == NULL)) {
2305                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2306                 return -EIO;
2307         }
2308         /* find where to start removing */
2309         ex = EXT_LAST_EXTENT(eh);
2310
2311         ex_ee_block = le32_to_cpu(ex->ee_block);
2312         ex_ee_len = ext4_ext_get_actual_len(ex);
2313
2314         while (ex >= EXT_FIRST_EXTENT(eh) &&
2315                         ex_ee_block + ex_ee_len > start) {
2316
2317                 if (ext4_ext_is_uninitialized(ex))
2318                         uninitialized = 1;
2319                 else
2320                         uninitialized = 0;
2321
2322                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2323                          uninitialized, ex_ee_len);
2324                 path[depth].p_ext = ex;
2325
2326                 a = ex_ee_block > start ? ex_ee_block : start;
2327                 b = ex_ee_block+ex_ee_len - 1 < end ?
2328                         ex_ee_block+ex_ee_len - 1 : end;
2329
2330                 ext_debug("  border %u:%u\n", a, b);
2331
2332                 /* If this extent is beyond the end of the hole, skip it */
2333                 if (end <= ex_ee_block) {
2334                         ex--;
2335                         ex_ee_block = le32_to_cpu(ex->ee_block);
2336                         ex_ee_len = ext4_ext_get_actual_len(ex);
2337                         continue;
2338                 } else if (a != ex_ee_block &&
2339                         b != ex_ee_block + ex_ee_len - 1) {
2340                         /*
2341                          * If this is a truncate, then this condition should
2342                          * never happen because at least one of the end points
2343                          * needs to be on the edge of the extent.
2344                          */
2345                         if (end == EXT_MAX_BLOCK) {
2346                                 ext_debug("  bad truncate %u:%u\n",
2347                                                 start, end);
2348                                 block = 0;
2349                                 num = 0;
2350                                 err = -EIO;
2351                                 goto out;
2352                         }
2353                         /*
2354                          * else this is a hole punch, so the extent needs to
2355                          * be split since neither edge of the hole is on the
2356                          * extent edge
2357                          */
2358                         else{
2359                                 map.m_pblk = ext4_ext_pblock(ex);
2360                                 map.m_lblk = ex_ee_block;
2361                                 map.m_len = b - ex_ee_block;
2362
2363                                 err = ext4_split_extent(handle,
2364                                         inode, path, &map, 0,
2365                                         EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2366                                         EXT4_GET_BLOCKS_PRE_IO);
2367
2368                                 if (err < 0)
2369                                         goto out;
2370
2371                                 ex_ee_len = ext4_ext_get_actual_len(ex);
2372
2373                                 b = ex_ee_block+ex_ee_len - 1 < end ?
2374                                         ex_ee_block+ex_ee_len - 1 : end;
2375
2376                                 /* Then remove tail of this extent */
2377                                 block = ex_ee_block;
2378                                 num = a - block;
2379                         }
2380                 } else if (a != ex_ee_block) {
2381                         /* remove tail of the extent */
2382                         block = ex_ee_block;
2383                         num = a - block;
2384                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2385                         /* remove head of the extent */
2386                         block = b;
2387                         num =  ex_ee_block + ex_ee_len - b;
2388
2389                         /*
2390                          * If this is a truncate, this condition
2391                          * should never happen
2392                          */
2393                         if (end == EXT_MAX_BLOCK) {
2394                                 ext_debug("  bad truncate %u:%u\n",
2395                                         start, end);
2396                                 err = -EIO;
2397                                 goto out;
2398                         }
2399                 } else {
2400                         /* remove whole extent: excellent! */
2401                         block = ex_ee_block;
2402                         num = 0;
2403                         if (a != ex_ee_block) {
2404                                 ext_debug("  bad truncate %u:%u\n",
2405                                         start, end);
2406                                 err = -EIO;
2407                                 goto out;
2408                         }
2409
2410                         if (b != ex_ee_block + ex_ee_len - 1) {
2411                                 ext_debug("  bad truncate %u:%u\n",
2412                                         start, end);
2413                                 err = -EIO;
2414                                 goto out;
2415                         }
2416                 }
2417
2418                 /*
2419                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2420                  * descriptor) for each block group; assume two block
2421                  * groups plus ex_ee_len/blocks_per_block_group for
2422                  * the worst case
2423                  */
2424                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2425                 if (ex == EXT_FIRST_EXTENT(eh)) {
2426                         correct_index = 1;
2427                         credits += (ext_depth(inode)) + 1;
2428                 }
2429                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2430
2431                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2432                 if (err)
2433                         goto out;
2434
2435                 err = ext4_ext_get_access(handle, inode, path + depth);
2436                 if (err)
2437                         goto out;
2438
2439                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2440                 if (err)
2441                         goto out;
2442
2443                 if (num == 0) {
2444                         /* this extent is removed; mark slot entirely unused */
2445                         ext4_ext_store_pblock(ex, 0);
2446                 } else if (block != ex_ee_block) {
2447                         /*
2448                          * If this was a head removal, then we need to update
2449                          * the physical block since it is now at a different
2450                          * location
2451                          */
2452                         ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2453                 }
2454
2455                 ex->ee_block = cpu_to_le32(block);
2456                 ex->ee_len = cpu_to_le16(num);
2457                 /*
2458                  * Do not mark uninitialized if all the blocks in the
2459                  * extent have been removed.
2460                  */
2461                 if (uninitialized && num)
2462                         ext4_ext_mark_uninitialized(ex);
2463
2464                 err = ext4_ext_dirty(handle, inode, path + depth);
2465                 if (err)
2466                         goto out;
2467
2468                 /*
2469                  * If the extent was completely released,
2470                  * we need to remove it from the leaf
2471                  */
2472                 if (num == 0) {
2473                         if (end != EXT_MAX_BLOCK) {
2474                                 /*
2475                                  * For hole punching, we need to scoot all the
2476                                  * extents up when an extent is removed so that
2477                                  * we dont have blank extents in the middle
2478                                  */
2479                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2480                                         sizeof(struct ext4_extent));
2481
2482                                 /* Now get rid of the one at the end */
2483                                 memset(EXT_LAST_EXTENT(eh), 0,
2484                                         sizeof(struct ext4_extent));
2485                         }
2486                         le16_add_cpu(&eh->eh_entries, -1);
2487                 }
2488
2489                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2490                                 ext4_ext_pblock(ex));
2491                 ex--;
2492                 ex_ee_block = le32_to_cpu(ex->ee_block);
2493                 ex_ee_len = ext4_ext_get_actual_len(ex);
2494         }
2495
2496         if (correct_index && eh->eh_entries)
2497                 err = ext4_ext_correct_indexes(handle, inode, path);
2498
2499         /* if this leaf is free, then we should
2500          * remove it from index block above */
2501         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2502                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2503
2504 out:
2505         return err;
2506 }
2507
2508 /*
2509  * ext4_ext_more_to_rm:
2510  * returns 1 if current index has to be freed (even partial)
2511  */
2512 static int
2513 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2514 {
2515         BUG_ON(path->p_idx == NULL);
2516
2517         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2518                 return 0;
2519
2520         /*
2521          * if truncate on deeper level happened, it wasn't partial,
2522          * so we have to consider current index for truncation
2523          */
2524         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2525                 return 0;
2526         return 1;
2527 }
2528
2529 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2530                                 ext4_lblk_t end)
2531 {
2532         struct super_block *sb = inode->i_sb;
2533         int depth = ext_depth(inode);
2534         struct ext4_ext_path *path;
2535         handle_t *handle;
2536         int i, err;
2537
2538         ext_debug("truncate since %u\n", start);
2539
2540         /* probably first extent we're gonna free will be last in block */
2541         handle = ext4_journal_start(inode, depth + 1);
2542         if (IS_ERR(handle))
2543                 return PTR_ERR(handle);
2544
2545 again:
2546         ext4_ext_invalidate_cache(inode);
2547
2548         /*
2549          * We start scanning from right side, freeing all the blocks
2550          * after i_size and walking into the tree depth-wise.
