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 /*
  48  * used by extent splitting.
  49  */
  50 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  51                                         due to ENOSPC */
  52 #define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
  53 #define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
  54
  55 #define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  56 #define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  57
  58 static __le32 ext4_extent_block_csum(struct inode *inode,
  59                                      struct ext4_extent_header *eh)
  60 {
  61         struct ext4_inode_info *ei = EXT4_I(inode);
  62         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  63         __u32 csum;
  64
  65         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  66                            EXT4_EXTENT_TAIL_OFFSET(eh));
  67         return cpu_to_le32(csum);
  68 }
  69
  70 static int ext4_extent_block_csum_verify(struct inode *inode,
  71                                          struct ext4_extent_header *eh)
  72 {
  73         struct ext4_extent_tail *et;
  74
  75         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  76                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  77                 return 1;
  78
  79         et = find_ext4_extent_tail(eh);
  80         if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  81                 return 0;
  82         return 1;
  83 }
  84
  85 static void ext4_extent_block_csum_set(struct inode *inode,
  86                                        struct ext4_extent_header *eh)
  87 {
  88         struct ext4_extent_tail *et;
  89
  90         if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  91                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  92                 return;
  93
  94         et = find_ext4_extent_tail(eh);
  95         et->et_checksum = ext4_extent_block_csum(inode, eh);
  96 }
  97
  98 static int ext4_split_extent(handle_t *handle,
  99                                 struct inode *inode,
 100                                 struct ext4_ext_path *path,
 101                                 struct ext4_map_blocks *map,
 102                                 int split_flag,
 103                                 int flags);
 104
 105 static int ext4_split_extent_at(handle_t *handle,
 106                              struct inode *inode,
 107                              struct ext4_ext_path *path,
 108                              ext4_lblk_t split,
 109                              int split_flag,
 110                              int flags);
 111
 112 static int ext4_ext_truncate_extend_restart(handle_t *handle,
 113                                             struct inode *inode,
 114                                             int needed)
 115 {
 116         int err;
 117
 118         if (!ext4_handle_valid(handle))
 119                 return 0;
 120         if (handle->h_buffer_credits > needed)
 121                 return 0;
 122         err = ext4_journal_extend(handle, needed);
 123         if (err <= 0)
 124                 return err;
 125         err = ext4_truncate_restart_trans(handle, inode, needed);
 126         if (err == 0)
 127                 err = -EAGAIN;
 128
 129         return err;
 130 }
 131
 132 /*
 133  * could return:
 134  *  - EROFS
 135  *  - ENOMEM
 136  */
 137 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 138                                 struct ext4_ext_path *path)
 139 {
 140         if (path->p_bh) {
 141                 /* path points to block */
 142                 return ext4_journal_get_write_access(handle, path->p_bh);
 143         }
 144         /* path points to leaf/index in inode body */
 145         /* we use in-core data, no need to protect them */
 146         return 0;
 147 }
 148
 149 /*
 150  * could return:
 151  *  - EROFS
 152  *  - ENOMEM
 153  *  - EIO
 154  */
 155 #define ext4_ext_dirty(handle, inode, path) \
 156                 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 157 static int __ext4_ext_dirty(const char *where, unsigned int line,
 158                             handle_t *handle, struct inode *inode,
 159                             struct ext4_ext_path *path)
 160 {
 161         int err;
 162         if (path->p_bh) {
 163                 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 164                 /* path points to block */
 165                 err = __ext4_handle_dirty_metadata(where, line, handle,
 166                                                    inode, path->p_bh);
 167         } else {
 168                 /* path points to leaf/index in inode body */
 169                 err = ext4_mark_inode_dirty(handle, inode);
 170         }
 171         return err;
 172 }
 173
 174 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 175                               struct ext4_ext_path *path,
 176                               ext4_lblk_t block)
 177 {
 178         if (path) {
 179                 int depth = path->p_depth;
 180                 struct ext4_extent *ex;
 181
 182                 /*
 183                  * Try to predict block placement assuming that we are
 184                  * filling in a file which will eventually be
 185                  * non-sparse --- i.e., in the case of libbfd writing
 186                  * an ELF object sections out-of-order but in a way
 187                  * the eventually results in a contiguous object or
 188                  * executable file, or some database extending a table
 189                  * space file.  However, this is actually somewhat
 190                  * non-ideal if we are writing a sparse file such as
 191                  * qemu or KVM writing a raw image file that is going
 192                  * to stay fairly sparse, since it will end up
 193                  * fragmenting the file system's free space.  Maybe we
 194                  * should have some hueristics or some way to allow
 195                  * userspace to pass a hint to file system,
 196                  * especially if the latter case turns out to be
 197                  * common.
 198                  */
 199                 ex = path[depth].p_ext;
 200                 if (ex) {
 201                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 202                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 203
 204                         if (block > ext_block)
 205                                 return ext_pblk + (block - ext_block);
 206                         else
 207                                 return ext_pblk - (ext_block - block);
 208                 }
 209
 210                 /* it looks like index is empty;
 211                  * try to find starting block from index itself */
 212                 if (path[depth].p_bh)
 213                         return path[depth].p_bh->b_blocknr;
 214         }
 215
 216         /* OK. use inode's group */
 217         return ext4_inode_to_goal_block(inode);
 218 }
 219
 220 /*
 221  * Allocation for a meta data block
 222  */
 223 static ext4_fsblk_t
 224 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 225                         struct ext4_ext_path *path,
 226                         struct ext4_extent *ex, int *err, unsigned int flags)
 227 {
 228         ext4_fsblk_t goal, newblock;
 229
 230         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 231         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 232                                         NULL, err);
 233         return newblock;
 234 }
 235
 236 static inline int ext4_ext_space_block(struct inode *inode, int check)
 237 {
 238         int size;
 239
 240         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 241                         / sizeof(struct ext4_extent);
 242 #ifdef AGGRESSIVE_TEST
 243         if (!check && size > 6)
 244                 size = 6;
 245 #endif
 246         return size;
 247 }
 248
 249 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 250 {
 251         int size;
 252
 253         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 254                         / sizeof(struct ext4_extent_idx);
 255 #ifdef AGGRESSIVE_TEST
 256         if (!check && size > 5)
 257                 size = 5;
 258 #endif
 259         return size;
 260 }
 261
 262 static inline int ext4_ext_space_root(struct inode *inode, int check)
 263 {
 264         int size;
 265
 266         size = sizeof(EXT4_I(inode)->i_data);
 267         size -= sizeof(struct ext4_extent_header);
 268         size /= sizeof(struct ext4_extent);
 269 #ifdef AGGRESSIVE_TEST
 270         if (!check && size > 3)
 271                 size = 3;
 272 #endif
 273         return size;
 274 }
 275
 276 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 277 {
 278         int size;
 279
 280         size = sizeof(EXT4_I(inode)->i_data);
 281         size -= sizeof(struct ext4_extent_header);
 282         size /= sizeof(struct ext4_extent_idx);
 283 #ifdef AGGRESSIVE_TEST
 284         if (!check && size > 4)
 285                 size = 4;
 286 #endif
 287         return size;
 288 }
 289
 290 /*
 291  * Calculate the number of metadata blocks needed
 292  * to allocate @blocks
 293  * Worse case is one block per extent
 294  */
 295 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 296 {
 297         struct ext4_inode_info *ei = EXT4_I(inode);
 298         int idxs;
 299
 300         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 301                 / sizeof(struct ext4_extent_idx));
 302
 303         /*
 304          * If the new delayed allocation block is contiguous with the
 305          * previous da block, it can share index blocks with the
 306          * previous block, so we only need to allocate a new index
 307          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 308          * an additional index block, and at ldxs**3 blocks, yet
 309          * another index blocks.
 310          */
 311         if (ei->i_da_metadata_calc_len &&
 312             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 313                 int num = 0;
 314
 315                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 316                         num++;
 317                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 318                         num++;
 319                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 320                         num++;
 321                         ei->i_da_metadata_calc_len = 0;
 322                 } else
 323                         ei->i_da_metadata_calc_len++;
 324                 ei->i_da_metadata_calc_last_lblock++;
 325                 return num;
 326         }
 327
 328         /*
 329          * In the worst case we need a new set of index blocks at
 330          * every level of the inode's extent tree.
 331          */
 332         ei->i_da_metadata_calc_len = 1;
 333         ei->i_da_metadata_calc_last_lblock = lblock;
 334         return ext_depth(inode) + 1;
 335 }
 336
 337 static int
 338 ext4_ext_max_entries(struct inode *inode, int depth)
 339 {
 340         int max;
 341
 342         if (depth == ext_depth(inode)) {
 343                 if (depth == 0)
 344                         max = ext4_ext_space_root(inode, 1);
 345                 else
 346                         max = ext4_ext_space_root_idx(inode, 1);
 347         } else {
 348                 if (depth == 0)
 349                         max = ext4_ext_space_block(inode, 1);
 350                 else
 351                         max = ext4_ext_space_block_idx(inode, 1);
 352         }
 353
 354         return max;
 355 }
 356
 357 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 358 {
 359         ext4_fsblk_t block = ext4_ext_pblock(ext);
 360         int len = ext4_ext_get_actual_len(ext);
 361
 362         if (len == 0)
 363                 return 0;
 364         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 365 }
 366
 367 static int ext4_valid_extent_idx(struct inode *inode,
 368                                 struct ext4_extent_idx *ext_idx)
 369 {
 370         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 371
 372         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 373 }
 374
 375 static int ext4_valid_extent_entries(struct inode *inode,
 376                                 struct ext4_extent_header *eh,
 377                                 int depth)
 378 {
 379         unsigned short entries;
 380         if (eh->eh_entries == 0)
 381                 return 1;
 382
 383         entries = le16_to_cpu(eh->eh_entries);
 384
 385         if (depth == 0) {
 386                 /* leaf entries */
 387                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 388                 while (entries) {
 389                         if (!ext4_valid_extent(inode, ext))
 390                                 return 0;
 391                         ext++;
 392                         entries--;
 393                 }
 394         } else {
 395                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 396                 while (entries) {
 397                         if (!ext4_valid_extent_idx(inode, ext_idx))
 398                                 return 0;
 399                         ext_idx++;
 400                         entries--;
 401                 }
 402         }
 403         return 1;
 404 }
 405
 406 static int __ext4_ext_check(const char *function, unsigned int line,
 407                             struct inode *inode, struct ext4_extent_header *eh,
 408                             int depth)
 409 {
 410         const char *error_msg;
 411         int max = 0;
 412
 413         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 414                 error_msg = "invalid magic";
 415                 goto corrupted;
 416         }
 417         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 418                 error_msg = "unexpected eh_depth";
 419                 goto corrupted;
 420         }
 421         if (unlikely(eh->eh_max == 0)) {
 422                 error_msg = "invalid eh_max";
 423                 goto corrupted;
 424         }
 425         max = ext4_ext_max_entries(inode, depth);
 426         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 427                 error_msg = "too large eh_max";
 428                 goto corrupted;
 429         }
 430         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 431                 error_msg = "invalid eh_entries";
 432                 goto corrupted;
 433         }
 434         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 435                 error_msg = "invalid extent entries";
 436                 goto corrupted;
 437         }
 438         /* Verify checksum on non-root extent tree nodes */
 439         if (ext_depth(inode) != depth &&
 440             !ext4_extent_block_csum_verify(inode, eh)) {
 441                 error_msg = "extent tree corrupted";
 442                 goto corrupted;
 443         }
 444         return 0;
 445
 446 corrupted:
 447         ext4_error_inode(inode, function, line, 0,
 448                         "bad header/extent: %s - magic %x, "
 449                         "entries %u, max %u(%u), depth %u(%u)",
 450                         error_msg, le16_to_cpu(eh->eh_magic),
 451                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 452                         max, le16_to_cpu(eh->eh_depth), depth);
 453
 454         return -EIO;
 455 }
 456
 457 #define ext4_ext_check(inode, eh, depth)        \
 458         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 459
 460 int ext4_ext_check_inode(struct inode *inode)
 461 {
 462         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 463 }
 464
 465 static int __ext4_ext_check_block(const char *function, unsigned int line,
 466                                   struct inode *inode,
 467                                   struct ext4_extent_header *eh,
 468                                   int depth,
 469                                   struct buffer_head *bh)
 470 {
 471         int ret;
 472
 473         if (buffer_verified(bh))
 474                 return 0;
 475         ret = ext4_ext_check(inode, eh, depth);
 476         if (ret)
 477                 return ret;
 478         set_buffer_verified(bh);
 479         return ret;
 480 }
 481
 482 #define ext4_ext_check_block(inode, eh, depth, bh)      \
 483         __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
 484
 485 #ifdef EXT_DEBUG
 486 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 487 {
 488         int k, l = path->p_depth;
 489
 490         ext_debug("path:");
 491         for (k = 0; k <= l; k++, path++) {
 492                 if (path->p_idx) {
 493                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 494                             ext4_idx_pblock(path->p_idx));
 495                 } else if (path->p_ext) {
 496                         ext_debug("  %d:[%d]%d:%llu ",
 497                                   le32_to_cpu(path->p_ext->ee_block),
 498                                   ext4_ext_is_uninitialized(path->p_ext),
 499                                   ext4_ext_get_actual_len(path->p_ext),
 500                                   ext4_ext_pblock(path->p_ext));
 501                 } else
 502                         ext_debug("  []");
 503         }
 504         ext_debug("\n");
 505 }
 506
 507 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 508 {
 509         int depth = ext_depth(inode);
 510         struct ext4_extent_header *eh;
 511         struct ext4_extent *ex;
 512         int i;
 513
 514         if (!path)
 515                 return;
 516
 517         eh = path[depth].p_hdr;
 518         ex = EXT_FIRST_EXTENT(eh);
 519
 520         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 521
 522         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 523                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 524                           ext4_ext_is_uninitialized(ex),
 525                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 526         }
 527         ext_debug("\n");
 528 }
 529
 530 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 531                         ext4_fsblk_t newblock, int level)
 532 {
 533         int depth = ext_depth(inode);
 534         struct ext4_extent *ex;
 535
 536         if (depth != level) {
 537                 struct ext4_extent_idx *idx;
 538                 idx = path[level].p_idx;
 539                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 540                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
 541                                         le32_to_cpu(idx->ei_block),
 542                                         ext4_idx_pblock(idx),
 543                                         newblock);
 544                         idx++;
 545                 }
 546
 547                 return;
 548         }
 549
 550         ex = path[depth].p_ext;
 551         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 552                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 553                                 le32_to_cpu(ex->ee_block),
 554                                 ext4_ext_pblock(ex),
 555                                 ext4_ext_is_uninitialized(ex),
 556                                 ext4_ext_get_actual_len(ex),
 557                                 newblock);
 558                 ex++;
 559         }
 560 }
 561
 562 #else
 563 #define ext4_ext_show_path(inode, path)
 564 #define ext4_ext_show_leaf(inode, path)
 565 #define ext4_ext_show_move(inode, path, newblock, level)
 566 #endif
 567
 568 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 569 {
 570         int depth = path->p_depth;
 571         int i;
 572
 573         for (i = 0; i <= depth; i++, path++)
 574                 if (path->p_bh) {
 575                         brelse(path->p_bh);
 576                         path->p_bh = NULL;
 577                 }
 578 }
 579
 580 /*
 581  * ext4_ext_binsearch_idx:
 582  * binary search for the closest index of the given block
 583  * the header must be checked before calling this
 584  */
 585 static void
 586 ext4_ext_binsearch_idx(struct inode *inode,
 587                         struct ext4_ext_path *path, ext4_lblk_t block)
 588 {
 589         struct ext4_extent_header *eh = path->p_hdr;
 590         struct ext4_extent_idx *r, *l, *m;
 591
 592
 593         ext_debug("binsearch for %u(idx):  ", block);
 594
 595         l = EXT_FIRST_INDEX(eh) + 1;
 596         r = EXT_LAST_INDEX(eh);
 597         while (l <= r) {
 598                 m = l + (r - l) / 2;
 599                 if (block < le32_to_cpu(m->ei_block))
 600                         r = m - 1;
 601                 else
 602                         l = m + 1;
 603                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 604                                 m, le32_to_cpu(m->ei_block),
 605                                 r, le32_to_cpu(r->ei_block));
 606         }
 607
 608         path->p_idx = l - 1;
 609         ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 610                   ext4_idx_pblock(path->p_idx));
 611
 612 #ifdef CHECK_BINSEARCH
 613         {
 614                 struct ext4_extent_idx *chix, *ix;
 615                 int k;
 616
 617                 chix = ix = EXT_FIRST_INDEX(eh);
 618                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 619                   if (k != 0 &&
 620                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 621                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 622                                        "first=0x%p\n", k,
 623                                        ix, EXT_FIRST_INDEX(eh));
 624                                 printk(KERN_DEBUG "%u <= %u\n",
 625                                        le32_to_cpu(ix->ei_block),
 626                                        le32_to_cpu(ix[-1].ei_block));
 627                         }
 628                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 629                                            <= le32_to_cpu(ix[-1].ei_block));
 630                         if (block < le32_to_cpu(ix->ei_block))
 631                                 break;
 632                         chix = ix;
 633                 }
 634                 BUG_ON(chix != path->p_idx);
 635         }
 636 #endif
 637
 638 }
 639
 640 /*
 641  * ext4_ext_binsearch:
 642  * binary search for closest extent of the given block
 643  * the header must be checked before calling this
 644  */
 645 static void
 646 ext4_ext_binsearch(struct inode *inode,
 647                 struct ext4_ext_path *path, ext4_lblk_t block)
 648 {
 649         struct ext4_extent_header *eh = path->p_hdr;
 650         struct ext4_extent *r, *l, *m;
 651
 652         if (eh->eh_entries == 0) {
 653                 /*
 654                  * this leaf is empty:
 655                  * we get such a leaf in split/add case
 656                  */
 657                 return;
 658         }
 659
 660         ext_debug("binsearch for %u:  ", block);
 661
 662         l = EXT_FIRST_EXTENT(eh) + 1;
 663         r = EXT_LAST_EXTENT(eh);
 664
 665         while (l <= r) {
 666                 m = l + (r - l) / 2;
 667                 if (block < le32_to_cpu(m->ee_block))
 668                         r = m - 1;
 669                 else
 670                         l = m + 1;
 671                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 672                                 m, le32_to_cpu(m->ee_block),
 673                                 r, le32_to_cpu(r->ee_block));
 674         }
 675
 676         path->p_ext = l - 1;
 677         ext_debug("  -> %d:%llu:[%d]%d ",
 678                         le32_to_cpu(path->p_ext->ee_block),
 679                         ext4_ext_pblock(path->p_ext),
 680                         ext4_ext_is_uninitialized(path->p_ext),
 681                         ext4_ext_get_actual_len(path->p_ext));
 682
 683 #ifdef CHECK_BINSEARCH
 684         {
 685                 struct ext4_extent *chex, *ex;
 686                 int k;
 687
 688                 chex = ex = EXT_FIRST_EXTENT(eh);
 689                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 690                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 691                                           <= le32_to_cpu(ex[-1].ee_block));
 692                         if (block < le32_to_cpu(ex->ee_block))
 693                                 break;
 694                         chex = ex;
 695                 }
 696                 BUG_ON(chex != path->p_ext);
 697         }
 698 #endif
 699
 700 }
 701
 702 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 703 {
 704         struct ext4_extent_header *eh;
 705
 706         eh = ext_inode_hdr(inode);
 707         eh->eh_depth = 0;
 708         eh->eh_entries = 0;
 709         eh->eh_magic = EXT4_EXT_MAGIC;
 710         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 711         ext4_mark_inode_dirty(handle, inode);
 712         ext4_ext_invalidate_cache(inode);
 713         return 0;
 714 }
 715
 716 struct ext4_ext_path *
 717 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 718                                         struct ext4_ext_path *path)
 719 {
 720         struct ext4_extent_header *eh;
 721         struct buffer_head *bh;
 722         short int depth, i, ppos = 0, alloc = 0;
 723
 724         eh = ext_inode_hdr(inode);
 725         depth = ext_depth(inode);
 726
 727         /* account possible depth increase */
 728         if (!path) {
 729                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 730                                 GFP_NOFS);
 731                 if (!path)
 732                         return ERR_PTR(-ENOMEM);
 733                 alloc = 1;
 734         }
 735         path[0].p_hdr = eh;
 736         path[0].p_bh = NULL;
 737
 738         i = depth;
 739         /* walk through the tree */
 740         while (i) {
 741                 ext_debug("depth %d: num %d, max %d\n",
 742                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 743
 744                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 745                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 746                 path[ppos].p_depth = i;
 747                 path[ppos].p_ext = NULL;
 748
 749                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 750                 if (unlikely(!bh))
 751                         goto err;
 752                 if (!bh_uptodate_or_lock(bh)) {
 753                         trace_ext4_ext_load_extent(inode, block,
 754                                                 path[ppos].p_block);
 755                         if (bh_submit_read(bh) < 0) {
 756                                 put_bh(bh);
 757                                 goto err;
 758                         }
 759                 }
 760                 eh = ext_block_hdr(bh);
 761                 ppos++;
 762                 if (unlikely(ppos > depth)) {
 763                         put_bh(bh);
 764                         EXT4_ERROR_INODE(inode,
 765                                          "ppos %d > depth %d", ppos, depth);
 766                         goto err;
 767                 }
 768                 path[ppos].p_bh = bh;
 769                 path[ppos].p_hdr = eh;
 770                 i--;
 771
 772                 if (ext4_ext_check_block(inode, eh, i, bh))
 773                         goto err;
 774         }
 775
 776         path[ppos].p_depth = i;
 777         path[ppos].p_ext = NULL;
 778         path[ppos].p_idx = NULL;
 779
 780         /* find extent */
 781         ext4_ext_binsearch(inode, path + ppos, block);
 782         /* if not an empty leaf */
 783         if (path[ppos].p_ext)
 784                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 785
 786         ext4_ext_show_path(inode, path);
 787
 788         return path;
 789
 790 err:
 791         ext4_ext_drop_refs(path);
 792         if (alloc)
 793                 kfree(path);
 794         return ERR_PTR(-EIO);
 795 }
 796
 797 /*
 798  * ext4_ext_insert_index:
 799  * insert new index [@logical;@ptr] into the block at @curp;
 800  * check where to insert: before @curp or after @curp
 801  */
 802 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 803                                  struct ext4_ext_path *curp,
 804                                  int logical, ext4_fsblk_t ptr)
 805 {
 806         struct ext4_extent_idx *ix;
 807         int len, err;
 808
 809         err = ext4_ext_get_access(handle, inode, curp);
 810         if (err)
 811                 return err;
 812
 813         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 814                 EXT4_ERROR_INODE(inode,
 815                                  "logical %d == ei_block %d!",
 816                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 817                 return -EIO;
 818         }
 819
 820         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 821                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 822                 EXT4_ERROR_INODE(inode,
 823                                  "eh_entries %d >= eh_max %d!",
 824                                  le16_to_cpu(curp->p_hdr->eh_entries),
 825                                  le16_to_cpu(curp->p_hdr->eh_max));
 826                 return -EIO;
 827         }
 828
 829         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 830                 /* insert after */
 831                 ext_debug("insert new index %d after: %llu\n", logical, ptr);
 832                 ix = curp->p_idx + 1;
 833         } else {
 834                 /* insert before */
 835                 ext_debug("insert new index %d before: %llu\n", logical, ptr);
 836                 ix = curp->p_idx;
 837         }
 838
 839         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 840         BUG_ON(len < 0);
 841         if (len > 0) {
 842                 ext_debug("insert new index %d: "
 843                                 "move %d indices from 0x%p to 0x%p\n",
 844                                 logical, len, ix, ix + 1);
 845                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 846         }
 847
 848         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 849                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 850                 return -EIO;
 851         }
 852
 853         ix->ei_block = cpu_to_le32(logical);
 854         ext4_idx_store_pblock(ix, ptr);
 855         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 856
 857         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 858                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 859                 return -EIO;
 860         }
 861
 862         err = ext4_ext_dirty(handle, inode, curp);
 863         ext4_std_error(inode->i_sb, err);
 864
 865         return err;
 866 }
 867
 868 /*
 869  * ext4_ext_split:
 870  * inserts new subtree into the path, using free index entry
 871  * at depth @at:
 872  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 873  * - makes decision where to split
 874  * - moves remaining extents and index entries (right to the split point)
 875  *   into the newly allocated blocks
 876  * - initializes subtree
 877  */
 878 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 879                           unsigned int flags,
 880                           struct ext4_ext_path *path,
 881                           struct ext4_extent *newext, int at)
 882 {
 883         struct buffer_head *bh = NULL;
 884         int depth = ext_depth(inode);
 885         struct ext4_extent_header *neh;
 886         struct ext4_extent_idx *fidx;
 887         int i = at, k, m, a;
 888         ext4_fsblk_t newblock, oldblock;
 889         __le32 border;
 890         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 891         int err = 0;
 892
 893         /* make decision: where to split? */
 894         /* FIXME: now decision is simplest: at current extent */
 895
 896         /* if current leaf will be split, then we should use
 897          * border from split point */
 898         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 899                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 900                 return -EIO;
 901         }
 902         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 903                 border = path[depth].p_ext[1].ee_block;
 904                 ext_debug("leaf will be split."
