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/fs.h>
  33 #include <linux/time.h>
  34 #include <linux/jbd2.h>
  35 #include <linux/highuid.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/string.h>
  39 #include <linux/slab.h>
  40 #include <linux/falloc.h>
  41 #include <asm/uaccess.h>
  42 #include <linux/fiemap.h>
  43 #include "ext4_jbd2.h"
  44
  45 #include <trace/events/ext4.h>
  46
  47 static int ext4_split_extent(handle_t *handle,
  48                                 struct inode *inode,
  49                                 struct ext4_ext_path *path,
  50                                 struct ext4_map_blocks *map,
  51                                 int split_flag,
  52                                 int flags);
  53
  54 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  55                                             struct inode *inode,
  56                                             int needed)
  57 {
  58         int err;
  59
  60         if (!ext4_handle_valid(handle))
  61                 return 0;
  62         if (handle->h_buffer_credits > needed)
  63                 return 0;
  64         err = ext4_journal_extend(handle, needed);
  65         if (err <= 0)
  66                 return err;
  67         err = ext4_truncate_restart_trans(handle, inode, needed);
  68         if (err == 0)
  69                 err = -EAGAIN;
  70
  71         return err;
  72 }
  73
  74 /*
  75  * could return:
  76  *  - EROFS
  77  *  - ENOMEM
  78  */
  79 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  80                                 struct ext4_ext_path *path)
  81 {
  82         if (path->p_bh) {
  83                 /* path points to block */
  84                 return ext4_journal_get_write_access(handle, path->p_bh);
  85         }
  86         /* path points to leaf/index in inode body */
  87         /* we use in-core data, no need to protect them */
  88         return 0;
  89 }
  90
  91 /*
  92  * could return:
  93  *  - EROFS
  94  *  - ENOMEM
  95  *  - EIO
  96  */
  97 #define ext4_ext_dirty(handle, inode, path) \
  98                 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
  99 static int __ext4_ext_dirty(const char *where, unsigned int line,
 100                             handle_t *handle, struct inode *inode,
 101                             struct ext4_ext_path *path)
 102 {
 103         int err;
 104         if (path->p_bh) {
 105                 /* path points to block */
 106                 err = __ext4_handle_dirty_metadata(where, line, handle,
 107                                                    inode, path->p_bh);
 108         } else {
 109                 /* path points to leaf/index in inode body */
 110                 err = ext4_mark_inode_dirty(handle, inode);
 111         }
 112         return err;
 113 }
 114
 115 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 116                               struct ext4_ext_path *path,
 117                               ext4_lblk_t block)
 118 {
 119         if (path) {
 120                 int depth = path->p_depth;
 121                 struct ext4_extent *ex;
 122
 123                 /*
 124                  * Try to predict block placement assuming that we are
 125                  * filling in a file which will eventually be
 126                  * non-sparse --- i.e., in the case of libbfd writing
 127                  * an ELF object sections out-of-order but in a way
 128                  * the eventually results in a contiguous object or
 129                  * executable file, or some database extending a table
 130                  * space file.  However, this is actually somewhat
 131                  * non-ideal if we are writing a sparse file such as
 132                  * qemu or KVM writing a raw image file that is going
 133                  * to stay fairly sparse, since it will end up
 134                  * fragmenting the file system's free space.  Maybe we
 135                  * should have some hueristics or some way to allow
 136                  * userspace to pass a hint to file system,
 137                  * especially if the latter case turns out to be
 138                  * common.
 139                  */
 140                 ex = path[depth].p_ext;
 141                 if (ex) {
 142                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 143                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 144
 145                         if (block > ext_block)
 146                                 return ext_pblk + (block - ext_block);
 147                         else
 148                                 return ext_pblk - (ext_block - block);
 149                 }
 150
 151                 /* it looks like index is empty;
 152                  * try to find starting block from index itself */
 153                 if (path[depth].p_bh)
 154                         return path[depth].p_bh->b_blocknr;
 155         }
 156
 157         /* OK. use inode's group */
 158         return ext4_inode_to_goal_block(inode);
 159 }
 160
 161 /*
 162  * Allocation for a meta data block
 163  */
 164 static ext4_fsblk_t
 165 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 166                         struct ext4_ext_path *path,
 167                         struct ext4_extent *ex, int *err, unsigned int flags)
 168 {
 169         ext4_fsblk_t goal, newblock;
 170
 171         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 172         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 173                                         NULL, err);
 174         return newblock;
 175 }
 176
 177 static inline int ext4_ext_space_block(struct inode *inode, int check)
 178 {
 179         int size;
 180
 181         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 182                         / sizeof(struct ext4_extent);
 183 #ifdef AGGRESSIVE_TEST
 184         if (!check && size > 6)
 185                 size = 6;
 186 #endif
 187         return size;
 188 }
 189
 190 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 191 {
 192         int size;
 193
 194         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 195                         / sizeof(struct ext4_extent_idx);
 196 #ifdef AGGRESSIVE_TEST
 197         if (!check && size > 5)
 198                 size = 5;
 199 #endif
 200         return size;
 201 }
 202
 203 static inline int ext4_ext_space_root(struct inode *inode, int check)
 204 {
 205         int size;
 206
 207         size = sizeof(EXT4_I(inode)->i_data);
 208         size -= sizeof(struct ext4_extent_header);
 209         size /= sizeof(struct ext4_extent);
 210 #ifdef AGGRESSIVE_TEST
 211         if (!check && size > 3)
 212                 size = 3;
 213 #endif
 214         return size;
 215 }
 216
 217 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 218 {
 219         int size;
 220
 221         size = sizeof(EXT4_I(inode)->i_data);
 222         size -= sizeof(struct ext4_extent_header);
 223         size /= sizeof(struct ext4_extent_idx);
 224 #ifdef AGGRESSIVE_TEST
 225         if (!check && size > 4)
 226                 size = 4;
 227 #endif
 228         return size;
 229 }
 230
 231 /*
 232  * Calculate the number of metadata blocks needed
 233  * to allocate @blocks
 234  * Worse case is one block per extent
 235  */
 236 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 237 {
 238         struct ext4_inode_info *ei = EXT4_I(inode);
 239         int idxs;
 240
 241         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 242                 / sizeof(struct ext4_extent_idx));
 243
 244         /*
 245          * If the new delayed allocation block is contiguous with the
 246          * previous da block, it can share index blocks with the
 247          * previous block, so we only need to allocate a new index
 248          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 249          * an additional index block, and at ldxs**3 blocks, yet
 250          * another index blocks.
 251          */
 252         if (ei->i_da_metadata_calc_len &&
 253             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 254                 int num = 0;
 255
 256                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 257                         num++;
 258                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 259                         num++;
 260                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 261                         num++;
 262                         ei->i_da_metadata_calc_len = 0;
 263                 } else
 264                         ei->i_da_metadata_calc_len++;
 265                 ei->i_da_metadata_calc_last_lblock++;
 266                 return num;
 267         }
 268
 269         /*
 270          * In the worst case we need a new set of index blocks at
 271          * every level of the inode's extent tree.
 272          */
 273         ei->i_da_metadata_calc_len = 1;
 274         ei->i_da_metadata_calc_last_lblock = lblock;
 275         return ext_depth(inode) + 1;
 276 }
 277
 278 static int
 279 ext4_ext_max_entries(struct inode *inode, int depth)
 280 {
 281         int max;
 282
 283         if (depth == ext_depth(inode)) {
 284                 if (depth == 0)
 285                         max = ext4_ext_space_root(inode, 1);
 286                 else
 287                         max = ext4_ext_space_root_idx(inode, 1);
 288         } else {
 289                 if (depth == 0)
 290                         max = ext4_ext_space_block(inode, 1);
 291                 else
 292                         max = ext4_ext_space_block_idx(inode, 1);
 293         }
 294
 295         return max;
 296 }
 297
 298 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 299 {
 300         ext4_fsblk_t block = ext4_ext_pblock(ext);
 301         int len = ext4_ext_get_actual_len(ext);
 302
 303         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 304 }
 305
 306 static int ext4_valid_extent_idx(struct inode *inode,
 307                                 struct ext4_extent_idx *ext_idx)
 308 {
 309         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 310
 311         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 312 }
 313
 314 static int ext4_valid_extent_entries(struct inode *inode,
 315                                 struct ext4_extent_header *eh,
 316                                 int depth)
 317 {
 318         unsigned short entries;
 319         if (eh->eh_entries == 0)
 320                 return 1;
 321
 322         entries = le16_to_cpu(eh->eh_entries);
 323
 324         if (depth == 0) {
 325                 /* leaf entries */
 326                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 327                 while (entries) {
 328                         if (!ext4_valid_extent(inode, ext))
 329                                 return 0;
 330                         ext++;
 331                         entries--;
 332                 }
 333         } else {
 334                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 335                 while (entries) {
 336                         if (!ext4_valid_extent_idx(inode, ext_idx))
 337                                 return 0;
 338                         ext_idx++;
 339                         entries--;
 340                 }
 341         }
 342         return 1;
 343 }
 344
 345 static int __ext4_ext_check(const char *function, unsigned int line,
 346                             struct inode *inode, struct ext4_extent_header *eh,
 347                             int depth)
 348 {
 349         const char *error_msg;
 350         int max = 0;
 351
 352         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 353                 error_msg = "invalid magic";
 354                 goto corrupted;
 355         }
 356         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 357                 error_msg = "unexpected eh_depth";
 358                 goto corrupted;
 359         }
 360         if (unlikely(eh->eh_max == 0)) {
 361                 error_msg = "invalid eh_max";
 362                 goto corrupted;
 363         }
 364         max = ext4_ext_max_entries(inode, depth);
 365         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 366                 error_msg = "too large eh_max";
 367                 goto corrupted;
 368         }
 369         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 370                 error_msg = "invalid eh_entries";
 371                 goto corrupted;
 372         }
 373         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 374                 error_msg = "invalid extent entries";
 375                 goto corrupted;
 376         }
 377         return 0;
 378
 379 corrupted:
 380         ext4_error_inode(inode, function, line, 0,
 381                         "bad header/extent: %s - magic %x, "
 382                         "entries %u, max %u(%u), depth %u(%u)",
 383                         error_msg, le16_to_cpu(eh->eh_magic),
 384                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 385                         max, le16_to_cpu(eh->eh_depth), depth);
 386
 387         return -EIO;
 388 }
 389
 390 #define ext4_ext_check(inode, eh, depth)        \
 391         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 392
 393 int ext4_ext_check_inode(struct inode *inode)
 394 {
 395         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 396 }
 397
 398 #ifdef EXT_DEBUG
 399 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 400 {
 401         int k, l = path->p_depth;
 402
 403         ext_debug("path:");
 404         for (k = 0; k <= l; k++, path++) {
 405                 if (path->p_idx) {
 406                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 407                             ext4_idx_pblock(path->p_idx));
 408                 } else if (path->p_ext) {
 409                         ext_debug("  %d:[%d]%d:%llu ",
 410                                   le32_to_cpu(path->p_ext->ee_block),
 411                                   ext4_ext_is_uninitialized(path->p_ext),
 412                                   ext4_ext_get_actual_len(path->p_ext),
 413                                   ext4_ext_pblock(path->p_ext));
 414                 } else
 415                         ext_debug("  []");
 416         }
 417         ext_debug("\n");
 418 }
 419
 420 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 421 {
 422         int depth = ext_depth(inode);
 423         struct ext4_extent_header *eh;
 424         struct ext4_extent *ex;
 425         int i;
 426
 427         if (!path)
 428                 return;
 429
 430         eh = path[depth].p_hdr;
 431         ex = EXT_FIRST_EXTENT(eh);
 432
 433         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 434
 435         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 436                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 437                           ext4_ext_is_uninitialized(ex),
 438                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 439         }
 440         ext_debug("\n");
 441 }
 442
 443 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 444                         ext4_fsblk_t newblock, int level)
 445 {
 446         int depth = ext_depth(inode);
 447         struct ext4_extent *ex;
 448
 449         if (depth != level) {
 450                 struct ext4_extent_idx *idx;
 451                 idx = path[level].p_idx;
 452                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 453                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
 454                                         le32_to_cpu(idx->ei_block),
 455                                         ext4_idx_pblock(idx),
 456                                         newblock);
 457                         idx++;
 458                 }
 459
 460                 return;
 461         }
 462
 463         ex = path[depth].p_ext;
 464         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 465                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 466                                 le32_to_cpu(ex->ee_block),
 467                                 ext4_ext_pblock(ex),
 468                                 ext4_ext_is_uninitialized(ex),
 469                                 ext4_ext_get_actual_len(ex),
 470                                 newblock);
 471                 ex++;
 472         }
 473 }
 474
 475 #else
 476 #define ext4_ext_show_path(inode, path)
 477 #define ext4_ext_show_leaf(inode, path)
 478 #define ext4_ext_show_move(inode, path, newblock, level)
 479 #endif
 480
 481 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 482 {
 483         int depth = path->p_depth;
 484         int i;
 485
 486         for (i = 0; i <= depth; i++, path++)
 487                 if (path->p_bh) {
 488                         brelse(path->p_bh);
 489                         path->p_bh = NULL;
 490                 }
 491 }
 492
 493 /*
 494  * ext4_ext_binsearch_idx:
 495  * binary search for the closest index of the given block
 496  * the header must be checked before calling this
 497  */
 498 static void
 499 ext4_ext_binsearch_idx(struct inode *inode,
 500                         struct ext4_ext_path *path, ext4_lblk_t block)
 501 {
 502         struct ext4_extent_header *eh = path->p_hdr;
 503         struct ext4_extent_idx *r, *l, *m;
 504
 505
 506         ext_debug("binsearch for %u(idx):  ", block);
 507
 508         l = EXT_FIRST_INDEX(eh) + 1;
 509         r = EXT_LAST_INDEX(eh);
 510         while (l <= r) {
 511                 m = l + (r - l) / 2;
 512                 if (block < le32_to_cpu(m->ei_block))
 513                         r = m - 1;
 514                 else
 515                         l = m + 1;
 516                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 517                                 m, le32_to_cpu(m->ei_block),
 518                                 r, le32_to_cpu(r->ei_block));
 519         }
 520
 521         path->p_idx = l - 1;
 522         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 523                   ext4_idx_pblock(path->p_idx));
 524
 525 #ifdef CHECK_BINSEARCH
 526         {
 527                 struct ext4_extent_idx *chix, *ix;
 528                 int k;
 529
 530                 chix = ix = EXT_FIRST_INDEX(eh);
 531                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 532                   if (k != 0 &&
 533                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 534                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 535                                        "first=0x%p\n", k,
 536                                        ix, EXT_FIRST_INDEX(eh));
 537                                 printk(KERN_DEBUG "%u <= %u\n",
 538                                        le32_to_cpu(ix->ei_block),
 539                                        le32_to_cpu(ix[-1].ei_block));
 540                         }
 541                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 542                                            <= le32_to_cpu(ix[-1].ei_block));
 543                         if (block < le32_to_cpu(ix->ei_block))
 544                                 break;
 545                         chix = ix;
 546                 }
 547                 BUG_ON(chix != path->p_idx);
 548         }
 549 #endif
 550
 551 }
 552
 553 /*
 554  * ext4_ext_binsearch:
 555  * binary search for closest extent of the given block
 556  * the header must be checked before calling this
 557  */
 558 static void
 559 ext4_ext_binsearch(struct inode *inode,
 560                 struct ext4_ext_path *path, ext4_lblk_t block)
 561 {
 562         struct ext4_extent_header *eh = path->p_hdr;
 563         struct ext4_extent *r, *l, *m;
 564
 565         if (eh->eh_entries == 0) {
 566                 /*
 567                  * this leaf is empty:
 568                  * we get such a leaf in split/add case
 569                  */
 570                 return;
 571         }
 572
 573         ext_debug("binsearch for %u:  ", block);
 574
 575         l = EXT_FIRST_EXTENT(eh) + 1;
 576         r = EXT_LAST_EXTENT(eh);
 577
 578         while (l <= r) {
 579                 m = l + (r - l) / 2;
 580                 if (block < le32_to_cpu(m->ee_block))
 581                         r = m - 1;
 582                 else
 583                         l = m + 1;
 584                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 585                                 m, le32_to_cpu(m->ee_block),
 586                                 r, le32_to_cpu(r->ee_block));
 587         }
 588
 589         path->p_ext = l - 1;
 590         ext_debug("  -> %d:%llu:[%d]%d ",
 591                         le32_to_cpu(path->p_ext->ee_block),
 592                         ext4_ext_pblock(path->p_ext),
 593                         ext4_ext_is_uninitialized(path->p_ext),
 594                         ext4_ext_get_actual_len(path->p_ext));
 595
 596 #ifdef CHECK_BINSEARCH
 597         {
 598                 struct ext4_extent *chex, *ex;
 599                 int k;
 600
 601                 chex = ex = EXT_FIRST_EXTENT(eh);
 602                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 603                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 604                                           <= le32_to_cpu(ex[-1].ee_block));
 605                         if (block < le32_to_cpu(ex->ee_block))
 606                                 break;
 607                         chex = ex;
 608                 }
 609                 BUG_ON(chex != path->p_ext);
 610         }
 611 #endif
 612
 613 }
 614
 615 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 616 {
 617         struct ext4_extent_header *eh;
 618
 619         eh = ext_inode_hdr(inode);
 620         eh->eh_depth = 0;
 621         eh->eh_entries = 0;
 622         eh->eh_magic = EXT4_EXT_MAGIC;
 623         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 624         ext4_mark_inode_dirty(handle, inode);
 625         ext4_ext_invalidate_cache(inode);
 626         return 0;
 627 }
 628
 629 struct ext4_ext_path *
 630 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 631                                         struct ext4_ext_path *path)
 632 {
 633         struct ext4_extent_header *eh;
 634         struct buffer_head *bh;
 635         short int depth, i, ppos = 0, alloc = 0;
 636
 637         eh = ext_inode_hdr(inode);
 638         depth = ext_depth(inode);
 639
 640         /* account possible depth increase */
 641         if (!path) {
 642                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 643                                 GFP_NOFS);
 644                 if (!path)
 645                         return ERR_PTR(-ENOMEM);
 646                 alloc = 1;
 647         }
 648         path[0].p_hdr = eh;
 649         path[0].p_bh = NULL;
 650
 651         i = depth;
 652         /* walk through the tree */
 653         while (i) {
 654                 int need_to_validate = 0;
 655
 656                 ext_debug("depth %d: num %d, max %d\n",
 657                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 658
 659                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 660                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 661                 path[ppos].p_depth = i;
 662                 path[ppos].p_ext = NULL;
 663
 664                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 665                 if (unlikely(!bh))
 666                         goto err;
 667                 if (!bh_uptodate_or_lock(bh)) {
 668                         trace_ext4_ext_load_extent(inode, block,
 669                                                 path[ppos].p_block);
 670                         if (bh_submit_read(bh) < 0) {
 671                                 put_bh(bh);
 672                                 goto err;
 673                         }
 674                         /* validate the extent entries */
 675                         need_to_validate = 1;
 676                 }
 677                 eh = ext_block_hdr(bh);
 678                 ppos++;
 679                 if (unlikely(ppos > depth)) {
 680                         put_bh(bh);
 681                         EXT4_ERROR_INODE(inode,
 682                                          "ppos %d > depth %d", ppos, depth);
 683                         goto err;
 684                 }
 685                 path[ppos].p_bh = bh;
 686                 path[ppos].p_hdr = eh;
 687                 i--;
 688
 689                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 690                         goto err;
 691         }
 692
 693         path[ppos].p_depth = i;
 694         path[ppos].p_ext = NULL;
 695         path[ppos].p_idx = NULL;
 696
 697         /* find extent */
 698         ext4_ext_binsearch(inode, path + ppos, block);
 699         /* if not an empty leaf */
 700         if (path[ppos].p_ext)
 701                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 702
 703         ext4_ext_show_path(inode, path);
 704
 705         return path;
 706
 707 err:
 708         ext4_ext_drop_refs(path);
 709         if (alloc)
 710                 kfree(path);
 711         return ERR_PTR(-EIO);
 712 }
 713
 714 /*
 715  * ext4_ext_insert_index:
 716  * insert new index [@logical;@ptr] into the block at @curp;
 717  * check where to insert: before @curp or after @curp
 718  */
 719 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 720                                  struct ext4_ext_path *curp,
 721                                  int logical, ext4_fsblk_t ptr)
 722 {
 723         struct ext4_extent_idx *ix;
 724         int len, err;
 725
 726         err = ext4_ext_get_access(handle, inode, curp);
 727         if (err)
 728                 return err;
 729
 730         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 731                 EXT4_ERROR_INODE(inode,
 732                                  "logical %d == ei_block %d!",
 733                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 734                 return -EIO;
 735         }
 736
 737         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 738                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 739                 EXT4_ERROR_INODE(inode,
 740                                  "eh_entries %d >= eh_max %d!",
 741                                  le16_to_cpu(curp->p_hdr->eh_entries),
 742                                  le16_to_cpu(curp->p_hdr->eh_max));
 743                 return -EIO;
 744         }
 745
 746         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 747                 /* insert after */
 748                 ext_debug("insert new index %d after: %llu\n", logical, ptr);
 749                 ix = curp->p_idx + 1;
 750         } else {
 751                 /* insert before */
 752                 ext_debug("insert new index %d before: %llu\n", logical, ptr);
 753                 ix = curp->p_idx;
 754         }
 755
 756         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 757         BUG_ON(len < 0);
 758         if (len > 0) {
 759                 ext_debug("insert new index %d: "
 760                                 "move %d indices from 0x%p to 0x%p\n",
 761                                 logical, len, ix, ix + 1);
 762                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 763         }
 764
 765         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 766                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 767                 return -EIO;
 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(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 775                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 776                 return -EIO;
 777         }
 778
 779         err = ext4_ext_dirty(handle, inode, curp);
 780         ext4_std_error(inode->i_sb, err);
 781
 782         return err;
 783 }
 784
 785 /*
 786  * ext4_ext_split:
 787  * inserts new subtree into the path, using free index entry
 788  * at depth @at:
 789  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 790  * - makes decision where to split
 791  * - moves remaining extents and index entries (right to the split point)
 792  *   into the newly allocated blocks
 793  * - initializes subtree
 794  */
 795 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 796                           unsigned int flags,
 797                           struct ext4_ext_path *path,
 798                           struct ext4_extent *newext, int at)
 799 {
 800         struct buffer_head *bh = NULL;
 801         int depth = ext_depth(inode);
 802         struct ext4_extent_header *neh;
 803         struct ext4_extent_idx *fidx;
 804         int i = at, k, m, a;
 805         ext4_fsblk_t newblock, oldblock;
 806         __le32 border;
 807         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 808         int err = 0;
 809
 810         /* make decision: where to split? */
 811         /* FIXME: now decision is simplest: at current extent */
 812
 813         /* if current leaf will be split, then we should use
 814          * border from split point */
 815         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 816                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 817                 return -EIO;
 818         }
 819         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 820                 border = path[depth].p_ext[1].ee_block;
 821                 ext_debug("leaf will be split."