2551          */
2552         depth = ext_depth(inode);
2553         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2554         if (path == NULL) {
2555                 ext4_journal_stop(handle);
2556                 return -ENOMEM;
2557         }
2558         path[0].p_depth = depth;
2559         path[0].p_hdr = ext_inode_hdr(inode);
2560         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2561                 err = -EIO;
2562                 goto out;
2563         }
2564         i = err = 0;
2565
2566         while (i >= 0 && err == 0) {
2567                 if (i == depth) {
2568                         /* this is leaf block */
2569                         err = ext4_ext_rm_leaf(handle, inode, path,
2570                                         start, end);
2571                         /* root level has p_bh == NULL, brelse() eats this */
2572                         brelse(path[i].p_bh);
2573                         path[i].p_bh = NULL;
2574                         i--;
2575                         continue;
2576                 }
2577
2578                 /* this is index block */
2579                 if (!path[i].p_hdr) {
2580                         ext_debug("initialize header\n");
2581                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2582                 }
2583
2584                 if (!path[i].p_idx) {
2585                         /* this level hasn't been touched yet */
2586                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2587                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2588                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2589                                   path[i].p_hdr,
2590                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2591                 } else {
2592                         /* we were already here, see at next index */
2593                         path[i].p_idx--;
2594                 }
2595
2596                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2597                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2598                                 path[i].p_idx);
2599                 if (ext4_ext_more_to_rm(path + i)) {
2600                         struct buffer_head *bh;
2601                         /* go to the next level */
2602                         ext_debug("move to level %d (block %llu)\n",
2603                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2604                         memset(path + i + 1, 0, sizeof(*path));
2605                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2606                         if (!bh) {
2607                                 /* should we reset i_size? */
2608                                 err = -EIO;
2609                                 break;
2610                         }
2611                         if (WARN_ON(i + 1 > depth)) {
2612                                 err = -EIO;
2613                                 break;
2614                         }
2615                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2616                                                         depth - i - 1)) {
2617                                 err = -EIO;
2618                                 break;
2619                         }
2620                         path[i + 1].p_bh = bh;
2621
2622                         /* save actual number of indexes since this
2623                          * number is changed at the next iteration */
2624                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2625                         i++;
2626                 } else {
2627                         /* we finished processing this index, go up */
2628                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2629                                 /* index is empty, remove it;
2630                                  * handle must be already prepared by the
2631                                  * truncatei_leaf() */
2632                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2633                         }
2634                         /* root level has p_bh == NULL, brelse() eats this */
2635                         brelse(path[i].p_bh);
2636                         path[i].p_bh = NULL;
2637                         i--;
2638                         ext_debug("return to level %d\n", i);
2639                 }
2640         }
2641
2642         /* TODO: flexible tree reduction should be here */
2643         if (path->p_hdr->eh_entries == 0) {
2644                 /*
2645                  * truncate to zero freed all the tree,
2646                  * so we need to correct eh_depth
2647                  */
2648                 err = ext4_ext_get_access(handle, inode, path);
2649                 if (err == 0) {
2650                         ext_inode_hdr(inode)->eh_depth = 0;
2651                         ext_inode_hdr(inode)->eh_max =
2652                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2653                         err = ext4_ext_dirty(handle, inode, path);
2654                 }
2655         }
2656 out:
2657         ext4_ext_drop_refs(path);
2658         kfree(path);
2659         if (err == -EAGAIN)
2660                 goto again;
2661         ext4_journal_stop(handle);
2662
2663         return err;
2664 }
2665
2666 /*
2667  * called at mount time
2668  */
2669 void ext4_ext_init(struct super_block *sb)
2670 {
2671         /*
2672          * possible initialization would be here
2673          */
2674
2675         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2676 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2677                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2678 #ifdef AGGRESSIVE_TEST
2679                 printk(", aggressive tests");
2680 #endif
2681 #ifdef CHECK_BINSEARCH
2682                 printk(", check binsearch");
2683 #endif
2684 #ifdef EXTENTS_STATS
2685                 printk(", stats");
2686 #endif
2687                 printk("\n");
2688 #endif
2689 #ifdef EXTENTS_STATS
2690                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2691                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2692                 EXT4_SB(sb)->s_ext_max = 0;
2693 #endif
2694         }
2695 }
2696
2697 /*
2698  * called at umount time
2699  */
2700 void ext4_ext_release(struct super_block *sb)
2701 {
2702         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2703                 return;
2704
2705 #ifdef EXTENTS_STATS
2706         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2707                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2708                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2709                         sbi->s_ext_blocks, sbi->s_ext_extents,
2710                         sbi->s_ext_blocks / sbi->s_ext_extents);
2711                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2712                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2713         }
2714 #endif
2715 }
2716
2717 /* FIXME!! we need to try to merge to left or right after zero-out  */
2718 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2719 {
2720         ext4_fsblk_t ee_pblock;
2721         unsigned int ee_len;
2722         int ret;
2723
2724         ee_len    = ext4_ext_get_actual_len(ex);
2725         ee_pblock = ext4_ext_pblock(ex);
2726
2727         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2728         if (ret > 0)
2729                 ret = 0;
2730
2731         return ret;
2732 }
2733
2734 /*
2735  * used by extent splitting.
2736  */
2737 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
2738                                         due to ENOSPC */
2739 #define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
2740 #define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
2741
2742 /*
2743  * ext4_split_extent_at() splits an extent at given block.
2744  *
2745  * @handle: the journal handle
2746  * @inode: the file inode
2747  * @path: the path to the extent
2748  * @split: the logical block where the extent is splitted.
2749  * @split_flags: indicates if the extent could be zeroout if split fails, and
2750  *               the states(init or uninit) of new extents.
2751  * @flags: flags used to insert new extent to extent tree.
2752  *
2753  *
2754  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2755  * of which are deterimined by split_flag.
2756  *
2757  * There are two cases:
2758  *  a> the extent are splitted into two extent.
2759  *  b> split is not needed, and just mark the extent.
2760  *
2761  * return 0 on success.
2762  */
2763 static int ext4_split_extent_at(handle_t *handle,
2764                              struct inode *inode,
2765                              struct ext4_ext_path *path,
2766                              ext4_lblk_t split,
2767                              int split_flag,
2768                              int flags)
2769 {
2770         ext4_fsblk_t newblock;
2771         ext4_lblk_t ee_block;
2772         struct ext4_extent *ex, newex, orig_ex;
2773         struct ext4_extent *ex2 = NULL;
2774         unsigned int ee_len, depth;
2775         int err = 0;
2776
2777         ext_debug("ext4_split_extents_at: inode %lu, logical"
2778                 "block %llu\n", inode->i_ino, (unsigned long long)split);
2779
2780         ext4_ext_show_leaf(inode, path);
2781
2782         depth = ext_depth(inode);
2783         ex = path[depth].p_ext;
2784         ee_block = le32_to_cpu(ex->ee_block);
2785         ee_len = ext4_ext_get_actual_len(ex);
2786         newblock = split - ee_block + ext4_ext_pblock(ex);
2787
2788         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2789
2790         err = ext4_ext_get_access(handle, inode, path + depth);
2791         if (err)
2792                 goto out;
2793
2794         if (split == ee_block) {
2795                 /*
2796                  * case b: block @split is the block that the extent begins with
2797                  * then we just change the state of the extent, and splitting
2798                  * is not needed.