 905                                 " next leaf starts at %d\n",
 906                                   le32_to_cpu(border));
 907         } else {
 908                 border = newext->ee_block;
 909                 ext_debug("leaf will be added."
 910                                 " next leaf starts at %d\n",
 911                                 le32_to_cpu(border));
 912         }
 913
 914         /*
 915          * If error occurs, then we break processing
 916          * and mark filesystem read-only. index won't
 917          * be inserted and tree will be in consistent
 918          * state. Next mount will repair buffers too.
 919          */
 920
 921         /*
 922          * Get array to track all allocated blocks.
 923          * We need this to handle errors and free blocks
 924          * upon them.
 925          */
 926         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 927         if (!ablocks)
 928                 return -ENOMEM;
 929
 930         /* allocate all needed blocks */
 931         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 932         for (a = 0; a < depth - at; a++) {
 933                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 934                                                    newext, &err, flags);
 935                 if (newblock == 0)
 936                         goto cleanup;
 937                 ablocks[a] = newblock;
 938         }
 939
 940         /* initialize new leaf */
 941         newblock = ablocks[--a];
 942         if (unlikely(newblock == 0)) {
 943                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 944                 err = -EIO;
 945                 goto cleanup;
 946         }
 947         bh = sb_getblk(inode->i_sb, newblock);
 948         if (!bh) {
 949                 err = -EIO;
 950                 goto cleanup;
 951         }
 952         lock_buffer(bh);
 953
 954         err = ext4_journal_get_create_access(handle, bh);
 955         if (err)
 956                 goto cleanup;
 957
 958         neh = ext_block_hdr(bh);
 959         neh->eh_entries = 0;
 960         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 961         neh->eh_magic = EXT4_EXT_MAGIC;
 962         neh->eh_depth = 0;
 963
 964         /* move remainder of path[depth] to the new leaf */
 965         if (unlikely(path[depth].p_hdr->eh_entries !=
 966                      path[depth].p_hdr->eh_max)) {
 967                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 968                                  path[depth].p_hdr->eh_entries,
 969                                  path[depth].p_hdr->eh_max);
 970                 err = -EIO;
 971                 goto cleanup;
 972         }
 973         /* start copy from next extent */
 974         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 975         ext4_ext_show_move(inode, path, newblock, depth);
 976         if (m) {
 977                 struct ext4_extent *ex;
 978                 ex = EXT_FIRST_EXTENT(neh);
 979                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 980                 le16_add_cpu(&neh->eh_entries, m);
 981         }
 982
 983         ext4_extent_block_csum_set(inode, neh);
 984         set_buffer_uptodate(bh);
 985         unlock_buffer(bh);
 986
 987         err = ext4_handle_dirty_metadata(handle, inode, bh);
 988         if (err)
 989                 goto cleanup;
 990         brelse(bh);
 991         bh = NULL;
 992
 993         /* correct old leaf */
 994         if (m) {
 995                 err = ext4_ext_get_access(handle, inode, path + depth);
 996                 if (err)
 997                         goto cleanup;
 998                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 999                 err = ext4_ext_dirty(handle, inode, path + depth);
1000                 if (err)
1001                         goto cleanup;
1002
1003         }
1004
1005         /* create intermediate indexes */
1006         k = depth - at - 1;
1007         if (unlikely(k < 0)) {
1008                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1009                 err = -EIO;
1010                 goto cleanup;
1011         }
1012         if (k)
1013                 ext_debug("create %d intermediate indices\n", k);
1014         /* insert new index into current index block */
1015         /* current depth stored in i var */
1016         i = depth - 1;
1017         while (k--) {
1018                 oldblock = newblock;
1019                 newblock = ablocks[--a];
1020                 bh = sb_getblk(inode->i_sb, newblock);
1021                 if (!bh) {
1022                         err = -EIO;
1023                         goto cleanup;
1024                 }
1025                 lock_buffer(bh);
1026
1027                 err = ext4_journal_get_create_access(handle, bh);
1028                 if (err)
1029                         goto cleanup;
1030
1031                 neh = ext_block_hdr(bh);
1032                 neh->eh_entries = cpu_to_le16(1);
1033                 neh->eh_magic = EXT4_EXT_MAGIC;
1034                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1035                 neh->eh_depth = cpu_to_le16(depth - i);
1036                 fidx = EXT_FIRST_INDEX(neh);
1037                 fidx->ei_block = border;
1038                 ext4_idx_store_pblock(fidx, oldblock);
1039
1040                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1041                                 i, newblock, le32_to_cpu(border), oldblock);
1042
1043                 /* move remainder of path[i] to the new index block */
1044                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1045                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1046                         EXT4_ERROR_INODE(inode,
1047                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1048                                          le32_to_cpu(path[i].p_ext->ee_block));
1049                         err = -EIO;
1050                         goto cleanup;
1051                 }
1052                 /* start copy indexes */
1053                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1054                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1055                                 EXT_MAX_INDEX(path[i].p_hdr));
1056                 ext4_ext_show_move(inode, path, newblock, i);
1057                 if (m) {
1058                         memmove(++fidx, path[i].p_idx,
1059                                 sizeof(struct ext4_extent_idx) * m);
1060                         le16_add_cpu(&neh->eh_entries, m);
1061                 }
1062                 ext4_extent_block_csum_set(inode, neh);
1063                 set_buffer_uptodate(bh);
1064                 unlock_buffer(bh);
1065
1066                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1067                 if (err)
1068                         goto cleanup;
1069                 brelse(bh);
1070                 bh = NULL;
1071
1072                 /* correct old index */
1073                 if (m) {
1074                         err = ext4_ext_get_access(handle, inode, path + i);
1075                         if (err)
1076                                 goto cleanup;
1077                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1078                         err = ext4_ext_dirty(handle, inode, path + i);
1079                         if (err)
1080                                 goto cleanup;
1081                 }
1082
1083                 i--;
1084         }
1085
1086         /* insert new index */
1087         err = ext4_ext_insert_index(handle, inode, path + at,
1088                                     le32_to_cpu(border), newblock);
1089
1090 cleanup:
1091         if (bh) {
1092                 if (buffer_locked(bh))
1093                         unlock_buffer(bh);
1094                 brelse(bh);
1095         }
1096
1097         if (err) {
1098                 /* free all allocated blocks in error case */
1099                 for (i = 0; i < depth; i++) {
1100                         if (!ablocks[i])
1101                                 continue;
1102                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1103                                          EXT4_FREE_BLOCKS_METADATA);
1104                 }
1105         }
1106         kfree(ablocks);
1107
1108         return err;
1109 }
1110
1111 /*
1112  * ext4_ext_grow_indepth:
1113  * implements tree growing procedure:
1114  * - allocates new block
1115  * - moves top-level data (index block or leaf) into the new block
1116  * - initializes new top-level, creating index that points to the
1117  *   just created block
1118  */
1119 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1120                                  unsigned int flags,
1121                                  struct ext4_extent *newext)
1122 {
1123         struct ext4_extent_header *neh;
1124         struct buffer_head *bh;
1125         ext4_fsblk_t newblock;
1126         int err = 0;
1127
1128         newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1129                 newext, &err, flags);
1130         if (newblock == 0)
1131                 return err;
1132
1133         bh = sb_getblk(inode->i_sb, newblock);
1134         if (!bh) {
1135                 err = -EIO;
1136                 ext4_std_error(inode->i_sb, err);
1137                 return err;
1138         }
1139         lock_buffer(bh);
1140
1141         err = ext4_journal_get_create_access(handle, bh);
1142         if (err) {
1143                 unlock_buffer(bh);
1144                 goto out;
1145         }
1146
1147         /* move top-level index/leaf into new block */
1148         memmove(bh->b_data, EXT4_I(inode)->i_data,
1149                 sizeof(EXT4_I(inode)->i_data));
1150
1151         /* set size of new block */
1152         neh = ext_block_hdr(bh);
1153         /* old root could have indexes or leaves
1154          * so calculate e_max right way */
1155         if (ext_depth(inode))
1156                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1157         else
1158                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1159         neh->eh_magic = EXT4_EXT_MAGIC;
1160         ext4_extent_block_csum_set(inode, neh);
1161         set_buffer_uptodate(bh);
1162         unlock_buffer(bh);
1163
1164         err = ext4_handle_dirty_metadata(handle, inode, bh);
1165         if (err)
1166                 goto out;
1167
1168         /* Update top-level index: num,max,pointer */
1169         neh = ext_inode_hdr(inode);
1170         neh->eh_entries = cpu_to_le16(1);
1171         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1172         if (neh->eh_depth == 0) {
1173                 /* Root extent block becomes index block */
1174                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1175                 EXT_FIRST_INDEX(neh)->ei_block =
1176                         EXT_FIRST_EXTENT(neh)->ee_block;
1177         }
1178         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1179                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1180                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1181                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1182
1183         le16_add_cpu(&neh->eh_depth, 1);
1184         ext4_mark_inode_dirty(handle, inode);
1185 out:
1186         brelse(bh);
1187
1188         return err;
1189 }
1190
1191 /*
1192  * ext4_ext_create_new_leaf:
1193  * finds empty index and adds new leaf.
1194  * if no free index is found, then it requests in-depth growing.
1195  */
1196 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1197                                     unsigned int flags,
1198                                     struct ext4_ext_path *path,
1199                                     struct ext4_extent *newext)
1200 {
1201         struct ext4_ext_path *curp;
1202         int depth, i, err = 0;
1203
1204 repeat:
1205         i = depth = ext_depth(inode);
1206
1207         /* walk up to the tree and look for free index entry */
1208         curp = path + depth;
1209         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1210                 i--;
1211                 curp--;
1212         }
1213
1214         /* we use already allocated block for index block,
1215          * so subsequent data blocks should be contiguous */
1216         if (EXT_HAS_FREE_INDEX(curp)) {
1217                 /* if we found index with free entry, then use that
1218                  * entry: create all needed subtree and add new leaf */
1219                 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1220                 if (err)
1221                         goto out;
1222
1223                 /* refill path */
1224                 ext4_ext_drop_refs(path);
1225                 path = ext4_ext_find_extent(inode,
1226                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1227                                     path);
1228                 if (IS_ERR(path))
1229                         err = PTR_ERR(path);
1230         } else {
1231                 /* tree is full, time to grow in depth */
1232                 err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1233                 if (err)
1234                         goto out;
1235
1236                 /* refill path */
1237                 ext4_ext_drop_refs(path);
1238                 path = ext4_ext_find_extent(inode,
1239                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1240                                     path);
1241                 if (IS_ERR(path)) {
1242                         err = PTR_ERR(path);
1243                         goto out;
1244                 }
1245
1246                 /*
1247                  * only first (depth 0 -> 1) produces free space;
1248                  * in all other cases we have to split the grown tree
1249                  */
1250                 depth = ext_depth(inode);
1251                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1252                         /* now we need to split */
1253                         goto repeat;
1254                 }
1255         }
1256
1257 out:
1258         return err;
1259 }
1260
1261 /*
1262  * search the closest allocated block to the left for *logical
1263  * and returns it at @logical + it's physical address at @phys
1264  * if *logical is the smallest allocated block, the function
1265  * returns 0 at @phys
1266  * return value contains 0 (success) or error code
1267  */
1268 static int ext4_ext_search_left(struct inode *inode,
1269                                 struct ext4_ext_path *path,
1270                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1271 {
1272         struct ext4_extent_idx *ix;
1273         struct ext4_extent *ex;
1274         int depth, ee_len;
1275
1276         if (unlikely(path == NULL)) {
1277                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1278                 return -EIO;
1279         }
1280         depth = path->p_depth;
1281         *phys = 0;
1282
1283         if (depth == 0 && path->p_ext == NULL)
1284                 return 0;
1285
1286         /* usually extent in the path covers blocks smaller
1287          * then *logical, but it can be that extent is the
1288          * first one in the file */
1289
1290         ex = path[depth].p_ext;
1291         ee_len = ext4_ext_get_actual_len(ex);
1292         if (*logical < le32_to_cpu(ex->ee_block)) {
1293                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1294                         EXT4_ERROR_INODE(inode,
1295                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1296                                          *logical, le32_to_cpu(ex->ee_block));
1297                         return -EIO;
1298                 }
1299                 while (--depth >= 0) {
1300                         ix = path[depth].p_idx;
1301                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1302                                 EXT4_ERROR_INODE(inode,
1303                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1304                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1305                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1306                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1307                                   depth);
1308                                 return -EIO;
1309                         }
1310                 }
1311                 return 0;
1312         }
1313
1314         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1315                 EXT4_ERROR_INODE(inode,
1316                                  "logical %d < ee_block %d + ee_len %d!",
1317                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1318                 return -EIO;
1319         }
1320
1321         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1322         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1323         return 0;
1324 }
1325
1326 /*
1327  * search the closest allocated block to the right for *logical
1328  * and returns it at @logical + it's physical address at @phys
1329  * if *logical is the largest allocated block, the function
1330  * returns 0 at @phys
1331  * return value contains 0 (success) or error code
1332  */
1333 static int ext4_ext_search_right(struct inode *inode,
1334                                  struct ext4_ext_path *path,
1335                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1336                                  struct ext4_extent **ret_ex)
1337 {
1338         struct buffer_head *bh = NULL;
1339         struct ext4_extent_header *eh;
1340         struct ext4_extent_idx *ix;
1341         struct ext4_extent *ex;
1342         ext4_fsblk_t block;
1343         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1344         int ee_len;
1345
1346         if (unlikely(path == NULL)) {
1347                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1348                 return -EIO;
1349         }
1350         depth = path->p_depth;
1351         *phys = 0;
1352
1353         if (depth == 0 && path->p_ext == NULL)
1354                 return 0;
1355
1356         /* usually extent in the path covers blocks smaller
1357          * then *logical, but it can be that extent is the
1358          * first one in the file */
1359
1360         ex = path[depth].p_ext;
1361         ee_len = ext4_ext_get_actual_len(ex);
1362         if (*logical < le32_to_cpu(ex->ee_block)) {
1363                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1364                         EXT4_ERROR_INODE(inode,
1365                                          "first_extent(path[%d].p_hdr) != ex",
1366                                          depth);
1367                         return -EIO;
1368                 }
1369                 while (--depth >= 0) {
1370                         ix = path[depth].p_idx;
1371                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1372                                 EXT4_ERROR_INODE(inode,
1373                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1374                                                  *logical);
1375                                 return -EIO;
1376                         }
1377                 }
1378                 goto found_extent;
1379         }
1380
1381         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1382                 EXT4_ERROR_INODE(inode,
1383                                  "logical %d < ee_block %d + ee_len %d!",
1384                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1385                 return -EIO;
1386         }
1387
1388         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1389                 /* next allocated block in this leaf */
1390                 ex++;
1391                 goto found_extent;
1392         }
1393
1394         /* go up and search for index to the right */
1395         while (--depth >= 0) {
1396                 ix = path[depth].p_idx;
1397                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1398                         goto got_index;
1399         }
1400
1401         /* we've gone up to the root and found no index to the right */
1402         return 0;
1403
1404 got_index:
1405         /* we've found index to the right, let's
1406          * follow it and find the closest allocated
1407          * block to the right */
1408         ix++;
1409         block = ext4_idx_pblock(ix);
1410         while (++depth < path->p_depth) {
1411                 bh = sb_bread(inode->i_sb, block);
1412                 if (bh == NULL)
1413                         return -EIO;
1414                 eh = ext_block_hdr(bh);
1415                 /* subtract from p_depth to get proper eh_depth */
1416                 if (ext4_ext_check_block(inode, eh,
1417                                          path->p_depth - depth, bh)) {
1418                         put_bh(bh);
1419                         return -EIO;
1420                 }
1421                 ix = EXT_FIRST_INDEX(eh);
1422                 block = ext4_idx_pblock(ix);
1423                 put_bh(bh);
1424         }
1425
1426         bh = sb_bread(inode->i_sb, block);
1427         if (bh == NULL)
1428                 return -EIO;
1429         eh = ext_block_hdr(bh);
1430         if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
1431                 put_bh(bh);
1432                 return -EIO;
1433         }
1434         ex = EXT_FIRST_EXTENT(eh);
1435 found_extent:
1436         *logical = le32_to_cpu(ex->ee_block);
1437         *phys = ext4_ext_pblock(ex);
1438         *ret_ex = ex;
1439         if (bh)
1440                 put_bh(bh);
1441         return 0;
1442 }
1443
1444 /*
1445  * ext4_ext_next_allocated_block:
1446  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1447  * NOTE: it considers block number from index entry as
1448  * allocated block. Thus, index entries have to be consistent
1449  * with leaves.