 822                                 " next leaf starts at %d\n",
 823                                   le32_to_cpu(border));
 824         } else {
 825                 border = newext->ee_block;
 826                 ext_debug("leaf will be added."
 827                                 " next leaf starts at %d\n",
 828                                 le32_to_cpu(border));
 829         }
 830
 831         /*
 832          * If error occurs, then we break processing
 833          * and mark filesystem read-only. index won't
 834          * be inserted and tree will be in consistent
 835          * state. Next mount will repair buffers too.
 836          */
 837
 838         /*
 839          * Get array to track all allocated blocks.
 840          * We need this to handle errors and free blocks
 841          * upon them.
 842          */
 843         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 844         if (!ablocks)
 845                 return -ENOMEM;
 846
 847         /* allocate all needed blocks */
 848         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 849         for (a = 0; a < depth - at; a++) {
 850                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 851                                                    newext, &err, flags);
 852                 if (newblock == 0)
 853                         goto cleanup;
 854                 ablocks[a] = newblock;
 855         }
 856
 857         /* initialize new leaf */
 858         newblock = ablocks[--a];
 859         if (unlikely(newblock == 0)) {
 860                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 861                 err = -EIO;
 862                 goto cleanup;
 863         }
 864         bh = sb_getblk(inode->i_sb, newblock);
 865         if (!bh) {
 866                 err = -EIO;
 867                 goto cleanup;
 868         }
 869         lock_buffer(bh);
 870
 871         err = ext4_journal_get_create_access(handle, bh);
 872         if (err)
 873                 goto cleanup;
 874
 875         neh = ext_block_hdr(bh);
 876         neh->eh_entries = 0;
 877         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 878         neh->eh_magic = EXT4_EXT_MAGIC;
 879         neh->eh_depth = 0;
 880
 881         /* move remainder of path[depth] to the new leaf */
 882         if (unlikely(path[depth].p_hdr->eh_entries !=
 883                      path[depth].p_hdr->eh_max)) {
 884                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 885                                  path[depth].p_hdr->eh_entries,
 886                                  path[depth].p_hdr->eh_max);
 887                 err = -EIO;
 888                 goto cleanup;
 889         }
 890         /* start copy from next extent */
 891         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 892         ext4_ext_show_move(inode, path, newblock, depth);
 893         if (m) {
 894                 struct ext4_extent *ex;
 895                 ex = EXT_FIRST_EXTENT(neh);
 896                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 897                 le16_add_cpu(&neh->eh_entries, m);
 898         }
 899
 900         set_buffer_uptodate(bh);
 901         unlock_buffer(bh);
 902
 903         err = ext4_handle_dirty_metadata(handle, inode, bh);
 904         if (err)
 905                 goto cleanup;
 906         brelse(bh);
 907         bh = NULL;
 908
 909         /* correct old leaf */
 910         if (m) {
 911                 err = ext4_ext_get_access(handle, inode, path + depth);
 912                 if (err)
 913                         goto cleanup;
 914                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 915                 err = ext4_ext_dirty(handle, inode, path + depth);
 916                 if (err)
 917                         goto cleanup;
 918
 919         }
 920
 921         /* create intermediate indexes */
 922         k = depth - at - 1;
 923         if (unlikely(k < 0)) {
 924                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 925                 err = -EIO;
 926                 goto cleanup;
 927         }
 928         if (k)
 929                 ext_debug("create %d intermediate indices\n", k);
 930         /* insert new index into current index block */
 931         /* current depth stored in i var */
 932         i = depth - 1;
 933         while (k--) {
 934                 oldblock = newblock;
 935                 newblock = ablocks[--a];
 936                 bh = sb_getblk(inode->i_sb, newblock);
 937                 if (!bh) {
 938                         err = -EIO;
 939                         goto cleanup;
 940                 }
 941                 lock_buffer(bh);
 942
 943                 err = ext4_journal_get_create_access(handle, bh);
 944                 if (err)
 945                         goto cleanup;
 946
 947                 neh = ext_block_hdr(bh);
 948                 neh->eh_entries = cpu_to_le16(1);
 949                 neh->eh_magic = EXT4_EXT_MAGIC;
 950                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 951                 neh->eh_depth = cpu_to_le16(depth - i);
 952                 fidx = EXT_FIRST_INDEX(neh);
 953                 fidx->ei_block = border;
 954                 ext4_idx_store_pblock(fidx, oldblock);
 955
 956                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 957                                 i, newblock, le32_to_cpu(border), oldblock);
 958
 959                 /* move remainder of path[i] to the new index block */
 960                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 961                                         EXT_LAST_INDEX(path[i].p_hdr))) {
 962                         EXT4_ERROR_INODE(inode,
 963                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 964                                          le32_to_cpu(path[i].p_ext->ee_block));
 965                         err = -EIO;
 966                         goto cleanup;
 967                 }
 968                 /* start copy indexes */
 969                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
 970                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 971                                 EXT_MAX_INDEX(path[i].p_hdr));
 972                 ext4_ext_show_move(inode, path, newblock, i);
 973                 if (m) {
 974                         memmove(++fidx, path[i].p_idx,
 975                                 sizeof(struct ext4_extent_idx) * m);
 976                         le16_add_cpu(&neh->eh_entries, m);
 977                 }
 978                 set_buffer_uptodate(bh);
 979                 unlock_buffer(bh);
 980
 981                 err = ext4_handle_dirty_metadata(handle, inode, bh);
 982                 if (err)
 983                         goto cleanup;
 984                 brelse(bh);
 985                 bh = NULL;
 986
 987                 /* correct old index */
 988                 if (m) {
 989                         err = ext4_ext_get_access(handle, inode, path + i);
 990                         if (err)
 991                                 goto cleanup;
 992                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
 993                         err = ext4_ext_dirty(handle, inode, path + i);
 994                         if (err)
 995                                 goto cleanup;
 996                 }
 997
 998                 i--;
 999         }
1000
1001         /* insert new index */
1002         err = ext4_ext_insert_index(handle, inode, path + at,
1003                                     le32_to_cpu(border), newblock);
1004
1005 cleanup:
1006         if (bh) {
1007                 if (buffer_locked(bh))
1008                         unlock_buffer(bh);
1009                 brelse(bh);
1010         }
1011
1012         if (err) {
1013                 /* free all allocated blocks in error case */
1014                 for (i = 0; i < depth; i++) {
1015                         if (!ablocks[i])
1016                                 continue;
1017                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1018                                          EXT4_FREE_BLOCKS_METADATA);
1019                 }
1020         }
1021         kfree(ablocks);
1022
1023         return err;
1024 }
1025
1026 /*
1027  * ext4_ext_grow_indepth:
1028  * implements tree growing procedure:
1029  * - allocates new block
1030  * - moves top-level data (index block or leaf) into the new block
1031  * - initializes new top-level, creating index that points to the
1032  *   just created block
1033  */
1034 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1035                                  unsigned int flags,
1036                                  struct ext4_extent *newext)
1037 {
1038         struct ext4_extent_header *neh;
1039         struct buffer_head *bh;
1040         ext4_fsblk_t newblock;
1041         int err = 0;
1042
1043         newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1044                 newext, &err, flags);
1045         if (newblock == 0)
1046                 return err;
1047
1048         bh = sb_getblk(inode->i_sb, newblock);
1049         if (!bh) {
1050                 err = -EIO;
1051                 ext4_std_error(inode->i_sb, err);
1052                 return err;
1053         }
1054         lock_buffer(bh);
1055
1056         err = ext4_journal_get_create_access(handle, bh);
1057         if (err) {
1058                 unlock_buffer(bh);
1059                 goto out;
1060         }
1061
1062         /* move top-level index/leaf into new block */
1063         memmove(bh->b_data, EXT4_I(inode)->i_data,
1064                 sizeof(EXT4_I(inode)->i_data));
1065
1066         /* set size of new block */
1067         neh = ext_block_hdr(bh);
1068         /* old root could have indexes or leaves
1069          * so calculate e_max right way */
1070         if (ext_depth(inode))
1071                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1072         else
1073                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1074         neh->eh_magic = EXT4_EXT_MAGIC;
1075         set_buffer_uptodate(bh);
1076         unlock_buffer(bh);
1077
1078         err = ext4_handle_dirty_metadata(handle, inode, bh);
1079         if (err)
1080                 goto out;
1081
1082         /* Update top-level index: num,max,pointer */
1083         neh = ext_inode_hdr(inode);
1084         neh->eh_entries = cpu_to_le16(1);
1085         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1086         if (neh->eh_depth == 0) {
1087                 /* Root extent block becomes index block */
1088                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1089                 EXT_FIRST_INDEX(neh)->ei_block =
1090                         EXT_FIRST_EXTENT(neh)->ee_block;
1091         }
1092         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1093                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1094                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1095                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1096
1097         neh->eh_depth = cpu_to_le16(neh->eh_depth + 1);
1098         ext4_mark_inode_dirty(handle, inode);
1099 out:
1100         brelse(bh);
1101
1102         return err;
1103 }
1104
1105 /*
1106  * ext4_ext_create_new_leaf:
1107  * finds empty index and adds new leaf.
1108  * if no free index is found, then it requests in-depth growing.
1109  */
1110 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1111                                     unsigned int flags,
1112                                     struct ext4_ext_path *path,
1113                                     struct ext4_extent *newext)
1114 {
1115         struct ext4_ext_path *curp;
1116         int depth, i, err = 0;
1117
1118 repeat:
1119         i = depth = ext_depth(inode);
1120
1121         /* walk up to the tree and look for free index entry */
1122         curp = path + depth;
1123         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1124                 i--;
1125                 curp--;
1126         }
1127
1128         /* we use already allocated block for index block,
1129          * so subsequent data blocks should be contiguous */
1130         if (EXT_HAS_FREE_INDEX(curp)) {
1131                 /* if we found index with free entry, then use that
1132                  * entry: create all needed subtree and add new leaf */
1133                 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1134                 if (err)
1135                         goto out;
1136
1137                 /* refill path */
1138                 ext4_ext_drop_refs(path);
1139                 path = ext4_ext_find_extent(inode,
1140                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1141                                     path);
1142                 if (IS_ERR(path))
1143                         err = PTR_ERR(path);
1144         } else {
1145                 /* tree is full, time to grow in depth */
1146                 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1147                 if (err)
1148                         goto out;
1149
1150                 /* refill path */
1151                 ext4_ext_drop_refs(path);
1152                 path = ext4_ext_find_extent(inode,
1153                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1154                                     path);
1155                 if (IS_ERR(path)) {
1156                         err = PTR_ERR(path);
1157                         goto out;
1158                 }
1159
1160                 /*
1161                  * only first (depth 0 -> 1) produces free space;
1162                  * in all other cases we have to split the grown tree
1163                  */
1164                 depth = ext_depth(inode);
1165                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1166                         /* now we need to split */
1167                         goto repeat;
1168                 }
1169         }
1170
1171 out:
1172         return err;
1173 }
1174
1175 /*
1176  * search the closest allocated block to the left for *logical
1177  * and returns it at @logical + it's physical address at @phys
1178  * if *logical is the smallest allocated block, the function
1179  * returns 0 at @phys
1180  * return value contains 0 (success) or error code
1181  */
1182 static int ext4_ext_search_left(struct inode *inode,
1183                                 struct ext4_ext_path *path,
1184                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1185 {
1186         struct ext4_extent_idx *ix;
1187         struct ext4_extent *ex;
1188         int depth, ee_len;
1189
1190         if (unlikely(path == NULL)) {
1191                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1192                 return -EIO;
1193         }
1194         depth = path->p_depth;
1195         *phys = 0;
1196
1197         if (depth == 0 && path->p_ext == NULL)
1198                 return 0;
1199
1200         /* usually extent in the path covers blocks smaller
1201          * then *logical, but it can be that extent is the
1202          * first one in the file */
1203
1204         ex = path[depth].p_ext;
1205         ee_len = ext4_ext_get_actual_len(ex);
1206         if (*logical < le32_to_cpu(ex->ee_block)) {
1207                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1208                         EXT4_ERROR_INODE(inode,
1209                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1210                                          *logical, le32_to_cpu(ex->ee_block));
1211                         return -EIO;
1212                 }
1213                 while (--depth >= 0) {
1214                         ix = path[depth].p_idx;
1215                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1216                                 EXT4_ERROR_INODE(inode,
1217                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1218                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1219                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1220                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1221                                   depth);
1222                                 return -EIO;
1223                         }
1224                 }
1225                 return 0;
1226         }
1227
1228         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1229                 EXT4_ERROR_INODE(inode,
1230                                  "logical %d < ee_block %d + ee_len %d!",
1231                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1232                 return -EIO;
1233         }
1234
1235         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1236         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1237         return 0;
1238 }
1239
1240 /*
1241  * search the closest allocated block to the right for *logical
1242  * and returns it at @logical + it's physical address at @phys
1243  * if *logical is the largest allocated block, the function
1244  * returns 0 at @phys
1245  * return value contains 0 (success) or error code
1246  */
1247 static int ext4_ext_search_right(struct inode *inode,
1248                                  struct ext4_ext_path *path,
1249                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1250                                  struct ext4_extent **ret_ex)
1251 {
1252         struct buffer_head *bh = NULL;
1253         struct ext4_extent_header *eh;
1254         struct ext4_extent_idx *ix;
1255         struct ext4_extent *ex;
1256         ext4_fsblk_t block;
1257         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1258         int ee_len;
1259
1260         if (unlikely(path == NULL)) {
1261                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1262                 return -EIO;
1263         }
1264         depth = path->p_depth;
1265         *phys = 0;
1266
1267         if (depth == 0 && path->p_ext == NULL)
1268                 return 0;
1269
1270         /* usually extent in the path covers blocks smaller
1271          * then *logical, but it can be that extent is the
1272          * first one in the file */
1273
1274         ex = path[depth].p_ext;
1275         ee_len = ext4_ext_get_actual_len(ex);
1276         if (*logical < le32_to_cpu(ex->ee_block)) {
1277                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1278                         EXT4_ERROR_INODE(inode,
1279                                          "first_extent(path[%d].p_hdr) != ex",
1280                                          depth);
1281                         return -EIO;
1282                 }
1283                 while (--depth >= 0) {
1284                         ix = path[depth].p_idx;
1285                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1286                                 EXT4_ERROR_INODE(inode,
1287                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1288                                                  *logical);
1289                                 return -EIO;
1290                         }
1291                 }
1292                 goto found_extent;
1293         }
1294
1295         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1296                 EXT4_ERROR_INODE(inode,
1297                                  "logical %d < ee_block %d + ee_len %d!",
1298                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1299                 return -EIO;
1300         }
1301
1302         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1303                 /* next allocated block in this leaf */
1304                 ex++;
1305                 goto found_extent;
1306         }
1307
1308         /* go up and search for index to the right */
1309         while (--depth >= 0) {
1310                 ix = path[depth].p_idx;
1311                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1312                         goto got_index;
1313         }
1314
1315         /* we've gone up to the root and found no index to the right */
1316         return 0;
1317
1318 got_index:
1319         /* we've found index to the right, let's
1320          * follow it and find the closest allocated
1321          * block to the right */
1322         ix++;
1323         block = ext4_idx_pblock(ix);
1324         while (++depth < path->p_depth) {
1325                 bh = sb_bread(inode->i_sb, block);
1326                 if (bh == NULL)
1327                         return -EIO;
1328                 eh = ext_block_hdr(bh);
1329                 /* subtract from p_depth to get proper eh_depth */
1330                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1331                         put_bh(bh);
1332                         return -EIO;
1333                 }
1334                 ix = EXT_FIRST_INDEX(eh);
1335                 block = ext4_idx_pblock(ix);
1336                 put_bh(bh);
1337         }
1338
1339         bh = sb_bread(inode->i_sb, block);
1340         if (bh == NULL)
1341                 return -EIO;
1342         eh = ext_block_hdr(bh);
1343         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1344                 put_bh(bh);
1345                 return -EIO;
1346         }
1347         ex = EXT_FIRST_EXTENT(eh);
1348 found_extent:
1349         *logical = le32_to_cpu(ex->ee_block);
1350         *phys = ext4_ext_pblock(ex);
1351         *ret_ex = ex;
1352         if (bh)
1353                 put_bh(bh);
1354         return 0;
1355 }
1356
1357 /*
1358  * ext4_ext_next_allocated_block:
1359  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1360  * NOTE: it considers block number from index entry as
1361  * allocated block. Thus, index entries have to be consistent
1362  * with leaves.