2799                  */
2800                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2801                         ext4_ext_mark_uninitialized(ex);
2802                 else
2803                         ext4_ext_mark_initialized(ex);
2804
2805                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2806                         ext4_ext_try_to_merge(inode, path, ex);
2807
2808                 err = ext4_ext_dirty(handle, inode, path + depth);
2809                 goto out;
2810         }
2811
2812         /* case a */
2813         memcpy(&orig_ex, ex, sizeof(orig_ex));
2814         ex->ee_len = cpu_to_le16(split - ee_block);
2815         if (split_flag & EXT4_EXT_MARK_UNINIT1)
2816                 ext4_ext_mark_uninitialized(ex);
2817
2818         /*
2819          * path may lead to new leaf, not to original leaf any more
2820          * after ext4_ext_insert_extent() returns,
2821          */
2822         err = ext4_ext_dirty(handle, inode, path + depth);
2823         if (err)
2824                 goto fix_extent_len;
2825
2826         ex2 = &newex;
2827         ex2->ee_block = cpu_to_le32(split);
2828         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2829         ext4_ext_store_pblock(ex2, newblock);
2830         if (split_flag & EXT4_EXT_MARK_UNINIT2)
2831                 ext4_ext_mark_uninitialized(ex2);
2832
2833         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2834         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2835                 err = ext4_ext_zeroout(inode, &orig_ex);
2836                 if (err)
2837                         goto fix_extent_len;
2838                 /* update the extent length and mark as initialized */
2839                 ex->ee_len = cpu_to_le32(ee_len);
2840                 ext4_ext_try_to_merge(inode, path, ex);
2841                 err = ext4_ext_dirty(handle, inode, path + depth);
2842                 goto out;
2843         } else if (err)
2844                 goto fix_extent_len;
2845
2846 out:
2847         ext4_ext_show_leaf(inode, path);
2848         return err;
2849
2850 fix_extent_len:
2851         ex->ee_len = orig_ex.ee_len;
2852         ext4_ext_dirty(handle, inode, path + depth);
2853         return err;
2854 }
2855
2856 /*
2857  * ext4_split_extents() splits an extent and mark extent which is covered
2858  * by @map as split_flags indicates
2859  *
2860  * It may result in splitting the extent into multiple extents (upto three)
2861  * There are three possibilities:
2862  *   a> There is no split required
2863  *   b> Splits in two extents: Split is happening at either end of the extent
2864  *   c> Splits in three extents: Somone is splitting in middle of the extent
2865  *
2866  */
2867 static int ext4_split_extent(handle_t *handle,
2868                               struct inode *inode,
2869                               struct ext4_ext_path *path,
2870                               struct ext4_map_blocks *map,
2871                               int split_flag,
2872                               int flags)
2873 {
2874         ext4_lblk_t ee_block;
2875         struct ext4_extent *ex;
2876         unsigned int ee_len, depth;
2877         int err = 0;
2878         int uninitialized;
2879         int split_flag1, flags1;
2880
2881         depth = ext_depth(inode);
2882         ex = path[depth].p_ext;
2883         ee_block = le32_to_cpu(ex->ee_block);
2884         ee_len = ext4_ext_get_actual_len(ex);
2885         uninitialized = ext4_ext_is_uninitialized(ex);
2886
2887         if (map->m_lblk + map->m_len < ee_block + ee_len) {
2888                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2889                               EXT4_EXT_MAY_ZEROOUT : 0;
2890                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2891                 if (uninitialized)
2892                         split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2893                                        EXT4_EXT_MARK_UNINIT2;
2894                 err = ext4_split_extent_at(handle, inode, path,
2895                                 map->m_lblk + map->m_len, split_flag1, flags1);
2896                 if (err)
2897                         goto out;
2898         }
2899
2900         ext4_ext_drop_refs(path);
2901         path = ext4_ext_find_extent(inode, map->m_lblk, path);
2902         if (IS_ERR(path))
2903                 return PTR_ERR(path);
2904
2905         if (map->m_lblk >= ee_block) {
2906                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2907                               EXT4_EXT_MAY_ZEROOUT : 0;
2908                 if (uninitialized)
2909                         split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2910                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2911                         split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2912                 err = ext4_split_extent_at(handle, inode, path,
2913                                 map->m_lblk, split_flag1, flags);
2914                 if (err)
2915                         goto out;
2916         }
2917
2918         ext4_ext_show_leaf(inode, path);
2919 out:
2920         return err ? err : map->m_len;
2921 }
2922
2923 #define EXT4_EXT_ZERO_LEN 7
2924 /*
2925  * This function is called by ext4_ext_map_blocks() if someone tries to write
2926  * to an uninitialized extent. It may result in splitting the uninitialized
2927  * extent into multiple extents (up to three - one initialized and two
2928  * uninitialized).
2929  * There are three possibilities:
2930  *   a> There is no split required: Entire extent should be initialized
2931  *   b> Splits in two extents: Write is happening at either end of the extent
2932  *   c> Splits in three extents: Somone is writing in middle of the extent
2933  */
2934 static int ext4_ext_convert_to_initialized(handle_t *handle,
2935                                            struct inode *inode,
2936                                            struct ext4_map_blocks *map,
2937                                            struct ext4_ext_path *path)
2938 {
2939         struct ext4_map_blocks split_map;
2940         struct ext4_extent zero_ex;
2941         struct ext4_extent *ex;
2942         ext4_lblk_t ee_block, eof_block;
2943         unsigned int allocated, ee_len, depth;
2944         int err = 0;
2945         int split_flag = 0;
2946
2947         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2948                 "block %llu, max_blocks %u\n", inode->i_ino,
2949                 (unsigned long long)map->m_lblk, map->m_len);
2950
2951         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2952                 inode->i_sb->s_blocksize_bits;
2953         if (eof_block < map->m_lblk + map->m_len)
2954                 eof_block = map->m_lblk + map->m_len;
2955
2956         depth = ext_depth(inode);
2957         ex = path[depth].p_ext;
2958         ee_block = le32_to_cpu(ex->ee_block);
2959         ee_len = ext4_ext_get_actual_len(ex);
2960         allocated = ee_len - (map->m_lblk - ee_block);
2961
2962         WARN_ON(map->m_lblk < ee_block);
2963         /*
2964          * It is safe to convert extent to initialized via explicit
2965          * zeroout only if extent is fully insde i_size or new_size.
2966          */
2967         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
2968
2969         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2970         if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2971             (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2972                 err = ext4_ext_zeroout(inode, ex);
2973                 if (err)
2974                         goto out;
2975
2976                 err = ext4_ext_get_access(handle, inode, path + depth);
2977                 if (err)
2978                         goto out;
2979                 ext4_ext_mark_initialized(ex);
2980                 ext4_ext_try_to_merge(inode, path, ex);
2981                 err = ext4_ext_dirty(handle, inode, path + depth);
2982                 goto out;
2983         }
2984
2985         /*
2986          * four cases:
2987          * 1. split the extent into three extents.
2988          * 2. split the extent into two extents, zeroout the first half.
2989          * 3. split the extent into two extents, zeroout the second half.
2990          * 4. split the extent into two extents with out zeroout.
2991          */
2992         split_map.m_lblk = map->m_lblk;
2993         split_map.m_len = map->m_len;
2994
2995         if (allocated > map->m_len) {
2996                 if (allocated <= EXT4_EXT_ZERO_LEN &&
2997                     (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2998                         /* case 3 */
2999                         zero_ex.ee_block =
3000                                          cpu_to_le32(map->m_lblk);
3001                         zero_ex.ee_len = cpu_to_le16(allocated);
3002                         ext4_ext_store_pblock(&zero_ex,
3003                                 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3004                         err = ext4_ext_zeroout(inode, &zero_ex);
3005                         if (err)
3006                                 goto out;
3007                         split_map.m_lblk = map->m_lblk;
3008                         split_map.m_len = allocated;
3009                 } else if ((map->m_lblk - ee_block + map->m_len <
3010                            EXT4_EXT_ZERO_LEN) &&
3011                            (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3012                         /* case 2 */
3013                         if (map->m_lblk != ee_block) {
3014                                 zero_ex.ee_block = ex->ee_block;
3015                                 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3016                                                         ee_block);
3017                                 ext4_ext_store_pblock(&zero_ex,
3018                                                       ext4_ext_pblock(ex));
3019                                 err = ext4_ext_zeroout(inode, &zero_ex);
3020                                 if (err)
3021                                         goto out;
3022                         }
3023
3024                         split_map.m_lblk = ee_block;
3025                         split_map.m_len = map->m_lblk - ee_block + map->m_len;
3026                         allocated = map->m_len;
3027                 }
3028         }
3029
3030         allocated = ext4_split_extent(handle, inode, path,
3031                                        &split_map, split_flag, 0);
3032         if (allocated < 0)
3033                 err = allocated;
3034
3035 out:
3036         return err ? err : allocated;
3037 }
3038
3039 /*
3040  * This function is called by ext4_ext_map_blocks() from
3041  * ext4_get_blocks_dio_write() when DIO to write
3042  * to an uninitialized extent.
3043  *
3044  * Writing to an uninitialized extent may result in splitting the uninitialized
3045  * extent into multiple /initialized uninitialized extents (up to three)
3046  * There are three possibilities:
3047  *   a> There is no split required: Entire extent should be uninitialized
3048  *   b> Splits in two extents: Write is happening at either end of the extent
3049  *   c> Splits in three extents: Somone is writing in middle of the extent
3050  *
3051  * One of more index blocks maybe needed if the extent tree grow after
3052  * the uninitialized extent split. To prevent ENOSPC occur at the IO
3053  * complete, we need to split the uninitialized extent before DIO submit
3054  * the IO. The uninitialized extent called at this time will be split
3055  * into three uninitialized extent(at most). After IO complete, the part
3056  * being filled will be convert to initialized by the end_io callback function
3057  * via ext4_convert_unwritten_extents().
3058  *
3059  * Returns the size of uninitialized extent to be written on success.