1450  */
1451 static ext4_lblk_t
1452 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1453 {
1454         int depth;
1455
1456         BUG_ON(path == NULL);
1457         depth = path->p_depth;
1458
1459         if (depth == 0 && path->p_ext == NULL)
1460                 return EXT_MAX_BLOCKS;
1461
1462         while (depth >= 0) {
1463                 if (depth == path->p_depth) {
1464                         /* leaf */
1465                         if (path[depth].p_ext &&
1466                                 path[depth].p_ext !=
1467                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1468                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1469                 } else {
1470                         /* index */
1471                         if (path[depth].p_idx !=
1472                                         EXT_LAST_INDEX(path[depth].p_hdr))
1473                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1474                 }
1475                 depth--;
1476         }
1477
1478         return EXT_MAX_BLOCKS;
1479 }
1480
1481 /*
1482  * ext4_ext_next_leaf_block:
1483  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1484  */
1485 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1486 {
1487         int depth;
1488
1489         BUG_ON(path == NULL);
1490         depth = path->p_depth;
1491
1492         /* zero-tree has no leaf blocks at all */
1493         if (depth == 0)
1494                 return EXT_MAX_BLOCKS;
1495
1496         /* go to index block */
1497         depth--;
1498
1499         while (depth >= 0) {
1500                 if (path[depth].p_idx !=
1501                                 EXT_LAST_INDEX(path[depth].p_hdr))
1502                         return (ext4_lblk_t)
1503                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1504                 depth--;
1505         }
1506
1507         return EXT_MAX_BLOCKS;
1508 }
1509
1510 /*
1511  * ext4_ext_correct_indexes:
1512  * if leaf gets modified and modified extent is first in the leaf,
1513  * then we have to correct all indexes above.
1514  * TODO: do we need to correct tree in all cases?
1515  */
1516 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1517                                 struct ext4_ext_path *path)
1518 {
1519         struct ext4_extent_header *eh;
1520         int depth = ext_depth(inode);
1521         struct ext4_extent *ex;
1522         __le32 border;
1523         int k, err = 0;
1524
1525         eh = path[depth].p_hdr;
1526         ex = path[depth].p_ext;
1527
1528         if (unlikely(ex == NULL || eh == NULL)) {
1529                 EXT4_ERROR_INODE(inode,
1530                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1531                 return -EIO;
1532         }
1533
1534         if (depth == 0) {
1535                 /* there is no tree at all */
1536                 return 0;
1537         }
1538
1539         if (ex != EXT_FIRST_EXTENT(eh)) {
1540                 /* we correct tree if first leaf got modified only */
1541                 return 0;
1542         }
1543
1544         /*
1545          * TODO: we need correction if border is smaller than current one
1546          */
1547         k = depth - 1;
1548         border = path[depth].p_ext->ee_block;
1549         err = ext4_ext_get_access(handle, inode, path + k);
1550         if (err)
1551                 return err;
1552         path[k].p_idx->ei_block = border;
1553         err = ext4_ext_dirty(handle, inode, path + k);
1554         if (err)
1555                 return err;
1556
1557         while (k--) {
1558                 /* change all left-side indexes */
1559                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1560                         break;
1561                 err = ext4_ext_get_access(handle, inode, path + k);
1562                 if (err)
1563                         break;
1564                 path[k].p_idx->ei_block = border;
1565                 err = ext4_ext_dirty(handle, inode, path + k);
1566                 if (err)
1567                         break;
1568         }
1569
1570         return err;
1571 }
1572
1573 int
1574 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1575                                 struct ext4_extent *ex2)
1576 {
1577         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1578
1579         /*
1580          * Make sure that either both extents are uninitialized, or
1581          * both are _not_.
1582          */
1583         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1584                 return 0;
1585
1586         if (ext4_ext_is_uninitialized(ex1))
1587                 max_len = EXT_UNINIT_MAX_LEN;
1588         else
1589                 max_len = EXT_INIT_MAX_LEN;
1590
1591         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1592         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1593
1594         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1595                         le32_to_cpu(ex2->ee_block))
1596                 return 0;
1597
1598         /*
1599          * To allow future support for preallocated extents to be added
1600          * as an RO_COMPAT feature, refuse to merge to extents if
1601          * this can result in the top bit of ee_len being set.
1602          */
1603         if (ext1_ee_len + ext2_ee_len > max_len)
1604                 return 0;
1605 #ifdef AGGRESSIVE_TEST
1606         if (ext1_ee_len >= 4)
1607                 return 0;
1608 #endif
1609
1610         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1611                 return 1;
1612         return 0;
1613 }
1614
1615 /*
1616  * This function tries to merge the "ex" extent to the next extent in the tree.
1617  * It always tries to merge towards right. If you want to merge towards
1618  * left, pass "ex - 1" as argument instead of "ex".
1619  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1620  * 1 if they got merged.
1621  */
1622 static int ext4_ext_try_to_merge_right(struct inode *inode,
1623                                  struct ext4_ext_path *path,
1624                                  struct ext4_extent *ex)
1625 {
1626         struct ext4_extent_header *eh;
1627         unsigned int depth, len;
1628         int merge_done = 0;
1629         int uninitialized = 0;
1630
1631         depth = ext_depth(inode);
1632         BUG_ON(path[depth].p_hdr == NULL);
1633         eh = path[depth].p_hdr;
1634
1635         while (ex < EXT_LAST_EXTENT(eh)) {
1636                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1637                         break;
1638                 /* merge with next extent! */
1639                 if (ext4_ext_is_uninitialized(ex))
1640                         uninitialized = 1;
1641                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1642                                 + ext4_ext_get_actual_len(ex + 1));
1643                 if (uninitialized)
1644                         ext4_ext_mark_uninitialized(ex);
1645
1646                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1647                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1648                                 * sizeof(struct ext4_extent);
1649                         memmove(ex + 1, ex + 2, len);
1650                 }
1651                 le16_add_cpu(&eh->eh_entries, -1);
1652                 merge_done = 1;
1653                 WARN_ON(eh->eh_entries == 0);
1654                 if (!eh->eh_entries)
1655                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1656         }
1657
1658         return merge_done;
1659 }
1660
1661 /*
1662  * This function does a very simple check to see if we can collapse
1663  * an extent tree with a single extent tree leaf block into the inode.
1664  */
1665 static void ext4_ext_try_to_merge_up(handle_t *handle,
1666                                      struct inode *inode,
1667                                      struct ext4_ext_path *path)
1668 {
1669         size_t s;
1670         unsigned max_root = ext4_ext_space_root(inode, 0);
1671         ext4_fsblk_t blk;
1672
1673         if ((path[0].p_depth != 1) ||
1674             (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1675             (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1676                 return;
1677
1678         /*
1679          * We need to modify the block allocation bitmap and the block
1680          * group descriptor to release the extent tree block.  If we
1681          * can't get the journal credits, give up.
1682          */
1683         if (ext4_journal_extend(handle, 2))
1684                 return;
1685
1686         /*
1687          * Copy the extent data up to the inode
1688          */
1689         blk = ext4_idx_pblock(path[0].p_idx);
1690         s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1691                 sizeof(struct ext4_extent_idx);
1692         s += sizeof(struct ext4_extent_header);
1693
1694         memcpy(path[0].p_hdr, path[1].p_hdr, s);
1695         path[0].p_depth = 0;
1696         path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1697                 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1698         path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1699
1700         brelse(path[1].p_bh);
1701         ext4_free_blocks(handle, inode, NULL, blk, 1,
1702                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1703 }
1704
1705 /*
1706  * This function tries to merge the @ex extent to neighbours in the tree.
1707  * return 1 if merge left else 0.
1708  */
1709 static void ext4_ext_try_to_merge(handle_t *handle,
1710                                   struct inode *inode,
1711                                   struct ext4_ext_path *path,
1712                                   struct ext4_extent *ex) {
1713         struct ext4_extent_header *eh;
1714         unsigned int depth;
1715         int merge_done = 0;
1716
1717         depth = ext_depth(inode);
1718         BUG_ON(path[depth].p_hdr == NULL);
1719         eh = path[depth].p_hdr;
1720
1721         if (ex > EXT_FIRST_EXTENT(eh))
1722                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1723
1724         if (!merge_done)
1725                 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1726
1727         ext4_ext_try_to_merge_up(handle, inode, path);
1728 }
1729
1730 /*
1731  * check if a portion of the "newext" extent overlaps with an
1732  * existing extent.
1733  *
1734  * If there is an overlap discovered, it updates the length of the newext
1735  * such that there will be no overlap, and then returns 1.
1736  * If there is no overlap found, it returns 0.
1737  */
1738 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1739                                            struct inode *inode,
1740                                            struct ext4_extent *newext,
1741                                            struct ext4_ext_path *path)
1742 {
1743         ext4_lblk_t b1, b2;
1744         unsigned int depth, len1;
1745         unsigned int ret = 0;
1746
1747         b1 = le32_to_cpu(newext->ee_block);
1748         len1 = ext4_ext_get_actual_len(newext);
1749         depth = ext_depth(inode);
1750         if (!path[depth].p_ext)
1751                 goto out;
1752         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1753         b2 &= ~(sbi->s_cluster_ratio - 1);
1754
1755         /*
1756          * get the next allocated block if the extent in the path
1757          * is before the requested block(s)
1758          */
1759         if (b2 < b1) {
1760                 b2 = ext4_ext_next_allocated_block(path);
1761                 if (b2 == EXT_MAX_BLOCKS)
1762                         goto out;
1763                 b2 &= ~(sbi->s_cluster_ratio - 1);
1764         }
1765
1766         /* check for wrap through zero on extent logical start block*/
1767         if (b1 + len1 < b1) {
1768                 len1 = EXT_MAX_BLOCKS - b1;
1769                 newext->ee_len = cpu_to_le16(len1);
1770                 ret = 1;
1771         }
1772
1773         /* check for overlap */
1774         if (b1 + len1 > b2) {
1775                 newext->ee_len = cpu_to_le16(b2 - b1);
1776                 ret = 1;
1777         }
1778 out:
1779         return ret;
1780 }
1781
1782 /*
1783  * ext4_ext_insert_extent:
1784  * tries to merge requsted extent into the existing extent or
1785  * inserts requested extent as new one into the tree,
1786  * creating new leaf in the no-space case.
1787  */
1788 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1789                                 struct ext4_ext_path *path,
1790                                 struct ext4_extent *newext, int flag)
1791 {
1792         struct ext4_extent_header *eh;
1793         struct ext4_extent *ex, *fex;
1794         struct ext4_extent *nearex; /* nearest extent */
1795         struct ext4_ext_path *npath = NULL;
1796         int depth, len, err;
1797         ext4_lblk_t next;
1798         unsigned uninitialized = 0;
1799         int flags = 0;
1800
1801         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1802                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1803                 return -EIO;
1804         }
1805         depth = ext_depth(inode);
1806         ex = path[depth].p_ext;
1807         if (unlikely(path[depth].p_hdr == NULL)) {
1808                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1809                 return -EIO;
1810         }
1811
1812         /* try to insert block into found extent and return */
1813         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1814                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1815                 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1816                           ext4_ext_is_uninitialized(newext),
1817                           ext4_ext_get_actual_len(newext),
1818                           le32_to_cpu(ex->ee_block),
1819                           ext4_ext_is_uninitialized(ex),
1820                           ext4_ext_get_actual_len(ex),
1821                           ext4_ext_pblock(ex));
1822                 err = ext4_ext_get_access(handle, inode, path + depth);
1823                 if (err)
1824                         return err;
1825
1826                 /*
1827                  * ext4_can_extents_be_merged should have checked that either
1828                  * both extents are uninitialized, or both aren't. Thus we
1829                  * need to check only one of them here.
1830                  */
1831                 if (ext4_ext_is_uninitialized(ex))
1832                         uninitialized = 1;
1833                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1834                                         + ext4_ext_get_actual_len(newext));
1835                 if (uninitialized)
1836                         ext4_ext_mark_uninitialized(ex);
1837                 eh = path[depth].p_hdr;
1838                 nearex = ex;
1839                 goto merge;
1840         }
1841
1842         depth = ext_depth(inode);
1843         eh = path[depth].p_hdr;
1844         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1845                 goto has_space;
1846
1847         /* probably next leaf has space for us? */
1848         fex = EXT_LAST_EXTENT(eh);
1849         next = EXT_MAX_BLOCKS;
1850         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1851                 next = ext4_ext_next_leaf_block(path);
1852         if (next != EXT_MAX_BLOCKS) {
1853                 ext_debug("next leaf block - %u\n", next);
1854                 BUG_ON(npath != NULL);
1855                 npath = ext4_ext_find_extent(inode, next, NULL);
1856                 if (IS_ERR(npath))
1857                         return PTR_ERR(npath);
1858                 BUG_ON(npath->p_depth != path->p_depth);
1859                 eh = npath[depth].p_hdr;
1860                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1861                         ext_debug("next leaf isn't full(%d)\n",
1862                                   le16_to_cpu(eh->eh_entries));
1863                         path = npath;
1864                         goto has_space;
1865                 }
1866                 ext_debug("next leaf has no free space(%d,%d)\n",
1867                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1868         }
1869
1870         /*
1871          * There is no free space in the found leaf.
1872          * We're gonna add a new leaf in the tree.
1873          */
1874         if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1875                 flags = EXT4_MB_USE_ROOT_BLOCKS;
1876         err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1877         if (err)
1878                 goto cleanup;
1879         depth = ext_depth(inode);
1880         eh = path[depth].p_hdr;
1881
1882 has_space:
1883         nearex = path[depth].p_ext;
1884
1885         err = ext4_ext_get_access(handle, inode, path + depth);
1886         if (err)
1887                 goto cleanup;
1888
1889         if (!nearex) {
1890                 /* there is no extent in this leaf, create first one */
1891                 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1892                                 le32_to_cpu(newext->ee_block),
1893                                 ext4_ext_pblock(newext),
1894                                 ext4_ext_is_uninitialized(newext),
1895                                 ext4_ext_get_actual_len(newext));
1896                 nearex = EXT_FIRST_EXTENT(eh);
1897         } else {
1898                 if (le32_to_cpu(newext->ee_block)
1899                            > le32_to_cpu(nearex->ee_block)) {
1900                         /* Insert after */
1901                         ext_debug("insert %u:%llu:[%d]%d before: "
1902                                         "nearest %p\n",
1903                                         le32_to_cpu(newext->ee_block),
1904                                         ext4_ext_pblock(newext),
1905                                         ext4_ext_is_uninitialized(newext),
1906                                         ext4_ext_get_actual_len(newext),
1907                                         nearex);
1908                         nearex++;
1909                 } else {
1910                         /* Insert before */
1911                         BUG_ON(newext->ee_block == nearex->ee_block);
1912                         ext_debug("insert %u:%llu:[%d]%d after: "
1913                                         "nearest %p\n",
1914                                         le32_to_cpu(newext->ee_block),
1915                                         ext4_ext_pblock(newext),
1916                                         ext4_ext_is_uninitialized(newext),
1917                                         ext4_ext_get_actual_len(newext),
1918                                         nearex);
1919                 }
1920                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
1921                 if (len > 0) {
1922                         ext_debug("insert %u:%llu:[%d]%d: "
1923                                         "move %d extents from 0x%p to 0x%p\n",
1924                                         le32_to_cpu(newext->ee_block),
1925                                         ext4_ext_pblock(newext),
1926                                         ext4_ext_is_uninitialized(newext),
1927                                         ext4_ext_get_actual_len(newext),
1928                                         len, nearex, nearex + 1);
1929                         memmove(nearex + 1, nearex,
1930                                 len * sizeof(struct ext4_extent));
1931                 }
1932         }
1933
1934         le16_add_cpu(&eh->eh_entries, 1);
1935         path[depth].p_ext = nearex;
1936         nearex->ee_block = newext->ee_block;
1937         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1938         nearex->ee_len = newext->ee_len;
1939
1940 merge:
1941         /* try to merge extents */
1942         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1943                 ext4_ext_try_to_merge(handle, inode, path, nearex);
1944
1945
1946         /* time to correct all indexes above */
1947         err = ext4_ext_correct_indexes(handle, inode, path);
1948         if (err)
1949                 goto cleanup;
1950
1951         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
1952
1953 cleanup:
1954         if (npath) {
1955                 ext4_ext_drop_refs(npath);
1956                 kfree(npath);
1957         }
1958         ext4_ext_invalidate_cache(inode);
1959         return err;
1960 }
1961
1962 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1963                                ext4_lblk_t num, ext_prepare_callback func,
1964                                void *cbdata)
1965 {
1966         struct ext4_ext_path *path = NULL;
1967         struct ext4_ext_cache cbex;
1968         struct ext4_extent *ex;
1969         ext4_lblk_t next, start = 0, end = 0;
1970         ext4_lblk_t last = block + num;
1971         int depth, exists, err = 0;
1972
1973         BUG_ON(func == NULL);
1974         BUG_ON(inode == NULL);
1975
1976         while (block < last && block != EXT_MAX_BLOCKS) {
1977                 num = last - block;
1978                 /* find extent for this block */
1979                 down_read(&EXT4_I(inode)->i_data_sem);
1980                 path = ext4_ext_find_extent(inode, block, path);
1981                 up_read(&EXT4_I(inode)->i_data_sem);
1982                 if (IS_ERR(path)) {
1983                         err = PTR_ERR(path);
1984                         path = NULL;
1985                         break;
1986                 }
1987
1988                 depth = ext_depth(inode);
1989                 if (unlikely(path[depth].p_hdr == NULL)) {
1990                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991                         err = -EIO;
1992                         break;
1993                 }
1994                 ex = path[depth].p_ext;
1995                 next = ext4_ext_next_allocated_block(path);
1996
1997                 exists = 0;
1998                 if (!ex) {
1999                         /* there is no extent yet, so try to allocate
2000                          * all requested space */
2001                         start = block;
2002                         end = block + num;
2003                 } else if (le32_to_cpu(ex->ee_block) > block) {
2004                         /* need to allocate space before found extent */
2005                         start = block;
2006                         end = le32_to_cpu(ex->ee_block);
2007                         if (block + num < end)
2008                                 end = block + num;
2009                 } else if (block >= le32_to_cpu(ex->ee_block)
2010                                         + ext4_ext_get_actual_len(ex)) {
2011                         /* need to allocate space after found extent */
2012                         start = block;
2013                         end = block + num;
2014                         if (end >= next)
2015                                 end = next;
2016                 } else if (block >= le32_to_cpu(ex->ee_block)) {
2017                         /*
2018                          * some part of requested space is covered
2019                          * by found extent
2020                          */
2021                         start = block;
2022                         end = le32_to_cpu(ex->ee_block)
2023                                 + ext4_ext_get_actual_len(ex);
2024                         if (block + num < end)
2025                                 end = block + num;
2026                         exists = 1;
2027                 } else {
2028                         BUG();
2029                 }
2030                 BUG_ON(end <= start);
2031
2032                 if (!exists) {
2033                         cbex.ec_block = start;
2034                         cbex.ec_len = end - start;
2035                         cbex.ec_start = 0;
2036                 } else {
2037                         cbex.ec_block = le32_to_cpu(ex->ee_block);
2038                         cbex.ec_len = ext4_ext_get_actual_len(ex);
2039                         cbex.ec_start = ext4_ext_pblock(ex);
2040                 }
2041
2042                 if (unlikely(cbex.ec_len == 0)) {
2043                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
2044                         err = -EIO;
2045                         break;
2046                 }
2047                 err = func(inode, next, &cbex, ex, cbdata);
2048                 ext4_ext_drop_refs(path);
2049
2050                 if (err < 0)
2051                         break;
2052
2053                 if (err == EXT_REPEAT)
2054                         continue;
2055                 else if (err == EXT_BREAK) {
2056                         err = 0;
2057                         break;
2058                 }
2059
2060                 if (ext_depth(inode) != depth) {
2061                         /* depth was changed. we have to realloc path */
2062                         kfree(path);
2063                         path = NULL;
2064                 }
2065
2066                 block = cbex.ec_block + cbex.ec_len;
2067         }
2068
2069         if (path) {
2070                 ext4_ext_drop_refs(path);
2071                 kfree(path);
2072         }
2073
2074         return err;
2075 }
2076
2077 static void
2078 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
2079                         __u32 len, ext4_fsblk_t start)
2080 {
2081         struct ext4_ext_cache *cex;
2082         BUG_ON(len == 0);
2083         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2084         trace_ext4_ext_put_in_cache(inode, block, len, start);
2085         cex = &EXT4_I(inode)->i_cached_extent;
2086         cex->ec_block = block;
2087         cex->ec_len = len;
2088         cex->ec_start = start;
2089         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2090 }
2091
2092 /*
2093  * ext4_ext_put_gap_in_cache:
2094  * calculate boundaries of the gap that the requested block fits into
2095  * and cache this gap
2096  */
2097 static void
2098 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2099                                 ext4_lblk_t block)
2100 {
2101         int depth = ext_depth(inode);
2102         unsigned long len;
2103         ext4_lblk_t lblock;
2104         struct ext4_extent *ex;
2105
2106         ex = path[depth].p_ext;
2107         if (ex == NULL) {
2108                 /* there is no extent yet, so gap is [0;-] */
2109                 lblock = 0;
2110                 len = EXT_MAX_BLOCKS;
2111                 ext_debug("cache gap(whole file):");
2112         } else if (block < le32_to_cpu(ex->ee_block)) {
2113                 lblock = block;
2114                 len = le32_to_cpu(ex->ee_block) - block;
2115                 ext_debug("cache gap(before): %u [%u:%u]",
2116                                 block,
2117                                 le32_to_cpu(ex->ee_block),
2118                                  ext4_ext_get_actual_len(ex));
2119         } else if (block >= le32_to_cpu(ex->ee_block)
2120                         + ext4_ext_get_actual_len(ex)) {
2121                 ext4_lblk_t next;
2122                 lblock = le32_to_cpu(ex->ee_block)
2123                         + ext4_ext_get_actual_len(ex);
2124
2125                 next = ext4_ext_next_allocated_block(path);
2126                 ext_debug("cache gap(after): [%u:%u] %u",
2127                                 le32_to_cpu(ex->ee_block),
2128                                 ext4_ext_get_actual_len(ex),
2129                                 block);
2130                 BUG_ON(next == lblock);
2131                 len = next - lblock;
2132         } else {
2133                 lblock = len = 0;
2134                 BUG();
2135         }
2136
2137         ext_debug(" -> %u:%lu\n", lblock, len);
2138         ext4_ext_put_in_cache(inode, lblock, len, 0);
2139 }
2140
2141 /*
2142  * ext4_ext_in_cache()
2143  * Checks to see if the given block is in the cache.