1363  */
1364 static ext4_lblk_t
1365 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1366 {
1367         int depth;
1368
1369         BUG_ON(path == NULL);
1370         depth = path->p_depth;
1371
1372         if (depth == 0 && path->p_ext == NULL)
1373                 return EXT_MAX_BLOCKS;
1374
1375         while (depth >= 0) {
1376                 if (depth == path->p_depth) {
1377                         /* leaf */
1378                         if (path[depth].p_ext &&
1379                                 path[depth].p_ext !=
1380                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1381                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1382                 } else {
1383                         /* index */
1384                         if (path[depth].p_idx !=
1385                                         EXT_LAST_INDEX(path[depth].p_hdr))
1386                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1387                 }
1388                 depth--;
1389         }
1390
1391         return EXT_MAX_BLOCKS;
1392 }
1393
1394 /*
1395  * ext4_ext_next_leaf_block:
1396  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1397  */
1398 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1399 {
1400         int depth;
1401
1402         BUG_ON(path == NULL);
1403         depth = path->p_depth;
1404
1405         /* zero-tree has no leaf blocks at all */
1406         if (depth == 0)
1407                 return EXT_MAX_BLOCKS;
1408
1409         /* go to index block */
1410         depth--;
1411
1412         while (depth >= 0) {
1413                 if (path[depth].p_idx !=
1414                                 EXT_LAST_INDEX(path[depth].p_hdr))
1415                         return (ext4_lblk_t)
1416                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1417                 depth--;
1418         }
1419
1420         return EXT_MAX_BLOCKS;
1421 }
1422
1423 /*
1424  * ext4_ext_correct_indexes:
1425  * if leaf gets modified and modified extent is first in the leaf,
1426  * then we have to correct all indexes above.
1427  * TODO: do we need to correct tree in all cases?
1428  */
1429 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1430                                 struct ext4_ext_path *path)
1431 {
1432         struct ext4_extent_header *eh;
1433         int depth = ext_depth(inode);
1434         struct ext4_extent *ex;
1435         __le32 border;
1436         int k, err = 0;
1437
1438         eh = path[depth].p_hdr;
1439         ex = path[depth].p_ext;
1440
1441         if (unlikely(ex == NULL || eh == NULL)) {
1442                 EXT4_ERROR_INODE(inode,
1443                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1444                 return -EIO;
1445         }
1446
1447         if (depth == 0) {
1448                 /* there is no tree at all */
1449                 return 0;
1450         }
1451
1452         if (ex != EXT_FIRST_EXTENT(eh)) {
1453                 /* we correct tree if first leaf got modified only */
1454                 return 0;
1455         }
1456
1457         /*
1458          * TODO: we need correction if border is smaller than current one
1459          */
1460         k = depth - 1;
1461         border = path[depth].p_ext->ee_block;
1462         err = ext4_ext_get_access(handle, inode, path + k);
1463         if (err)
1464                 return err;
1465         path[k].p_idx->ei_block = border;
1466         err = ext4_ext_dirty(handle, inode, path + k);
1467         if (err)
1468                 return err;
1469
1470         while (k--) {
1471                 /* change all left-side indexes */
1472                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1473                         break;
1474                 err = ext4_ext_get_access(handle, inode, path + k);
1475                 if (err)
1476                         break;
1477                 path[k].p_idx->ei_block = border;
1478                 err = ext4_ext_dirty(handle, inode, path + k);
1479                 if (err)
1480                         break;
1481         }
1482
1483         return err;
1484 }
1485
1486 int
1487 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1488                                 struct ext4_extent *ex2)
1489 {
1490         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1491
1492         /*
1493          * Make sure that either both extents are uninitialized, or
1494          * both are _not_.
1495          */
1496         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1497                 return 0;
1498
1499         if (ext4_ext_is_uninitialized(ex1))
1500                 max_len = EXT_UNINIT_MAX_LEN;
1501         else
1502                 max_len = EXT_INIT_MAX_LEN;
1503
1504         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1505         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1506
1507         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1508                         le32_to_cpu(ex2->ee_block))
1509                 return 0;
1510
1511         /*
1512          * To allow future support for preallocated extents to be added
1513          * as an RO_COMPAT feature, refuse to merge to extents if
1514          * this can result in the top bit of ee_len being set.
1515          */
1516         if (ext1_ee_len + ext2_ee_len > max_len)
1517                 return 0;
1518 #ifdef AGGRESSIVE_TEST
1519         if (ext1_ee_len >= 4)
1520                 return 0;
1521 #endif
1522
1523         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1524                 return 1;
1525         return 0;
1526 }
1527
1528 /*
1529  * This function tries to merge the "ex" extent to the next extent in the tree.
1530  * It always tries to merge towards right. If you want to merge towards
1531  * left, pass "ex - 1" as argument instead of "ex".
1532  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1533  * 1 if they got merged.
1534  */
1535 static int ext4_ext_try_to_merge_right(struct inode *inode,
1536                                  struct ext4_ext_path *path,
1537                                  struct ext4_extent *ex)
1538 {
1539         struct ext4_extent_header *eh;
1540         unsigned int depth, len;
1541         int merge_done = 0;
1542         int uninitialized = 0;
1543
1544         depth = ext_depth(inode);
1545         BUG_ON(path[depth].p_hdr == NULL);
1546         eh = path[depth].p_hdr;
1547
1548         while (ex < EXT_LAST_EXTENT(eh)) {
1549                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1550                         break;
1551                 /* merge with next extent! */
1552                 if (ext4_ext_is_uninitialized(ex))
1553                         uninitialized = 1;
1554                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1555                                 + ext4_ext_get_actual_len(ex + 1));
1556                 if (uninitialized)
1557                         ext4_ext_mark_uninitialized(ex);
1558
1559                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1560                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1561                                 * sizeof(struct ext4_extent);
1562                         memmove(ex + 1, ex + 2, len);
1563                 }
1564                 le16_add_cpu(&eh->eh_entries, -1);
1565                 merge_done = 1;
1566                 WARN_ON(eh->eh_entries == 0);
1567                 if (!eh->eh_entries)
1568                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1569         }
1570
1571         return merge_done;
1572 }
1573
1574 /*
1575  * This function tries to merge the @ex extent to neighbours in the tree.
1576  * return 1 if merge left else 0.
1577  */
1578 static int ext4_ext_try_to_merge(struct inode *inode,
1579                                   struct ext4_ext_path *path,
1580                                   struct ext4_extent *ex) {
1581         struct ext4_extent_header *eh;
1582         unsigned int depth;
1583         int merge_done = 0;
1584         int ret = 0;
1585
1586         depth = ext_depth(inode);
1587         BUG_ON(path[depth].p_hdr == NULL);
1588         eh = path[depth].p_hdr;
1589
1590         if (ex > EXT_FIRST_EXTENT(eh))
1591                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1592
1593         if (!merge_done)
1594                 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1595
1596         return ret;
1597 }
1598
1599 /*
1600  * check if a portion of the "newext" extent overlaps with an
1601  * existing extent.
1602  *
1603  * If there is an overlap discovered, it updates the length of the newext
1604  * such that there will be no overlap, and then returns 1.
1605  * If there is no overlap found, it returns 0.
1606  */
1607 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1608                                            struct inode *inode,
1609                                            struct ext4_extent *newext,
1610                                            struct ext4_ext_path *path)
1611 {
1612         ext4_lblk_t b1, b2;
1613         unsigned int depth, len1;
1614         unsigned int ret = 0;
1615
1616         b1 = le32_to_cpu(newext->ee_block);
1617         len1 = ext4_ext_get_actual_len(newext);
1618         depth = ext_depth(inode);
1619         if (!path[depth].p_ext)
1620                 goto out;
1621         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1622         b2 &= ~(sbi->s_cluster_ratio - 1);
1623
1624         /*
1625          * get the next allocated block if the extent in the path
1626          * is before the requested block(s)
1627          */
1628         if (b2 < b1) {
1629                 b2 = ext4_ext_next_allocated_block(path);
1630                 if (b2 == EXT_MAX_BLOCKS)
1631                         goto out;
1632                 b2 &= ~(sbi->s_cluster_ratio - 1);
1633         }
1634
1635         /* check for wrap through zero on extent logical start block*/
1636         if (b1 + len1 < b1) {
1637                 len1 = EXT_MAX_BLOCKS - b1;
1638                 newext->ee_len = cpu_to_le16(len1);
1639                 ret = 1;
1640         }
1641
1642         /* check for overlap */
1643         if (b1 + len1 > b2) {
1644                 newext->ee_len = cpu_to_le16(b2 - b1);
1645                 ret = 1;
1646         }
1647 out:
1648         return ret;
1649 }
1650
1651 /*
1652  * ext4_ext_insert_extent:
1653  * tries to merge requsted extent into the existing extent or
1654  * inserts requested extent as new one into the tree,
1655  * creating new leaf in the no-space case.
1656  */
1657 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1658                                 struct ext4_ext_path *path,
1659                                 struct ext4_extent *newext, int flag)
1660 {
1661         struct ext4_extent_header *eh;
1662         struct ext4_extent *ex, *fex;
1663         struct ext4_extent *nearex; /* nearest extent */
1664         struct ext4_ext_path *npath = NULL;
1665         int depth, len, err;
1666         ext4_lblk_t next;
1667         unsigned uninitialized = 0;
1668         int flags = 0;
1669
1670         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1671                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1672                 return -EIO;
1673         }
1674         depth = ext_depth(inode);
1675         ex = path[depth].p_ext;
1676         if (unlikely(path[depth].p_hdr == NULL)) {
1677                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1678                 return -EIO;
1679         }
1680
1681         /* try to insert block into found extent and return */
1682         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1683                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1684                 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1685                           ext4_ext_is_uninitialized(newext),
1686                           ext4_ext_get_actual_len(newext),
1687                           le32_to_cpu(ex->ee_block),
1688                           ext4_ext_is_uninitialized(ex),
1689                           ext4_ext_get_actual_len(ex),
1690                           ext4_ext_pblock(ex));
1691                 err = ext4_ext_get_access(handle, inode, path + depth);
1692                 if (err)
1693                         return err;
1694
1695                 /*
1696                  * ext4_can_extents_be_merged should have checked that either
1697                  * both extents are uninitialized, or both aren't. Thus we
1698                  * need to check only one of them here.
1699                  */
1700                 if (ext4_ext_is_uninitialized(ex))
1701                         uninitialized = 1;
1702                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1703                                         + ext4_ext_get_actual_len(newext));
1704                 if (uninitialized)
1705                         ext4_ext_mark_uninitialized(ex);
1706                 eh = path[depth].p_hdr;
1707                 nearex = ex;
1708                 goto merge;
1709         }
1710
1711         depth = ext_depth(inode);
1712         eh = path[depth].p_hdr;
1713         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1714                 goto has_space;
1715
1716         /* probably next leaf has space for us? */
1717         fex = EXT_LAST_EXTENT(eh);
1718         next = EXT_MAX_BLOCKS;
1719         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1720                 next = ext4_ext_next_leaf_block(path);
1721         if (next != EXT_MAX_BLOCKS) {
1722                 ext_debug("next leaf block - %u\n", next);
1723                 BUG_ON(npath != NULL);
1724                 npath = ext4_ext_find_extent(inode, next, NULL);
1725                 if (IS_ERR(npath))
1726                         return PTR_ERR(npath);
1727                 BUG_ON(npath->p_depth != path->p_depth);
1728                 eh = npath[depth].p_hdr;
1729                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1730                         ext_debug("next leaf isn't full(%d)\n",
1731                                   le16_to_cpu(eh->eh_entries));
1732                         path = npath;
1733                         goto has_space;
1734                 }
1735                 ext_debug("next leaf has no free space(%d,%d)\n",
1736                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1737         }
1738
1739         /*
1740          * There is no free space in the found leaf.
1741          * We're gonna add a new leaf in the tree.
1742          */
1743         if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1744                 flags = EXT4_MB_USE_ROOT_BLOCKS;
1745         err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1746         if (err)
1747                 goto cleanup;
1748         depth = ext_depth(inode);
1749         eh = path[depth].p_hdr;
1750
1751 has_space:
1752         nearex = path[depth].p_ext;
1753
1754         err = ext4_ext_get_access(handle, inode, path + depth);
1755         if (err)
1756                 goto cleanup;
1757
1758         if (!nearex) {
1759                 /* there is no extent in this leaf, create first one */
1760                 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1761                                 le32_to_cpu(newext->ee_block),
1762                                 ext4_ext_pblock(newext),
1763                                 ext4_ext_is_uninitialized(newext),
1764                                 ext4_ext_get_actual_len(newext));
1765                 nearex = EXT_FIRST_EXTENT(eh);
1766         } else {
1767                 if (le32_to_cpu(newext->ee_block)
1768                            > le32_to_cpu(nearex->ee_block)) {
1769                         /* Insert after */
1770                         ext_debug("insert %u:%llu:[%d]%d before: "
1771                                         "nearest %p\n",
1772                                         le32_to_cpu(newext->ee_block),
1773                                         ext4_ext_pblock(newext),
1774                                         ext4_ext_is_uninitialized(newext),
1775                                         ext4_ext_get_actual_len(newext),
1776                                         nearex);
1777                         nearex++;
1778                 } else {
1779                         /* Insert before */
1780                         BUG_ON(newext->ee_block == nearex->ee_block);
1781                         ext_debug("insert %u:%llu:[%d]%d after: "
1782                                         "nearest %p\n",
1783                                         le32_to_cpu(newext->ee_block),
1784                                         ext4_ext_pblock(newext),
1785                                         ext4_ext_is_uninitialized(newext),
1786                                         ext4_ext_get_actual_len(newext),
1787                                         nearex);
1788                 }
1789                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1790                 if (len > 0) {
1791                         ext_debug("insert %u:%llu:[%d]%d: "
1792                                         "move %d extents from 0x%p to 0x%p\n",
1793                                         le32_to_cpu(newext->ee_block),
1794                                         ext4_ext_pblock(newext),
1795                                         ext4_ext_is_uninitialized(newext),
1796                                         ext4_ext_get_actual_len(newext),
1797                                         len, nearex, nearex + 1);
1798                         memmove(nearex + 1, nearex,
1799                                 len * sizeof(struct ext4_extent));
1800                 }
1801         }
1802
1803         le16_add_cpu(&eh->eh_entries, 1);
1804         path[depth].p_ext = nearex;
1805         nearex->ee_block = newext->ee_block;
1806         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1807         nearex->ee_len = newext->ee_len;
1808
1809 merge:
1810         /* try to merge extents to the right */
1811         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1812                 ext4_ext_try_to_merge(inode, path, nearex);
1813
1814         /* try to merge extents to the left */
1815
1816         /* time to correct all indexes above */
1817         err = ext4_ext_correct_indexes(handle, inode, path);
1818         if (err)
1819                 goto cleanup;
1820
1821         err = ext4_ext_dirty(handle, inode, path + depth);
1822
1823 cleanup:
1824         if (npath) {
1825                 ext4_ext_drop_refs(npath);
1826                 kfree(npath);
1827         }
1828         ext4_ext_invalidate_cache(inode);
1829         return err;
1830 }
1831
1832 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1833                                ext4_lblk_t num, ext_prepare_callback func,
1834                                void *cbdata)
1835 {
1836         struct ext4_ext_path *path = NULL;
1837         struct ext4_ext_cache cbex;
1838         struct ext4_extent *ex;
1839         ext4_lblk_t next, start = 0, end = 0;
1840         ext4_lblk_t last = block + num;
1841         int depth, exists, err = 0;
1842
1843         BUG_ON(func == NULL);
1844         BUG_ON(inode == NULL);
1845
1846         while (block < last && block != EXT_MAX_BLOCKS) {
1847                 num = last - block;
1848                 /* find extent for this block */
1849                 down_read(&EXT4_I(inode)->i_data_sem);
1850                 path = ext4_ext_find_extent(inode, block, path);
1851                 up_read(&EXT4_I(inode)->i_data_sem);
1852                 if (IS_ERR(path)) {
1853                         err = PTR_ERR(path);
1854                         path = NULL;
1855                         break;
1856                 }
1857
1858                 depth = ext_depth(inode);
1859                 if (unlikely(path[depth].p_hdr == NULL)) {
1860                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1861                         err = -EIO;
1862                         break;
1863                 }
1864                 ex = path[depth].p_ext;
1865                 next = ext4_ext_next_allocated_block(path);
1866
1867                 exists = 0;
1868                 if (!ex) {
1869                         /* there is no extent yet, so try to allocate
1870                          * all requested space */
1871                         start = block;
1872                         end = block + num;
1873                 } else if (le32_to_cpu(ex->ee_block) > block) {
1874                         /* need to allocate space before found extent */
1875                         start = block;
1876                         end = le32_to_cpu(ex->ee_block);
1877                         if (block + num < end)
1878                                 end = block + num;
1879                 } else if (block >= le32_to_cpu(ex->ee_block)
1880                                         + ext4_ext_get_actual_len(ex)) {
1881                         /* need to allocate space after found extent */
1882                         start = block;
1883                         end = block + num;
1884                         if (end >= next)
1885                                 end = next;
1886                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1887                         /*
1888                          * some part of requested space is covered
1889                          * by found extent
1890                          */
1891                         start = block;
1892                         end = le32_to_cpu(ex->ee_block)
1893                                 + ext4_ext_get_actual_len(ex);
1894                         if (block + num < end)
1895                                 end = block + num;
1896                         exists = 1;
1897                 } else {
1898                         BUG();
1899                 }
1900                 BUG_ON(end <= start);
1901
1902                 if (!exists) {
1903                         cbex.ec_block = start;
1904                         cbex.ec_len = end - start;
1905                         cbex.ec_start = 0;
1906                 } else {
1907                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1908                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1909                         cbex.ec_start = ext4_ext_pblock(ex);
1910                 }
1911
1912                 if (unlikely(cbex.ec_len == 0)) {
1913                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1914                         err = -EIO;
1915                         break;
1916                 }
1917                 err = func(inode, next, &cbex, ex, cbdata);
1918                 ext4_ext_drop_refs(path);
1919
1920                 if (err < 0)
1921                         break;
1922
1923                 if (err == EXT_REPEAT)
1924                         continue;
1925                 else if (err == EXT_BREAK) {
1926                         err = 0;
1927                         break;
1928                 }
1929
1930                 if (ext_depth(inode) != depth) {
1931                         /* depth was changed. we have to realloc path */
1932                         kfree(path);
1933                         path = NULL;
1934                 }
1935
1936                 block = cbex.ec_block + cbex.ec_len;
1937         }
1938
1939         if (path) {
1940                 ext4_ext_drop_refs(path);
1941                 kfree(path);
1942         }
1943
1944         return err;
1945 }
1946
1947 static void
1948 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1949                         __u32 len, ext4_fsblk_t start)
1950 {
1951         struct ext4_ext_cache *cex;
1952         BUG_ON(len == 0);
1953         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1954         trace_ext4_ext_put_in_cache(inode, block, len, start);
1955         cex = &EXT4_I(inode)->i_cached_extent;
1956         cex->ec_block = block;
1957         cex->ec_len = len;
1958         cex->ec_start = start;
1959         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1960 }
1961
1962 /*
1963  * ext4_ext_put_gap_in_cache:
1964  * calculate boundaries of the gap that the requested block fits into
1965  * and cache this gap
1966  */
1967 static void
1968 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1969                                 ext4_lblk_t block)
1970 {
1971         int depth = ext_depth(inode);
1972         unsigned long len;
1973         ext4_lblk_t lblock;
1974         struct ext4_extent *ex;
1975
1976         ex = path[depth].p_ext;
1977         if (ex == NULL) {
1978                 /* there is no extent yet, so gap is [0;-] */
1979                 lblock = 0;
1980                 len = EXT_MAX_BLOCKS;
1981                 ext_debug("cache gap(whole file):");
1982         } else if (block < le32_to_cpu(ex->ee_block)) {
1983                 lblock = block;
1984                 len = le32_to_cpu(ex->ee_block) - block;
1985                 ext_debug("cache gap(before): %u [%u:%u]",
1986                                 block,
1987                                 le32_to_cpu(ex->ee_block),
1988                                  ext4_ext_get_actual_len(ex));
1989         } else if (block >= le32_to_cpu(ex->ee_block)
1990                         + ext4_ext_get_actual_len(ex)) {
1991                 ext4_lblk_t next;
1992                 lblock = le32_to_cpu(ex->ee_block)
1993                         + ext4_ext_get_actual_len(ex);
1994
1995                 next = ext4_ext_next_allocated_block(path);
1996                 ext_debug("cache gap(after): [%u:%u] %u",
1997                                 le32_to_cpu(ex->ee_block),
1998                                 ext4_ext_get_actual_len(ex),
1999                                 block);
2000                 BUG_ON(next == lblock);
2001                 len = next - lblock;
2002         } else {
2003                 lblock = len = 0;
2004                 BUG();
2005         }
2006
2007         ext_debug(" -> %u:%lu\n", lblock, len);
2008         ext4_ext_put_in_cache(inode, lblock, len, 0);
2009 }
2010
2011 /*
2012  * ext4_ext_check_cache()
2013  * Checks to see if the given block is in the cache.