3060  */
3061 static int ext4_split_unwritten_extents(handle_t *handle,
3062                                         struct inode *inode,
3063                                         struct ext4_map_blocks *map,
3064                                         struct ext4_ext_path *path,
3065                                         int flags)
3066 {
3067         ext4_lblk_t eof_block;
3068         ext4_lblk_t ee_block;
3069         struct ext4_extent *ex;
3070         unsigned int ee_len;
3071         int split_flag = 0, depth;
3072
3073         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3074                 "block %llu, max_blocks %u\n", inode->i_ino,
3075                 (unsigned long long)map->m_lblk, map->m_len);
3076
3077         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3078                 inode->i_sb->s_blocksize_bits;
3079         if (eof_block < map->m_lblk + map->m_len)
3080                 eof_block = map->m_lblk + map->m_len;
3081         /*
3082          * It is safe to convert extent to initialized via explicit
3083          * zeroout only if extent is fully insde i_size or new_size.
3084          */
3085         depth = ext_depth(inode);
3086         ex = path[depth].p_ext;
3087         ee_block = le32_to_cpu(ex->ee_block);
3088         ee_len = ext4_ext_get_actual_len(ex);
3089
3090         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3091         split_flag |= EXT4_EXT_MARK_UNINIT2;
3092
3093         flags |= EXT4_GET_BLOCKS_PRE_IO;
3094         return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3095 }
3096
3097 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3098                                               struct inode *inode,
3099                                               struct ext4_ext_path *path)
3100 {
3101         struct ext4_extent *ex;
3102         struct ext4_extent_header *eh;
3103         int depth;
3104         int err = 0;
3105
3106         depth = ext_depth(inode);
3107         eh = path[depth].p_hdr;
3108         ex = path[depth].p_ext;
3109
3110         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3111                 "block %llu, max_blocks %u\n", inode->i_ino,
3112                 (unsigned long long)le32_to_cpu(ex->ee_block),
3113                 ext4_ext_get_actual_len(ex));
3114
3115         err = ext4_ext_get_access(handle, inode, path + depth);
3116         if (err)
3117                 goto out;
3118         /* first mark the extent as initialized */
3119         ext4_ext_mark_initialized(ex);
3120
3121         /* note: ext4_ext_correct_indexes() isn't needed here because
3122          * borders are not changed
3123          */
3124         ext4_ext_try_to_merge(inode, path, ex);
3125
3126         /* Mark modified extent as dirty */
3127         err = ext4_ext_dirty(handle, inode, path + depth);
3128 out:
3129         ext4_ext_show_leaf(inode, path);
3130         return err;
3131 }
3132
3133 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3134                         sector_t block, int count)
3135 {
3136         int i;
3137         for (i = 0; i < count; i++)
3138                 unmap_underlying_metadata(bdev, block + i);
3139 }
3140
3141 /*
3142  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3143  */
3144 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3145                               ext4_lblk_t lblk,
3146                               struct ext4_ext_path *path,
3147                               unsigned int len)
3148 {
3149         int i, depth;
3150         struct ext4_extent_header *eh;
3151         struct ext4_extent *last_ex;
3152
3153         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3154                 return 0;
3155
3156         depth = ext_depth(inode);
3157         eh = path[depth].p_hdr;
3158
3159         if (unlikely(!eh->eh_entries)) {
3160                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3161                                  "EOFBLOCKS_FL set");
3162                 return -EIO;
3163         }
3164         last_ex = EXT_LAST_EXTENT(eh);
3165         /*
3166          * We should clear the EOFBLOCKS_FL flag if we are writing the
3167          * last block in the last extent in the file.  We test this by
3168          * first checking to see if the caller to
3169          * ext4_ext_get_blocks() was interested in the last block (or
3170          * a block beyond the last block) in the current extent.  If
3171          * this turns out to be false, we can bail out from this
3172          * function immediately.
3173          */
3174         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3175             ext4_ext_get_actual_len(last_ex))
3176                 return 0;
3177         /*
3178          * If the caller does appear to be planning to write at or
3179          * beyond the end of the current extent, we then test to see
3180          * if the current extent is the last extent in the file, by
3181          * checking to make sure it was reached via the rightmost node
3182          * at each level of the tree.
3183          */
3184         for (i = depth-1; i >= 0; i--)
3185                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3186                         return 0;
3187         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3188         return ext4_mark_inode_dirty(handle, inode);
3189 }
3190
3191 static int
3192 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3193                         struct ext4_map_blocks *map,
3194                         struct ext4_ext_path *path, int flags,
3195                         unsigned int allocated, ext4_fsblk_t newblock)
3196 {
3197         int ret = 0;
3198         int err = 0;
3199         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3200
3201         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3202                   "block %llu, max_blocks %u, flags %d, allocated %u",
3203                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3204                   flags, allocated);
3205         ext4_ext_show_leaf(inode, path);
3206
3207         /* get_block() before submit the IO, split the extent */
3208         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3209                 ret = ext4_split_unwritten_extents(handle, inode, map,
3210                                                    path, flags);
3211                 /*
3212                  * Flag the inode(non aio case) or end_io struct (aio case)
3213                  * that this IO needs to conversion to written when IO is
3214                  * completed
3215                  */
3216                 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3217                         io->flag = EXT4_IO_END_UNWRITTEN;
3218                         atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3219                 } else
3220                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3221                 if (ext4_should_dioread_nolock(inode))
3222                         map->m_flags |= EXT4_MAP_UNINIT;
3223                 goto out;
3224         }
3225         /* IO end_io complete, convert the filled extent to written */
3226         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3227                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3228                                                         path);
3229                 if (ret >= 0) {
3230                         ext4_update_inode_fsync_trans(handle, inode, 1);
3231                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3232                                                  path, map->m_len);
3233                 } else
3234                         err = ret;
3235                 goto out2;
3236         }
3237         /* buffered IO case */
3238         /*
3239          * repeat fallocate creation request
3240          * we already have an unwritten extent
3241          */
3242         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3243                 goto map_out;
3244
3245         /* buffered READ or buffered write_begin() lookup */
3246         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3247                 /*
3248                  * We have blocks reserved already.  We
3249                  * return allocated blocks so that delalloc
3250                  * won't do block reservation for us.  But
3251                  * the buffer head will be unmapped so that
3252                  * a read from the block returns 0s.
3253                  */
3254                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3255                 goto out1;
3256         }
3257
3258         /* buffered write, writepage time, convert*/
3259         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3260         if (ret >= 0) {
3261                 ext4_update_inode_fsync_trans(handle, inode, 1);
3262                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3263                                          map->m_len);
3264                 if (err < 0)
3265                         goto out2;
3266         }
3267
3268 out:
3269         if (ret <= 0) {
3270                 err = ret;
3271                 goto out2;
3272         } else
3273                 allocated = ret;
3274         map->m_flags |= EXT4_MAP_NEW;
3275         /*
3276          * if we allocated more blocks than requested
3277          * we need to make sure we unmap the extra block
3278          * allocated. The actual needed block will get
3279          * unmapped later when we find the buffer_head marked
3280          * new.
3281          */
3282         if (allocated > map->m_len) {
3283                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3284                                         newblock + map->m_len,
3285                                         allocated - map->m_len);
3286                 allocated = map->m_len;
3287         }
3288
3289         /*
3290          * If we have done fallocate with the offset that is already
3291          * delayed allocated, we would have block reservation
3292          * and quota reservation done in the delayed write path.
3293          * But fallocate would have already updated quota and block
3294          * count for this offset. So cancel these reservation
3295          */
3296         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3297                 ext4_da_update_reserve_space(inode, allocated, 0);
3298
3299 map_out:
3300         map->m_flags |= EXT4_MAP_MAPPED;
3301 out1:
3302         if (allocated > map->m_len)
3303                 allocated = map->m_len;
3304         ext4_ext_show_leaf(inode, path);
3305         map->m_pblk = newblock;
3306         map->m_len = allocated;
3307 out2:
3308         if (path) {
3309                 ext4_ext_drop_refs(path);
3310                 kfree(path);
3311         }
3312         return err ? err : allocated;
3313 }
3314
3315 /*
3316  * Block allocation/map/preallocation routine for extents based files
3317  *
3318  *
3319  * Need to be called with
3320  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3321  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3322  *
3323  * return > 0, number of of blocks already mapped/allocated
3324  *          if create == 0 and these are pre-allocated blocks
3325  *              buffer head is unmapped
3326  *          otherwise blocks are mapped
3327  *
3328  * return = 0, if plain look up failed (blocks have not been allocated)
3329  *          buffer head is unmapped
3330  *
3331  * return < 0, error case.