2144  * If it is, the cached extent is stored in the given
2145  * cache extent pointer.
2146  *
2147  * @inode: The files inode
2148  * @block: The block to look for in the cache
2149  * @ex:    Pointer where the cached extent will be stored
2150  *         if it contains block
2151  *
2152  * Return 0 if cache is invalid; 1 if the cache is valid
2153  */
2154 static int
2155 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2156                   struct ext4_extent *ex)
2157 {
2158         struct ext4_ext_cache *cex;
2159         struct ext4_sb_info *sbi;
2160         int ret = 0;
2161
2162         /*
2163          * We borrow i_block_reservation_lock to protect i_cached_extent
2164          */
2165         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2166         cex = &EXT4_I(inode)->i_cached_extent;
2167         sbi = EXT4_SB(inode->i_sb);
2168
2169         /* has cache valid data? */
2170         if (cex->ec_len == 0)
2171                 goto errout;
2172
2173         if (in_range(block, cex->ec_block, cex->ec_len)) {
2174                 ex->ee_block = cpu_to_le32(cex->ec_block);
2175                 ext4_ext_store_pblock(ex, cex->ec_start);
2176                 ex->ee_len = cpu_to_le16(cex->ec_len);
2177                 ext_debug("%u cached by %u:%u:%llu\n",
2178                                 block,
2179                                 cex->ec_block, cex->ec_len, cex->ec_start);
2180                 ret = 1;
2181         }
2182 errout:
2183         trace_ext4_ext_in_cache(inode, block, ret);
2184         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2185         return ret;
2186 }
2187
2188 /*
2189  * ext4_ext_rm_idx:
2190  * removes index from the index block.
2191  */
2192 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2193                         struct ext4_ext_path *path)
2194 {
2195         int err;
2196         ext4_fsblk_t leaf;
2197
2198         /* free index block */
2199         path--;
2200         leaf = ext4_idx_pblock(path->p_idx);
2201         if (unlikely(path->p_hdr->eh_entries == 0)) {
2202                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2203                 return -EIO;
2204         }
2205         err = ext4_ext_get_access(handle, inode, path);
2206         if (err)
2207                 return err;
2208
2209         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2210                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2211                 len *= sizeof(struct ext4_extent_idx);
2212                 memmove(path->p_idx, path->p_idx + 1, len);
2213         }
2214
2215         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2216         err = ext4_ext_dirty(handle, inode, path);
2217         if (err)
2218                 return err;
2219         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2220         trace_ext4_ext_rm_idx(inode, leaf);
2221
2222         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2223                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2224         return err;
2225 }
2226
2227 /*
2228  * ext4_ext_calc_credits_for_single_extent:
2229  * This routine returns max. credits that needed to insert an extent
2230  * to the extent tree.
2231  * When pass the actual path, the caller should calculate credits
2232  * under i_data_sem.
2233  */
2234 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2235                                                 struct ext4_ext_path *path)
2236 {
2237         if (path) {
2238                 int depth = ext_depth(inode);
2239                 int ret = 0;
2240
2241                 /* probably there is space in leaf? */
2242                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2243                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2244
2245                         /*
2246                          *  There are some space in the leaf tree, no
2247                          *  need to account for leaf block credit
2248                          *
2249                          *  bitmaps and block group descriptor blocks
2250                          *  and other metadata blocks still need to be
2251                          *  accounted.
2252                          */
2253                         /* 1 bitmap, 1 block group descriptor */
2254                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2255                         return ret;
2256                 }
2257         }
2258
2259         return ext4_chunk_trans_blocks(inode, nrblocks);
2260 }
2261
2262 /*
2263  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2264  *
2265  * if nrblocks are fit in a single extent (chunk flag is 1), then
2266  * in the worse case, each tree level index/leaf need to be changed
2267  * if the tree split due to insert a new extent, then the old tree
2268  * index/leaf need to be updated too
2269  *
2270  * If the nrblocks are discontiguous, they could cause
2271  * the whole tree split more than once, but this is really rare.
2272  */
2273 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2274 {
2275         int index;
2276         int depth = ext_depth(inode);
2277
2278         if (chunk)
2279                 index = depth * 2;
2280         else
2281                 index = depth * 3;
2282
2283         return index;
2284 }
2285
2286 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2287                               struct ext4_extent *ex,
2288                               ext4_fsblk_t *partial_cluster,
2289                               ext4_lblk_t from, ext4_lblk_t to)
2290 {
2291         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2292         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2293         ext4_fsblk_t pblk;
2294         int flags = 0;
2295
2296         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2297                 flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2298         else if (ext4_should_journal_data(inode))
2299                 flags |= EXT4_FREE_BLOCKS_FORGET;
2300
2301         /*
2302          * For bigalloc file systems, we never free a partial cluster
2303          * at the beginning of the extent.  Instead, we make a note
2304          * that we tried freeing the cluster, and check to see if we
2305          * need to free it on a subsequent call to ext4_remove_blocks,
2306          * or at the end of the ext4_truncate() operation.
2307          */
2308         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2309
2310         trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2311         /*
2312          * If we have a partial cluster, and it's different from the
2313          * cluster of the last block, we need to explicitly free the
2314          * partial cluster here.
2315          */
2316         pblk = ext4_ext_pblock(ex) + ee_len - 1;
2317         if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2318                 ext4_free_blocks(handle, inode, NULL,
2319                                  EXT4_C2B(sbi, *partial_cluster),
2320                                  sbi->s_cluster_ratio, flags);
2321                 *partial_cluster = 0;
2322         }
2323
2324 #ifdef EXTENTS_STATS
2325         {
2326                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2327                 spin_lock(&sbi->s_ext_stats_lock);
2328                 sbi->s_ext_blocks += ee_len;
2329                 sbi->s_ext_extents++;
2330                 if (ee_len < sbi->s_ext_min)
2331                         sbi->s_ext_min = ee_len;
2332                 if (ee_len > sbi->s_ext_max)
2333                         sbi->s_ext_max = ee_len;
2334                 if (ext_depth(inode) > sbi->s_depth_max)
2335                         sbi->s_depth_max = ext_depth(inode);
2336                 spin_unlock(&sbi->s_ext_stats_lock);
2337         }
2338 #endif
2339         if (from >= le32_to_cpu(ex->ee_block)
2340             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2341                 /* tail removal */
2342                 ext4_lblk_t num;
2343
2344                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2345                 pblk = ext4_ext_pblock(ex) + ee_len - num;
2346                 ext_debug("free last %u blocks starting %llu\n", num, pblk);
2347                 ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2348                 /*
2349                  * If the block range to be freed didn't start at the
2350                  * beginning of a cluster, and we removed the entire
2351                  * extent, save the partial cluster here, since we
2352                  * might need to delete if we determine that the
2353                  * truncate operation has removed all of the blocks in
2354                  * the cluster.
2355                  */
2356                 if (pblk & (sbi->s_cluster_ratio - 1) &&
2357                     (ee_len == num))
2358                         *partial_cluster = EXT4_B2C(sbi, pblk);
2359                 else
2360                         *partial_cluster = 0;
2361         } else if (from == le32_to_cpu(ex->ee_block)
2362                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2363                 /* head removal */
2364                 ext4_lblk_t num;
2365                 ext4_fsblk_t start;
2366
2367                 num = to - from;
2368                 start = ext4_ext_pblock(ex);
2369
2370                 ext_debug("free first %u blocks starting %llu\n", num, start);
2371                 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2372
2373         } else {
2374                 printk(KERN_INFO "strange request: removal(2) "
2375                                 "%u-%u from %u:%u\n",
2376                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2377         }
2378         return 0;
2379 }
2380
2381
2382 /*
2383  * ext4_ext_rm_leaf() Removes the extents associated with the
2384  * blocks appearing between "start" and "end", and splits the extents
2385  * if "start" and "end" appear in the same extent
2386  *
2387  * @handle: The journal handle
2388  * @inode:  The files inode
2389  * @path:   The path to the leaf
2390  * @start:  The first block to remove
2391  * @end:   The last block to remove
2392  */
2393 static int
2394 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2395                  struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2396                  ext4_lblk_t start, ext4_lblk_t end)
2397 {
2398         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2399         int err = 0, correct_index = 0;
2400         int depth = ext_depth(inode), credits;
2401         struct ext4_extent_header *eh;
2402         ext4_lblk_t a, b;
2403         unsigned num;
2404         ext4_lblk_t ex_ee_block;
2405         unsigned short ex_ee_len;
2406         unsigned uninitialized = 0;
2407         struct ext4_extent *ex;
2408
2409         /* the header must be checked already in ext4_ext_remove_space() */
2410         ext_debug("truncate since %u in leaf to %u\n", start, end);
2411         if (!path[depth].p_hdr)
2412                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2413         eh = path[depth].p_hdr;
2414         if (unlikely(path[depth].p_hdr == NULL)) {
2415                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2416                 return -EIO;
2417         }
2418         /* find where to start removing */
2419         ex = EXT_LAST_EXTENT(eh);
2420
2421         ex_ee_block = le32_to_cpu(ex->ee_block);
2422         ex_ee_len = ext4_ext_get_actual_len(ex);
2423
2424         trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2425
2426         while (ex >= EXT_FIRST_EXTENT(eh) &&
2427                         ex_ee_block + ex_ee_len > start) {
2428
2429                 if (ext4_ext_is_uninitialized(ex))
2430                         uninitialized = 1;
2431                 else
2432                         uninitialized = 0;
2433
2434                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2435                          uninitialized, ex_ee_len);
2436                 path[depth].p_ext = ex;
2437
2438                 a = ex_ee_block > start ? ex_ee_block : start;
2439                 b = ex_ee_block+ex_ee_len - 1 < end ?
2440                         ex_ee_block+ex_ee_len - 1 : end;
2441
2442                 ext_debug("  border %u:%u\n", a, b);
2443
2444                 /* If this extent is beyond the end of the hole, skip it */
2445                 if (end < ex_ee_block) {
2446                         ex--;
2447                         ex_ee_block = le32_to_cpu(ex->ee_block);
2448                         ex_ee_len = ext4_ext_get_actual_len(ex);
2449                         continue;
2450                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2451                         EXT4_ERROR_INODE(inode,
2452                                          "can not handle truncate %u:%u "
2453                                          "on extent %u:%u",
2454                                          start, end, ex_ee_block,
2455                                          ex_ee_block + ex_ee_len - 1);
2456                         err = -EIO;
2457                         goto out;
2458                 } else if (a != ex_ee_block) {
2459                         /* remove tail of the extent */
2460                         num = a - ex_ee_block;
2461                 } else {
2462                         /* remove whole extent: excellent! */
2463                         num = 0;
2464                 }
2465                 /*
2466                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2467                  * descriptor) for each block group; assume two block
2468                  * groups plus ex_ee_len/blocks_per_block_group for
2469                  * the worst case
2470                  */
2471                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2472                 if (ex == EXT_FIRST_EXTENT(eh)) {
2473                         correct_index = 1;
2474                         credits += (ext_depth(inode)) + 1;
2475                 }
2476                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2477
2478                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2479                 if (err)
2480                         goto out;
2481
2482                 err = ext4_ext_get_access(handle, inode, path + depth);
2483                 if (err)
2484                         goto out;
2485
2486                 err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2487                                          a, b);
2488                 if (err)
2489                         goto out;
2490
2491                 if (num == 0)
2492                         /* this extent is removed; mark slot entirely unused */
2493                         ext4_ext_store_pblock(ex, 0);
2494
2495                 ex->ee_len = cpu_to_le16(num);
2496                 /*
2497                  * Do not mark uninitialized if all the blocks in the
2498                  * extent have been removed.
2499                  */
2500                 if (uninitialized && num)
2501                         ext4_ext_mark_uninitialized(ex);
2502                 /*
2503                  * If the extent was completely released,
2504                  * we need to remove it from the leaf
2505                  */
2506                 if (num == 0) {
2507                         if (end != EXT_MAX_BLOCKS - 1) {
2508                                 /*
2509                                  * For hole punching, we need to scoot all the
2510                                  * extents up when an extent is removed so that
2511                                  * we dont have blank extents in the middle
2512                                  */
2513                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2514                                         sizeof(struct ext4_extent));
2515
2516                                 /* Now get rid of the one at the end */
2517                                 memset(EXT_LAST_EXTENT(eh), 0,
2518                                         sizeof(struct ext4_extent));
2519                         }
2520                         le16_add_cpu(&eh->eh_entries, -1);
2521                 } else
2522                         *partial_cluster = 0;
2523
2524                 err = ext4_ext_dirty(handle, inode, path + depth);
2525                 if (err)
2526                         goto out;
2527
2528                 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2529                                 ext4_ext_pblock(ex));
2530                 ex--;
2531                 ex_ee_block = le32_to_cpu(ex->ee_block);
2532                 ex_ee_len = ext4_ext_get_actual_len(ex);
2533         }
2534
2535         if (correct_index && eh->eh_entries)
2536                 err = ext4_ext_correct_indexes(handle, inode, path);
2537
2538         /*
2539          * If there is still a entry in the leaf node, check to see if
2540          * it references the partial cluster.  This is the only place
2541          * where it could; if it doesn't, we can free the cluster.
2542          */
2543         if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2544             (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2545              *partial_cluster)) {
2546                 int flags = EXT4_FREE_BLOCKS_FORGET;
2547
2548                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2549                         flags |= EXT4_FREE_BLOCKS_METADATA;
2550
2551                 ext4_free_blocks(handle, inode, NULL,
2552                                  EXT4_C2B(sbi, *partial_cluster),
2553                                  sbi->s_cluster_ratio, flags);
2554                 *partial_cluster = 0;
2555         }
2556
2557         /* if this leaf is free, then we should
2558          * remove it from index block above */
2559         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2560                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2561
2562 out:
2563         return err;
2564 }
2565
2566 /*
2567  * ext4_ext_more_to_rm:
2568  * returns 1 if current index has to be freed (even partial)
2569  */
2570 static int
2571 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2572 {
2573         BUG_ON(path->p_idx == NULL);
2574
2575         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2576                 return 0;
2577
2578         /*
2579          * if truncate on deeper level happened, it wasn't partial,
2580          * so we have to consider current index for truncation
2581          */
2582         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2583                 return 0;
2584         return 1;
2585 }
2586
2587 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2588                                  ext4_lblk_t end)
2589 {
2590         struct super_block *sb = inode->i_sb;
2591         int depth = ext_depth(inode);
2592         struct ext4_ext_path *path = NULL;
2593         ext4_fsblk_t partial_cluster = 0;
2594         handle_t *handle;
2595         int i = 0, err = 0;
2596
2597         ext_debug("truncate since %u to %u\n", start, end);
2598
2599         /* probably first extent we're gonna free will be last in block */
2600         handle = ext4_journal_start(inode, depth + 1);
2601         if (IS_ERR(handle))
2602                 return PTR_ERR(handle);
2603
2604 again:
2605         ext4_ext_invalidate_cache(inode);
2606
2607         trace_ext4_ext_remove_space(inode, start, depth);
2608
2609         /*
2610          * Check if we are removing extents inside the extent tree. If that
2611          * is the case, we are going to punch a hole inside the extent tree
2612          * so we have to check whether we need to split the extent covering
2613          * the last block to remove so we can easily remove the part of it
2614          * in ext4_ext_rm_leaf().