2014  * If it is, the cached extent is stored in the given
2015  * cache extent pointer.  If the cached extent is a hole,
2016  * this routine should be used instead of
2017  * ext4_ext_in_cache if the calling function needs to
2018  * know the size of the hole.
2019  *
2020  * @inode: The files inode
2021  * @block: The block to look for in the cache
2022  * @ex:    Pointer where the cached extent will be stored
2023  *         if it contains block
2024  *
2025  * Return 0 if cache is invalid; 1 if the cache is valid
2026  */
2027 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2028         struct ext4_ext_cache *ex){
2029         struct ext4_ext_cache *cex;
2030         struct ext4_sb_info *sbi;
2031         int ret = 0;
2032
2033         /*
2034          * We borrow i_block_reservation_lock to protect i_cached_extent
2035          */
2036         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2037         cex = &EXT4_I(inode)->i_cached_extent;
2038         sbi = EXT4_SB(inode->i_sb);
2039
2040         /* has cache valid data? */
2041         if (cex->ec_len == 0)
2042                 goto errout;
2043
2044         if (in_range(block, cex->ec_block, cex->ec_len)) {
2045                 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2046                 ext_debug("%u cached by %u:%u:%llu\n",
2047                                 block,
2048                                 cex->ec_block, cex->ec_len, cex->ec_start);
2049                 ret = 1;
2050         }
2051 errout:
2052         if (!ret)
2053                 sbi->extent_cache_misses++;
2054         else
2055                 sbi->extent_cache_hits++;
2056         trace_ext4_ext_in_cache(inode, block, ret);
2057         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2058         return ret;
2059 }
2060
2061 /*
2062  * ext4_ext_in_cache()
2063  * Checks to see if the given block is in the cache.
2064  * If it is, the cached extent is stored in the given
2065  * extent pointer.
2066  *
2067  * @inode: The files inode
2068  * @block: The block to look for in the cache
2069  * @ex:    Pointer where the cached extent will be stored
2070  *         if it contains block
2071  *
2072  * Return 0 if cache is invalid; 1 if the cache is valid
2073  */
2074 static int
2075 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2076                         struct ext4_extent *ex)
2077 {
2078         struct ext4_ext_cache cex;
2079         int ret = 0;
2080
2081         if (ext4_ext_check_cache(inode, block, &cex)) {
2082                 ex->ee_block = cpu_to_le32(cex.ec_block);
2083                 ext4_ext_store_pblock(ex, cex.ec_start);
2084                 ex->ee_len = cpu_to_le16(cex.ec_len);
2085                 ret = 1;
2086         }
2087
2088         return ret;
2089 }
2090
2091
2092 /*
2093  * ext4_ext_rm_idx:
2094  * removes index from the index block.
2095  */
2096 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2097                         struct ext4_ext_path *path)
2098 {
2099         int err;
2100         ext4_fsblk_t leaf;
2101
2102         /* free index block */
2103         path--;
2104         leaf = ext4_idx_pblock(path->p_idx);
2105         if (unlikely(path->p_hdr->eh_entries == 0)) {
2106                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2107                 return -EIO;
2108         }
2109         err = ext4_ext_get_access(handle, inode, path);
2110         if (err)
2111                 return err;
2112
2113         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2114                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2115                 len *= sizeof(struct ext4_extent_idx);
2116                 memmove(path->p_idx, path->p_idx + 1, len);
2117         }
2118
2119         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2120         err = ext4_ext_dirty(handle, inode, path);
2121         if (err)
2122                 return err;
2123         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2124         trace_ext4_ext_rm_idx(inode, leaf);
2125
2126         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2127                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2128         return err;
2129 }
2130
2131 /*
2132  * ext4_ext_calc_credits_for_single_extent:
2133  * This routine returns max. credits that needed to insert an extent
2134  * to the extent tree.
2135  * When pass the actual path, the caller should calculate credits
2136  * under i_data_sem.
2137  */
2138 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2139                                                 struct ext4_ext_path *path)
2140 {
2141         if (path) {
2142                 int depth = ext_depth(inode);
2143                 int ret = 0;
2144
2145                 /* probably there is space in leaf? */
2146                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2147                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2148
2149                         /*
2150                          *  There are some space in the leaf tree, no
2151                          *  need to account for leaf block credit
2152                          *
2153                          *  bitmaps and block group descriptor blocks
2154                          *  and other metadata blocks still need to be
2155                          *  accounted.
2156                          */
2157                         /* 1 bitmap, 1 block group descriptor */
2158                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2159                         return ret;
2160                 }
2161         }
2162
2163         return ext4_chunk_trans_blocks(inode, nrblocks);
2164 }
2165
2166 /*
2167  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2168  *
2169  * if nrblocks are fit in a single extent (chunk flag is 1), then
2170  * in the worse case, each tree level index/leaf need to be changed
2171  * if the tree split due to insert a new extent, then the old tree
2172  * index/leaf need to be updated too
2173  *
2174  * If the nrblocks are discontiguous, they could cause
2175  * the whole tree split more than once, but this is really rare.
2176  */
2177 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2178 {
2179         int index;
2180         int depth = ext_depth(inode);
2181
2182         if (chunk)
2183                 index = depth * 2;
2184         else
2185                 index = depth * 3;
2186
2187         return index;
2188 }
2189
2190 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2191                               struct ext4_extent *ex,
2192                               ext4_fsblk_t *partial_cluster,
2193                               ext4_lblk_t from, ext4_lblk_t to)
2194 {
2195         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2196         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2197         ext4_fsblk_t pblk;
2198         int flags = EXT4_FREE_BLOCKS_FORGET;
2199
2200         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2201                 flags |= EXT4_FREE_BLOCKS_METADATA;
2202         /*
2203          * For bigalloc file systems, we never free a partial cluster
2204          * at the beginning of the extent.  Instead, we make a note
2205          * that we tried freeing the cluster, and check to see if we
2206          * need to free it on a subsequent call to ext4_remove_blocks,
2207          * or at the end of the ext4_truncate() operation.
2208          */
2209         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2210
2211         trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2212         /*
2213          * If we have a partial cluster, and it's different from the
2214          * cluster of the last block, we need to explicitly free the
2215          * partial cluster here.
2216          */
2217         pblk = ext4_ext_pblock(ex) + ee_len - 1;
2218         if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2219                 ext4_free_blocks(handle, inode, NULL,
2220                                  EXT4_C2B(sbi, *partial_cluster),
2221                                  sbi->s_cluster_ratio, flags);
2222                 *partial_cluster = 0;
2223         }
2224
2225 #ifdef EXTENTS_STATS
2226         {
2227                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2228                 spin_lock(&sbi->s_ext_stats_lock);
2229                 sbi->s_ext_blocks += ee_len;
2230                 sbi->s_ext_extents++;
2231                 if (ee_len < sbi->s_ext_min)
2232                         sbi->s_ext_min = ee_len;
2233                 if (ee_len > sbi->s_ext_max)
2234                         sbi->s_ext_max = ee_len;
2235                 if (ext_depth(inode) > sbi->s_depth_max)
2236                         sbi->s_depth_max = ext_depth(inode);
2237                 spin_unlock(&sbi->s_ext_stats_lock);
2238         }
2239 #endif
2240         if (from >= le32_to_cpu(ex->ee_block)
2241             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2242                 /* tail removal */
2243                 ext4_lblk_t num;
2244
2245                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2246                 pblk = ext4_ext_pblock(ex) + ee_len - num;
2247                 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2248                 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2249                 /*
2250                  * If the block range to be freed didn't start at the
2251                  * beginning of a cluster, and we removed the entire
2252                  * extent, save the partial cluster here, since we
2253                  * might need to delete if we determine that the
2254                  * truncate operation has removed all of the blocks in
2255                  * the cluster.
2256                  */
2257                 if (pblk & (sbi->s_cluster_ratio - 1) &&
2258                     (ee_len == num))
2259                         *partial_cluster = EXT4_B2C(sbi, pblk);
2260                 else
2261                         *partial_cluster = 0;
2262         } else if (from == le32_to_cpu(ex->ee_block)
2263                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2264                 /* head removal */
2265                 ext4_lblk_t num;
2266                 ext4_fsblk_t start;
2267
2268                 num = to - from;
2269                 start = ext4_ext_pblock(ex);
2270
2271                 ext_debug("free first %u blocks starting %llu\n", num, start);
2272                 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2273
2274         } else {
2275                 printk(KERN_INFO "strange request: removal(2) "
2276                                 "%u-%u from %u:%u\n",
2277                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2278         }
2279         return 0;
2280 }
2281
2282
2283 /*
2284  * ext4_ext_rm_leaf() Removes the extents associated with the
2285  * blocks appearing between "start" and "end", and splits the extents
2286  * if "start" and "end" appear in the same extent
2287  *
2288  * @handle: The journal handle
2289  * @inode:  The files inode
2290  * @path:   The path to the leaf
2291  * @start:  The first block to remove
2292  * @end:   The last block to remove
2293  */
2294 static int
2295 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2296                  struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2297                  ext4_lblk_t start, ext4_lblk_t end)
2298 {
2299         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2300         int err = 0, correct_index = 0;
2301         int depth = ext_depth(inode), credits;
2302         struct ext4_extent_header *eh;
2303         ext4_lblk_t a, b;
2304         unsigned num;
2305         ext4_lblk_t ex_ee_block;
2306         unsigned short ex_ee_len;
2307         unsigned uninitialized = 0;
2308         struct ext4_extent *ex;
2309
2310         /* the header must be checked already in ext4_ext_remove_space() */
2311         ext_debug("truncate since %u in leaf\n", start);
2312         if (!path[depth].p_hdr)
2313                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2314         eh = path[depth].p_hdr;
2315         if (unlikely(path[depth].p_hdr == NULL)) {
2316                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2317                 return -EIO;
2318         }
2319         /* find where to start removing */
2320         ex = EXT_LAST_EXTENT(eh);
2321
2322         ex_ee_block = le32_to_cpu(ex->ee_block);
2323         ex_ee_len = ext4_ext_get_actual_len(ex);
2324
2325         trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2326
2327         while (ex >= EXT_FIRST_EXTENT(eh) &&
2328                         ex_ee_block + ex_ee_len > start) {
2329
2330                 if (ext4_ext_is_uninitialized(ex))
2331                         uninitialized = 1;
2332                 else
2333                         uninitialized = 0;
2334
2335                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2336                          uninitialized, ex_ee_len);
2337                 path[depth].p_ext = ex;
2338
2339                 a = ex_ee_block > start ? ex_ee_block : start;
2340                 b = ex_ee_block+ex_ee_len - 1 < end ?
2341                         ex_ee_block+ex_ee_len - 1 : end;
2342
2343                 ext_debug("  border %u:%u\n", a, b);
2344
2345                 /* If this extent is beyond the end of the hole, skip it */
2346                 if (end <= ex_ee_block) {
2347                         ex--;
2348                         ex_ee_block = le32_to_cpu(ex->ee_block);
2349                         ex_ee_len = ext4_ext_get_actual_len(ex);
2350                         continue;
2351                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2352                         EXT4_ERROR_INODE(inode,"  bad truncate %u:%u\n",
2353                                          start, end);
2354                         err = -EIO;
2355                         goto out;
2356                 } else if (a != ex_ee_block) {
2357                         /* remove tail of the extent */
2358                         num = a - ex_ee_block;
2359                 } else {
2360                         /* remove whole extent: excellent! */
2361                         num = 0;
2362                 }
2363                 /*
2364                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2365                  * descriptor) for each block group; assume two block
2366                  * groups plus ex_ee_len/blocks_per_block_group for
2367                  * the worst case
2368                  */
2369                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2370                 if (ex == EXT_FIRST_EXTENT(eh)) {
2371                         correct_index = 1;
2372                         credits += (ext_depth(inode)) + 1;
2373                 }
2374                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2375
2376                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2377                 if (err)
2378                         goto out;
2379
2380                 err = ext4_ext_get_access(handle, inode, path + depth);
2381                 if (err)
2382                         goto out;
2383
2384                 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2385                                          a, b);
2386                 if (err)
2387                         goto out;
2388
2389                 if (num == 0)
2390                         /* this extent is removed; mark slot entirely unused */
2391                         ext4_ext_store_pblock(ex, 0);
2392
2393                 ex->ee_len = cpu_to_le16(num);
2394                 /*
2395                  * Do not mark uninitialized if all the blocks in the
2396                  * extent have been removed.
2397                  */
2398                 if (uninitialized && num)
2399                         ext4_ext_mark_uninitialized(ex);
2400                 /*
2401                  * If the extent was completely released,
2402                  * we need to remove it from the leaf
2403                  */
2404                 if (num == 0) {
2405                         if (end != EXT_MAX_BLOCKS - 1) {
2406                                 /*
2407                                  * For hole punching, we need to scoot all the
2408                                  * extents up when an extent is removed so that
2409                                  * we dont have blank extents in the middle
2410                                  */
2411                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2412                                         sizeof(struct ext4_extent));
2413
2414                                 /* Now get rid of the one at the end */
2415                                 memset(EXT_LAST_EXTENT(eh), 0,
2416                                         sizeof(struct ext4_extent));
2417                         }
2418                         le16_add_cpu(&eh->eh_entries, -1);
2419                 } else
2420                         *partial_cluster = 0;
2421
2422                 err = ext4_ext_dirty(handle, inode, path + depth);
2423                 if (err)
2424                         goto out;
2425
2426                 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2427                                 ext4_ext_pblock(ex));
2428                 ex--;
2429                 ex_ee_block = le32_to_cpu(ex->ee_block);
2430                 ex_ee_len = ext4_ext_get_actual_len(ex);
2431         }
2432
2433         if (correct_index && eh->eh_entries)
2434                 err = ext4_ext_correct_indexes(handle, inode, path);
2435
2436         /*
2437          * If there is still a entry in the leaf node, check to see if
2438          * it references the partial cluster.  This is the only place
2439          * where it could; if it doesn't, we can free the cluster.
2440          */
2441         if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2442             (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2443              *partial_cluster)) {
2444                 int flags = EXT4_FREE_BLOCKS_FORGET;
2445
2446                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2447                         flags |= EXT4_FREE_BLOCKS_METADATA;
2448
2449                 ext4_free_blocks(handle, inode, NULL,
2450                                  EXT4_C2B(sbi, *partial_cluster),
2451                                  sbi->s_cluster_ratio, flags);
2452                 *partial_cluster = 0;
2453         }
2454
2455         /* if this leaf is free, then we should
2456          * remove it from index block above */
2457         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2458                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2459
2460 out:
2461         return err;
2462 }
2463
2464 /*
2465  * ext4_ext_more_to_rm:
2466  * returns 1 if current index has to be freed (even partial)
2467  */
2468 static int
2469 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2470 {
2471         BUG_ON(path->p_idx == NULL);
2472
2473         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2474                 return 0;
2475
2476         /*
2477          * if truncate on deeper level happened, it wasn't partial,
2478          * so we have to consider current index for truncation
2479          */
2480         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2481                 return 0;
2482         return 1;
2483 }
2484
2485 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2486 {
2487         struct super_block *sb = inode->i_sb;
2488         int depth = ext_depth(inode);
2489         struct ext4_ext_path *path;
2490         ext4_fsblk_t partial_cluster = 0;
2491         handle_t *handle;
2492         int i, err;
2493
2494         ext_debug("truncate since %u\n", start);
2495
2496         /* probably first extent we're gonna free will be last in block */
2497         handle = ext4_journal_start(inode, depth + 1);
2498         if (IS_ERR(handle))
2499                 return PTR_ERR(handle);
2500
2501 again:
2502         ext4_ext_invalidate_cache(inode);
2503
2504         trace_ext4_ext_remove_space(inode, start, depth);
2505
2506         /*
2507          * We start scanning from right side, freeing all the blocks
2508          * after i_size and walking into the tree depth-wise.