3332  */
3333 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3334                         struct ext4_map_blocks *map, int flags)
3335 {
3336         struct ext4_ext_path *path = NULL;
3337         struct ext4_extent newex, *ex;
3338         ext4_fsblk_t newblock = 0;
3339         int err = 0, depth, ret;
3340         unsigned int allocated = 0;
3341         unsigned int punched_out = 0;
3342         unsigned int result = 0;
3343         struct ext4_allocation_request ar;
3344         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3345         struct ext4_map_blocks punch_map;
3346
3347         ext_debug("blocks %u/%u requested for inode %lu\n",
3348                   map->m_lblk, map->m_len, inode->i_ino);
3349         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3350
3351         /* check in cache */
3352         if (ext4_ext_in_cache(inode, map->m_lblk, &newex) &&
3353                 ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0)) {
3354                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3355                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3356                                 /*
3357                                  * block isn't allocated yet and
3358                                  * user doesn't want to allocate it
3359                                  */
3360                                 goto out2;
3361                         }
3362                         /* we should allocate requested block */
3363                 } else {
3364                         /* block is already allocated */
3365                         newblock = map->m_lblk
3366                                    - le32_to_cpu(newex.ee_block)
3367                                    + ext4_ext_pblock(&newex);
3368                         /* number of remaining blocks in the extent */
3369                         allocated = ext4_ext_get_actual_len(&newex) -
3370                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3371                         goto out;
3372                 }
3373         }
3374
3375         /* find extent for this block */
3376         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3377         if (IS_ERR(path)) {
3378                 err = PTR_ERR(path);
3379                 path = NULL;
3380                 goto out2;
3381         }
3382
3383         depth = ext_depth(inode);
3384
3385         /*
3386          * consistent leaf must not be empty;
3387          * this situation is possible, though, _during_ tree modification;
3388          * this is why assert can't be put in ext4_ext_find_extent()
3389          */
3390         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3391                 EXT4_ERROR_INODE(inode, "bad extent address "
3392                                  "lblock: %lu, depth: %d pblock %lld",
3393                                  (unsigned long) map->m_lblk, depth,
3394                                  path[depth].p_block);
3395                 err = -EIO;
3396                 goto out2;
3397         }
3398
3399         ex = path[depth].p_ext;
3400         if (ex) {
3401                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3402                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3403                 unsigned short ee_len;
3404
3405                 /*
3406                  * Uninitialized extents are treated as holes, except that
3407                  * we split out initialized portions during a write.
3408                  */
3409                 ee_len = ext4_ext_get_actual_len(ex);
3410                 /* if found extent covers block, simply return it */
3411                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3412                         newblock = map->m_lblk - ee_block + ee_start;
3413                         /* number of remaining blocks in the extent */
3414                         allocated = ee_len - (map->m_lblk - ee_block);
3415                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3416                                   ee_block, ee_len, newblock);
3417
3418                         if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3419                                 /*
3420                                  * Do not put uninitialized extent
3421                                  * in the cache
3422                                  */
3423                                 if (!ext4_ext_is_uninitialized(ex)) {
3424                                         ext4_ext_put_in_cache(inode, ee_block,
3425                                                 ee_len, ee_start);
3426                                         goto out;
3427                                 }
3428                                 ret = ext4_ext_handle_uninitialized_extents(
3429                                         handle, inode, map, path, flags,
3430                                         allocated, newblock);
3431                                 return ret;
3432                         }
3433
3434                         /*
3435                          * Punch out the map length, but only to the
3436                          * end of the extent
3437                          */
3438                         punched_out = allocated < map->m_len ?
3439                                 allocated : map->m_len;
3440
3441                         /*
3442                          * Sense extents need to be converted to
3443                          * uninitialized, they must fit in an
3444                          * uninitialized extent
3445                          */
3446                         if (punched_out > EXT_UNINIT_MAX_LEN)
3447                                 punched_out = EXT_UNINIT_MAX_LEN;
3448
3449                         punch_map.m_lblk = map->m_lblk;
3450                         punch_map.m_pblk = newblock;
3451                         punch_map.m_len = punched_out;
3452                         punch_map.m_flags = 0;
3453
3454                         /* Check to see if the extent needs to be split */
3455                         if (punch_map.m_len != ee_len ||
3456                                 punch_map.m_lblk != ee_block) {
3457
3458                                 ret = ext4_split_extent(handle, inode,
3459                                 path, &punch_map, 0,
3460                                 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3461                                 EXT4_GET_BLOCKS_PRE_IO);
3462
3463                                 if (ret < 0) {
3464                                         err = ret;
3465                                         goto out2;
3466                                 }
3467                                 /*
3468                                  * find extent for the block at
3469                                  * the start of the hole
3470                                  */
3471                                 ext4_ext_drop_refs(path);
3472                                 kfree(path);
3473
3474                                 path = ext4_ext_find_extent(inode,
3475                                 map->m_lblk, NULL);
3476                                 if (IS_ERR(path)) {
3477                                         err = PTR_ERR(path);
3478                                         path = NULL;
3479                                         goto out2;
3480                                 }
3481
3482                                 depth = ext_depth(inode);
3483                                 ex = path[depth].p_ext;
3484                                 ee_len = ext4_ext_get_actual_len(ex);
3485                                 ee_block = le32_to_cpu(ex->ee_block);
3486                                 ee_start = ext4_ext_pblock(ex);
3487
3488                         }
3489
3490                         ext4_ext_mark_uninitialized(ex);
3491
3492                         err = ext4_ext_remove_space(inode, map->m_lblk,
3493                                 map->m_lblk + punched_out);
3494
3495                         goto out2;
3496                 }
3497         }
3498
3499         /*
3500          * requested block isn't allocated yet;
3501          * we couldn't try to create block if create flag is zero
3502          */
3503         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3504                 /*
3505                  * put just found gap into cache to speed up
3506                  * subsequent requests
3507                  */
3508                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3509                 goto out2;
3510         }
3511         /*
3512          * Okay, we need to do block allocation.
3513          */
3514
3515         /* find neighbour allocated blocks */
3516         ar.lleft = map->m_lblk;
3517         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3518         if (err)
3519                 goto out2;
3520         ar.lright = map->m_lblk;
3521         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3522         if (err)
3523                 goto out2;
3524
3525         /*
3526          * See if request is beyond maximum number of blocks we can have in
3527          * a single extent. For an initialized extent this limit is
3528          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3529          * EXT_UNINIT_MAX_LEN.
3530          */
3531         if (map->m_len > EXT_INIT_MAX_LEN &&
3532             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3533                 map->m_len = EXT_INIT_MAX_LEN;
3534         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3535                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3536                 map->m_len = EXT_UNINIT_MAX_LEN;
3537
3538         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3539         newex.ee_block = cpu_to_le32(map->m_lblk);
3540         newex.ee_len = cpu_to_le16(map->m_len);
3541         err = ext4_ext_check_overlap(inode, &newex, path);
3542         if (err)
3543                 allocated = ext4_ext_get_actual_len(&newex);
3544         else
3545                 allocated = map->m_len;
3546
3547         /* allocate new block */
3548         ar.inode = inode;
3549         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3550         ar.logical = map->m_lblk;
3551         ar.len = allocated;
3552         if (S_ISREG(inode->i_mode))
3553                 ar.flags = EXT4_MB_HINT_DATA;
3554         else
3555                 /* disable in-core preallocation for non-regular files */
3556                 ar.flags = 0;
3557         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3558         if (!newblock)
3559                 goto out2;
3560         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3561                   ar.goal, newblock, allocated);
3562
3563         /* try to insert new extent into found leaf and return */
3564         ext4_ext_store_pblock(&newex, newblock);
3565         newex.ee_len = cpu_to_le16(ar.len);
3566         /* Mark uninitialized */
3567         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3568                 ext4_ext_mark_uninitialized(&newex);
3569                 /*
3570                  * io_end structure was created for every IO write to an
3571                  * uninitialized extent. To avoid unnecessary conversion,
3572                  * here we flag the IO that really needs the conversion.
3573                  * For non asycn direct IO case, flag the inode state
3574                  * that we need to perform conversion when IO is done.
3575                  */
3576                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3577                         if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3578                                 io->flag = EXT4_IO_END_UNWRITTEN;
3579                                 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3580                         } else
3581                                 ext4_set_inode_state(inode,
3582                                                      EXT4_STATE_DIO_UNWRITTEN);
3583                 }
3584                 if (ext4_should_dioread_nolock(inode))
3585                         map->m_flags |= EXT4_MAP_UNINIT;
3586         }
3587
3588         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3589         if (err)
3590                 goto out2;
3591
3592         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3593         if (err) {
3594                 /* free data blocks we just allocated */
3595                 /* not a good idea to call discard here directly,
3596                  * but otherwise we'd need to call it every free() */
3597                 ext4_discard_preallocations(inode);
3598                 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3599                                  ext4_ext_get_actual_len(&newex), 0);
3600                 goto out2;
3601         }
3602
3603         /* previous routine could use block we allocated */
3604         newblock = ext4_ext_pblock(&newex);
3605         allocated = ext4_ext_get_actual_len(&newex);
3606         if (allocated > map->m_len)
3607                 allocated = map->m_len;
3608         map->m_flags |= EXT4_MAP_NEW;
3609
3610         /*
3611          * Update reserved blocks/metadata blocks after successful
3612          * block allocation which had been deferred till now.