2615          */
2616         if (end < EXT_MAX_BLOCKS - 1) {
2617                 struct ext4_extent *ex;
2618                 ext4_lblk_t ee_block;
2619
2620                 /* find extent for this block */
2621                 path = ext4_ext_find_extent(inode, end, NULL);
2622                 if (IS_ERR(path)) {
2623                         ext4_journal_stop(handle);
2624                         return PTR_ERR(path);
2625                 }
2626                 depth = ext_depth(inode);
2627                 /* Leaf not may not exist only if inode has no blocks at all */
2628                 ex = path[depth].p_ext;
2629                 if (!ex) {
2630                         if (depth) {
2631                                 EXT4_ERROR_INODE(inode,
2632                                                  "path[%d].p_hdr == NULL",
2633                                                  depth);
2634                                 err = -EIO;
2635                         }
2636                         goto out;
2637                 }
2638
2639                 ee_block = le32_to_cpu(ex->ee_block);
2640
2641                 /*
2642                  * See if the last block is inside the extent, if so split
2643                  * the extent at 'end' block so we can easily remove the
2644                  * tail of the first part of the split extent in
2645                  * ext4_ext_rm_leaf().
2646                  */
2647                 if (end >= ee_block &&
2648                     end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2649                         int split_flag = 0;
2650
2651                         if (ext4_ext_is_uninitialized(ex))
2652                                 split_flag = EXT4_EXT_MARK_UNINIT1 |
2653                                              EXT4_EXT_MARK_UNINIT2;
2654
2655                         /*
2656                          * Split the extent in two so that 'end' is the last
2657                          * block in the first new extent
2658                          */
2659                         err = ext4_split_extent_at(handle, inode, path,
2660                                                 end + 1, split_flag,
2661                                                 EXT4_GET_BLOCKS_PRE_IO |
2662                                                 EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2663
2664                         if (err < 0)
2665                                 goto out;
2666                 }
2667         }
2668         /*
2669          * We start scanning from right side, freeing all the blocks
2670          * after i_size and walking into the tree depth-wise.
2671          */
2672         depth = ext_depth(inode);
2673         if (path) {
2674                 int k = i = depth;
2675                 while (--k > 0)
2676                         path[k].p_block =
2677                                 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2678         } else {
2679                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2680                                GFP_NOFS);
2681                 if (path == NULL) {
2682                         ext4_journal_stop(handle);
2683                         return -ENOMEM;
2684                 }
2685                 path[0].p_depth = depth;
2686                 path[0].p_hdr = ext_inode_hdr(inode);
2687                 i = 0;
2688
2689                 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2690                         err = -EIO;
2691                         goto out;
2692                 }
2693         }
2694         err = 0;
2695
2696         while (i >= 0 && err == 0) {
2697                 if (i == depth) {
2698                         /* this is leaf block */
2699                         err = ext4_ext_rm_leaf(handle, inode, path,
2700                                                &partial_cluster, start,
2701                                                end);
2702                         /* root level has p_bh == NULL, brelse() eats this */
2703                         brelse(path[i].p_bh);
2704                         path[i].p_bh = NULL;
2705                         i--;
2706                         continue;
2707                 }
2708
2709                 /* this is index block */
2710                 if (!path[i].p_hdr) {
2711                         ext_debug("initialize header\n");
2712                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2713                 }
2714
2715                 if (!path[i].p_idx) {
2716                         /* this level hasn't been touched yet */
2717                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2718                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2719                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2720                                   path[i].p_hdr,
2721                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2722                 } else {
2723                         /* we were already here, see at next index */
2724                         path[i].p_idx--;
2725                 }
2726
2727                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2728                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2729                                 path[i].p_idx);
2730                 if (ext4_ext_more_to_rm(path + i)) {
2731                         struct buffer_head *bh;
2732                         /* go to the next level */
2733                         ext_debug("move to level %d (block %llu)\n",
2734                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2735                         memset(path + i + 1, 0, sizeof(*path));
2736                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2737                         if (!bh) {
2738                                 /* should we reset i_size? */
2739                                 err = -EIO;
2740                                 break;
2741                         }
2742                         if (WARN_ON(i + 1 > depth)) {
2743                                 err = -EIO;
2744                                 break;
2745                         }
2746                         if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2747                                                         depth - i - 1, bh)) {
2748                                 err = -EIO;
2749                                 break;
2750                         }
2751                         path[i + 1].p_bh = bh;
2752
2753                         /* save actual number of indexes since this
2754                          * number is changed at the next iteration */
2755                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2756                         i++;
2757                 } else {
2758                         /* we finished processing this index, go up */
2759                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2760                                 /* index is empty, remove it;
2761                                  * handle must be already prepared by the
2762                                  * truncatei_leaf() */
2763                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2764                         }
2765                         /* root level has p_bh == NULL, brelse() eats this */
2766                         brelse(path[i].p_bh);
2767                         path[i].p_bh = NULL;
2768                         i--;
2769                         ext_debug("return to level %d\n", i);
2770                 }
2771         }
2772
2773         trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2774                         path->p_hdr->eh_entries);
2775
2776         /* If we still have something in the partial cluster and we have removed
2777          * even the first extent, then we should free the blocks in the partial
2778          * cluster as well. */
2779         if (partial_cluster && path->p_hdr->eh_entries == 0) {
2780                 int flags = EXT4_FREE_BLOCKS_FORGET;
2781
2782                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2783                         flags |= EXT4_FREE_BLOCKS_METADATA;
2784
2785                 ext4_free_blocks(handle, inode, NULL,
2786                                  EXT4_C2B(EXT4_SB(sb), partial_cluster),
2787                                  EXT4_SB(sb)->s_cluster_ratio, flags);
2788                 partial_cluster = 0;
2789         }
2790
2791         /* TODO: flexible tree reduction should be here */
2792         if (path->p_hdr->eh_entries == 0) {
2793                 /*
2794                  * truncate to zero freed all the tree,
2795                  * so we need to correct eh_depth
2796                  */
2797                 err = ext4_ext_get_access(handle, inode, path);
2798                 if (err == 0) {
2799                         ext_inode_hdr(inode)->eh_depth = 0;
2800                         ext_inode_hdr(inode)->eh_max =
2801                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2802                         err = ext4_ext_dirty(handle, inode, path);
2803                 }
2804         }
2805 out:
2806         ext4_ext_drop_refs(path);
2807         kfree(path);
2808         if (err == -EAGAIN) {
2809                 path = NULL;
2810                 goto again;
2811         }
2812         ext4_journal_stop(handle);
2813
2814         return err;
2815 }
2816
2817 /*
2818  * called at mount time
2819  */
2820 void ext4_ext_init(struct super_block *sb)
2821 {
2822         /*
2823          * possible initialization would be here
2824          */
2825
2826         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2827 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2828                 printk(KERN_INFO "EXT4-fs: file extents enabled"
2829 #ifdef AGGRESSIVE_TEST
2830                        ", aggressive tests"
2831 #endif
2832 #ifdef CHECK_BINSEARCH
2833                        ", check binsearch"
2834 #endif
2835 #ifdef EXTENTS_STATS
2836                        ", stats"
2837 #endif
2838                        "\n");
2839 #endif
2840 #ifdef EXTENTS_STATS
2841                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2842                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2843                 EXT4_SB(sb)->s_ext_max = 0;
2844 #endif
2845         }
2846 }
2847
2848 /*
2849  * called at umount time
2850  */
2851 void ext4_ext_release(struct super_block *sb)
2852 {
2853         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2854                 return;
2855
2856 #ifdef EXTENTS_STATS
2857         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2858                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2859                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2860                         sbi->s_ext_blocks, sbi->s_ext_extents,
2861                         sbi->s_ext_blocks / sbi->s_ext_extents);
2862                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2863                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2864         }
2865 #endif
2866 }
2867
2868 /* FIXME!! we need to try to merge to left or right after zero-out  */
2869 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2870 {
2871         ext4_fsblk_t ee_pblock;
2872         unsigned int ee_len;
2873         int ret;
2874
2875         ee_len    = ext4_ext_get_actual_len(ex);
2876         ee_pblock = ext4_ext_pblock(ex);
2877
2878         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2879         if (ret > 0)
2880                 ret = 0;
2881
2882         return ret;
2883 }
2884
2885 /*
2886  * ext4_split_extent_at() splits an extent at given block.
2887  *
2888  * @handle: the journal handle
2889  * @inode: the file inode
2890  * @path: the path to the extent
2891  * @split: the logical block where the extent is splitted.
2892  * @split_flags: indicates if the extent could be zeroout if split fails, and
2893  *               the states(init or uninit) of new extents.
2894  * @flags: flags used to insert new extent to extent tree.
2895  *
2896  *
2897  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2898  * of which are deterimined by split_flag.
2899  *
2900  * There are two cases:
2901  *  a> the extent are splitted into two extent.
2902  *  b> split is not needed, and just mark the extent.
2903  *
2904  * return 0 on success.
2905  */
2906 static int ext4_split_extent_at(handle_t *handle,
2907                              struct inode *inode,
2908                              struct ext4_ext_path *path,
2909                              ext4_lblk_t split,
2910                              int split_flag,
2911                              int flags)
2912 {
2913         ext4_fsblk_t newblock;
2914         ext4_lblk_t ee_block;
2915         struct ext4_extent *ex, newex, orig_ex;
2916         struct ext4_extent *ex2 = NULL;
2917         unsigned int ee_len, depth;
2918         int err = 0;
2919
2920         BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2921                (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2922
2923         ext_debug("ext4_split_extents_at: inode %lu, logical"
2924                 "block %llu\n", inode->i_ino, (unsigned long long)split);
2925
2926         ext4_ext_show_leaf(inode, path);
2927
2928         depth = ext_depth(inode);
2929         ex = path[depth].p_ext;
2930         ee_block = le32_to_cpu(ex->ee_block);
2931         ee_len = ext4_ext_get_actual_len(ex);
2932         newblock = split - ee_block + ext4_ext_pblock(ex);
2933
2934         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2935
2936         err = ext4_ext_get_access(handle, inode, path + depth);
2937         if (err)
2938                 goto out;
2939
2940         if (split == ee_block) {
2941                 /*
2942                  * case b: block @split is the block that the extent begins with
2943                  * then we just change the state of the extent, and splitting
2944                  * is not needed.
2945                  */
2946                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2947                         ext4_ext_mark_uninitialized(ex);
2948                 else
2949                         ext4_ext_mark_initialized(ex);
2950
2951                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2952                         ext4_ext_try_to_merge(handle, inode, path, ex);
2953
2954                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2955                 goto out;
2956         }
2957
2958         /* case a */
2959         memcpy(&orig_ex, ex, sizeof(orig_ex));
2960         ex->ee_len = cpu_to_le16(split - ee_block);
2961         if (split_flag & EXT4_EXT_MARK_UNINIT1)
2962                 ext4_ext_mark_uninitialized(ex);
2963
2964         /*
2965          * path may lead to new leaf, not to original leaf any more
2966          * after ext4_ext_insert_extent() returns,
2967          */
2968         err = ext4_ext_dirty(handle, inode, path + depth);
2969         if (err)
2970                 goto fix_extent_len;
2971
2972         ex2 = &newex;
2973         ex2->ee_block = cpu_to_le32(split);
2974         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2975         ext4_ext_store_pblock(ex2, newblock);
2976         if (split_flag & EXT4_EXT_MARK_UNINIT2)
2977                 ext4_ext_mark_uninitialized(ex2);
2978
2979         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2980         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2981                 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2982                         if (split_flag & EXT4_EXT_DATA_VALID1)
2983                                 err = ext4_ext_zeroout(inode, ex2);
2984                         else
2985                                 err = ext4_ext_zeroout(inode, ex);
2986                 } else
2987                         err = ext4_ext_zeroout(inode, &orig_ex);
2988
2989                 if (err)
2990                         goto fix_extent_len;
2991                 /* update the extent length and mark as initialized */
2992                 ex->ee_len = cpu_to_le16(ee_len);
2993                 ext4_ext_try_to_merge(handle, inode, path, ex);
2994                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2995                 goto out;
2996         } else if (err)
2997                 goto fix_extent_len;
2998
2999 out:
3000         ext4_ext_show_leaf(inode, path);
3001         return err;
3002
3003 fix_extent_len:
3004         ex->ee_len = orig_ex.ee_len;
3005         ext4_ext_dirty(handle, inode, path + depth);
3006         return err;
3007 }
3008
3009 /*
3010  * ext4_split_extents() splits an extent and mark extent which is covered
3011  * by @map as split_flags indicates
3012  *
3013  * It may result in splitting the extent into multiple extents (upto three)
3014  * There are three possibilities:
3015  *   a> There is no split required
3016  *   b> Splits in two extents: Split is happening at either end of the extent
3017  *   c> Splits in three extents: Somone is splitting in middle of the extent
3018  *
3019  */
3020 static int ext4_split_extent(handle_t *handle,
3021                               struct inode *inode,
3022                               struct ext4_ext_path *path,
3023                               struct ext4_map_blocks *map,
3024                               int split_flag,
3025                               int flags)
3026 {
3027         ext4_lblk_t ee_block;
3028         struct ext4_extent *ex;
3029         unsigned int ee_len, depth;
3030         int err = 0;
3031         int uninitialized;
3032         int split_flag1, flags1;
3033
3034         depth = ext_depth(inode);
3035         ex = path[depth].p_ext;
3036         ee_block = le32_to_cpu(ex->ee_block);
3037         ee_len = ext4_ext_get_actual_len(ex);
3038         uninitialized = ext4_ext_is_uninitialized(ex);
3039
3040         if (map->m_lblk + map->m_len < ee_block + ee_len) {
3041                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3042                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3043                 if (uninitialized)
3044                         split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3045                                        EXT4_EXT_MARK_UNINIT2;
3046                 if (split_flag & EXT4_EXT_DATA_VALID2)
3047                         split_flag1 |= EXT4_EXT_DATA_VALID1;
3048                 err = ext4_split_extent_at(handle, inode, path,
3049                                 map->m_lblk + map->m_len, split_flag1, flags1);
3050                 if (err)
3051                         goto out;
3052         }
3053
3054         ext4_ext_drop_refs(path);
3055         path = ext4_ext_find_extent(inode, map->m_lblk, path);
3056         if (IS_ERR(path))
3057                 return PTR_ERR(path);
3058
3059         if (map->m_lblk >= ee_block) {
3060                 split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
3061                                             EXT4_EXT_DATA_VALID2);
3062                 if (uninitialized)
3063                         split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3064                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
3065                         split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3066                 err = ext4_split_extent_at(handle, inode, path,
3067                                 map->m_lblk, split_flag1, flags);
3068                 if (err)
3069                         goto out;
3070         }
3071
3072         ext4_ext_show_leaf(inode, path);
3073 out:
3074         return err ? err : map->m_len;
3075 }
3076
3077 /*
3078  * This function is called by ext4_ext_map_blocks() if someone tries to write
3079  * to an uninitialized extent. It may result in splitting the uninitialized
3080  * extent into multiple extents (up to three - one initialized and two
3081  * uninitialized).
3082  * There are three possibilities:
3083  *   a> There is no split required: Entire extent should be initialized
3084  *   b> Splits in two extents: Write is happening at either end of the extent
3085  *   c> Splits in three extents: Somone is writing in middle of the extent
3086  *
3087  * Pre-conditions:
3088  *  - The extent pointed to by 'path' is uninitialized.
3089  *  - The extent pointed to by 'path' contains a superset
3090  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3091  *
3092  * Post-conditions on success:
3093  *  - the returned value is the number of blocks beyond map->l_lblk
3094  *    that are allocated and initialized.
3095  *    It is guaranteed to be >= map->m_len.
3096  */
3097 static int ext4_ext_convert_to_initialized(handle_t *handle,
3098                                            struct inode *inode,
3099                                            struct ext4_map_blocks *map,
3100                                            struct ext4_ext_path *path)
3101 {
3102         struct ext4_sb_info *sbi;
3103         struct ext4_extent_header *eh;
3104         struct ext4_map_blocks split_map;
3105         struct ext4_extent zero_ex;
3106         struct ext4_extent *ex;
3107         ext4_lblk_t ee_block, eof_block;
3108         unsigned int ee_len, depth;
3109         int allocated, max_zeroout = 0;
3110         int err = 0;
3111         int split_flag = 0;
3112
3113         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3114                 "block %llu, max_blocks %u\n", inode->i_ino,
3115                 (unsigned long long)map->m_lblk, map->m_len);
3116
3117         sbi = EXT4_SB(inode->i_sb);
3118         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3119                 inode->i_sb->s_blocksize_bits;
3120         if (eof_block < map->m_lblk + map->m_len)
3121                 eof_block = map->m_lblk + map->m_len;
3122
3123         depth = ext_depth(inode);
3124         eh = path[depth].p_hdr;
3125         ex = path[depth].p_ext;
3126         ee_block = le32_to_cpu(ex->ee_block);
3127         ee_len = ext4_ext_get_actual_len(ex);
3128         allocated = ee_len - (map->m_lblk - ee_block);
3129
3130         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3131
3132         /* Pre-conditions */
3133         BUG_ON(!ext4_ext_is_uninitialized(ex));
3134         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3135
3136         /*
3137          * Attempt to transfer newly initialized blocks from the currently
3138          * uninitialized extent to its left neighbor. This is much cheaper
3139          * than an insertion followed by a merge as those involve costly
3140          * memmove() calls. This is the common case in steady state for
3141          * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3142          * writes.
3143          *
3144          * Limitations of the current logic:
3145          *  - L1: we only deal with writes at the start of the extent.
3146          *    The approach could be extended to writes at the end
3147          *    of the extent but this scenario was deemed less common.
3148          *  - L2: we do not deal with writes covering the whole extent.
3149          *    This would require removing the extent if the transfer
3150          *    is possible.
3151          *  - L3: we only attempt to merge with an extent stored in the
3152          *    same extent tree node.
3153          */
3154         if ((map->m_lblk == ee_block) &&        /*L1*/
3155                 (map->m_len < ee_len) &&        /*L2*/
3156                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L3*/
3157                 struct ext4_extent *prev_ex;
3158                 ext4_lblk_t prev_lblk;
3159                 ext4_fsblk_t prev_pblk, ee_pblk;
3160                 unsigned int prev_len, write_len;
3161
3162                 prev_ex = ex - 1;
3163                 prev_lblk = le32_to_cpu(prev_ex->ee_block);
3164                 prev_len = ext4_ext_get_actual_len(prev_ex);
3165                 prev_pblk = ext4_ext_pblock(prev_ex);
3166                 ee_pblk = ext4_ext_pblock(ex);
3167                 write_len = map->m_len;
3168
3169                 /*
3170                  * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3171                  * upon those conditions:
3172                  * - C1: prev_ex is initialized,
3173                  * - C2: prev_ex is logically abutting ex,
3174                  * - C3: prev_ex is physically abutting ex,
3175                  * - C4: prev_ex can receive the additional blocks without
3176                  *   overflowing the (initialized) length limit.
3177                  */
3178                 if ((!ext4_ext_is_uninitialized(prev_ex)) &&            /*C1*/
3179                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3180                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3181                         (prev_len < (EXT_INIT_MAX_LEN - write_len))) {  /*C4*/
3182                         err = ext4_ext_get_access(handle, inode, path + depth);
3183                         if (err)
3184                                 goto out;
3185
3186                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3187                                 map, ex, prev_ex);
3188
3189                         /* Shift the start of ex by 'write_len' blocks */
3190                         ex->ee_block = cpu_to_le32(ee_block + write_len);
3191                         ext4_ext_store_pblock(ex, ee_pblk + write_len);
3192                         ex->ee_len = cpu_to_le16(ee_len - write_len);
3193                         ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3194
3195                         /* Extend prev_ex by 'write_len' blocks */
3196                         prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3197
3198                         /* Mark the block containing both extents as dirty */
3199                         ext4_ext_dirty(handle, inode, path + depth);
3200
3201                         /* Update path to point to the right extent */
3202                         path[depth].p_ext = prev_ex;
3203
3204                         /* Result: number of initialized blocks past m_lblk */
3205                         allocated = write_len;
3206                         goto out;
3207                 }
3208         }
3209
3210         WARN_ON(map->m_lblk < ee_block);
3211         /*
3212          * It is safe to convert extent to initialized via explicit
3213          * zeroout only if extent is fully insde i_size or new_size.