2509          */
2510         depth = ext_depth(inode);
2511         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2512         if (path == NULL) {
2513                 ext4_journal_stop(handle);
2514                 return -ENOMEM;
2515         }
2516         path[0].p_depth = depth;
2517         path[0].p_hdr = ext_inode_hdr(inode);
2518         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2519                 err = -EIO;
2520                 goto out;
2521         }
2522         i = err = 0;
2523
2524         while (i >= 0 && err == 0) {
2525                 if (i == depth) {
2526                         /* this is leaf block */
2527                         err = ext4_ext_rm_leaf(handle, inode, path,
2528                                                &partial_cluster, start,
2529                                                EXT_MAX_BLOCKS - 1);
2530                         /* root level has p_bh == NULL, brelse() eats this */
2531                         brelse(path[i].p_bh);
2532                         path[i].p_bh = NULL;
2533                         i--;
2534                         continue;
2535                 }
2536
2537                 /* this is index block */
2538                 if (!path[i].p_hdr) {
2539                         ext_debug("initialize header\n");
2540                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2541                 }
2542
2543                 if (!path[i].p_idx) {
2544                         /* this level hasn't been touched yet */
2545                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2546                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2547                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2548                                   path[i].p_hdr,
2549                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2550                 } else {
2551                         /* we were already here, see at next index */
2552                         path[i].p_idx--;
2553                 }
2554
2555                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2556                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2557                                 path[i].p_idx);
2558                 if (ext4_ext_more_to_rm(path + i)) {
2559                         struct buffer_head *bh;
2560                         /* go to the next level */
2561                         ext_debug("move to level %d (block %llu)\n",
2562                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2563                         memset(path + i + 1, 0, sizeof(*path));
2564                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2565                         if (!bh) {
2566                                 /* should we reset i_size? */
2567                                 err = -EIO;
2568                                 break;
2569                         }
2570                         if (WARN_ON(i + 1 > depth)) {
2571                                 err = -EIO;
2572                                 break;
2573                         }
2574                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2575                                                         depth - i - 1)) {
2576                                 err = -EIO;
2577                                 break;
2578                         }
2579                         path[i + 1].p_bh = bh;
2580
2581                         /* save actual number of indexes since this
2582                          * number is changed at the next iteration */
2583                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2584                         i++;
2585                 } else {
2586                         /* we finished processing this index, go up */
2587                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2588                                 /* index is empty, remove it;
2589                                  * handle must be already prepared by the
2590                                  * truncatei_leaf() */
2591                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2592                         }
2593                         /* root level has p_bh == NULL, brelse() eats this */
2594                         brelse(path[i].p_bh);
2595                         path[i].p_bh = NULL;
2596                         i--;
2597                         ext_debug("return to level %d\n", i);
2598                 }
2599         }
2600
2601         trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2602                         path->p_hdr->eh_entries);
2603
2604         /* If we still have something in the partial cluster and we have removed
2605          * even the first extent, then we should free the blocks in the partial
2606          * cluster as well. */
2607         if (partial_cluster && path->p_hdr->eh_entries == 0) {
2608                 int flags = EXT4_FREE_BLOCKS_FORGET;
2609
2610                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2611                         flags |= EXT4_FREE_BLOCKS_METADATA;
2612
2613                 ext4_free_blocks(handle, inode, NULL,
2614                                  EXT4_C2B(EXT4_SB(sb), partial_cluster),
2615                                  EXT4_SB(sb)->s_cluster_ratio, flags);
2616                 partial_cluster = 0;
2617         }
2618
2619         /* TODO: flexible tree reduction should be here */
2620         if (path->p_hdr->eh_entries == 0) {
2621                 /*
2622                  * truncate to zero freed all the tree,
2623                  * so we need to correct eh_depth
2624                  */
2625                 err = ext4_ext_get_access(handle, inode, path);
2626                 if (err == 0) {
2627                         ext_inode_hdr(inode)->eh_depth = 0;
2628                         ext_inode_hdr(inode)->eh_max =
2629                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2630                         err = ext4_ext_dirty(handle, inode, path);
2631                 }
2632         }
2633 out:
2634         ext4_ext_drop_refs(path);
2635         kfree(path);
2636         if (err == -EAGAIN)
2637                 goto again;
2638         ext4_journal_stop(handle);
2639
2640         return err;
2641 }
2642
2643 /*
2644  * called at mount time
2645  */
2646 void ext4_ext_init(struct super_block *sb)
2647 {
2648         /*
2649          * possible initialization would be here
2650          */
2651
2652         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2653 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2654                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2655 #ifdef AGGRESSIVE_TEST
2656                 printk(", aggressive tests");
2657 #endif
2658 #ifdef CHECK_BINSEARCH
2659                 printk(", check binsearch");
2660 #endif
2661 #ifdef EXTENTS_STATS
2662                 printk(", stats");
2663 #endif
2664                 printk("\n");
2665 #endif
2666 #ifdef EXTENTS_STATS
2667                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2668                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2669                 EXT4_SB(sb)->s_ext_max = 0;
2670 #endif
2671         }
2672 }
2673
2674 /*
2675  * called at umount time
2676  */
2677 void ext4_ext_release(struct super_block *sb)
2678 {
2679         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2680                 return;
2681
2682 #ifdef EXTENTS_STATS
2683         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2684                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2685                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2686                         sbi->s_ext_blocks, sbi->s_ext_extents,
2687                         sbi->s_ext_blocks / sbi->s_ext_extents);
2688                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2689                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2690         }
2691 #endif
2692 }
2693
2694 /* FIXME!! we need to try to merge to left or right after zero-out  */
2695 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2696 {
2697         ext4_fsblk_t ee_pblock;
2698         unsigned int ee_len;
2699         int ret;
2700
2701         ee_len    = ext4_ext_get_actual_len(ex);
2702         ee_pblock = ext4_ext_pblock(ex);
2703
2704         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2705         if (ret > 0)
2706                 ret = 0;
2707
2708         return ret;
2709 }
2710
2711 /*
2712  * used by extent splitting.
2713  */
2714 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
2715                                         due to ENOSPC */
2716 #define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
2717 #define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
2718
2719 /*
2720  * ext4_split_extent_at() splits an extent at given block.
2721  *
2722  * @handle: the journal handle
2723  * @inode: the file inode
2724  * @path: the path to the extent
2725  * @split: the logical block where the extent is splitted.
2726  * @split_flags: indicates if the extent could be zeroout if split fails, and
2727  *               the states(init or uninit) of new extents.
2728  * @flags: flags used to insert new extent to extent tree.
2729  *
2730  *
2731  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2732  * of which are deterimined by split_flag.
2733  *
2734  * There are two cases:
2735  *  a> the extent are splitted into two extent.
2736  *  b> split is not needed, and just mark the extent.
2737  *
2738  * return 0 on success.
2739  */
2740 static int ext4_split_extent_at(handle_t *handle,
2741                              struct inode *inode,
2742                              struct ext4_ext_path *path,
2743                              ext4_lblk_t split,
2744                              int split_flag,
2745                              int flags)
2746 {
2747         ext4_fsblk_t newblock;
2748         ext4_lblk_t ee_block;
2749         struct ext4_extent *ex, newex, orig_ex;
2750         struct ext4_extent *ex2 = NULL;
2751         unsigned int ee_len, depth;
2752         int err = 0;
2753
2754         ext_debug("ext4_split_extents_at: inode %lu, logical"
2755                 "block %llu\n", inode->i_ino, (unsigned long long)split);
2756
2757         ext4_ext_show_leaf(inode, path);
2758
2759         depth = ext_depth(inode);
2760         ex = path[depth].p_ext;
2761         ee_block = le32_to_cpu(ex->ee_block);
2762         ee_len = ext4_ext_get_actual_len(ex);
2763         newblock = split - ee_block + ext4_ext_pblock(ex);
2764
2765         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2766
2767         err = ext4_ext_get_access(handle, inode, path + depth);
2768         if (err)
2769                 goto out;
2770
2771         if (split == ee_block) {
2772                 /*
2773                  * case b: block @split is the block that the extent begins with
2774                  * then we just change the state of the extent, and splitting
2775                  * is not needed.
2776                  */
2777                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2778                         ext4_ext_mark_uninitialized(ex);
2779                 else
2780                         ext4_ext_mark_initialized(ex);
2781
2782                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2783                         ext4_ext_try_to_merge(inode, path, ex);
2784
2785                 err = ext4_ext_dirty(handle, inode, path + depth);
2786                 goto out;
2787         }
2788
2789         /* case a */
2790         memcpy(&orig_ex, ex, sizeof(orig_ex));
2791         ex->ee_len = cpu_to_le16(split - ee_block);
2792         if (split_flag & EXT4_EXT_MARK_UNINIT1)
2793                 ext4_ext_mark_uninitialized(ex);
2794
2795         /*
2796          * path may lead to new leaf, not to original leaf any more
2797          * after ext4_ext_insert_extent() returns,
2798          */
2799         err = ext4_ext_dirty(handle, inode, path + depth);
2800         if (err)
2801                 goto fix_extent_len;
2802
2803         ex2 = &newex;
2804         ex2->ee_block = cpu_to_le32(split);
2805         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2806         ext4_ext_store_pblock(ex2, newblock);
2807         if (split_flag & EXT4_EXT_MARK_UNINIT2)
2808                 ext4_ext_mark_uninitialized(ex2);
2809
2810         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2811         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2812                 err = ext4_ext_zeroout(inode, &orig_ex);
2813                 if (err)
2814                         goto fix_extent_len;
2815                 /* update the extent length and mark as initialized */
2816                 ex->ee_len = cpu_to_le32(ee_len);
2817                 ext4_ext_try_to_merge(inode, path, ex);
2818                 err = ext4_ext_dirty(handle, inode, path + depth);
2819                 goto out;
2820         } else if (err)
2821                 goto fix_extent_len;
2822
2823 out:
2824         ext4_ext_show_leaf(inode, path);
2825         return err;
2826
2827 fix_extent_len:
2828         ex->ee_len = orig_ex.ee_len;
2829         ext4_ext_dirty(handle, inode, path + depth);
2830         return err;
2831 }
2832
2833 /*
2834  * ext4_split_extents() splits an extent and mark extent which is covered
2835  * by @map as split_flags indicates
2836  *
2837  * It may result in splitting the extent into multiple extents (upto three)
2838  * There are three possibilities:
2839  *   a> There is no split required
2840  *   b> Splits in two extents: Split is happening at either end of the extent
2841  *   c> Splits in three extents: Somone is splitting in middle of the extent
2842  *
2843  */
2844 static int ext4_split_extent(handle_t *handle,
2845                               struct inode *inode,
2846                               struct ext4_ext_path *path,
2847                               struct ext4_map_blocks *map,
2848                               int split_flag,
2849                               int flags)
2850 {
2851         ext4_lblk_t ee_block;
2852         struct ext4_extent *ex;
2853         unsigned int ee_len, depth;
2854         int err = 0;
2855         int uninitialized;
2856         int split_flag1, flags1;
2857
2858         depth = ext_depth(inode);
2859         ex = path[depth].p_ext;
2860         ee_block = le32_to_cpu(ex->ee_block);
2861         ee_len = ext4_ext_get_actual_len(ex);
2862         uninitialized = ext4_ext_is_uninitialized(ex);
2863
2864         if (map->m_lblk + map->m_len < ee_block + ee_len) {
2865                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2866                               EXT4_EXT_MAY_ZEROOUT : 0;
2867                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2868                 if (uninitialized)
2869                         split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2870                                        EXT4_EXT_MARK_UNINIT2;
2871                 err = ext4_split_extent_at(handle, inode, path,
2872                                 map->m_lblk + map->m_len, split_flag1, flags1);
2873                 if (err)
2874                         goto out;
2875         }
2876
2877         ext4_ext_drop_refs(path);
2878         path = ext4_ext_find_extent(inode, map->m_lblk, path);
2879         if (IS_ERR(path))
2880                 return PTR_ERR(path);
2881
2882         if (map->m_lblk >= ee_block) {
2883                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2884                               EXT4_EXT_MAY_ZEROOUT : 0;
2885                 if (uninitialized)
2886                         split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2887                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2888                         split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2889                 err = ext4_split_extent_at(handle, inode, path,
2890                                 map->m_lblk, split_flag1, flags);
2891                 if (err)
2892                         goto out;
2893         }
2894
2895         ext4_ext_show_leaf(inode, path);
2896 out:
2897         return err ? err : map->m_len;
2898 }
2899
2900 #define EXT4_EXT_ZERO_LEN 7
2901 /*
2902  * This function is called by ext4_ext_map_blocks() if someone tries to write
2903  * to an uninitialized extent. It may result in splitting the uninitialized
2904  * extent into multiple extents (up to three - one initialized and two
2905  * uninitialized).
2906  * There are three possibilities:
2907  *   a> There is no split required: Entire extent should be initialized
2908  *   b> Splits in two extents: Write is happening at either end of the extent
2909  *   c> Splits in three extents: Somone is writing in middle of the extent
2910  *
2911  * Pre-conditions:
2912  *  - The extent pointed to by 'path' is uninitialized.
2913  *  - The extent pointed to by 'path' contains a superset
2914  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
2915  *
2916  * Post-conditions on success:
2917  *  - the returned value is the number of blocks beyond map->l_lblk
2918  *    that are allocated and initialized.
2919  *    It is guaranteed to be >= map->m_len.
2920  */
2921 static int ext4_ext_convert_to_initialized(handle_t *handle,
2922                                            struct inode *inode,
2923                                            struct ext4_map_blocks *map,
2924                                            struct ext4_ext_path *path)
2925 {
2926         struct ext4_extent_header *eh;
2927         struct ext4_map_blocks split_map;
2928         struct ext4_extent zero_ex;
2929         struct ext4_extent *ex;
2930         ext4_lblk_t ee_block, eof_block;
2931         unsigned int ee_len, depth;
2932         int allocated;
2933         int err = 0;
2934         int split_flag = 0;
2935
2936         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2937                 "block %llu, max_blocks %u\n", inode->i_ino,
2938                 (unsigned long long)map->m_lblk, map->m_len);
2939
2940         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2941                 inode->i_sb->s_blocksize_bits;
2942         if (eof_block < map->m_lblk + map->m_len)
2943                 eof_block = map->m_lblk + map->m_len;
2944
2945         depth = ext_depth(inode);
2946         eh = path[depth].p_hdr;
2947         ex = path[depth].p_ext;
2948         ee_block = le32_to_cpu(ex->ee_block);
2949         ee_len = ext4_ext_get_actual_len(ex);
2950         allocated = ee_len - (map->m_lblk - ee_block);
2951
2952         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
2953
2954         /* Pre-conditions */
2955         BUG_ON(!ext4_ext_is_uninitialized(ex));
2956         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
2957         BUG_ON(map->m_lblk + map->m_len > ee_block + ee_len);
2958
2959         /*
2960          * Attempt to transfer newly initialized blocks from the currently
2961          * uninitialized extent to its left neighbor. This is much cheaper
2962          * than an insertion followed by a merge as those involve costly
2963          * memmove() calls. This is the common case in steady state for
2964          * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
2965          * writes.
2966          *
2967          * Limitations of the current logic:
2968          *  - L1: we only deal with writes at the start of the extent.
2969          *    The approach could be extended to writes at the end
2970          *    of the extent but this scenario was deemed less common.
2971          *  - L2: we do not deal with writes covering the whole extent.
2972          *    This would require removing the extent if the transfer
2973          *    is possible.
2974          *  - L3: we only attempt to merge with an extent stored in the
2975          *    same extent tree node.
2976          */
2977         if ((map->m_lblk == ee_block) &&        /*L1*/
2978                 (map->m_len < ee_len) &&        /*L2*/
2979                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L3*/
2980                 struct ext4_extent *prev_ex;
2981                 ext4_lblk_t prev_lblk;
2982                 ext4_fsblk_t prev_pblk, ee_pblk;
2983                 unsigned int prev_len, write_len;
2984
2985                 prev_ex = ex - 1;
2986                 prev_lblk = le32_to_cpu(prev_ex->ee_block);
2987                 prev_len = ext4_ext_get_actual_len(prev_ex);
2988                 prev_pblk = ext4_ext_pblock(prev_ex);
2989                 ee_pblk = ext4_ext_pblock(ex);
2990                 write_len = map->m_len;
2991
2992                 /*
2993                  * A transfer of blocks from 'ex' to 'prev_ex' is allowed
2994                  * upon those conditions:
2995                  * - C1: prev_ex is initialized,
2996                  * - C2: prev_ex is logically abutting ex,
2997                  * - C3: prev_ex is physically abutting ex,
2998                  * - C4: prev_ex can receive the additional blocks without
2999                  *   overflowing the (initialized) length limit.
3000                  */
3001                 if ((!ext4_ext_is_uninitialized(prev_ex)) &&            /*C1*/
3002                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3003                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3004                         (prev_len < (EXT_INIT_MAX_LEN - write_len))) {  /*C4*/
3005                         err = ext4_ext_get_access(handle, inode, path + depth);
3006                         if (err)
3007                                 goto out;
3008
3009                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3010                                 map, ex, prev_ex);
3011
3012                         /* Shift the start of ex by 'write_len' blocks */
3013                         ex->ee_block = cpu_to_le32(ee_block + write_len);
3014                         ext4_ext_store_pblock(ex, ee_pblk + write_len);
3015                         ex->ee_len = cpu_to_le16(ee_len - write_len);
3016                         ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3017
3018                         /* Extend prev_ex by 'write_len' blocks */
3019                         prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3020
3021                         /* Mark the block containing both extents as dirty */
3022                         ext4_ext_dirty(handle, inode, path + depth);
3023
3024                         /* Update path to point to the right extent */
3025                         path[depth].p_ext = prev_ex;
3026
3027                         /* Result: number of initialized blocks past m_lblk */
3028                         allocated = write_len;
3029                         goto out;
3030                 }
3031         }
3032
3033         WARN_ON(map->m_lblk < ee_block);
3034         /*
3035          * It is safe to convert extent to initialized via explicit
3036          * zeroout only if extent is fully insde i_size or new_size.
3037          */
3038         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3039
3040         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3041         if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3042             (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3043                 err = ext4_ext_zeroout(inode, ex);
3044                 if (err)
3045                         goto out;
3046
3047                 err = ext4_ext_get_access(handle, inode, path + depth);
3048                 if (err)
3049                         goto out;
3050                 ext4_ext_mark_initialized(ex);
3051                 ext4_ext_try_to_merge(inode, path, ex);
3052                 err = ext4_ext_dirty(handle, inode, path + depth);
3053                 goto out;
3054         }
3055
3056         /*
3057          * four cases:
3058          * 1. split the extent into three extents.
3059          * 2. split the extent into two extents, zeroout the first half.
3060          * 3. split the extent into two extents, zeroout the second half.
3061          * 4. split the extent into two extents with out zeroout.
3062          */
3063         split_map.m_lblk = map->m_lblk;
3064         split_map.m_len = map->m_len;
3065
3066         if (allocated > map->m_len) {
3067                 if (allocated <= EXT4_EXT_ZERO_LEN &&
3068                     (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3069                         /* case 3 */
3070                         zero_ex.ee_block =
3071                                          cpu_to_le32(map->m_lblk);
3072                         zero_ex.ee_len = cpu_to_le16(allocated);
3073                         ext4_ext_store_pblock(&zero_ex,
3074                                 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3075                         err = ext4_ext_zeroout(inode, &zero_ex);
3076                         if (err)
3077                                 goto out;
3078                         split_map.m_lblk = map->m_lblk;
3079                         split_map.m_len = allocated;
3080                 } else if ((map->m_lblk - ee_block + map->m_len <
3081                            EXT4_EXT_ZERO_LEN) &&
3082                            (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3083                         /* case 2 */
3084                         if (map->m_lblk != ee_block) {
3085                                 zero_ex.ee_block = ex->ee_block;
3086                                 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3087                                                         ee_block);
3088                                 ext4_ext_store_pblock(&zero_ex,
3089                                                       ext4_ext_pblock(ex));
3090                                 err = ext4_ext_zeroout(inode, &zero_ex);
3091                                 if (err)
3092                                         goto out;
3093                         }
3094
3095                         split_map.m_lblk = ee_block;
3096                         split_map.m_len = map->m_lblk - ee_block + map->m_len;
3097                         allocated = map->m_len;
3098                 }
3099         }
3100
3101         allocated = ext4_split_extent(handle, inode, path,
3102                                        &split_map, split_flag, 0);
3103         if (allocated < 0)
3104                 err = allocated;
3105
3106 out:
3107         return err ? err : allocated;
3108 }
3109
3110 /*
3111  * This function is called by ext4_ext_map_blocks() from
3112  * ext4_get_blocks_dio_write() when DIO to write
3113  * to an uninitialized extent.
3114  *
3115  * Writing to an uninitialized extent may result in splitting the uninitialized
3116  * extent into multiple /initialized uninitialized extents (up to three)
3117  * There are three possibilities:
3118  *   a> There is no split required: Entire extent should be uninitialized
3119  *   b> Splits in two extents: Write is happening at either end of the extent
3120  *   c> Splits in three extents: Somone is writing in middle of the extent
3121  *
3122  * One of more index blocks maybe needed if the extent tree grow after
3123  * the uninitialized extent split. To prevent ENOSPC occur at the IO
3124  * complete, we need to split the uninitialized extent before DIO submit
3125  * the IO. The uninitialized extent called at this time will be split
3126  * into three uninitialized extent(at most). After IO complete, the part
3127  * being filled will be convert to initialized by the end_io callback function
3128  * via ext4_convert_unwritten_extents().
3129  *
3130  * Returns the size of uninitialized extent to be written on success.
3131  */
3132 static int ext4_split_unwritten_extents(handle_t *handle,
3133                                         struct inode *inode,
3134                                         struct ext4_map_blocks *map,
3135                                         struct ext4_ext_path *path,
3136                                         int flags)
3137 {
3138         ext4_lblk_t eof_block;
3139         ext4_lblk_t ee_block;
3140         struct ext4_extent *ex;
3141         unsigned int ee_len;
3142         int split_flag = 0, depth;
3143
3144         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3145                 "block %llu, max_blocks %u\n", inode->i_ino,
3146                 (unsigned long long)map->m_lblk, map->m_len);
3147
3148         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3149                 inode->i_sb->s_blocksize_bits;
3150         if (eof_block < map->m_lblk + map->m_len)
3151                 eof_block = map->m_lblk + map->m_len;
3152         /*
3153          * It is safe to convert extent to initialized via explicit
3154          * zeroout only if extent is fully insde i_size or new_size.