3613          */
3614         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3615                 ext4_da_update_reserve_space(inode, allocated, 1);
3616
3617         /*
3618          * Cache the extent and update transaction to commit on fdatasync only
3619          * when it is _not_ an uninitialized extent.
3620          */
3621         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3622                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3623                 ext4_update_inode_fsync_trans(handle, inode, 1);
3624         } else
3625                 ext4_update_inode_fsync_trans(handle, inode, 0);
3626 out:
3627         if (allocated > map->m_len)
3628                 allocated = map->m_len;
3629         ext4_ext_show_leaf(inode, path);
3630         map->m_flags |= EXT4_MAP_MAPPED;
3631         map->m_pblk = newblock;
3632         map->m_len = allocated;
3633 out2:
3634         if (path) {
3635                 ext4_ext_drop_refs(path);
3636                 kfree(path);
3637         }
3638         trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3639                 newblock, map->m_len, err ? err : allocated);
3640
3641         result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3642                         punched_out : allocated;
3643
3644         return err ? err : result;
3645 }
3646
3647 void ext4_ext_truncate(struct inode *inode)
3648 {
3649         struct address_space *mapping = inode->i_mapping;
3650         struct super_block *sb = inode->i_sb;
3651         ext4_lblk_t last_block;
3652         handle_t *handle;
3653         int err = 0;
3654
3655         /*
3656          * finish any pending end_io work so we won't run the risk of
3657          * converting any truncated blocks to initialized later
3658          */
3659         ext4_flush_completed_IO(inode);
3660
3661         /*
3662          * probably first extent we're gonna free will be last in block
3663          */
3664         err = ext4_writepage_trans_blocks(inode);
3665         handle = ext4_journal_start(inode, err);
3666         if (IS_ERR(handle))
3667                 return;
3668
3669         if (inode->i_size & (sb->s_blocksize - 1))
3670                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3671
3672         if (ext4_orphan_add(handle, inode))
3673                 goto out_stop;
3674
3675         down_write(&EXT4_I(inode)->i_data_sem);
3676         ext4_ext_invalidate_cache(inode);
3677
3678         ext4_discard_preallocations(inode);
3679
3680         /*
3681          * TODO: optimization is possible here.
3682          * Probably we need not scan at all,
3683          * because page truncation is enough.
3684          */
3685
3686         /* we have to know where to truncate from in crash case */
3687         EXT4_I(inode)->i_disksize = inode->i_size;
3688         ext4_mark_inode_dirty(handle, inode);
3689
3690         last_block = (inode->i_size + sb->s_blocksize - 1)
3691                         >> EXT4_BLOCK_SIZE_BITS(sb);
3692         err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCK);
3693
3694         /* In a multi-transaction truncate, we only make the final
3695          * transaction synchronous.
3696          */
3697         if (IS_SYNC(inode))
3698                 ext4_handle_sync(handle);
3699
3700         up_write(&EXT4_I(inode)->i_data_sem);
3701
3702 out_stop:
3703         /*
3704          * If this was a simple ftruncate() and the file will remain alive,
3705          * then we need to clear up the orphan record which we created above.
3706          * However, if this was a real unlink then we were called by
3707          * ext4_delete_inode(), and we allow that function to clean up the
3708          * orphan info for us.
3709          */
3710         if (inode->i_nlink)
3711                 ext4_orphan_del(handle, inode);
3712
3713         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3714         ext4_mark_inode_dirty(handle, inode);
3715         ext4_journal_stop(handle);
3716 }
3717
3718 static void ext4_falloc_update_inode(struct inode *inode,
3719                                 int mode, loff_t new_size, int update_ctime)
3720 {
3721         struct timespec now;
3722
3723         if (update_ctime) {
3724                 now = current_fs_time(inode->i_sb);
3725                 if (!timespec_equal(&inode->i_ctime, &now))
3726                         inode->i_ctime = now;
3727         }
3728         /*
3729          * Update only when preallocation was requested beyond
3730          * the file size.
3731          */
3732         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3733                 if (new_size > i_size_read(inode))
3734                         i_size_write(inode, new_size);
3735                 if (new_size > EXT4_I(inode)->i_disksize)
3736                         ext4_update_i_disksize(inode, new_size);
3737         } else {
3738                 /*
3739                  * Mark that we allocate beyond EOF so the subsequent truncate
3740                  * can proceed even if the new size is the same as i_size.
3741                  */
3742                 if (new_size > i_size_read(inode))
3743                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3744         }
3745
3746 }
3747
3748 /*
3749  * preallocate space for a file. This implements ext4's fallocate file
3750  * operation, which gets called from sys_fallocate system call.
3751  * For block-mapped files, posix_fallocate should fall back to the method
3752  * of writing zeroes to the required new blocks (the same behavior which is
3753  * expected for file systems which do not support fallocate() system call).
3754  */
3755 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3756 {
3757         struct inode *inode = file->f_path.dentry->d_inode;
3758         handle_t *handle;
3759         loff_t new_size;
3760         unsigned int max_blocks;
3761         int ret = 0;
3762         int ret2 = 0;
3763         int retries = 0;
3764         struct ext4_map_blocks map;
3765         unsigned int credits, blkbits = inode->i_blkbits;
3766
3767         /*
3768          * currently supporting (pre)allocate mode for extent-based
3769          * files _only_
3770          */
3771         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3772                 return -EOPNOTSUPP;
3773
3774         /* Return error if mode is not supported */
3775         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3776                 return -EOPNOTSUPP;
3777
3778         if (mode & FALLOC_FL_PUNCH_HOLE)
3779                 return ext4_punch_hole(file, offset, len);
3780
3781         trace_ext4_fallocate_enter(inode, offset, len, mode);
3782         map.m_lblk = offset >> blkbits;
3783         /*
3784          * We can't just convert len to max_blocks because
3785          * If blocksize = 4096 offset = 3072 and len = 2048
3786          */
3787         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3788                 - map.m_lblk;
3789         /*
3790          * credits to insert 1 extent into extent tree
3791          */
3792         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3793         mutex_lock(&inode->i_mutex);
3794         ret = inode_newsize_ok(inode, (len + offset));
3795         if (ret) {
3796                 mutex_unlock(&inode->i_mutex);
3797                 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3798                 return ret;
3799         }
3800 retry:
3801         while (ret >= 0 && ret < max_blocks) {
3802                 map.m_lblk = map.m_lblk + ret;
3803                 map.m_len = max_blocks = max_blocks - ret;
3804                 handle = ext4_journal_start(inode, credits);
3805                 if (IS_ERR(handle)) {
3806                         ret = PTR_ERR(handle);
3807                         break;
3808                 }
3809                 ret = ext4_map_blocks(handle, inode, &map,
3810                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3811                 if (ret <= 0) {
3812 #ifdef EXT4FS_DEBUG
3813                         WARN_ON(ret <= 0);
3814                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3815                                     "returned error inode#%lu, block=%u, "
3816                                     "max_blocks=%u", __func__,
3817                                     inode->i_ino, map.m_lblk, max_blocks);
3818 #endif
3819                         ext4_mark_inode_dirty(handle, inode);
3820                         ret2 = ext4_journal_stop(handle);
3821                         break;
3822                 }
3823                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3824                                                 blkbits) >> blkbits))
3825                         new_size = offset + len;
3826                 else
3827                         new_size = (map.m_lblk + ret) << blkbits;
3828
3829                 ext4_falloc_update_inode(inode, mode, new_size,
3830                                          (map.m_flags & EXT4_MAP_NEW));
3831                 ext4_mark_inode_dirty(handle, inode);
3832                 ret2 = ext4_journal_stop(handle);
3833                 if (ret2)
3834                         break;
3835         }
3836         if (ret == -ENOSPC &&
3837                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3838                 ret = 0;
3839                 goto retry;
3840         }
3841         mutex_unlock(&inode->i_mutex);
3842         trace_ext4_fallocate_exit(inode, offset, max_blocks,
3843                                 ret > 0 ? ret2 : ret);
3844         return ret > 0 ? ret2 : ret;
3845 }
3846
3847 /*
3848  * This function convert a range of blocks to written extents
3849  * The caller of this function will pass the start offset and the size.