3214          */
3215         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3216
3217         if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3218                 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3219                         inode->i_sb->s_blocksize_bits;
3220
3221         /* If extent is less than s_max_zeroout_kb, zeroout directly */
3222         if (max_zeroout && (ee_len <= max_zeroout)) {
3223                 err = ext4_ext_zeroout(inode, ex);
3224                 if (err)
3225                         goto out;
3226
3227                 err = ext4_ext_get_access(handle, inode, path + depth);
3228                 if (err)
3229                         goto out;
3230                 ext4_ext_mark_initialized(ex);
3231                 ext4_ext_try_to_merge(handle, inode, path, ex);
3232                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3233                 goto out;
3234         }
3235
3236         /*
3237          * four cases:
3238          * 1. split the extent into three extents.
3239          * 2. split the extent into two extents, zeroout the first half.
3240          * 3. split the extent into two extents, zeroout the second half.
3241          * 4. split the extent into two extents with out zeroout.
3242          */
3243         split_map.m_lblk = map->m_lblk;
3244         split_map.m_len = map->m_len;
3245
3246         if (max_zeroout && (allocated > map->m_len)) {
3247                 if (allocated <= max_zeroout) {
3248                         /* case 3 */
3249                         zero_ex.ee_block =
3250                                          cpu_to_le32(map->m_lblk);
3251                         zero_ex.ee_len = cpu_to_le16(allocated);
3252                         ext4_ext_store_pblock(&zero_ex,
3253                                 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3254                         err = ext4_ext_zeroout(inode, &zero_ex);
3255                         if (err)
3256                                 goto out;
3257                         split_map.m_lblk = map->m_lblk;
3258                         split_map.m_len = allocated;
3259                 } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3260                         /* case 2 */
3261                         if (map->m_lblk != ee_block) {
3262                                 zero_ex.ee_block = ex->ee_block;
3263                                 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3264                                                         ee_block);
3265                                 ext4_ext_store_pblock(&zero_ex,
3266                                                       ext4_ext_pblock(ex));
3267                                 err = ext4_ext_zeroout(inode, &zero_ex);
3268                                 if (err)
3269                                         goto out;
3270                         }
3271
3272                         split_map.m_lblk = ee_block;
3273                         split_map.m_len = map->m_lblk - ee_block + map->m_len;
3274                         allocated = map->m_len;
3275                 }
3276         }
3277
3278         allocated = ext4_split_extent(handle, inode, path,
3279                                       &split_map, split_flag, 0);
3280         if (allocated < 0)
3281                 err = allocated;
3282
3283 out:
3284         return err ? err : allocated;
3285 }
3286
3287 /*
3288  * This function is called by ext4_ext_map_blocks() from
3289  * ext4_get_blocks_dio_write() when DIO to write
3290  * to an uninitialized extent.
3291  *
3292  * Writing to an uninitialized extent may result in splitting the uninitialized
3293  * extent into multiple initialized/uninitialized extents (up to three)
3294  * There are three possibilities:
3295  *   a> There is no split required: Entire extent should be uninitialized
3296  *   b> Splits in two extents: Write is happening at either end of the extent
3297  *   c> Splits in three extents: Somone is writing in middle of the extent
3298  *
3299  * One of more index blocks maybe needed if the extent tree grow after
3300  * the uninitialized extent split. To prevent ENOSPC occur at the IO
3301  * complete, we need to split the uninitialized extent before DIO submit
3302  * the IO. The uninitialized extent called at this time will be split
3303  * into three uninitialized extent(at most). After IO complete, the part
3304  * being filled will be convert to initialized by the end_io callback function
3305  * via ext4_convert_unwritten_extents().
3306  *
3307  * Returns the size of uninitialized extent to be written on success.
3308  */
3309 static int ext4_split_unwritten_extents(handle_t *handle,
3310                                         struct inode *inode,
3311                                         struct ext4_map_blocks *map,
3312                                         struct ext4_ext_path *path,
3313                                         int flags)
3314 {
3315         ext4_lblk_t eof_block;
3316         ext4_lblk_t ee_block;
3317         struct ext4_extent *ex;
3318         unsigned int ee_len;
3319         int split_flag = 0, depth;
3320
3321         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3322                 "block %llu, max_blocks %u\n", inode->i_ino,
3323                 (unsigned long long)map->m_lblk, map->m_len);
3324
3325         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3326                 inode->i_sb->s_blocksize_bits;
3327         if (eof_block < map->m_lblk + map->m_len)
3328                 eof_block = map->m_lblk + map->m_len;
3329         /*
3330          * It is safe to convert extent to initialized via explicit
3331          * zeroout only if extent is fully insde i_size or new_size.
3332          */
3333         depth = ext_depth(inode);
3334         ex = path[depth].p_ext;
3335         ee_block = le32_to_cpu(ex->ee_block);
3336         ee_len = ext4_ext_get_actual_len(ex);
3337
3338         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3339         split_flag |= EXT4_EXT_MARK_UNINIT2;
3340         if (flags & EXT4_GET_BLOCKS_CONVERT)
3341                 split_flag |= EXT4_EXT_DATA_VALID2;
3342         flags |= EXT4_GET_BLOCKS_PRE_IO;
3343         return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3344 }
3345
3346 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3347                                                 struct inode *inode,
3348                                                 struct ext4_map_blocks *map,
3349                                                 struct ext4_ext_path *path)
3350 {
3351         struct ext4_extent *ex;
3352         ext4_lblk_t ee_block;
3353         unsigned int ee_len;
3354         int depth;
3355         int err = 0;
3356
3357         depth = ext_depth(inode);
3358         ex = path[depth].p_ext;
3359         ee_block = le32_to_cpu(ex->ee_block);
3360         ee_len = ext4_ext_get_actual_len(ex);
3361
3362         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3363                 "block %llu, max_blocks %u\n", inode->i_ino,
3364                   (unsigned long long)ee_block, ee_len);
3365
3366         /* If extent is larger than requested then split is required */
3367         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3368                 err = ext4_split_unwritten_extents(handle, inode, map, path,
3369                                                    EXT4_GET_BLOCKS_CONVERT);
3370                 if (err < 0)
3371                         goto out;
3372                 ext4_ext_drop_refs(path);
3373                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3374                 if (IS_ERR(path)) {
3375                         err = PTR_ERR(path);
3376                         goto out;
3377                 }
3378                 depth = ext_depth(inode);
3379                 ex = path[depth].p_ext;
3380         }
3381
3382         err = ext4_ext_get_access(handle, inode, path + depth);
3383         if (err)
3384                 goto out;
3385         /* first mark the extent as initialized */
3386         ext4_ext_mark_initialized(ex);
3387
3388         /* note: ext4_ext_correct_indexes() isn't needed here because
3389          * borders are not changed
3390          */
3391         ext4_ext_try_to_merge(handle, inode, path, ex);
3392
3393         /* Mark modified extent as dirty */
3394         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3395 out:
3396         ext4_ext_show_leaf(inode, path);
3397         return err;
3398 }
3399
3400 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3401                         sector_t block, int count)
3402 {
3403         int i;
3404         for (i = 0; i < count; i++)
3405                 unmap_underlying_metadata(bdev, block + i);
3406 }
3407
3408 /*
3409  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3410  */
3411 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3412                               ext4_lblk_t lblk,
3413                               struct ext4_ext_path *path,
3414                               unsigned int len)
3415 {
3416         int i, depth;
3417         struct ext4_extent_header *eh;
3418         struct ext4_extent *last_ex;
3419
3420         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3421                 return 0;
3422
3423         depth = ext_depth(inode);
3424         eh = path[depth].p_hdr;
3425
3426         /*
3427          * We're going to remove EOFBLOCKS_FL entirely in future so we
3428          * do not care for this case anymore. Simply remove the flag
3429          * if there are no extents.
3430          */
3431         if (unlikely(!eh->eh_entries))
3432                 goto out;
3433         last_ex = EXT_LAST_EXTENT(eh);
3434         /*
3435          * We should clear the EOFBLOCKS_FL flag if we are writing the
3436          * last block in the last extent in the file.  We test this by
3437          * first checking to see if the caller to
3438          * ext4_ext_get_blocks() was interested in the last block (or
3439          * a block beyond the last block) in the current extent.  If
3440          * this turns out to be false, we can bail out from this
3441          * function immediately.
3442          */
3443         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3444             ext4_ext_get_actual_len(last_ex))
3445                 return 0;
3446         /*
3447          * If the caller does appear to be planning to write at or
3448          * beyond the end of the current extent, we then test to see
3449          * if the current extent is the last extent in the file, by
3450          * checking to make sure it was reached via the rightmost node
3451          * at each level of the tree.
3452          */
3453         for (i = depth-1; i >= 0; i--)
3454                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3455                         return 0;
3456 out:
3457         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3458         return ext4_mark_inode_dirty(handle, inode);
3459 }
3460
3461 /**
3462  * ext4_find_delalloc_range: find delayed allocated block in the given range.
3463  *
3464  * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3465  * whether there are any buffers marked for delayed allocation. It returns '1'
3466  * on the first delalloc'ed buffer head found. If no buffer head in the given
3467  * range is marked for delalloc, it returns 0.
3468  * lblk_start should always be <= lblk_end.
3469  * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3470  * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3471  * block sooner). This is useful when blocks are truncated sequentially from
3472  * lblk_start towards lblk_end.
3473  */
3474 static int ext4_find_delalloc_range(struct inode *inode,
3475                                     ext4_lblk_t lblk_start,
3476                                     ext4_lblk_t lblk_end,
3477                                     int search_hint_reverse)
3478 {
3479         struct address_space *mapping = inode->i_mapping;
3480         struct buffer_head *head, *bh = NULL;
3481         struct page *page;
3482         ext4_lblk_t i, pg_lblk;
3483         pgoff_t index;
3484
3485         if (!test_opt(inode->i_sb, DELALLOC))
3486                 return 0;
3487
3488         /* reverse search wont work if fs block size is less than page size */
3489         if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3490                 search_hint_reverse = 0;
3491
3492         if (search_hint_reverse)
3493                 i = lblk_end;
3494         else
3495                 i = lblk_start;
3496
3497         index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3498
3499         while ((i >= lblk_start) && (i <= lblk_end)) {
3500                 page = find_get_page(mapping, index);
3501                 if (!page)
3502                         goto nextpage;
3503
3504                 if (!page_has_buffers(page))
3505                         goto nextpage;
3506
3507                 head = page_buffers(page);
3508                 if (!head)
3509                         goto nextpage;
3510
3511                 bh = head;
3512                 pg_lblk = index << (PAGE_CACHE_SHIFT -
3513                                                 inode->i_blkbits);
3514                 do {
3515                         if (unlikely(pg_lblk < lblk_start)) {
3516                                 /*
3517                                  * This is possible when fs block size is less
3518                                  * than page size and our cluster starts/ends in
3519                                  * middle of the page. So we need to skip the
3520                                  * initial few blocks till we reach the 'lblk'
3521                                  */
3522                                 pg_lblk++;
3523                                 continue;
3524                         }
3525
3526                         /* Check if the buffer is delayed allocated and that it
3527                          * is not yet mapped. (when da-buffers are mapped during
3528                          * their writeout, their da_mapped bit is set.)
3529                          */
3530                         if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3531                                 page_cache_release(page);
3532                                 trace_ext4_find_delalloc_range(inode,
3533                                                 lblk_start, lblk_end,
3534                                                 search_hint_reverse,
3535                                                 1, i);
3536                                 return 1;
3537                         }
3538                         if (search_hint_reverse)
3539                                 i--;
3540                         else
3541                                 i++;
3542                 } while ((i >= lblk_start) && (i <= lblk_end) &&
3543                                 ((bh = bh->b_this_page) != head));
3544 nextpage:
3545                 if (page)
3546                         page_cache_release(page);
3547                 /*
3548                  * Move to next page. 'i' will be the first lblk in the next
3549                  * page.
3550                  */
3551                 if (search_hint_reverse)
3552                         index--;
3553                 else
3554                         index++;
3555                 i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3556         }
3557
3558         trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3559                                         search_hint_reverse, 0, 0);
3560         return 0;
3561 }
3562
3563 int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3564                                int search_hint_reverse)
3565 {
3566         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3567         ext4_lblk_t lblk_start, lblk_end;
3568         lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3569         lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3570
3571         return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3572                                         search_hint_reverse);
3573 }
3574
3575 /**
3576  * Determines how many complete clusters (out of those specified by the 'map')
3577  * are under delalloc and were reserved quota for.
3578  * This function is called when we are writing out the blocks that were
3579  * originally written with their allocation delayed, but then the space was
3580  * allocated using fallocate() before the delayed allocation could be resolved.
3581  * The cases to look for are:
3582  * ('=' indicated delayed allocated blocks
3583  *  '-' indicates non-delayed allocated blocks)
3584  * (a) partial clusters towards beginning and/or end outside of allocated range
3585  *     are not delalloc'ed.
3586  *      Ex:
3587  *      |----c---=|====c====|====c====|===-c----|
3588  *               |++++++ allocated ++++++|
3589  *      ==> 4 complete clusters in above example
3590  *
3591  * (b) partial cluster (outside of allocated range) towards either end is
3592  *     marked for delayed allocation. In this case, we will exclude that
3593  *     cluster.
3594  *      Ex:
3595  *      |----====c========|========c========|
3596  *           |++++++ allocated ++++++|
3597  *      ==> 1 complete clusters in above example
3598  *
3599  *      Ex:
3600  *      |================c================|
3601  *            |++++++ allocated ++++++|
3602  *      ==> 0 complete clusters in above example
3603  *
3604  * The ext4_da_update_reserve_space will be called only if we
3605  * determine here that there were some "entire" clusters that span
3606  * this 'allocated' range.
3607  * In the non-bigalloc case, this function will just end up returning num_blks
3608  * without ever calling ext4_find_delalloc_range.
3609  */
3610 static unsigned int
3611 get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3612                            unsigned int num_blks)
3613 {
3614         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3615         ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3616         ext4_lblk_t lblk_from, lblk_to, c_offset;
3617         unsigned int allocated_clusters = 0;
3618
3619         alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3620         alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3621
3622         /* max possible clusters for this allocation */
3623         allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3624
3625         trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3626
3627         /* Check towards left side */
3628         c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3629         if (c_offset) {
3630                 lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3631                 lblk_to = lblk_from + c_offset - 1;
3632
3633                 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3634                         allocated_clusters--;
3635         }
3636
3637         /* Now check towards right. */
3638         c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3639         if (allocated_clusters && c_offset) {
3640                 lblk_from = lblk_start + num_blks;
3641                 lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3642
3643                 if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3644                         allocated_clusters--;
3645         }
3646
3647         return allocated_clusters;
3648 }
3649
3650 static int
3651 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3652                         struct ext4_map_blocks *map,
3653                         struct ext4_ext_path *path, int flags,
3654                         unsigned int allocated, ext4_fsblk_t newblock)
3655 {
3656         int ret = 0;
3657         int err = 0;
3658         ext4_io_end_t *io = ext4_inode_aio(inode);
3659
3660         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3661                   "block %llu, max_blocks %u, flags %x, allocated %u\n",
3662                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3663                   flags, allocated);
3664         ext4_ext_show_leaf(inode, path);
3665
3666         trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3667                                                     newblock);
3668
3669         /* get_block() before submit the IO, split the extent */
3670         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3671                 ret = ext4_split_unwritten_extents(handle, inode, map,
3672                                                    path, flags);
3673                 if (ret <= 0)
3674                         goto out;
3675                 /*
3676                  * Flag the inode(non aio case) or end_io struct (aio case)
3677                  * that this IO needs to conversion to written when IO is
3678                  * completed
3679                  */
3680                 if (io)
3681                         ext4_set_io_unwritten_flag(inode, io);
3682                 else
3683                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3684                 if (ext4_should_dioread_nolock(inode))
3685                         map->m_flags |= EXT4_MAP_UNINIT;
3686                 goto out;
3687         }
3688         /* IO end_io complete, convert the filled extent to written */
3689         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3690                 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3691                                                         path);
3692                 if (ret >= 0) {
3693                         ext4_update_inode_fsync_trans(handle, inode, 1);
3694                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3695                                                  path, map->m_len);
3696                 } else
3697                         err = ret;
3698                 goto out2;
3699         }
3700         /* buffered IO case */
3701         /*
3702          * repeat fallocate creation request
3703          * we already have an unwritten extent
3704          */
3705         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3706                 goto map_out;
3707
3708         /* buffered READ or buffered write_begin() lookup */
3709         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3710                 /*
3711                  * We have blocks reserved already.  We
3712                  * return allocated blocks so that delalloc
3713                  * won't do block reservation for us.  But
3714                  * the buffer head will be unmapped so that
3715                  * a read from the block returns 0s.
3716                  */
3717                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3718                 goto out1;
3719         }
3720
3721         /* buffered write, writepage time, convert*/
3722         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3723         if (ret >= 0)
3724                 ext4_update_inode_fsync_trans(handle, inode, 1);
3725 out:
3726         if (ret <= 0) {
3727                 err = ret;
3728                 goto out2;
3729         } else
3730                 allocated = ret;
3731         map->m_flags |= EXT4_MAP_NEW;
3732         /*
3733          * if we allocated more blocks than requested
3734          * we need to make sure we unmap the extra block
3735          * allocated. The actual needed block will get
3736          * unmapped later when we find the buffer_head marked
3737          * new.
3738          */
3739         if (allocated > map->m_len) {
3740                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3741                                         newblock + map->m_len,
3742                                         allocated - map->m_len);
3743                 allocated = map->m_len;
3744         }
3745
3746         /*
3747          * If we have done fallocate with the offset that is already
3748          * delayed allocated, we would have block reservation
3749          * and quota reservation done in the delayed write path.
3750          * But fallocate would have already updated quota and block
3751          * count for this offset. So cancel these reservation
3752          */
3753         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3754                 unsigned int reserved_clusters;
3755                 reserved_clusters = get_reserved_cluster_alloc(inode,
3756                                 map->m_lblk, map->m_len);
3757                 if (reserved_clusters)
3758                         ext4_da_update_reserve_space(inode,
3759                                                      reserved_clusters,
3760                                                      0);
3761         }
3762
3763 map_out:
3764         map->m_flags |= EXT4_MAP_MAPPED;
3765         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3766                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3767                                          map->m_len);
3768                 if (err < 0)
3769                         goto out2;
3770         }
3771 out1:
3772         if (allocated > map->m_len)
3773                 allocated = map->m_len;
3774         ext4_ext_show_leaf(inode, path);
3775         map->m_pblk = newblock;
3776         map->m_len = allocated;
3777 out2:
3778         if (path) {
3779                 ext4_ext_drop_refs(path);
3780                 kfree(path);
3781         }
3782         return err ? err : allocated;
3783 }
3784
3785 /*
3786  * get_implied_cluster_alloc - check to see if the requested
3787  * allocation (in the map structure) overlaps with a cluster already
3788  * allocated in an extent.
3789  *      @sb     The filesystem superblock structure
3790  *      @map    The requested lblk->pblk mapping
3791  *      @ex     The extent structure which might contain an implied
3792  *                      cluster allocation
3793  *
3794  * This function is called by ext4_ext_map_blocks() after we failed to
3795  * find blocks that were already in the inode's extent tree.  Hence,
3796  * we know that the beginning of the requested region cannot overlap
3797  * the extent from the inode's extent tree.  There are three cases we
3798  * want to catch.  The first is this case:
3799  *
3800  *               |--- cluster # N--|
3801  *    |--- extent ---|  |---- requested region ---|
3802  *                      |==========|
3803  *
3804  * The second case that we need to test for is this one:
3805  *
3806  *   |--------- cluster # N ----------------|
3807  *         |--- requested region --|   |------- extent ----|
3808  *         |=======================|
3809  *
3810  * The third case is when the requested region lies between two extents
3811  * within the same cluster:
3812  *          |------------- cluster # N-------------|
3813  * |----- ex -----|                  |---- ex_right ----|
3814  *                  |------ requested region ------|
3815  *                  |================|
3816  *
3817  * In each of the above cases, we need to set the map->m_pblk and
3818  * map->m_len so it corresponds to the return the extent labelled as
3819  * "|====|" from cluster #N, since it is already in use for data in
3820  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3821  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3822  * as a new "allocated" block region.  Otherwise, we will return 0 and
3823  * ext4_ext_map_blocks() will then allocate one or more new clusters
3824  * by calling ext4_mb_new_blocks().