3155          */
3156         depth = ext_depth(inode);
3157         ex = path[depth].p_ext;
3158         ee_block = le32_to_cpu(ex->ee_block);
3159         ee_len = ext4_ext_get_actual_len(ex);
3160
3161         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3162         split_flag |= EXT4_EXT_MARK_UNINIT2;
3163
3164         flags |= EXT4_GET_BLOCKS_PRE_IO;
3165         return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3166 }
3167
3168 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3169                                               struct inode *inode,
3170                                               struct ext4_ext_path *path)
3171 {
3172         struct ext4_extent *ex;
3173         int depth;
3174         int err = 0;
3175
3176         depth = ext_depth(inode);
3177         ex = path[depth].p_ext;
3178
3179         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3180                 "block %llu, max_blocks %u\n", inode->i_ino,
3181                 (unsigned long long)le32_to_cpu(ex->ee_block),
3182                 ext4_ext_get_actual_len(ex));
3183
3184         err = ext4_ext_get_access(handle, inode, path + depth);
3185         if (err)
3186                 goto out;
3187         /* first mark the extent as initialized */
3188         ext4_ext_mark_initialized(ex);
3189
3190         /* note: ext4_ext_correct_indexes() isn't needed here because
3191          * borders are not changed
3192          */
3193         ext4_ext_try_to_merge(inode, path, ex);
3194
3195         /* Mark modified extent as dirty */
3196         err = ext4_ext_dirty(handle, inode, path + depth);
3197 out:
3198         ext4_ext_show_leaf(inode, path);
3199         return err;
3200 }
3201
3202 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3203                         sector_t block, int count)
3204 {
3205         int i;
3206         for (i = 0; i < count; i++)
3207                 unmap_underlying_metadata(bdev, block + i);
3208 }
3209
3210 /*
3211  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3212  */
3213 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3214                               ext4_lblk_t lblk,
3215                               struct ext4_ext_path *path,
3216                               unsigned int len)
3217 {
3218         int i, depth;
3219         struct ext4_extent_header *eh;
3220         struct ext4_extent *last_ex;
3221
3222         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3223                 return 0;
3224
3225         depth = ext_depth(inode);
3226         eh = path[depth].p_hdr;
3227
3228         if (unlikely(!eh->eh_entries)) {
3229                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3230                                  "EOFBLOCKS_FL set");
3231                 return -EIO;
3232         }
3233         last_ex = EXT_LAST_EXTENT(eh);
3234         /*
3235          * We should clear the EOFBLOCKS_FL flag if we are writing the
3236          * last block in the last extent in the file.  We test this by
3237          * first checking to see if the caller to
3238          * ext4_ext_get_blocks() was interested in the last block (or
3239          * a block beyond the last block) in the current extent.  If
3240          * this turns out to be false, we can bail out from this
3241          * function immediately.
3242          */
3243         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3244             ext4_ext_get_actual_len(last_ex))
3245                 return 0;
3246         /*
3247          * If the caller does appear to be planning to write at or
3248          * beyond the end of the current extent, we then test to see
3249          * if the current extent is the last extent in the file, by
3250          * checking to make sure it was reached via the rightmost node
3251          * at each level of the tree.
3252          */
3253         for (i = depth-1; i >= 0; i--)
3254                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3255                         return 0;
3256         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3257         return ext4_mark_inode_dirty(handle, inode);
3258 }
3259
3260 /**
3261  * ext4_find_delalloc_range: find delayed allocated block in the given range.
3262  *
3263  * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3264  * whether there are any buffers marked for delayed allocation. It returns '1'
3265  * on the first delalloc'ed buffer head found. If no buffer head in the given
3266  * range is marked for delalloc, it returns 0.
3267  * lblk_start should always be <= lblk_end.
3268  * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3269  * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3270  * block sooner). This is useful when blocks are truncated sequentially from
3271  * lblk_start towards lblk_end.
3272  */
3273 static int ext4_find_delalloc_range(struct inode *inode,
3274                                     ext4_lblk_t lblk_start,
3275                                     ext4_lblk_t lblk_end,
3276                                     int search_hint_reverse)
3277 {
3278         struct address_space *mapping = inode->i_mapping;
3279         struct buffer_head *head, *bh = NULL;
3280         struct page *page;
3281         ext4_lblk_t i, pg_lblk;
3282         pgoff_t index;
3283
3284         /* reverse search wont work if fs block size is less than page size */
3285         if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3286                 search_hint_reverse = 0;
3287
3288         if (search_hint_reverse)
3289                 i = lblk_end;
3290         else
3291                 i = lblk_start;
3292
3293         index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3294
3295         while ((i >= lblk_start) && (i <= lblk_end)) {
3296                 page = find_get_page(mapping, index);
3297                 if (!page)
3298                         goto nextpage;
3299
3300                 if (!page_has_buffers(page))
3301                         goto nextpage;
3302
3303                 head = page_buffers(page);
3304                 if (!head)
3305                         goto nextpage;
3306
3307                 bh = head;
3308                 pg_lblk = index << (PAGE_CACHE_SHIFT -
3309                                                 inode->i_blkbits);
3310                 do {
3311                         if (unlikely(pg_lblk < lblk_start)) {
3312                                 /*
3313                                  * This is possible when fs block size is less
3314                                  * than page size and our cluster starts/ends in
3315                                  * middle of the page. So we need to skip the
3316                                  * initial few blocks till we reach the 'lblk'
3317                                  */
3318                                 pg_lblk++;
3319                                 continue;
3320                         }
3321
3322                         /* Check if the buffer is delayed allocated and that it
3323                          * is not yet mapped. (when da-buffers are mapped during
3324                          * their writeout, their da_mapped bit is set.)
3325                          */
3326                         if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3327                                 page_cache_release(page);
3328                                 trace_ext4_find_delalloc_range(inode,
3329                                                 lblk_start, lblk_end,
3330                                                 search_hint_reverse,
3331                                                 1, i);
3332                                 return 1;
3333                         }
3334                         if (search_hint_reverse)
3335                                 i--;
3336                         else
3337                                 i++;
3338                 } while ((i >= lblk_start) && (i <= lblk_end) &&
3339                                 ((bh = bh->b_this_page) != head));
3340 nextpage:
3341                 if (page)
3342                         page_cache_release(page);
3343                 /*
3344                  * Move to next page. 'i' will be the first lblk in the next
3345                  * page.
3346                  */
3347                 if (search_hint_reverse)
3348                         index--;
3349                 else
3350                         index++;
3351                 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3352         }
3353
3354         trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3355                                         search_hint_reverse, 0, 0);
3356         return 0;
3357 }
3358
3359 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3360                                int search_hint_reverse)
3361 {
3362         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3363         ext4_lblk_t lblk_start, lblk_end;
3364         lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3365         lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3366
3367         return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3368                                         search_hint_reverse);
3369 }
3370
3371 /**
3372  * Determines how many complete clusters (out of those specified by the 'map')
3373  * are under delalloc and were reserved quota for.
3374  * This function is called when we are writing out the blocks that were
3375  * originally written with their allocation delayed, but then the space was
3376  * allocated using fallocate() before the delayed allocation could be resolved.
3377  * The cases to look for are:
3378  * ('=' indicated delayed allocated blocks
3379  *  '-' indicates non-delayed allocated blocks)
3380  * (a) partial clusters towards beginning and/or end outside of allocated range
3381  *     are not delalloc'ed.
3382  *      Ex:
3383  *      |----c---=|====c====|====c====|===-c----|
3384  *               |++++++ allocated ++++++|
3385  *      ==> 4 complete clusters in above example
3386  *
3387  * (b) partial cluster (outside of allocated range) towards either end is
3388  *     marked for delayed allocation. In this case, we will exclude that
3389  *     cluster.
3390  *      Ex:
3391  *      |----====c========|========c========|
3392  *           |++++++ allocated ++++++|
3393  *      ==> 1 complete clusters in above example
3394  *
3395  *      Ex:
3396  *      |================c================|
3397  *            |++++++ allocated ++++++|
3398  *      ==> 0 complete clusters in above example
3399  *
3400  * The ext4_da_update_reserve_space will be called only if we
3401  * determine here that there were some "entire" clusters that span
3402  * this 'allocated' range.
3403  * In the non-bigalloc case, this function will just end up returning num_blks
3404  * without ever calling ext4_find_delalloc_range.
3405  */
3406 static unsigned int
3407 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3408                            unsigned int num_blks)
3409 {
3410         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3411         ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3412         ext4_lblk_t lblk_from, lblk_to, c_offset;
3413         unsigned int allocated_clusters = 0;
3414
3415         alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3416         alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3417
3418         /* max possible clusters for this allocation */
3419         allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3420
3421         trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3422
3423         /* Check towards left side */
3424         c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3425         if (c_offset) {
3426                 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3427                 lblk_to = lblk_from + c_offset - 1;
3428
3429                 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3430                         allocated_clusters--;
3431         }
3432
3433         /* Now check towards right. */
3434         c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3435         if (allocated_clusters && c_offset) {
3436                 lblk_from = lblk_start + num_blks;
3437                 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3438
3439                 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3440                         allocated_clusters--;
3441         }
3442
3443         return allocated_clusters;
3444 }
3445
3446 static int
3447 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3448                         struct ext4_map_blocks *map,
3449                         struct ext4_ext_path *path, int flags,
3450                         unsigned int allocated, ext4_fsblk_t newblock)
3451 {
3452         int ret = 0;
3453         int err = 0;
3454         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3455
3456         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3457                   "block %llu, max_blocks %u, flags %d, allocated %u",
3458                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3459                   flags, allocated);
3460         ext4_ext_show_leaf(inode, path);
3461
3462         trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3463                                                     newblock);
3464
3465         /* get_block() before submit the IO, split the extent */
3466         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3467                 ret = ext4_split_unwritten_extents(handle, inode, map,
3468                                                    path, flags);
3469                 /*
3470                  * Flag the inode(non aio case) or end_io struct (aio case)
3471                  * that this IO needs to conversion to written when IO is
3472                  * completed
3473                  */
3474                 if (io)
3475                         ext4_set_io_unwritten_flag(inode, io);
3476                 else
3477                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3478                 if (ext4_should_dioread_nolock(inode))
3479                         map->m_flags |= EXT4_MAP_UNINIT;
3480                 goto out;
3481         }
3482         /* IO end_io complete, convert the filled extent to written */
3483         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3484                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3485                                                         path);
3486                 if (ret >= 0) {
3487                         ext4_update_inode_fsync_trans(handle, inode, 1);
3488                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3489                                                  path, map->m_len);
3490                 } else
3491                         err = ret;
3492                 goto out2;
3493         }
3494         /* buffered IO case */
3495         /*
3496          * repeat fallocate creation request
3497          * we already have an unwritten extent
3498          */
3499         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3500                 goto map_out;
3501
3502         /* buffered READ or buffered write_begin() lookup */
3503         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3504                 /*
3505                  * We have blocks reserved already.  We
3506                  * return allocated blocks so that delalloc
3507                  * won't do block reservation for us.  But
3508                  * the buffer head will be unmapped so that
3509                  * a read from the block returns 0s.
3510                  */
3511                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3512                 goto out1;
3513         }
3514
3515         /* buffered write, writepage time, convert*/
3516         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3517         if (ret >= 0)
3518                 ext4_update_inode_fsync_trans(handle, inode, 1);
3519 out:
3520         if (ret <= 0) {
3521                 err = ret;
3522                 goto out2;
3523         } else
3524                 allocated = ret;
3525         map->m_flags |= EXT4_MAP_NEW;
3526         /*
3527          * if we allocated more blocks than requested
3528          * we need to make sure we unmap the extra block
3529          * allocated. The actual needed block will get
3530          * unmapped later when we find the buffer_head marked
3531          * new.
3532          */
3533         if (allocated > map->m_len) {
3534                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3535                                         newblock + map->m_len,
3536                                         allocated - map->m_len);
3537                 allocated = map->m_len;
3538         }
3539
3540         /*
3541          * If we have done fallocate with the offset that is already
3542          * delayed allocated, we would have block reservation
3543          * and quota reservation done in the delayed write path.
3544          * But fallocate would have already updated quota and block
3545          * count for this offset. So cancel these reservation
3546          */
3547         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3548                 unsigned int reserved_clusters;
3549                 reserved_clusters = get_reserved_cluster_alloc(inode,
3550                                 map->m_lblk, map->m_len);
3551                 if (reserved_clusters)
3552                         ext4_da_update_reserve_space(inode,
3553                                                      reserved_clusters,
3554                                                      0);
3555         }
3556
3557 map_out:
3558         map->m_flags |= EXT4_MAP_MAPPED;
3559         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3560                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3561                                          map->m_len);
3562                 if (err < 0)
3563                         goto out2;
3564         }
3565 out1:
3566         if (allocated > map->m_len)
3567                 allocated = map->m_len;
3568         ext4_ext_show_leaf(inode, path);
3569         map->m_pblk = newblock;
3570         map->m_len = allocated;
3571 out2:
3572         if (path) {
3573                 ext4_ext_drop_refs(path);
3574                 kfree(path);
3575         }
3576         return err ? err : allocated;
3577 }
3578
3579 /*
3580  * get_implied_cluster_alloc - check to see if the requested
3581  * allocation (in the map structure) overlaps with a cluster already
3582  * allocated in an extent.
3583  *      @sb     The filesystem superblock structure
3584  *      @map    The requested lblk->pblk mapping
3585  *      @ex     The extent structure which might contain an implied
3586  *                      cluster allocation
3587  *
3588  * This function is called by ext4_ext_map_blocks() after we failed to
3589  * find blocks that were already in the inode's extent tree.  Hence,
3590  * we know that the beginning of the requested region cannot overlap
3591  * the extent from the inode's extent tree.  There are three cases we
3592  * want to catch.  The first is this case:
3593  *
3594  *               |--- cluster # N--|
3595  *    |--- extent ---|  |---- requested region ---|
3596  *                      |==========|
3597  *
3598  * The second case that we need to test for is this one:
3599  *
3600  *   |--------- cluster # N ----------------|
3601  *         |--- requested region --|   |------- extent ----|
3602  *         |=======================|
3603  *
3604  * The third case is when the requested region lies between two extents
3605  * within the same cluster:
3606  *          |------------- cluster # N-------------|
3607  * |----- ex -----|                  |---- ex_right ----|
3608  *                  |------ requested region ------|
3609  *                  |================|
3610  *
3611  * In each of the above cases, we need to set the map->m_pblk and
3612  * map->m_len so it corresponds to the return the extent labelled as
3613  * "|====|" from cluster #N, since it is already in use for data in
3614  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3615  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3616  * as a new "allocated" block region.  Otherwise, we will return 0 and
3617  * ext4_ext_map_blocks() will then allocate one or more new clusters
3618  * by calling ext4_mb_new_blocks().
3619  */
3620 static int get_implied_cluster_alloc(struct super_block *sb,
3621                                      struct ext4_map_blocks *map,
3622                                      struct ext4_extent *ex,
3623                                      struct ext4_ext_path *path)
3624 {
3625         struct ext4_sb_info *sbi = EXT4_SB(sb);
3626         ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3627         ext4_lblk_t ex_cluster_start, ex_cluster_end;
3628         ext4_lblk_t rr_cluster_start, rr_cluster_end;
3629         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3630         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3631         unsigned short ee_len = ext4_ext_get_actual_len(ex);
3632
3633         /* The extent passed in that we are trying to match */
3634         ex_cluster_start = EXT4_B2C(sbi, ee_block);
3635         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3636
3637         /* The requested region passed into ext4_map_blocks() */
3638         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3639         rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3640
3641         if ((rr_cluster_start == ex_cluster_end) ||
3642             (rr_cluster_start == ex_cluster_start)) {
3643                 if (rr_cluster_start == ex_cluster_end)
3644                         ee_start += ee_len - 1;
3645                 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3646                         c_offset;
3647                 map->m_len = min(map->m_len,
3648                                  (unsigned) sbi->s_cluster_ratio - c_offset);
3649                 /*
3650                  * Check for and handle this case:
3651                  *
3652                  *   |--------- cluster # N-------------|
3653                  *                     |------- extent ----|
3654                  *         |--- requested region ---|
3655                  *         |===========|
3656                  */
3657
3658                 if (map->m_lblk < ee_block)
3659                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
3660
3661                 /*
3662                  * Check for the case where there is already another allocated
3663                  * block to the right of 'ex' but before the end of the cluster.
3664                  *
3665                  *          |------------- cluster # N-------------|
3666                  * |----- ex -----|                  |---- ex_right ----|
3667                  *                  |------ requested region ------|
3668                  *                  |================|
3669                  */
3670                 if (map->m_lblk > ee_block) {
3671                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3672                         map->m_len = min(map->m_len, next - map->m_lblk);
3673                 }
3674
3675                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3676                 return 1;
3677         }
3678
3679         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3680         return 0;
3681 }
3682
3683
3684 /*
3685  * Block allocation/map/preallocation routine for extents based files
3686  *
3687  *
3688  * Need to be called with
3689  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3690  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3691  *
3692  * return > 0, number of of blocks already mapped/allocated
3693  *          if create == 0 and these are pre-allocated blocks
3694  *              buffer head is unmapped
3695  *          otherwise blocks are mapped
3696  *
3697  * return = 0, if plain look up failed (blocks have not been allocated)
3698  *          buffer head is unmapped
3699  *
3700  * return < 0, error case.
3701  */
3702 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3703                         struct ext4_map_blocks *map, int flags)
3704 {
3705         struct ext4_ext_path *path = NULL;
3706         struct ext4_extent newex, *ex, *ex2;
3707         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3708         ext4_fsblk_t newblock = 0;
3709         int free_on_err = 0, err = 0, depth, ret;
3710         unsigned int allocated = 0, offset = 0;
3711         unsigned int allocated_clusters = 0;
3712         unsigned int punched_out = 0;
3713         unsigned int result = 0;
3714         struct ext4_allocation_request ar;
3715         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3716         ext4_lblk_t cluster_offset;
3717
3718         ext_debug("blocks %u/%u requested for inode %lu\n",
3719                   map->m_lblk, map->m_len, inode->i_ino);
3720         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3721
3722         /* check in cache */
3723         if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3724                 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3725                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3726                         if ((sbi->s_cluster_ratio > 1) &&
3727                             ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3728                                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3729
3730                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3731                                 /*
3732                                  * block isn't allocated yet and
3733                                  * user doesn't want to allocate it
3734                                  */
3735                                 goto out2;
3736                         }
3737                         /* we should allocate requested block */
3738                 } else {
3739                         /* block is already allocated */
3740                         if (sbi->s_cluster_ratio > 1)
3741                                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3742                         newblock = map->m_lblk
3743                                    - le32_to_cpu(newex.ee_block)
3744                                    + ext4_ext_pblock(&newex);
3745                         /* number of remaining blocks in the extent */
3746                         allocated = ext4_ext_get_actual_len(&newex) -
3747                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3748                         goto out;
3749                 }
3750         }
3751
3752         /* find extent for this block */
3753         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3754         if (IS_ERR(path)) {
3755                 err = PTR_ERR(path);
3756                 path = NULL;
3757                 goto out2;
3758         }
3759
3760         depth = ext_depth(inode);
3761
3762         /*
3763          * consistent leaf must not be empty;
3764          * this situation is possible, though, _during_ tree modification;
3765          * this is why assert can't be put in ext4_ext_find_extent()
3766          */
3767         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3768                 EXT4_ERROR_INODE(inode, "bad extent address "
3769                                  "lblock: %lu, depth: %d pblock %lld",
3770                                  (unsigned long) map->m_lblk, depth,
3771                                  path[depth].p_block);
3772                 err = -EIO;
3773                 goto out2;
3774         }
3775
3776         ex = path[depth].p_ext;
3777         if (ex) {
3778                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3779                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3780                 unsigned short ee_len;
3781
3782                 /*
3783                  * Uninitialized extents are treated as holes, except that
3784                  * we split out initialized portions during a write.