3850  * all unwritten extents within this range will be converted to
3851  * written extents.
3852  *
3853  * This function is called from the direct IO end io call back
3854  * function, to convert the fallocated extents after IO is completed.
3855  * Returns 0 on success.
3856  */
3857 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3858                                     ssize_t len)
3859 {
3860         handle_t *handle;
3861         unsigned int max_blocks;
3862         int ret = 0;
3863         int ret2 = 0;
3864         struct ext4_map_blocks map;
3865         unsigned int credits, blkbits = inode->i_blkbits;
3866
3867         map.m_lblk = offset >> blkbits;
3868         /*
3869          * We can't just convert len to max_blocks because
3870          * If blocksize = 4096 offset = 3072 and len = 2048
3871          */
3872         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3873                       map.m_lblk);
3874         /*
3875          * credits to insert 1 extent into extent tree
3876          */
3877         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3878         while (ret >= 0 && ret < max_blocks) {
3879                 map.m_lblk += ret;
3880                 map.m_len = (max_blocks -= ret);
3881                 handle = ext4_journal_start(inode, credits);
3882                 if (IS_ERR(handle)) {
3883                         ret = PTR_ERR(handle);
3884                         break;
3885                 }
3886                 ret = ext4_map_blocks(handle, inode, &map,
3887                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3888                 if (ret <= 0) {
3889                         WARN_ON(ret <= 0);
3890                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3891                                     "returned error inode#%lu, block=%u, "
3892                                     "max_blocks=%u", __func__,
3893                                     inode->i_ino, map.m_lblk, map.m_len);
3894                 }
3895                 ext4_mark_inode_dirty(handle, inode);
3896                 ret2 = ext4_journal_stop(handle);
3897                 if (ret <= 0 || ret2 )
3898                         break;
3899         }
3900         return ret > 0 ? ret2 : ret;
3901 }
3902
3903 /*
3904  * Callback function called for each extent to gather FIEMAP information.
3905  */
3906 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3907                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3908                        void *data)
3909 {
3910         __u64   logical;
3911         __u64   physical;
3912         __u64   length;
3913         loff_t  size;
3914         __u32   flags = 0;
3915         int             ret = 0;
3916         struct fiemap_extent_info *fieinfo = data;
3917         unsigned char blksize_bits;
3918
3919         blksize_bits = inode->i_sb->s_blocksize_bits;
3920         logical = (__u64)newex->ec_block << blksize_bits;
3921
3922         if (newex->ec_start == 0) {
3923                 /*
3924                  * No extent in extent-tree contains block @newex->ec_start,
3925                  * then the block may stay in 1)a hole or 2)delayed-extent.
3926                  *
3927                  * Holes or delayed-extents are processed as follows.
3928                  * 1. lookup dirty pages with specified range in pagecache.
3929                  *    If no page is got, then there is no delayed-extent and
3930                  *    return with EXT_CONTINUE.
3931                  * 2. find the 1st mapped buffer,
3932                  * 3. check if the mapped buffer is both in the request range
3933                  *    and a delayed buffer. If not, there is no delayed-extent,
3934                  *    then return.
3935                  * 4. a delayed-extent is found, the extent will be collected.
3936                  */
3937                 ext4_lblk_t     end = 0;
3938                 pgoff_t         last_offset;
3939                 pgoff_t         offset;
3940                 pgoff_t         index;
3941                 pgoff_t         start_index = 0;
3942                 struct page     **pages = NULL;
3943                 struct buffer_head *bh = NULL;
3944                 struct buffer_head *head = NULL;
3945                 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
3946
3947                 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
3948                 if (pages == NULL)
3949                         return -ENOMEM;
3950
3951                 offset = logical >> PAGE_SHIFT;
3952 repeat:
3953                 last_offset = offset;
3954                 head = NULL;
3955                 ret = find_get_pages_tag(inode->i_mapping, &offset,
3956                                         PAGECACHE_TAG_DIRTY, nr_pages, pages);
3957
3958                 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3959                         /* First time, try to find a mapped buffer. */
3960                         if (ret == 0) {
3961 out:
3962                                 for (index = 0; index < ret; index++)
3963                                         page_cache_release(pages[index]);
3964                                 /* just a hole. */
3965                                 kfree(pages);
3966                                 return EXT_CONTINUE;
3967                         }
3968                         index = 0;
3969
3970 next_page:
3971                         /* Try to find the 1st mapped buffer. */
3972                         end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
3973                                   blksize_bits;
3974                         if (!page_has_buffers(pages[index]))
3975                                 goto out;
3976                         head = page_buffers(pages[index]);
3977                         if (!head)
3978                                 goto out;
3979
3980                         index++;
3981                         bh = head;
3982                         do {
3983                                 if (end >= newex->ec_block +
3984                                         newex->ec_len)
3985                                         /* The buffer is out of
3986                                          * the request range.
3987                                          */
3988                                         goto out;
3989
3990                                 if (buffer_mapped(bh) &&
3991                                     end >= newex->ec_block) {
3992                                         start_index = index - 1;
3993                                         /* get the 1st mapped buffer. */
3994                                         goto found_mapped_buffer;
3995                                 }
3996
3997                                 bh = bh->b_this_page;
3998                                 end++;
3999                         } while (bh != head);
4000
4001                         /* No mapped buffer in the range found in this page,
4002                          * We need to look up next page.
4003                          */
4004                         if (index >= ret) {
4005                                 /* There is no page left, but we need to limit
4006                                  * newex->ec_len.
4007                                  */
4008                                 newex->ec_len = end - newex->ec_block;
4009                                 goto out;
4010                         }
4011                         goto next_page;
4012                 } else {
4013                         /*Find contiguous delayed buffers. */
4014                         if (ret > 0 && pages[0]->index == last_offset)
4015                                 head = page_buffers(pages[0]);
4016                         bh = head;
4017                         index = 1;
4018                         start_index = 0;
4019                 }
4020
4021 found_mapped_buffer:
4022                 if (bh != NULL && buffer_delay(bh)) {
4023                         /* 1st or contiguous delayed buffer found. */
4024                         if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4025                                 /*
4026                                  * 1st delayed buffer found, record
4027                                  * the start of extent.