3825  */
3826 static int get_implied_cluster_alloc(struct super_block *sb,
3827                                      struct ext4_map_blocks *map,
3828                                      struct ext4_extent *ex,
3829                                      struct ext4_ext_path *path)
3830 {
3831         struct ext4_sb_info *sbi = EXT4_SB(sb);
3832         ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3833         ext4_lblk_t ex_cluster_start, ex_cluster_end;
3834         ext4_lblk_t rr_cluster_start;
3835         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3836         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3837         unsigned short ee_len = ext4_ext_get_actual_len(ex);
3838
3839         /* The extent passed in that we are trying to match */
3840         ex_cluster_start = EXT4_B2C(sbi, ee_block);
3841         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3842
3843         /* The requested region passed into ext4_map_blocks() */
3844         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3845
3846         if ((rr_cluster_start == ex_cluster_end) ||
3847             (rr_cluster_start == ex_cluster_start)) {
3848                 if (rr_cluster_start == ex_cluster_end)
3849                         ee_start += ee_len - 1;
3850                 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3851                         c_offset;
3852                 map->m_len = min(map->m_len,
3853                                  (unsigned) sbi->s_cluster_ratio - c_offset);
3854                 /*
3855                  * Check for and handle this case:
3856                  *
3857                  *   |--------- cluster # N-------------|
3858                  *                     |------- extent ----|
3859                  *         |--- requested region ---|
3860                  *         |===========|
3861                  */
3862
3863                 if (map->m_lblk < ee_block)
3864                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
3865
3866                 /*
3867                  * Check for the case where there is already another allocated
3868                  * block to the right of 'ex' but before the end of the cluster.
3869                  *
3870                  *          |------------- cluster # N-------------|
3871                  * |----- ex -----|                  |---- ex_right ----|
3872                  *                  |------ requested region ------|
3873                  *                  |================|
3874                  */
3875                 if (map->m_lblk > ee_block) {
3876                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3877                         map->m_len = min(map->m_len, next - map->m_lblk);
3878                 }
3879
3880                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3881                 return 1;
3882         }
3883
3884         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3885         return 0;
3886 }
3887
3888
3889 /*
3890  * Block allocation/map/preallocation routine for extents based files
3891  *
3892  *
3893  * Need to be called with
3894  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3895  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3896  *
3897  * return > 0, number of of blocks already mapped/allocated
3898  *          if create == 0 and these are pre-allocated blocks
3899  *              buffer head is unmapped
3900  *          otherwise blocks are mapped
3901  *
3902  * return = 0, if plain look up failed (blocks have not been allocated)
3903  *          buffer head is unmapped
3904  *
3905  * return < 0, error case.
3906  */
3907 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3908                         struct ext4_map_blocks *map, int flags)
3909 {
3910         struct ext4_ext_path *path = NULL;
3911         struct ext4_extent newex, *ex, *ex2;
3912         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3913         ext4_fsblk_t newblock = 0;
3914         int free_on_err = 0, err = 0, depth, ret;
3915         unsigned int allocated = 0, offset = 0;
3916         unsigned int allocated_clusters = 0;
3917         struct ext4_allocation_request ar;
3918         ext4_io_end_t *io = ext4_inode_aio(inode);
3919         ext4_lblk_t cluster_offset;
3920         int set_unwritten = 0;
3921
3922         ext_debug("blocks %u/%u requested for inode %lu\n",
3923                   map->m_lblk, map->m_len, inode->i_ino);
3924         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3925
3926         /* check in cache */
3927         if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3928                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3929                         if ((sbi->s_cluster_ratio > 1) &&
3930                             ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3931                                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3932
3933                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3934                                 /*
3935                                  * block isn't allocated yet and
3936                                  * user doesn't want to allocate it
3937                                  */
3938                                 goto out2;
3939                         }
3940                         /* we should allocate requested block */
3941                 } else {
3942                         /* block is already allocated */
3943                         if (sbi->s_cluster_ratio > 1)
3944                                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3945                         newblock = map->m_lblk
3946                                    - le32_to_cpu(newex.ee_block)
3947                                    + ext4_ext_pblock(&newex);
3948                         /* number of remaining blocks in the extent */
3949                         allocated = ext4_ext_get_actual_len(&newex) -
3950                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3951                         goto out;
3952                 }
3953         }
3954
3955         /* find extent for this block */
3956         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3957         if (IS_ERR(path)) {
3958                 err = PTR_ERR(path);
3959                 path = NULL;
3960                 goto out2;
3961         }
3962
3963         depth = ext_depth(inode);
3964
3965         /*
3966          * consistent leaf must not be empty;
3967          * this situation is possible, though, _during_ tree modification;
3968          * this is why assert can't be put in ext4_ext_find_extent()
3969          */
3970         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3971                 EXT4_ERROR_INODE(inode, "bad extent address "
3972                                  "lblock: %lu, depth: %d pblock %lld",
3973                                  (unsigned long) map->m_lblk, depth,
3974                                  path[depth].p_block);
3975                 err = -EIO;
3976                 goto out2;
3977         }
3978
3979         ex = path[depth].p_ext;
3980         if (ex) {
3981                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3982                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3983                 unsigned short ee_len;
3984
3985                 /*
3986                  * Uninitialized extents are treated as holes, except that
3987                  * we split out initialized portions during a write.
3988                  */
3989                 ee_len = ext4_ext_get_actual_len(ex);
3990
3991                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3992
3993                 /* if found extent covers block, simply return it */
3994                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3995                         newblock = map->m_lblk - ee_block + ee_start;
3996                         /* number of remaining blocks in the extent */
3997                         allocated = ee_len - (map->m_lblk - ee_block);
3998                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3999                                   ee_block, ee_len, newblock);
4000
4001                         /*
4002                          * Do not put uninitialized extent
4003                          * in the cache
4004                          */
4005                         if (!ext4_ext_is_uninitialized(ex)) {
4006                                 ext4_ext_put_in_cache(inode, ee_block,
4007                                         ee_len, ee_start);
4008                                 goto out;
4009                         }
4010                         ret = ext4_ext_handle_uninitialized_extents(
4011                                 handle, inode, map, path, flags,
4012                                 allocated, newblock);
4013                         return ret;
4014                 }
4015         }
4016
4017         if ((sbi->s_cluster_ratio > 1) &&
4018             ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4019                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4020
4021         /*
4022          * requested block isn't allocated yet;
4023          * we couldn't try to create block if create flag is zero
4024          */
4025         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4026                 /*
4027                  * put just found gap into cache to speed up
4028                  * subsequent requests
4029                  */
4030                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4031                 goto out2;
4032         }
4033
4034         /*
4035          * Okay, we need to do block allocation.
4036          */
4037         map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4038         newex.ee_block = cpu_to_le32(map->m_lblk);
4039         cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4040
4041         /*
4042          * If we are doing bigalloc, check to see if the extent returned
4043          * by ext4_ext_find_extent() implies a cluster we can use.
4044          */
4045         if (cluster_offset && ex &&
4046             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4047                 ar.len = allocated = map->m_len;
4048                 newblock = map->m_pblk;
4049                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4050                 goto got_allocated_blocks;
4051         }
4052
4053         /* find neighbour allocated blocks */
4054         ar.lleft = map->m_lblk;
4055         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4056         if (err)
4057                 goto out2;
4058         ar.lright = map->m_lblk;
4059         ex2 = NULL;
4060         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4061         if (err)
4062                 goto out2;
4063
4064         /* Check if the extent after searching to the right implies a
4065          * cluster we can use. */
4066         if ((sbi->s_cluster_ratio > 1) && ex2 &&
4067             get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4068                 ar.len = allocated = map->m_len;
4069                 newblock = map->m_pblk;
4070                 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4071                 goto got_allocated_blocks;
4072         }
4073
4074         /*
4075          * See if request is beyond maximum number of blocks we can have in
4076          * a single extent. For an initialized extent this limit is
4077          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4078          * EXT_UNINIT_MAX_LEN.
4079          */
4080         if (map->m_len > EXT_INIT_MAX_LEN &&
4081             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4082                 map->m_len = EXT_INIT_MAX_LEN;
4083         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4084                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4085                 map->m_len = EXT_UNINIT_MAX_LEN;
4086
4087         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4088         newex.ee_len = cpu_to_le16(map->m_len);
4089         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4090         if (err)
4091                 allocated = ext4_ext_get_actual_len(&newex);
4092         else
4093                 allocated = map->m_len;
4094
4095         /* allocate new block */
4096         ar.inode = inode;
4097         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4098         ar.logical = map->m_lblk;
4099         /*
4100          * We calculate the offset from the beginning of the cluster
4101          * for the logical block number, since when we allocate a
4102          * physical cluster, the physical block should start at the
4103          * same offset from the beginning of the cluster.  This is
4104          * needed so that future calls to get_implied_cluster_alloc()
4105          * work correctly.
4106          */
4107         offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4108         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4109         ar.goal -= offset;
4110         ar.logical -= offset;
4111         if (S_ISREG(inode->i_mode))
4112                 ar.flags = EXT4_MB_HINT_DATA;
4113         else
4114                 /* disable in-core preallocation for non-regular files */
4115                 ar.flags = 0;
4116         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4117                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4118         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4119         if (!newblock)
4120                 goto out2;
4121         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4122                   ar.goal, newblock, allocated);
4123         free_on_err = 1;
4124         allocated_clusters = ar.len;
4125         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4126         if (ar.len > allocated)
4127                 ar.len = allocated;
4128
4129 got_allocated_blocks:
4130         /* try to insert new extent into found leaf and return */
4131         ext4_ext_store_pblock(&newex, newblock + offset);
4132         newex.ee_len = cpu_to_le16(ar.len);
4133         /* Mark uninitialized */
4134         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4135                 ext4_ext_mark_uninitialized(&newex);
4136                 /*
4137                  * io_end structure was created for every IO write to an
4138                  * uninitialized extent. To avoid unnecessary conversion,
4139                  * here we flag the IO that really needs the conversion.
4140                  * For non asycn direct IO case, flag the inode state
4141                  * that we need to perform conversion when IO is done.
4142                  */
4143                 if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4144                         set_unwritten = 1;
4145                 if (ext4_should_dioread_nolock(inode))
4146                         map->m_flags |= EXT4_MAP_UNINIT;
4147         }
4148
4149         err = 0;
4150         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4151                 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4152                                          path, ar.len);
4153         if (!err)
4154                 err = ext4_ext_insert_extent(handle, inode, path,
4155                                              &newex, flags);
4156
4157         if (!err && set_unwritten) {
4158                 if (io)
4159                         ext4_set_io_unwritten_flag(inode, io);
4160                 else
4161                         ext4_set_inode_state(inode,
4162                                              EXT4_STATE_DIO_UNWRITTEN);
4163         }
4164
4165         if (err && free_on_err) {
4166                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4167                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4168                 /* free data blocks we just allocated */
4169                 /* not a good idea to call discard here directly,
4170                  * but otherwise we'd need to call it every free() */
4171                 ext4_discard_preallocations(inode);
4172                 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4173                                  ext4_ext_get_actual_len(&newex), fb_flags);
4174                 goto out2;
4175         }
4176
4177         /* previous routine could use block we allocated */
4178         newblock = ext4_ext_pblock(&newex);
4179         allocated = ext4_ext_get_actual_len(&newex);
4180         if (allocated > map->m_len)
4181                 allocated = map->m_len;
4182         map->m_flags |= EXT4_MAP_NEW;
4183
4184         /*
4185          * Update reserved blocks/metadata blocks after successful
4186          * block allocation which had been deferred till now.
4187          */
4188         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4189                 unsigned int reserved_clusters;
4190                 /*
4191                  * Check how many clusters we had reserved this allocated range
4192                  */
4193                 reserved_clusters = get_reserved_cluster_alloc(inode,
4194                                                 map->m_lblk, allocated);
4195                 if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4196                         if (reserved_clusters) {
4197                                 /*
4198                                  * We have clusters reserved for this range.
4199                                  * But since we are not doing actual allocation
4200                                  * and are simply using blocks from previously
4201                                  * allocated cluster, we should release the
4202                                  * reservation and not claim quota.
4203                                  */
4204                                 ext4_da_update_reserve_space(inode,
4205                                                 reserved_clusters, 0);
4206                         }
4207                 } else {
4208                         BUG_ON(allocated_clusters < reserved_clusters);
4209                         /* We will claim quota for all newly allocated blocks.*/
4210                         ext4_da_update_reserve_space(inode, allocated_clusters,
4211                                                         1);
4212                         if (reserved_clusters < allocated_clusters) {
4213                                 struct ext4_inode_info *ei = EXT4_I(inode);
4214                                 int reservation = allocated_clusters -
4215                                                   reserved_clusters;
4216                                 /*
4217                                  * It seems we claimed few clusters outside of
4218                                  * the range of this allocation. We should give
4219                                  * it back to the reservation pool. This can
4220                                  * happen in the following case:
4221                                  *
4222                                  * * Suppose s_cluster_ratio is 4 (i.e., each
4223                                  *   cluster has 4 blocks. Thus, the clusters
4224                                  *   are [0-3],[4-7],[8-11]...
4225                                  * * First comes delayed allocation write for
4226                                  *   logical blocks 10 & 11. Since there were no
4227                                  *   previous delayed allocated blocks in the
4228                                  *   range [8-11], we would reserve 1 cluster
4229                                  *   for this write.
4230                                  * * Next comes write for logical blocks 3 to 8.
4231                                  *   In this case, we will reserve 2 clusters
4232                                  *   (for [0-3] and [4-7]; and not for [8-11] as
4233                                  *   that range has a delayed allocated blocks.
4234                                  *   Thus total reserved clusters now becomes 3.
4235                                  * * Now, during the delayed allocation writeout
4236                                  *   time, we will first write blocks [3-8] and
4237                                  *   allocate 3 clusters for writing these
4238                                  *   blocks. Also, we would claim all these
4239                                  *   three clusters above.
4240                                  * * Now when we come here to writeout the
4241                                  *   blocks [10-11], we would expect to claim
4242                                  *   the reservation of 1 cluster we had made
4243                                  *   (and we would claim it since there are no
4244                                  *   more delayed allocated blocks in the range
4245                                  *   [8-11]. But our reserved cluster count had
4246                                  *   already gone to 0.
4247                                  *
4248                                  *   Thus, at the step 4 above when we determine
4249                                  *   that there are still some unwritten delayed
4250                                  *   allocated blocks outside of our current
4251                                  *   block range, we should increment the
4252                                  *   reserved clusters count so that when the
4253                                  *   remaining blocks finally gets written, we
4254                                  *   could claim them.
4255                                  */
4256                                 dquot_reserve_block(inode,
4257                                                 EXT4_C2B(sbi, reservation));
4258                                 spin_lock(&ei->i_block_reservation_lock);
4259                                 ei->i_reserved_data_blocks += reservation;
4260                                 spin_unlock(&ei->i_block_reservation_lock);
4261                         }
4262                 }
4263         }
4264
4265         /*
4266          * Cache the extent and update transaction to commit on fdatasync only
4267          * when it is _not_ an uninitialized extent.
4268          */
4269         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4270                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4271                 ext4_update_inode_fsync_trans(handle, inode, 1);
4272         } else
4273                 ext4_update_inode_fsync_trans(handle, inode, 0);
4274 out:
4275         if (allocated > map->m_len)
4276                 allocated = map->m_len;
4277         ext4_ext_show_leaf(inode, path);
4278         map->m_flags |= EXT4_MAP_MAPPED;
4279         map->m_pblk = newblock;
4280         map->m_len = allocated;
4281 out2:
4282         if (path) {
4283                 ext4_ext_drop_refs(path);
4284                 kfree(path);
4285         }
4286
4287         trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4288                 newblock, map->m_len, err ? err : allocated);
4289
4290         return err ? err : allocated;
4291 }
4292
4293 void ext4_ext_truncate(struct inode *inode)
4294 {
4295         struct address_space *mapping = inode->i_mapping;
4296         struct super_block *sb = inode->i_sb;
4297         ext4_lblk_t last_block;
4298         handle_t *handle;
4299         loff_t page_len;
4300         int err = 0;
4301
4302         /*
4303          * finish any pending end_io work so we won't run the risk of
4304          * converting any truncated blocks to initialized later
4305          */
4306         ext4_flush_unwritten_io(inode);
4307
4308         /*
4309          * probably first extent we're gonna free will be last in block
4310          */
4311         err = ext4_writepage_trans_blocks(inode);
4312         handle = ext4_journal_start(inode, err);
4313         if (IS_ERR(handle))
4314                 return;
4315
4316         if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4317                 page_len = PAGE_CACHE_SIZE -
4318                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
4319
4320                 err = ext4_discard_partial_page_buffers(handle,
4321                         mapping, inode->i_size, page_len, 0);
4322
4323                 if (err)
4324                         goto out_stop;
4325         }
4326
4327         if (ext4_orphan_add(handle, inode))
4328                 goto out_stop;
4329
4330         down_write(&EXT4_I(inode)->i_data_sem);
4331         ext4_ext_invalidate_cache(inode);
4332
4333         ext4_discard_preallocations(inode);
4334
4335         /*
4336          * TODO: optimization is possible here.
4337          * Probably we need not scan at all,
4338          * because page truncation is enough.
4339          */
4340
4341         /* we have to know where to truncate from in crash case */
4342         EXT4_I(inode)->i_disksize = inode->i_size;
4343         ext4_mark_inode_dirty(handle, inode);
4344
4345         last_block = (inode->i_size + sb->s_blocksize - 1)
4346                         >> EXT4_BLOCK_SIZE_BITS(sb);
4347         err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4348
4349         /* In a multi-transaction truncate, we only make the final
4350          * transaction synchronous.
4351          */
4352         if (IS_SYNC(inode))
4353                 ext4_handle_sync(handle);
4354
4355         up_write(&EXT4_I(inode)->i_data_sem);
4356
4357 out_stop:
4358         /*
4359          * If this was a simple ftruncate() and the file will remain alive,
4360          * then we need to clear up the orphan record which we created above.
4361          * However, if this was a real unlink then we were called by
4362          * ext4_delete_inode(), and we allow that function to clean up the
4363          * orphan info for us.
4364          */
4365         if (inode->i_nlink)
4366                 ext4_orphan_del(handle, inode);
4367
4368         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4369         ext4_mark_inode_dirty(handle, inode);
4370         ext4_journal_stop(handle);
4371 }
4372
4373 static void ext4_falloc_update_inode(struct inode *inode,
4374                                 int mode, loff_t new_size, int update_ctime)
4375 {
4376         struct timespec now;
4377
4378         if (update_ctime) {
4379                 now = current_fs_time(inode->i_sb);
4380                 if (!timespec_equal(&inode->i_ctime, &now))
4381                         inode->i_ctime = now;
4382         }
4383         /*
4384          * Update only when preallocation was requested beyond
4385          * the file size.
4386          */
4387         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4388                 if (new_size > i_size_read(inode))
4389                         i_size_write(inode, new_size);
4390                 if (new_size > EXT4_I(inode)->i_disksize)
4391                         ext4_update_i_disksize(inode, new_size);
4392         } else {
4393                 /*
4394                  * Mark that we allocate beyond EOF so the subsequent truncate
4395                  * can proceed even if the new size is the same as i_size.