3785                  */
3786                 ee_len = ext4_ext_get_actual_len(ex);
3787
3788                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3789
3790                 /* if found extent covers block, simply return it */
3791                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3792                         struct ext4_map_blocks punch_map;
3793                         ext4_fsblk_t partial_cluster = 0;
3794
3795                         newblock = map->m_lblk - ee_block + ee_start;
3796                         /* number of remaining blocks in the extent */
3797                         allocated = ee_len - (map->m_lblk - ee_block);
3798                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3799                                   ee_block, ee_len, newblock);
3800
3801                         if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3802                                 /*
3803                                  * Do not put uninitialized extent
3804                                  * in the cache
3805                                  */
3806                                 if (!ext4_ext_is_uninitialized(ex)) {
3807                                         ext4_ext_put_in_cache(inode, ee_block,
3808                                                 ee_len, ee_start);
3809                                         goto out;
3810                                 }
3811                                 ret = ext4_ext_handle_uninitialized_extents(
3812                                         handle, inode, map, path, flags,
3813                                         allocated, newblock);
3814                                 return ret;
3815                         }
3816
3817                         /*
3818                          * Punch out the map length, but only to the
3819                          * end of the extent
3820                          */
3821                         punched_out = allocated < map->m_len ?
3822                                 allocated : map->m_len;
3823
3824                         /*
3825                          * Sense extents need to be converted to
3826                          * uninitialized, they must fit in an
3827                          * uninitialized extent
3828                          */
3829                         if (punched_out > EXT_UNINIT_MAX_LEN)
3830                                 punched_out = EXT_UNINIT_MAX_LEN;
3831
3832                         punch_map.m_lblk = map->m_lblk;
3833                         punch_map.m_pblk = newblock;
3834                         punch_map.m_len = punched_out;
3835                         punch_map.m_flags = 0;
3836
3837                         /* Check to see if the extent needs to be split */
3838                         if (punch_map.m_len != ee_len ||
3839                                 punch_map.m_lblk != ee_block) {
3840
3841                                 ret = ext4_split_extent(handle, inode,
3842                                 path, &punch_map, 0,
3843                                 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3844                                 EXT4_GET_BLOCKS_PRE_IO);
3845
3846                                 if (ret < 0) {
3847                                         err = ret;
3848                                         goto out2;
3849                                 }
3850                                 /*
3851                                  * find extent for the block at
3852                                  * the start of the hole
3853                                  */
3854                                 ext4_ext_drop_refs(path);
3855                                 kfree(path);
3856
3857                                 path = ext4_ext_find_extent(inode,
3858                                 map->m_lblk, NULL);
3859                                 if (IS_ERR(path)) {
3860                                         err = PTR_ERR(path);
3861                                         path = NULL;
3862                                         goto out2;
3863                                 }
3864
3865                                 depth = ext_depth(inode);
3866                                 ex = path[depth].p_ext;
3867                                 ee_len = ext4_ext_get_actual_len(ex);
3868                                 ee_block = le32_to_cpu(ex->ee_block);
3869                                 ee_start = ext4_ext_pblock(ex);
3870
3871                         }
3872
3873                         ext4_ext_mark_uninitialized(ex);
3874
3875                         ext4_ext_invalidate_cache(inode);
3876
3877                         err = ext4_ext_rm_leaf(handle, inode, path,
3878                                                &partial_cluster, map->m_lblk,
3879                                                map->m_lblk + punched_out);
3880
3881                         if (!err && path->p_hdr->eh_entries == 0) {
3882                                 /*
3883                                  * Punch hole freed all of this sub tree,
3884                                  * so we need to correct eh_depth
3885                                  */
3886                                 err = ext4_ext_get_access(handle, inode, path);
3887                                 if (err == 0) {
3888                                         ext_inode_hdr(inode)->eh_depth = 0;
3889                                         ext_inode_hdr(inode)->eh_max =
3890                                         cpu_to_le16(ext4_ext_space_root(
3891                                                 inode, 0));
3892
3893                                         err = ext4_ext_dirty(
3894                                                 handle, inode, path);
3895                                 }
3896                         }
3897
3898                         goto out2;
3899                 }
3900         }
3901
3902         if ((sbi->s_cluster_ratio > 1) &&
3903             ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3904                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3905
3906         /*
3907          * requested block isn't allocated yet;
3908          * we couldn't try to create block if create flag is zero
3909          */
3910         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3911                 /*
3912                  * put just found gap into cache to speed up
3913                  * subsequent requests
3914                  */
3915                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3916                 goto out2;
3917         }
3918
3919         /*
3920          * Okay, we need to do block allocation.
3921          */
3922         map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3923         newex.ee_block = cpu_to_le32(map->m_lblk);
3924         cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3925
3926         /*
3927          * If we are doing bigalloc, check to see if the extent returned
3928          * by ext4_ext_find_extent() implies a cluster we can use.
3929          */
3930         if (cluster_offset && ex &&
3931             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3932                 ar.len = allocated = map->m_len;
3933                 newblock = map->m_pblk;
3934                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3935                 goto got_allocated_blocks;
3936         }
3937
3938         /* find neighbour allocated blocks */
3939         ar.lleft = map->m_lblk;
3940         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3941         if (err)
3942                 goto out2;
3943         ar.lright = map->m_lblk;
3944         ex2 = NULL;
3945         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3946         if (err)
3947                 goto out2;
3948
3949         /* Check if the extent after searching to the right implies a
3950          * cluster we can use. */
3951         if ((sbi->s_cluster_ratio > 1) && ex2 &&
3952             get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
3953                 ar.len = allocated = map->m_len;
3954                 newblock = map->m_pblk;
3955                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3956                 goto got_allocated_blocks;
3957         }
3958
3959         /*
3960          * See if request is beyond maximum number of blocks we can have in
3961          * a single extent. For an initialized extent this limit is
3962          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3963          * EXT_UNINIT_MAX_LEN.
3964          */
3965         if (map->m_len > EXT_INIT_MAX_LEN &&
3966             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3967                 map->m_len = EXT_INIT_MAX_LEN;
3968         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3969                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3970                 map->m_len = EXT_UNINIT_MAX_LEN;
3971
3972         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3973         newex.ee_len = cpu_to_le16(map->m_len);
3974         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
3975         if (err)
3976                 allocated = ext4_ext_get_actual_len(&newex);
3977         else
3978                 allocated = map->m_len;
3979
3980         /* allocate new block */
3981         ar.inode = inode;
3982         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3983         ar.logical = map->m_lblk;
3984         /*
3985          * We calculate the offset from the beginning of the cluster
3986          * for the logical block number, since when we allocate a
3987          * physical cluster, the physical block should start at the
3988          * same offset from the beginning of the cluster.  This is
3989          * needed so that future calls to get_implied_cluster_alloc()
3990          * work correctly.
3991          */
3992         offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3993         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3994         ar.goal -= offset;
3995         ar.logical -= offset;
3996         if (S_ISREG(inode->i_mode))
3997                 ar.flags = EXT4_MB_HINT_DATA;
3998         else
3999                 /* disable in-core preallocation for non-regular files */
4000                 ar.flags = 0;
4001         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4002                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4003         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4004         if (!newblock)
4005                 goto out2;
4006         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4007                   ar.goal, newblock, allocated);
4008         free_on_err = 1;
4009         allocated_clusters = ar.len;
4010         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4011         if (ar.len > allocated)
4012                 ar.len = allocated;
4013
4014 got_allocated_blocks:
4015         /* try to insert new extent into found leaf and return */
4016         ext4_ext_store_pblock(&newex, newblock + offset);
4017         newex.ee_len = cpu_to_le16(ar.len);
4018         /* Mark uninitialized */
4019         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4020                 ext4_ext_mark_uninitialized(&newex);
4021                 /*
4022                  * io_end structure was created for every IO write to an
4023                  * uninitialized extent. To avoid unnecessary conversion,
4024                  * here we flag the IO that really needs the conversion.
4025                  * For non asycn direct IO case, flag the inode state
4026                  * that we need to perform conversion when IO is done.
4027                  */
4028                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4029                         if (io)
4030                                 ext4_set_io_unwritten_flag(inode, io);
4031                         else
4032                                 ext4_set_inode_state(inode,
4033                                                      EXT4_STATE_DIO_UNWRITTEN);
4034                 }
4035                 if (ext4_should_dioread_nolock(inode))
4036                         map->m_flags |= EXT4_MAP_UNINIT;
4037         }
4038
4039         err = 0;
4040         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4041                 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4042                                          path, ar.len);
4043         if (!err)
4044                 err = ext4_ext_insert_extent(handle, inode, path,
4045                                              &newex, flags);
4046         if (err && free_on_err) {
4047                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4048                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4049                 /* free data blocks we just allocated */
4050                 /* not a good idea to call discard here directly,
4051                  * but otherwise we'd need to call it every free() */
4052                 ext4_discard_preallocations(inode);
4053                 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4054                                  ext4_ext_get_actual_len(&newex), fb_flags);
4055                 goto out2;
4056         }
4057
4058         /* previous routine could use block we allocated */
4059         newblock = ext4_ext_pblock(&newex);
4060         allocated = ext4_ext_get_actual_len(&newex);
4061         if (allocated > map->m_len)
4062                 allocated = map->m_len;
4063         map->m_flags |= EXT4_MAP_NEW;
4064
4065         /*
4066          * Update reserved blocks/metadata blocks after successful
4067          * block allocation which had been deferred till now.
4068          */
4069         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4070                 unsigned int reserved_clusters;
4071                 /*
4072                  * Check how many clusters we had reserved this allocated range
4073                  */
4074                 reserved_clusters = get_reserved_cluster_alloc(inode,
4075                                                 map->m_lblk, allocated);
4076                 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4077                         if (reserved_clusters) {
4078                                 /*
4079                                  * We have clusters reserved for this range.
4080                                  * But since we are not doing actual allocation
4081                                  * and are simply using blocks from previously
4082                                  * allocated cluster, we should release the
4083                                  * reservation and not claim quota.
4084                                  */
4085                                 ext4_da_update_reserve_space(inode,
4086                                                 reserved_clusters, 0);
4087                         }
4088                 } else {
4089                         BUG_ON(allocated_clusters < reserved_clusters);
4090                         /* We will claim quota for all newly allocated blocks.*/
4091                         ext4_da_update_reserve_space(inode, allocated_clusters,
4092                                                         1);
4093                         if (reserved_clusters < allocated_clusters) {
4094                                 struct ext4_inode_info *ei = EXT4_I(inode);
4095                                 int reservation = allocated_clusters -
4096                                                   reserved_clusters;
4097                                 /*
4098                                  * It seems we claimed few clusters outside of
4099                                  * the range of this allocation. We should give
4100                                  * it back to the reservation pool. This can
4101                                  * happen in the following case:
4102                                  *
4103                                  * * Suppose s_cluster_ratio is 4 (i.e., each
4104                                  *   cluster has 4 blocks. Thus, the clusters
4105                                  *   are [0-3],[4-7],[8-11]...
4106                                  * * First comes delayed allocation write for
4107                                  *   logical blocks 10 & 11. Since there were no
4108                                  *   previous delayed allocated blocks in the
4109                                  *   range [8-11], we would reserve 1 cluster
4110                                  *   for this write.
4111                                  * * Next comes write for logical blocks 3 to 8.
4112                                  *   In this case, we will reserve 2 clusters
4113                                  *   (for [0-3] and [4-7]; and not for [8-11] as
4114                                  *   that range has a delayed allocated blocks.
4115                                  *   Thus total reserved clusters now becomes 3.
4116                                  * * Now, during the delayed allocation writeout
4117                                  *   time, we will first write blocks [3-8] and
4118                                  *   allocate 3 clusters for writing these
4119                                  *   blocks. Also, we would claim all these
4120                                  *   three clusters above.
4121                                  * * Now when we come here to writeout the
4122                                  *   blocks [10-11], we would expect to claim
4123                                  *   the reservation of 1 cluster we had made
4124                                  *   (and we would claim it since there are no
4125                                  *   more delayed allocated blocks in the range
4126                                  *   [8-11]. But our reserved cluster count had
4127                                  *   already gone to 0.
4128                                  *
4129                                  *   Thus, at the step 4 above when we determine
4130                                  *   that there are still some unwritten delayed
4131                                  *   allocated blocks outside of our current
4132                                  *   block range, we should increment the
4133                                  *   reserved clusters count so that when the
4134                                  *   remaining blocks finally gets written, we
4135                                  *   could claim them.
4136                                  */
4137                                 dquot_reserve_block(inode,
4138                                                 EXT4_C2B(sbi, reservation));
4139                                 spin_lock(&ei->i_block_reservation_lock);
4140                                 ei->i_reserved_data_blocks += reservation;
4141                                 spin_unlock(&ei->i_block_reservation_lock);
4142                         }
4143                 }
4144         }
4145
4146         /*
4147          * Cache the extent and update transaction to commit on fdatasync only
4148          * when it is _not_ an uninitialized extent.
4149          */
4150         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4151                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4152                 ext4_update_inode_fsync_trans(handle, inode, 1);
4153         } else
4154                 ext4_update_inode_fsync_trans(handle, inode, 0);
4155 out:
4156         if (allocated > map->m_len)
4157                 allocated = map->m_len;
4158         ext4_ext_show_leaf(inode, path);
4159         map->m_flags |= EXT4_MAP_MAPPED;
4160         map->m_pblk = newblock;
4161         map->m_len = allocated;
4162 out2:
4163         if (path) {
4164                 ext4_ext_drop_refs(path);
4165                 kfree(path);
4166         }
4167         result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4168                         punched_out : allocated;
4169
4170         trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4171                 newblock, map->m_len, err ? err : result);
4172
4173         return err ? err : result;
4174 }
4175
4176 void ext4_ext_truncate(struct inode *inode)
4177 {
4178         struct address_space *mapping = inode->i_mapping;
4179         struct super_block *sb = inode->i_sb;
4180         ext4_lblk_t last_block;
4181         handle_t *handle;
4182         loff_t page_len;
4183         int err = 0;
4184
4185         /*
4186          * finish any pending end_io work so we won't run the risk of
4187          * converting any truncated blocks to initialized later
4188          */
4189         ext4_flush_completed_IO(inode);
4190
4191         /*
4192          * probably first extent we're gonna free will be last in block
4193          */
4194         err = ext4_writepage_trans_blocks(inode);
4195         handle = ext4_journal_start(inode, err);
4196         if (IS_ERR(handle))
4197                 return;
4198
4199         if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4200                 page_len = PAGE_CACHE_SIZE -
4201                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
4202
4203                 err = ext4_discard_partial_page_buffers(handle,
4204                         mapping, inode->i_size, page_len, 0);
4205
4206                 if (err)
4207                         goto out_stop;
4208         }
4209
4210         if (ext4_orphan_add(handle, inode))
4211                 goto out_stop;
4212
4213         down_write(&EXT4_I(inode)->i_data_sem);
4214         ext4_ext_invalidate_cache(inode);
4215
4216         ext4_discard_preallocations(inode);
4217
4218         /*
4219          * TODO: optimization is possible here.
4220          * Probably we need not scan at all,
4221          * because page truncation is enough.
4222          */
4223
4224         /* we have to know where to truncate from in crash case */
4225         EXT4_I(inode)->i_disksize = inode->i_size;
4226         ext4_mark_inode_dirty(handle, inode);
4227
4228         last_block = (inode->i_size + sb->s_blocksize - 1)
4229                         >> EXT4_BLOCK_SIZE_BITS(sb);
4230         err = ext4_ext_remove_space(inode, last_block);
4231
4232         /* In a multi-transaction truncate, we only make the final
4233          * transaction synchronous.
4234          */
4235         if (IS_SYNC(inode))
4236                 ext4_handle_sync(handle);
4237
4238         up_write(&EXT4_I(inode)->i_data_sem);
4239
4240 out_stop:
4241         /*
4242          * If this was a simple ftruncate() and the file will remain alive,
4243          * then we need to clear up the orphan record which we created above.
4244          * However, if this was a real unlink then we were called by
4245          * ext4_delete_inode(), and we allow that function to clean up the
4246          * orphan info for us.
4247          */
4248         if (inode->i_nlink)
4249                 ext4_orphan_del(handle, inode);
4250
4251         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4252         ext4_mark_inode_dirty(handle, inode);
4253         ext4_journal_stop(handle);
4254 }
4255
4256 static void ext4_falloc_update_inode(struct inode *inode,
4257                                 int mode, loff_t new_size, int update_ctime)
4258 {
4259         struct timespec now;
4260
4261         if (update_ctime) {
4262                 now = current_fs_time(inode->i_sb);
4263                 if (!timespec_equal(&inode->i_ctime, &now))
4264                         inode->i_ctime = now;
4265         }
4266         /*
4267          * Update only when preallocation was requested beyond
4268          * the file size.
4269          */
4270         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4271                 if (new_size > i_size_read(inode))
4272                         i_size_write(inode, new_size);
4273                 if (new_size > EXT4_I(inode)->i_disksize)
4274                         ext4_update_i_disksize(inode, new_size);
4275         } else {
4276                 /*
4277                  * Mark that we allocate beyond EOF so the subsequent truncate
4278                  * can proceed even if the new size is the same as i_size.
4279                  */
4280                 if (new_size > i_size_read(inode))
4281                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4282         }
4283
4284 }
4285
4286 /*
4287  * preallocate space for a file. This implements ext4's fallocate file
4288  * operation, which gets called from sys_fallocate system call.
4289  * For block-mapped files, posix_fallocate should fall back to the method
4290  * of writing zeroes to the required new blocks (the same behavior which is
4291  * expected for file systems which do not support fallocate() system call).
4292  */
4293 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4294 {
4295         struct inode *inode = file->f_path.dentry->d_inode;
4296         handle_t *handle;
4297         loff_t new_size;
4298         unsigned int max_blocks;
4299         int ret = 0;
4300         int ret2 = 0;
4301         int retries = 0;
4302         int flags;
4303         struct ext4_map_blocks map;
4304         unsigned int credits, blkbits = inode->i_blkbits;
4305
4306         /*
4307          * currently supporting (pre)allocate mode for extent-based
4308          * files _only_
4309          */
4310         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4311                 return -EOPNOTSUPP;
4312
4313         /* Return error if mode is not supported */
4314         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4315                 return -EOPNOTSUPP;
4316
4317         if (mode & FALLOC_FL_PUNCH_HOLE)
4318                 return ext4_punch_hole(file, offset, len);
4319
4320         trace_ext4_fallocate_enter(inode, offset, len, mode);
4321         map.m_lblk = offset >> blkbits;
4322         /*
4323          * We can't just convert len to max_blocks because
4324          * If blocksize = 4096 offset = 3072 and len = 2048
4325          */
4326         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4327                 - map.m_lblk;
4328         /*
4329          * credits to insert 1 extent into extent tree
4330          */
4331         credits = ext4_chunk_trans_blocks(inode, max_blocks);
4332         mutex_lock(&inode->i_mutex);
4333         ret = inode_newsize_ok(inode, (len + offset));
4334         if (ret) {
4335                 mutex_unlock(&inode->i_mutex);
4336                 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4337                 return ret;
4338         }
4339         flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4340         if (mode & FALLOC_FL_KEEP_SIZE)
4341                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4342         /*
4343          * Don't normalize the request if it can fit in one extent so
4344          * that it doesn't get unnecessarily split into multiple
4345          * extents.