4028                                  */
4029                                 flags |= FIEMAP_EXTENT_DELALLOC;
4030                                 newex->ec_block = end;
4031                                 logical = (__u64)end << blksize_bits;
4032                         }
4033                         /* Find contiguous delayed buffers. */
4034                         do {
4035                                 if (!buffer_delay(bh))
4036                                         goto found_delayed_extent;
4037                                 bh = bh->b_this_page;
4038                                 end++;
4039                         } while (bh != head);
4040
4041                         for (; index < ret; index++) {
4042                                 if (!page_has_buffers(pages[index])) {
4043                                         bh = NULL;
4044                                         break;
4045                                 }
4046                                 head = page_buffers(pages[index]);
4047                                 if (!head) {
4048                                         bh = NULL;
4049                                         break;
4050                                 }
4051
4052                                 if (pages[index]->index !=
4053                                     pages[start_index]->index + index
4054                                     - start_index) {
4055                                         /* Blocks are not contiguous. */
4056                                         bh = NULL;
4057                                         break;
4058                                 }
4059                                 bh = head;
4060                                 do {
4061                                         if (!buffer_delay(bh))
4062                                                 /* Delayed-extent ends. */
4063                                                 goto found_delayed_extent;
4064                                         bh = bh->b_this_page;
4065                                         end++;
4066                                 } while (bh != head);
4067                         }
4068                 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4069                         /* a hole found. */
4070                         goto out;
4071
4072 found_delayed_extent:
4073                 newex->ec_len = min(end - newex->ec_block,
4074                                                 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4075                 if (ret == nr_pages && bh != NULL &&
4076                         newex->ec_len < EXT_INIT_MAX_LEN &&
4077                         buffer_delay(bh)) {
4078                         /* Have not collected an extent and continue. */
4079                         for (index = 0; index < ret; index++)
4080                                 page_cache_release(pages[index]);
4081                         goto repeat;
4082                 }
4083
4084                 for (index = 0; index < ret; index++)
4085                         page_cache_release(pages[index]);
4086                 kfree(pages);
4087         }
4088
4089         physical = (__u64)newex->ec_start << blksize_bits;
4090         length =   (__u64)newex->ec_len << blksize_bits;
4091
4092         if (ex && ext4_ext_is_uninitialized(ex))
4093                 flags |= FIEMAP_EXTENT_UNWRITTEN;
4094
4095         size = i_size_read(inode);
4096         if (logical + length >= size)
4097                 flags |= FIEMAP_EXTENT_LAST;
4098
4099         ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4100                                         length, flags);
4101         if (ret < 0)
4102                 return ret;
4103         if (ret == 1)
4104                 return EXT_BREAK;
4105         return EXT_CONTINUE;
4106 }
4107
4108 /* fiemap flags we can handle specified here */
4109 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4110
4111 static int ext4_xattr_fiemap(struct inode *inode,
4112                                 struct fiemap_extent_info *fieinfo)
4113 {
4114         __u64 physical = 0;
4115         __u64 length;
4116         __u32 flags = FIEMAP_EXTENT_LAST;
4117         int blockbits = inode->i_sb->s_blocksize_bits;
4118         int error = 0;
4119
4120         /* in-inode? */
4121         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4122                 struct ext4_iloc iloc;
4123                 int offset;     /* offset of xattr in inode */
4124
4125                 error = ext4_get_inode_loc(inode, &iloc);
4126                 if (error)
4127                         return error;
4128                 physical = iloc.bh->b_blocknr << blockbits;
4129                 offset = EXT4_GOOD_OLD_INODE_SIZE +
4130                                 EXT4_I(inode)->i_extra_isize;
4131                 physical += offset;
4132                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4133                 flags |= FIEMAP_EXTENT_DATA_INLINE;
4134                 brelse(iloc.bh);
4135         } else { /* external block */
4136                 physical = EXT4_I(inode)->i_file_acl << blockbits;
4137                 length = inode->i_sb->s_blocksize;
4138         }
4139
4140         if (physical)
4141                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4142                                                 length, flags);
4143         return (error < 0 ? error : 0);
4144 }
4145
4146 /*
4147  * ext4_ext_punch_hole
4148  *
4149  * Punches a hole of "length" bytes in a file starting
4150  * at byte "offset"
4151  *
4152  * @inode:  The inode of the file to punch a hole in
4153  * @offset: The starting byte offset of the hole
4154  * @length: The length of the hole
4155  *
4156  * Returns the number of blocks removed or negative on err
4157  */
4158 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4159 {
4160         struct inode *inode = file->f_path.dentry->d_inode;
4161         struct super_block *sb = inode->i_sb;
4162         struct ext4_ext_cache cache_ex;
4163         ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4164         struct address_space *mapping = inode->i_mapping;
4165         struct ext4_map_blocks map;
4166         handle_t *handle;
4167         loff_t first_block_offset, last_block_offset, block_len;
4168         loff_t first_page, last_page, first_page_offset, last_page_offset;
4169         int ret, credits, blocks_released, err = 0;
4170
4171         first_block = (offset + sb->s_blocksize - 1) >>
4172                 EXT4_BLOCK_SIZE_BITS(sb);
4173         last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4174
4175         first_block_offset = first_block << EXT4_BLOCK_SIZE_BITS(sb);
4176         last_block_offset = last_block << EXT4_BLOCK_SIZE_BITS(sb);
4177
4178         first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4179         last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4180
4181         first_page_offset = first_page << PAGE_CACHE_SHIFT;
4182         last_page_offset = last_page << PAGE_CACHE_SHIFT;
4183
4184         /*
4185          * Write out all dirty pages to avoid race conditions
4186          * Then release them.
4187          */
4188         if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4189                 err = filemap_write_and_wait_range(mapping,
4190                         first_page_offset == 0 ? 0 : first_page_offset-1,
4191                         last_page_offset);
4192
4193                         if (err)
4194                                 return err;
4195         }
4196
4197         /* Now release the pages */
4198         if (last_page_offset > first_page_offset) {
4199                 truncate_inode_pages_range(mapping, first_page_offset,
4200                                            last_page_offset-1);
4201         }
4202
4203         /* finish any pending end_io work */
4204         ext4_flush_completed_IO(inode);
4205
4206         credits = ext4_writepage_trans_blocks(inode);
4207         handle = ext4_journal_start(inode, credits);
4208         if (IS_ERR(handle))
4209                 return PTR_ERR(handle);
4210
4211         err = ext4_orphan_add(handle, inode);
4212         if (err)
4213                 goto out;
4214
4215         /*
4216          * Now we need to zero out the un block aligned data.
4217          * If the file is smaller than a block, just
4218          * zero out the middle
4219          */
4220         if (first_block > last_block)
4221                 ext4_block_zero_page_range(handle, mapping, offset, length);
4222         else {
4223                 /* zero out the head of the hole before the first block */
4224                 block_len  = first_block_offset - offset;
4225                 if (block_len > 0)
4226                         ext4_block_zero_page_range(handle, mapping,
4227                                                    offset, block_len);
4228
4229                 /* zero out the tail of the hole after the last block */
4230                 block_len = offset + length - last_block_offset;
4231                 if (block_len > 0) {
4232                         ext4_block_zero_page_range(handle, mapping,
4233                                         last_block_offset, block_len);
4234                 }
4235         }
4236
4237         /* If there are no blocks to remove, return now */
4238         if (first_block >= last_block)
4239                 goto out;
4240
4241         down_write(&EXT4_I(inode)->i_data_sem);
4242         ext4_ext_invalidate_cache(inode);
4243         ext4_discard_preallocations(inode);
4244
4245         /*
4246          * Loop over all the blocks and identify blocks
4247          * that need to be punched out
4248          */
4249         iblock = first_block;
4250         blocks_released = 0;
4251         while (iblock < last_block) {
4252                 max_blocks = last_block - iblock;
4253                 num_blocks = 1;
4254                 memset(&map, 0, sizeof(map));
4255                 map.m_lblk = iblock;
4256                 map.m_len = max_blocks;
4257                 ret = ext4_ext_map_blocks(handle, inode, &map,
4258                         EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4259
4260                 if (ret > 0) {
4261                         blocks_released += ret;
4262                         num_blocks = ret;
4263                 } else if (ret == 0) {
4264                         /*
4265                          * If map blocks could not find the block,
4266                          * then it is in a hole.  If the hole was
4267                          * not already cached, then map blocks should
4268                          * put it in the cache.  So we can get the hole
4269                          * out of the cache
4270                          */
4271                         memset(&cache_ex, 0, sizeof(cache_ex));
4272                         if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4273                                 !cache_ex.ec_start) {
4274
4275                                 /* The hole is cached */
4276                                 num_blocks = cache_ex.ec_block +
4277                                 cache_ex.ec_len - iblock;
4278
4279                         } else {
4280                                 /* The block could not be identified */
4281                                 err = -EIO;
4282                                 break;
4283                         }
4284                 } else {
4285                         /* Map blocks error */
4286                         err = ret;
4287                         break;
4288                 }
4289
4290                 if (num_blocks == 0) {
4291                         /* This condition should never happen */
4292                         ext_debug("Block lookup failed");
4293                         err = -EIO;
4294                         break;
4295                 }
4296
4297                 iblock += num_blocks;
4298         }
4299
4300         if (blocks_released > 0) {
4301                 ext4_ext_invalidate_cache(inode);
4302                 ext4_discard_preallocations(inode);
4303         }
4304
4305         if (IS_SYNC(inode))
4306                 ext4_handle_sync(handle);
4307
4308         up_write(&EXT4_I(inode)->i_data_sem);
4309
4310 out:
4311         ext4_orphan_del(handle, inode);
4312         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4313         ext4_mark_inode_dirty(handle, inode);
4314         ext4_journal_stop(handle);
4315         return err;
4316 }
4317 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4318                 __u64 start, __u64 len)
4319 {
4320         ext4_lblk_t start_blk;
4321         int error = 0;
4322
4323         /* fallback to generic here if not in extents fmt */
4324         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4325                 return generic_block_fiemap(inode, fieinfo, start, len,
4326                         ext4_get_block);
4327
4328         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4329                 return -EBADR;
4330
4331         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4332                 error = ext4_xattr_fiemap(inode, fieinfo);
4333         } else {
4334                 ext4_lblk_t len_blks;
4335                 __u64 last_blk;
4336
4337                 start_blk = start >> inode->i_sb->s_blocksize_bits;
4338                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4339                 if (last_blk >= EXT_MAX_BLOCK)
4340                         last_blk = EXT_MAX_BLOCK-1;
4341                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4342
4343                 /*
4344                  * Walk the extent tree gathering extent information.
4345                  * ext4_ext_fiemap_cb will push extents back to user.
4346                  */
4347                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4348                                           ext4_ext_fiemap_cb, fieinfo);
4349         }
4350
4351         return error;
4352 }