4346          */
4347         if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4348                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4349 retry:
4350         while (ret >= 0 && ret < max_blocks) {
4351                 map.m_lblk = map.m_lblk + ret;
4352                 map.m_len = max_blocks = max_blocks - ret;
4353                 handle = ext4_journal_start(inode, credits);
4354                 if (IS_ERR(handle)) {
4355                         ret = PTR_ERR(handle);
4356                         break;
4357                 }
4358                 ret = ext4_map_blocks(handle, inode, &map, flags);
4359                 if (ret <= 0) {
4360 #ifdef EXT4FS_DEBUG
4361                         WARN_ON(ret <= 0);
4362                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
4363                                     "returned error inode#%lu, block=%u, "
4364                                     "max_blocks=%u", __func__,
4365                                     inode->i_ino, map.m_lblk, max_blocks);
4366 #endif
4367                         ext4_mark_inode_dirty(handle, inode);
4368                         ret2 = ext4_journal_stop(handle);
4369                         break;
4370                 }
4371                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4372                                                 blkbits) >> blkbits))
4373                         new_size = offset + len;
4374                 else
4375                         new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4376
4377                 ext4_falloc_update_inode(inode, mode, new_size,
4378                                          (map.m_flags & EXT4_MAP_NEW));
4379                 ext4_mark_inode_dirty(handle, inode);
4380                 ret2 = ext4_journal_stop(handle);
4381                 if (ret2)
4382                         break;
4383         }
4384         if (ret == -ENOSPC &&
4385                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4386                 ret = 0;
4387                 goto retry;
4388         }
4389         mutex_unlock(&inode->i_mutex);
4390         trace_ext4_fallocate_exit(inode, offset, max_blocks,
4391                                 ret > 0 ? ret2 : ret);
4392         return ret > 0 ? ret2 : ret;
4393 }
4394
4395 /*
4396  * This function convert a range of blocks to written extents
4397  * The caller of this function will pass the start offset and the size.
4398  * all unwritten extents within this range will be converted to
4399  * written extents.
4400  *
4401  * This function is called from the direct IO end io call back
4402  * function, to convert the fallocated extents after IO is completed.
4403  * Returns 0 on success.
4404  */
4405 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4406                                     ssize_t len)
4407 {
4408         handle_t *handle;
4409         unsigned int max_blocks;
4410         int ret = 0;
4411         int ret2 = 0;
4412         struct ext4_map_blocks map;
4413         unsigned int credits, blkbits = inode->i_blkbits;
4414
4415         map.m_lblk = offset >> blkbits;
4416         /*
4417          * We can't just convert len to max_blocks because
4418          * If blocksize = 4096 offset = 3072 and len = 2048
4419          */
4420         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4421                       map.m_lblk);
4422         /*
4423          * credits to insert 1 extent into extent tree
4424          */
4425         credits = ext4_chunk_trans_blocks(inode, max_blocks);
4426         while (ret >= 0 && ret < max_blocks) {
4427                 map.m_lblk += ret;
4428                 map.m_len = (max_blocks -= ret);
4429                 handle = ext4_journal_start(inode, credits);
4430                 if (IS_ERR(handle)) {
4431                         ret = PTR_ERR(handle);
4432                         break;
4433                 }
4434                 ret = ext4_map_blocks(handle, inode, &map,
4435                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4436                 if (ret <= 0) {
4437                         WARN_ON(ret <= 0);
4438                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
4439                                     "returned error inode#%lu, block=%u, "
4440                                     "max_blocks=%u", __func__,
4441                                     inode->i_ino, map.m_lblk, map.m_len);
4442                 }
4443                 ext4_mark_inode_dirty(handle, inode);
4444                 ret2 = ext4_journal_stop(handle);
4445                 if (ret <= 0 || ret2 )
4446                         break;
4447         }
4448         return ret > 0 ? ret2 : ret;
4449 }
4450
4451 /*
4452  * Callback function called for each extent to gather FIEMAP information.
4453  */
4454 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4455                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
4456                        void *data)
4457 {
4458         __u64   logical;
4459         __u64   physical;
4460         __u64   length;
4461         __u32   flags = 0;
4462         int             ret = 0;
4463         struct fiemap_extent_info *fieinfo = data;
4464         unsigned char blksize_bits;
4465
4466         blksize_bits = inode->i_sb->s_blocksize_bits;
4467         logical = (__u64)newex->ec_block << blksize_bits;
4468
4469         if (newex->ec_start == 0) {
4470                 /*
4471                  * No extent in extent-tree contains block @newex->ec_start,
4472                  * then the block may stay in 1)a hole or 2)delayed-extent.
4473                  *
4474                  * Holes or delayed-extents are processed as follows.
4475                  * 1. lookup dirty pages with specified range in pagecache.
4476                  *    If no page is got, then there is no delayed-extent and
4477                  *    return with EXT_CONTINUE.
4478                  * 2. find the 1st mapped buffer,
4479                  * 3. check if the mapped buffer is both in the request range
4480                  *    and a delayed buffer. If not, there is no delayed-extent,
4481                  *    then return.
4482                  * 4. a delayed-extent is found, the extent will be collected.
4483                  */
4484                 ext4_lblk_t     end = 0;
4485                 pgoff_t         last_offset;
4486                 pgoff_t         offset;
4487                 pgoff_t         index;
4488                 pgoff_t         start_index = 0;
4489                 struct page     **pages = NULL;
4490                 struct buffer_head *bh = NULL;
4491                 struct buffer_head *head = NULL;
4492                 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4493
4494                 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4495                 if (pages == NULL)
4496                         return -ENOMEM;
4497
4498                 offset = logical >> PAGE_SHIFT;
4499 repeat:
4500                 last_offset = offset;
4501                 head = NULL;
4502                 ret = find_get_pages_tag(inode->i_mapping, &offset,
4503                                         PAGECACHE_TAG_DIRTY, nr_pages, pages);
4504
4505                 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4506                         /* First time, try to find a mapped buffer. */
4507                         if (ret == 0) {
4508 out:
4509                                 for (index = 0; index < ret; index++)
4510                                         page_cache_release(pages[index]);
4511                                 /* just a hole. */
4512                                 kfree(pages);
4513                                 return EXT_CONTINUE;
4514                         }
4515                         index = 0;
4516
4517 next_page:
4518                         /* Try to find the 1st mapped buffer. */
4519                         end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4520                                   blksize_bits;
4521                         if (!page_has_buffers(pages[index]))
4522                                 goto out;
4523                         head = page_buffers(pages[index]);
4524                         if (!head)
4525                                 goto out;
4526
4527                         index++;
4528                         bh = head;
4529                         do {
4530                                 if (end >= newex->ec_block +
4531                                         newex->ec_len)
4532                                         /* The buffer is out of
4533                                          * the request range.
4534                                          */
4535                                         goto out;
4536
4537                                 if (buffer_mapped(bh) &&
4538                                     end >= newex->ec_block) {
4539                                         start_index = index - 1;
4540                                         /* get the 1st mapped buffer. */
4541                                         goto found_mapped_buffer;
4542                                 }
4543
4544                                 bh = bh->b_this_page;
4545                                 end++;
4546                         } while (bh != head);
4547
4548                         /* No mapped buffer in the range found in this page,
4549                          * We need to look up next page.
4550                          */
4551                         if (index >= ret) {
4552                                 /* There is no page left, but we need to limit
4553                                  * newex->ec_len.
4554                                  */
4555                                 newex->ec_len = end - newex->ec_block;
4556                                 goto out;
4557                         }
4558                         goto next_page;
4559                 } else {
4560                         /*Find contiguous delayed buffers. */
4561                         if (ret > 0 && pages[0]->index == last_offset)
4562                                 head = page_buffers(pages[0]);
4563                         bh = head;
4564                         index = 1;
4565                         start_index = 0;
4566                 }
4567
4568 found_mapped_buffer:
4569                 if (bh != NULL && buffer_delay(bh)) {
4570                         /* 1st or contiguous delayed buffer found. */
4571                         if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4572                                 /*
4573                                  * 1st delayed buffer found, record
4574                                  * the start of extent.
4575                                  */
4576                                 flags |= FIEMAP_EXTENT_DELALLOC;
4577                                 newex->ec_block = end;
4578                                 logical = (__u64)end << blksize_bits;
4579                         }
4580                         /* Find contiguous delayed buffers. */
4581                         do {
4582                                 if (!buffer_delay(bh))
4583                                         goto found_delayed_extent;
4584                                 bh = bh->b_this_page;
4585                                 end++;
4586                         } while (bh != head);
4587
4588                         for (; index < ret; index++) {
4589                                 if (!page_has_buffers(pages[index])) {
4590                                         bh = NULL;
4591                                         break;
4592                                 }
4593                                 head = page_buffers(pages[index]);
4594                                 if (!head) {
4595                                         bh = NULL;
4596                                         break;
4597                                 }
4598
4599                                 if (pages[index]->index !=
4600                                     pages[start_index]->index + index
4601                                     - start_index) {
4602                                         /* Blocks are not contiguous. */
4603                                         bh = NULL;
4604                                         break;
4605                                 }
4606                                 bh = head;
4607                                 do {
4608                                         if (!buffer_delay(bh))
4609                                                 /* Delayed-extent ends. */
4610                                                 goto found_delayed_extent;
4611                                         bh = bh->b_this_page;
4612                                         end++;
4613                                 } while (bh != head);
4614                         }
4615                 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4616                         /* a hole found. */
4617                         goto out;
4618
4619 found_delayed_extent:
4620                 newex->ec_len = min(end - newex->ec_block,
4621                                                 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4622                 if (ret == nr_pages && bh != NULL &&
4623                         newex->ec_len < EXT_INIT_MAX_LEN &&
4624                         buffer_delay(bh)) {
4625                         /* Have not collected an extent and continue. */
4626                         for (index = 0; index < ret; index++)
4627                                 page_cache_release(pages[index]);
4628                         goto repeat;
4629                 }
4630
4631                 for (index = 0; index < ret; index++)
4632                         page_cache_release(pages[index]);
4633                 kfree(pages);
4634         }
4635
4636         physical = (__u64)newex->ec_start << blksize_bits;
4637         length =   (__u64)newex->ec_len << blksize_bits;
4638
4639         if (ex && ext4_ext_is_uninitialized(ex))
4640                 flags |= FIEMAP_EXTENT_UNWRITTEN;
4641
4642         if (next == EXT_MAX_BLOCKS)
4643                 flags |= FIEMAP_EXTENT_LAST;
4644
4645         ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4646                                         length, flags);
4647         if (ret < 0)
4648                 return ret;
4649         if (ret == 1)
4650                 return EXT_BREAK;
4651         return EXT_CONTINUE;
4652 }
4653 /* fiemap flags we can handle specified here */
4654 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4655
4656 static int ext4_xattr_fiemap(struct inode *inode,
4657                                 struct fiemap_extent_info *fieinfo)
4658 {
4659         __u64 physical = 0;
4660         __u64 length;
4661         __u32 flags = FIEMAP_EXTENT_LAST;
4662         int blockbits = inode->i_sb->s_blocksize_bits;
4663         int error = 0;
4664
4665         /* in-inode? */
4666         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4667                 struct ext4_iloc iloc;
4668                 int offset;     /* offset of xattr in inode */
4669
4670                 error = ext4_get_inode_loc(inode, &iloc);
4671                 if (error)
4672                         return error;
4673                 physical = iloc.bh->b_blocknr << blockbits;
4674                 offset = EXT4_GOOD_OLD_INODE_SIZE +
4675                                 EXT4_I(inode)->i_extra_isize;
4676                 physical += offset;
4677                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4678                 flags |= FIEMAP_EXTENT_DATA_INLINE;
4679                 brelse(iloc.bh);
4680         } else { /* external block */
4681                 physical = EXT4_I(inode)->i_file_acl << blockbits;
4682                 length = inode->i_sb->s_blocksize;
4683         }
4684
4685         if (physical)
4686                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4687                                                 length, flags);
4688         return (error < 0 ? error : 0);
4689 }
4690
4691 /*
4692  * ext4_ext_punch_hole
4693  *
4694  * Punches a hole of "length" bytes in a file starting
4695  * at byte "offset"
4696  *
4697  * @inode:  The inode of the file to punch a hole in
4698  * @offset: The starting byte offset of the hole
4699  * @length: The length of the hole
4700  *
4701  * Returns the number of blocks removed or negative on err
4702  */
4703 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4704 {
4705         struct inode *inode = file->f_path.dentry->d_inode;
4706         struct super_block *sb = inode->i_sb;
4707         struct ext4_ext_cache cache_ex;
4708         ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4709         struct address_space *mapping = inode->i_mapping;
4710         struct ext4_map_blocks map;
4711         handle_t *handle;
4712         loff_t first_page, last_page, page_len;
4713         loff_t first_page_offset, last_page_offset;
4714         int ret, credits, blocks_released, err = 0;
4715
4716         /* No need to punch hole beyond i_size */
4717         if (offset >= inode->i_size)
4718                 return 0;
4719
4720         /*
4721          * If the hole extends beyond i_size, set the hole
4722          * to end after the page that contains i_size
4723          */
4724         if (offset + length > inode->i_size) {
4725                 length = inode->i_size +
4726                    PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4727                    offset;
4728         }
4729
4730         first_block = (offset + sb->s_blocksize - 1) >>
4731                 EXT4_BLOCK_SIZE_BITS(sb);
4732         last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4733
4734         first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4735         last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4736
4737         first_page_offset = first_page << PAGE_CACHE_SHIFT;
4738         last_page_offset = last_page << PAGE_CACHE_SHIFT;
4739
4740         /*
4741          * Write out all dirty pages to avoid race conditions
4742          * Then release them.
4743          */
4744         if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4745                 err = filemap_write_and_wait_range(mapping,
4746                         offset, offset + length - 1);
4747
4748                 if (err)
4749                         return err;
4750         }
4751
4752         /* Now release the pages */
4753         if (last_page_offset > first_page_offset) {
4754                 truncate_inode_pages_range(mapping, first_page_offset,
4755                                            last_page_offset-1);
4756         }
4757
4758         /* finish any pending end_io work */
4759         ext4_flush_completed_IO(inode);
4760
4761         credits = ext4_writepage_trans_blocks(inode);
4762         handle = ext4_journal_start(inode, credits);
4763         if (IS_ERR(handle))
4764                 return PTR_ERR(handle);
4765
4766         err = ext4_orphan_add(handle, inode);
4767         if (err)
4768                 goto out;
4769
4770         /*
4771          * Now we need to zero out the non-page-aligned data in the
4772          * pages at the start and tail of the hole, and unmap the buffer
4773          * heads for the block aligned regions of the page that were
4774          * completely zeroed.
4775          */
4776         if (first_page > last_page) {
4777                 /*
4778                  * If the file space being truncated is contained within a page
4779                  * just zero out and unmap the middle of that page
4780                  */
4781                 err = ext4_discard_partial_page_buffers(handle,
4782                         mapping, offset, length, 0);
4783
4784                 if (err)
4785                         goto out;
4786         } else {
4787                 /*
4788                  * zero out and unmap the partial page that contains
4789                  * the start of the hole
4790                  */
4791                 page_len  = first_page_offset - offset;
4792                 if (page_len > 0) {
4793                         err = ext4_discard_partial_page_buffers(handle, mapping,
4794                                                    offset, page_len, 0);
4795                         if (err)
4796                                 goto out;
4797                 }
4798
4799                 /*
4800                  * zero out and unmap the partial page that contains
4801                  * the end of the hole
4802                  */
4803                 page_len = offset + length - last_page_offset;
4804                 if (page_len > 0) {
4805                         err = ext4_discard_partial_page_buffers(handle, mapping,
4806                                         last_page_offset, page_len, 0);
4807                         if (err)
4808                                 goto out;
4809                 }
4810         }
4811
4812
4813         /*
4814          * If i_size is contained in the last page, we need to
4815          * unmap and zero the partial page after i_size
4816          */
4817         if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4818            inode->i_size % PAGE_CACHE_SIZE != 0) {
4819
4820                 page_len = PAGE_CACHE_SIZE -
4821                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
4822
4823                 if (page_len > 0) {
4824                         err = ext4_discard_partial_page_buffers(handle,
4825                           mapping, inode->i_size, page_len, 0);
4826
4827                         if (err)
4828                                 goto out;
4829                 }
4830         }
4831
4832         /* If there are no blocks to remove, return now */
4833         if (first_block >= last_block)
4834                 goto out;
4835
4836         down_write(&EXT4_I(inode)->i_data_sem);
4837         ext4_ext_invalidate_cache(inode);
4838         ext4_discard_preallocations(inode);
4839
4840         /*
4841          * Loop over all the blocks and identify blocks
4842          * that need to be punched out
4843          */
4844         iblock = first_block;
4845         blocks_released = 0;
4846         while (iblock < last_block) {
4847                 max_blocks = last_block - iblock;
4848                 num_blocks = 1;
4849                 memset(&map, 0, sizeof(map));
4850                 map.m_lblk = iblock;
4851                 map.m_len = max_blocks;
4852                 ret = ext4_ext_map_blocks(handle, inode, &map,
4853                         EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4854
4855                 if (ret > 0) {
4856                         blocks_released += ret;
4857                         num_blocks = ret;
4858                 } else if (ret == 0) {
4859                         /*
4860                          * If map blocks could not find the block,
4861                          * then it is in a hole.  If the hole was
4862                          * not already cached, then map blocks should
4863                          * put it in the cache.  So we can get the hole
4864                          * out of the cache
4865                          */
4866                         memset(&cache_ex, 0, sizeof(cache_ex));
4867                         if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4868                                 !cache_ex.ec_start) {
4869
4870                                 /* The hole is cached */
4871                                 num_blocks = cache_ex.ec_block +
4872                                 cache_ex.ec_len - iblock;
4873
4874                         } else {
4875                                 /* The block could not be identified */
4876                                 err = -EIO;
4877                                 break;
4878                         }
4879                 } else {
4880                         /* Map blocks error */
4881                         err = ret;
4882                         break;
4883                 }
4884
4885                 if (num_blocks == 0) {
4886                         /* This condition should never happen */
4887                         ext_debug("Block lookup failed");
4888                         err = -EIO;
4889                         break;
4890                 }
4891
4892                 iblock += num_blocks;
4893         }
4894
4895         if (blocks_released > 0) {
4896                 ext4_ext_invalidate_cache(inode);
4897                 ext4_discard_preallocations(inode);
4898         }
4899
4900         if (IS_SYNC(inode))
4901                 ext4_handle_sync(handle);
4902
4903         up_write(&EXT4_I(inode)->i_data_sem);
4904
4905 out:
4906         ext4_orphan_del(handle, inode);
4907         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4908         ext4_mark_inode_dirty(handle, inode);
4909         ext4_journal_stop(handle);
4910         return err;
4911 }
4912 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4913                 __u64 start, __u64 len)
4914 {
4915         ext4_lblk_t start_blk;
4916         int error = 0;
4917
4918         /* fallback to generic here if not in extents fmt */
4919         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4920                 return generic_block_fiemap(inode, fieinfo, start, len,
4921                         ext4_get_block);
4922
4923         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4924                 return -EBADR;
4925
4926         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4927                 error = ext4_xattr_fiemap(inode, fieinfo);
4928         } else {
4929                 ext4_lblk_t len_blks;
4930                 __u64 last_blk;
4931
4932                 start_blk = start >> inode->i_sb->s_blocksize_bits;
4933                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4934                 if (last_blk >= EXT_MAX_BLOCKS)
4935                         last_blk = EXT_MAX_BLOCKS-1;
4936                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4937
4938                 /*
4939                  * Walk the extent tree gathering extent information.
4940                  * ext4_ext_fiemap_cb will push extents back to user.
4941                  */
4942                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4943                                           ext4_ext_fiemap_cb, fieinfo);
4944         }
4945
4946         return error;
4947 }