4396                  */
4397                 if (new_size > i_size_read(inode))
4398                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4399         }
4400
4401 }
4402
4403 /*
4404  * preallocate space for a file. This implements ext4's fallocate file
4405  * operation, which gets called from sys_fallocate system call.
4406  * For block-mapped files, posix_fallocate should fall back to the method
4407  * of writing zeroes to the required new blocks (the same behavior which is
4408  * expected for file systems which do not support fallocate() system call).
4409  */
4410 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4411 {
4412         struct inode *inode = file->f_path.dentry->d_inode;
4413         handle_t *handle;
4414         loff_t new_size;
4415         unsigned int max_blocks;
4416         int ret = 0;
4417         int ret2 = 0;
4418         int retries = 0;
4419         int flags;
4420         struct ext4_map_blocks map;
4421         unsigned int credits, blkbits = inode->i_blkbits;
4422
4423         /*
4424          * currently supporting (pre)allocate mode for extent-based
4425          * files _only_
4426          */
4427         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4428                 return -EOPNOTSUPP;
4429
4430         /* Return error if mode is not supported */
4431         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4432                 return -EOPNOTSUPP;
4433
4434         if (mode & FALLOC_FL_PUNCH_HOLE)
4435                 return ext4_punch_hole(file, offset, len);
4436
4437         trace_ext4_fallocate_enter(inode, offset, len, mode);
4438         map.m_lblk = offset >> blkbits;
4439         /*
4440          * We can't just convert len to max_blocks because
4441          * If blocksize = 4096 offset = 3072 and len = 2048
4442          */
4443         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4444                 - map.m_lblk;
4445         /*
4446          * credits to insert 1 extent into extent tree
4447          */
4448         credits = ext4_chunk_trans_blocks(inode, max_blocks);
4449         mutex_lock(&inode->i_mutex);
4450         ret = inode_newsize_ok(inode, (len + offset));
4451         if (ret) {
4452                 mutex_unlock(&inode->i_mutex);
4453                 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4454                 return ret;
4455         }
4456         flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4457         if (mode & FALLOC_FL_KEEP_SIZE)
4458                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4459         /*
4460          * Don't normalize the request if it can fit in one extent so
4461          * that it doesn't get unnecessarily split into multiple
4462          * extents.
4463          */
4464         if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4465                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4466
4467         /* Prevent race condition between unwritten */
4468         ext4_flush_unwritten_io(inode);
4469 retry:
4470         while (ret >= 0 && ret < max_blocks) {
4471                 map.m_lblk = map.m_lblk + ret;
4472                 map.m_len = max_blocks = max_blocks - ret;
4473                 handle = ext4_journal_start(inode, credits);
4474                 if (IS_ERR(handle)) {
4475                         ret = PTR_ERR(handle);
4476                         break;
4477                 }
4478                 ret = ext4_map_blocks(handle, inode, &map, flags);
4479                 if (ret <= 0) {
4480 #ifdef EXT4FS_DEBUG
4481                         WARN_ON(ret <= 0);
4482                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
4483                                     "returned error inode#%lu, block=%u, "
4484                                     "max_blocks=%u", __func__,
4485                                     inode->i_ino, map.m_lblk, max_blocks);
4486 #endif
4487                         ext4_mark_inode_dirty(handle, inode);
4488                         ret2 = ext4_journal_stop(handle);
4489                         break;
4490                 }
4491                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4492                                                 blkbits) >> blkbits))
4493                         new_size = offset + len;
4494                 else
4495                         new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4496
4497                 ext4_falloc_update_inode(inode, mode, new_size,
4498                                          (map.m_flags & EXT4_MAP_NEW));
4499                 ext4_mark_inode_dirty(handle, inode);
4500                 if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4501                         ext4_handle_sync(handle);
4502                 ret2 = ext4_journal_stop(handle);
4503                 if (ret2)
4504                         break;
4505         }
4506         if (ret == -ENOSPC &&
4507                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4508                 ret = 0;
4509                 goto retry;
4510         }
4511         mutex_unlock(&inode->i_mutex);
4512         trace_ext4_fallocate_exit(inode, offset, max_blocks,
4513                                 ret > 0 ? ret2 : ret);
4514         return ret > 0 ? ret2 : ret;
4515 }
4516
4517 /*
4518  * This function convert a range of blocks to written extents
4519  * The caller of this function will pass the start offset and the size.
4520  * all unwritten extents within this range will be converted to
4521  * written extents.
4522  *
4523  * This function is called from the direct IO end io call back
4524  * function, to convert the fallocated extents after IO is completed.
4525  * Returns 0 on success.
4526  */
4527 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4528                                     ssize_t len)
4529 {
4530         handle_t *handle;
4531         unsigned int max_blocks;
4532         int ret = 0;
4533         int ret2 = 0;
4534         struct ext4_map_blocks map;
4535         unsigned int credits, blkbits = inode->i_blkbits;
4536
4537         map.m_lblk = offset >> blkbits;
4538         /*
4539          * We can't just convert len to max_blocks because
4540          * If blocksize = 4096 offset = 3072 and len = 2048
4541          */
4542         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4543                       map.m_lblk);
4544         /*
4545          * credits to insert 1 extent into extent tree
4546          */
4547         credits = ext4_chunk_trans_blocks(inode, max_blocks);
4548         while (ret >= 0 && ret < max_blocks) {
4549                 map.m_lblk += ret;
4550                 map.m_len = (max_blocks -= ret);
4551                 handle = ext4_journal_start(inode, credits);
4552                 if (IS_ERR(handle)) {
4553                         ret = PTR_ERR(handle);
4554                         break;
4555                 }
4556                 ret = ext4_map_blocks(handle, inode, &map,
4557                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4558                 if (ret <= 0) {
4559                         WARN_ON(ret <= 0);
4560                         ext4_msg(inode->i_sb, KERN_ERR,
4561                                  "%s:%d: inode #%lu: block %u: len %u: "
4562                                  "ext4_ext_map_blocks returned %d",
4563                                  __func__, __LINE__, inode->i_ino, map.m_lblk,
4564                                  map.m_len, ret);
4565                 }
4566                 ext4_mark_inode_dirty(handle, inode);
4567                 ret2 = ext4_journal_stop(handle);
4568                 if (ret <= 0 || ret2 )
4569                         break;
4570         }
4571         return ret > 0 ? ret2 : ret;
4572 }
4573
4574 /*
4575  * Callback function called for each extent to gather FIEMAP information.
4576  */
4577 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4578                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
4579                        void *data)
4580 {
4581         __u64   logical;
4582         __u64   physical;
4583         __u64   length;
4584         __u32   flags = 0;
4585         int             ret = 0;
4586         struct fiemap_extent_info *fieinfo = data;
4587         unsigned char blksize_bits;
4588
4589         blksize_bits = inode->i_sb->s_blocksize_bits;
4590         logical = (__u64)newex->ec_block << blksize_bits;
4591
4592         if (newex->ec_start == 0) {
4593                 /*
4594                  * No extent in extent-tree contains block @newex->ec_start,
4595                  * then the block may stay in 1)a hole or 2)delayed-extent.
4596                  *
4597                  * Holes or delayed-extents are processed as follows.
4598                  * 1. lookup dirty pages with specified range in pagecache.
4599                  *    If no page is got, then there is no delayed-extent and
4600                  *    return with EXT_CONTINUE.
4601                  * 2. find the 1st mapped buffer,
4602                  * 3. check if the mapped buffer is both in the request range
4603                  *    and a delayed buffer. If not, there is no delayed-extent,
4604                  *    then return.
4605                  * 4. a delayed-extent is found, the extent will be collected.
4606                  */
4607                 ext4_lblk_t     end = 0;
4608                 pgoff_t         last_offset;
4609                 pgoff_t         offset;
4610                 pgoff_t         index;
4611                 pgoff_t         start_index = 0;
4612                 struct page     **pages = NULL;
4613                 struct buffer_head *bh = NULL;
4614                 struct buffer_head *head = NULL;
4615                 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4616
4617                 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4618                 if (pages == NULL)
4619                         return -ENOMEM;
4620
4621                 offset = logical >> PAGE_SHIFT;
4622 repeat:
4623                 last_offset = offset;
4624                 head = NULL;
4625                 ret = find_get_pages_tag(inode->i_mapping, &offset,
4626                                         PAGECACHE_TAG_DIRTY, nr_pages, pages);
4627
4628                 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4629                         /* First time, try to find a mapped buffer. */
4630                         if (ret == 0) {
4631 out:
4632                                 for (index = 0; index < ret; index++)
4633                                         page_cache_release(pages[index]);
4634                                 /* just a hole. */
4635                                 kfree(pages);
4636                                 return EXT_CONTINUE;
4637                         }
4638                         index = 0;
4639
4640 next_page:
4641                         /* Try to find the 1st mapped buffer. */
4642                         end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4643                                   blksize_bits;
4644                         if (!page_has_buffers(pages[index]))
4645                                 goto out;
4646                         head = page_buffers(pages[index]);
4647                         if (!head)
4648                                 goto out;
4649
4650                         index++;
4651                         bh = head;
4652                         do {
4653                                 if (end >= newex->ec_block +
4654                                         newex->ec_len)
4655                                         /* The buffer is out of
4656                                          * the request range.
4657                                          */
4658                                         goto out;
4659
4660                                 if (buffer_mapped(bh) &&
4661                                     end >= newex->ec_block) {
4662                                         start_index = index - 1;
4663                                         /* get the 1st mapped buffer. */
4664                                         goto found_mapped_buffer;
4665                                 }
4666
4667                                 bh = bh->b_this_page;
4668                                 end++;
4669                         } while (bh != head);
4670
4671                         /* No mapped buffer in the range found in this page,
4672                          * We need to look up next page.
4673                          */
4674                         if (index >= ret) {
4675                                 /* There is no page left, but we need to limit
4676                                  * newex->ec_len.
4677                                  */
4678                                 newex->ec_len = end - newex->ec_block;
4679                                 goto out;
4680                         }
4681                         goto next_page;
4682                 } else {
4683                         /*Find contiguous delayed buffers. */
4684                         if (ret > 0 && pages[0]->index == last_offset)
4685                                 head = page_buffers(pages[0]);
4686                         bh = head;
4687                         index = 1;
4688                         start_index = 0;
4689                 }
4690
4691 found_mapped_buffer:
4692                 if (bh != NULL && buffer_delay(bh)) {
4693                         /* 1st or contiguous delayed buffer found. */
4694                         if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4695                                 /*
4696                                  * 1st delayed buffer found, record
4697                                  * the start of extent.
4698                                  */
4699                                 flags |= FIEMAP_EXTENT_DELALLOC;
4700                                 newex->ec_block = end;
4701                                 logical = (__u64)end << blksize_bits;
4702                         }
4703                         /* Find contiguous delayed buffers. */
4704                         do {
4705                                 if (!buffer_delay(bh))
4706                                         goto found_delayed_extent;
4707                                 bh = bh->b_this_page;
4708                                 end++;
4709                         } while (bh != head);
4710
4711                         for (; index < ret; index++) {
4712                                 if (!page_has_buffers(pages[index])) {
4713                                         bh = NULL;
4714                                         break;
4715                                 }
4716                                 head = page_buffers(pages[index]);
4717                                 if (!head) {
4718                                         bh = NULL;
4719                                         break;
4720                                 }
4721
4722                                 if (pages[index]->index !=
4723                                     pages[start_index]->index + index
4724                                     - start_index) {
4725                                         /* Blocks are not contiguous. */
4726                                         bh = NULL;
4727                                         break;
4728                                 }
4729                                 bh = head;
4730                                 do {
4731                                         if (!buffer_delay(bh))
4732                                                 /* Delayed-extent ends. */
4733                                                 goto found_delayed_extent;
4734                                         bh = bh->b_this_page;
4735                                         end++;
4736                                 } while (bh != head);
4737                         }
4738                 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4739                         /* a hole found. */
4740                         goto out;
4741
4742 found_delayed_extent:
4743                 newex->ec_len = min(end - newex->ec_block,
4744                                                 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4745                 if (ret == nr_pages && bh != NULL &&
4746                         newex->ec_len < EXT_INIT_MAX_LEN &&
4747                         buffer_delay(bh)) {
4748                         /* Have not collected an extent and continue. */
4749                         for (index = 0; index < ret; index++)
4750                                 page_cache_release(pages[index]);
4751                         goto repeat;
4752                 }
4753
4754                 for (index = 0; index < ret; index++)
4755                         page_cache_release(pages[index]);
4756                 kfree(pages);
4757         }
4758
4759         physical = (__u64)newex->ec_start << blksize_bits;
4760         length =   (__u64)newex->ec_len << blksize_bits;
4761
4762         if (ex && ext4_ext_is_uninitialized(ex))
4763                 flags |= FIEMAP_EXTENT_UNWRITTEN;
4764
4765         if (next == EXT_MAX_BLOCKS)
4766                 flags |= FIEMAP_EXTENT_LAST;
4767
4768         ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4769                                         length, flags);
4770         if (ret < 0)
4771                 return ret;
4772         if (ret == 1)
4773                 return EXT_BREAK;
4774         return EXT_CONTINUE;
4775 }
4776 /* fiemap flags we can handle specified here */
4777 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4778
4779 static int ext4_xattr_fiemap(struct inode *inode,
4780                                 struct fiemap_extent_info *fieinfo)
4781 {
4782         __u64 physical = 0;
4783         __u64 length;
4784         __u32 flags = FIEMAP_EXTENT_LAST;
4785         int blockbits = inode->i_sb->s_blocksize_bits;
4786         int error = 0;
4787
4788         /* in-inode? */
4789         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4790                 struct ext4_iloc iloc;
4791                 int offset;     /* offset of xattr in inode */
4792
4793                 error = ext4_get_inode_loc(inode, &iloc);
4794                 if (error)
4795                         return error;
4796                 physical = iloc.bh->b_blocknr << blockbits;
4797                 offset = EXT4_GOOD_OLD_INODE_SIZE +
4798                                 EXT4_I(inode)->i_extra_isize;
4799                 physical += offset;
4800                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4801                 flags |= FIEMAP_EXTENT_DATA_INLINE;
4802                 brelse(iloc.bh);
4803         } else { /* external block */
4804                 physical = EXT4_I(inode)->i_file_acl << blockbits;
4805                 length = inode->i_sb->s_blocksize;
4806         }
4807
4808         if (physical)
4809                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4810                                                 length, flags);
4811         return (error < 0 ? error : 0);
4812 }
4813
4814 /*
4815  * ext4_ext_punch_hole
4816  *
4817  * Punches a hole of "length" bytes in a file starting
4818  * at byte "offset"
4819  *
4820  * @inode:  The inode of the file to punch a hole in
4821  * @offset: The starting byte offset of the hole
4822  * @length: The length of the hole
4823  *
4824  * Returns the number of blocks removed or negative on err
4825  */
4826 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4827 {
4828         struct inode *inode = file->f_path.dentry->d_inode;
4829         struct super_block *sb = inode->i_sb;
4830         ext4_lblk_t first_block, stop_block;
4831         struct address_space *mapping = inode->i_mapping;
4832         handle_t *handle;
4833         loff_t first_page, last_page, page_len;
4834         loff_t first_page_offset, last_page_offset;
4835         int credits, err = 0;
4836
4837         /*
4838          * Write out all dirty pages to avoid race conditions
4839          * Then release them.
4840          */
4841         if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4842                 err = filemap_write_and_wait_range(mapping,
4843                         offset, offset + length - 1);
4844
4845                 if (err)
4846                         return err;
4847         }
4848
4849         mutex_lock(&inode->i_mutex);
4850         /* It's not possible punch hole on append only file */
4851         if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4852                 err = -EPERM;
4853                 goto out_mutex;
4854         }
4855         if (IS_SWAPFILE(inode)) {
4856                 err = -ETXTBSY;
4857                 goto out_mutex;
4858         }
4859
4860         /* No need to punch hole beyond i_size */
4861         if (offset >= inode->i_size)
4862                 goto out_mutex;
4863
4864         /*
4865          * If the hole extends beyond i_size, set the hole
4866          * to end after the page that contains i_size
4867          */
4868         if (offset + length > inode->i_size) {
4869                 length = inode->i_size +
4870                    PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4871                    offset;
4872         }
4873
4874         first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4875         last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4876
4877         first_page_offset = first_page << PAGE_CACHE_SHIFT;
4878         last_page_offset = last_page << PAGE_CACHE_SHIFT;
4879
4880         /* Now release the pages */
4881         if (last_page_offset > first_page_offset) {
4882                 truncate_pagecache_range(inode, first_page_offset,
4883                                          last_page_offset - 1);
4884         }
4885
4886         /* Wait all existing dio workers, newcomers will block on i_mutex */
4887         ext4_inode_block_unlocked_dio(inode);
4888         err = ext4_flush_unwritten_io(inode);
4889         if (err)
4890                 goto out_dio;
4891         inode_dio_wait(inode);
4892
4893         credits = ext4_writepage_trans_blocks(inode);
4894         handle = ext4_journal_start(inode, credits);
4895         if (IS_ERR(handle)) {
4896                 err = PTR_ERR(handle);
4897                 goto out_dio;
4898         }
4899
4900
4901         /*
4902          * Now we need to zero out the non-page-aligned data in the
4903          * pages at the start and tail of the hole, and unmap the buffer
4904          * heads for the block aligned regions of the page that were
4905          * completely zeroed.
4906          */
4907         if (first_page > last_page) {
4908                 /*
4909                  * If the file space being truncated is contained within a page
4910                  * just zero out and unmap the middle of that page
4911                  */
4912                 err = ext4_discard_partial_page_buffers(handle,
4913                         mapping, offset, length, 0);
4914
4915                 if (err)
4916                         goto out;
4917         } else {
4918                 /*
4919                  * zero out and unmap the partial page that contains
4920                  * the start of the hole
4921                  */
4922                 page_len  = first_page_offset - offset;
4923                 if (page_len > 0) {
4924                         err = ext4_discard_partial_page_buffers(handle, mapping,
4925                                                    offset, page_len, 0);
4926                         if (err)
4927                                 goto out;
4928                 }
4929
4930                 /*
4931                  * zero out and unmap the partial page that contains
4932                  * the end of the hole
4933                  */
4934                 page_len = offset + length - last_page_offset;
4935                 if (page_len > 0) {
4936                         err = ext4_discard_partial_page_buffers(handle, mapping,
4937                                         last_page_offset, page_len, 0);
4938                         if (err)
4939                                 goto out;
4940                 }
4941         }
4942
4943         /*
4944          * If i_size is contained in the last page, we need to
4945          * unmap and zero the partial page after i_size
4946          */
4947         if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4948            inode->i_size % PAGE_CACHE_SIZE != 0) {
4949
4950                 page_len = PAGE_CACHE_SIZE -
4951                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
4952
4953                 if (page_len > 0) {
4954                         err = ext4_discard_partial_page_buffers(handle,
4955                           mapping, inode->i_size, page_len, 0);
4956
4957                         if (err)
4958                                 goto out;
4959                 }
4960         }
4961
4962         first_block = (offset + sb->s_blocksize - 1) >>
4963                 EXT4_BLOCK_SIZE_BITS(sb);
4964         stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4965
4966         /* If there are no blocks to remove, return now */
4967         if (first_block >= stop_block)
4968                 goto out;
4969
4970         down_write(&EXT4_I(inode)->i_data_sem);
4971         ext4_ext_invalidate_cache(inode);
4972         ext4_discard_preallocations(inode);
4973
4974         err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
4975
4976         ext4_ext_invalidate_cache(inode);
4977         ext4_discard_preallocations(inode);
4978
4979         if (IS_SYNC(inode))
4980                 ext4_handle_sync(handle);
4981
4982         up_write(&EXT4_I(inode)->i_data_sem);
4983
4984 out:
4985         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4986         ext4_mark_inode_dirty(handle, inode);
4987         ext4_journal_stop(handle);
4988 out_dio:
4989         ext4_inode_resume_unlocked_dio(inode);
4990 out_mutex:
4991         mutex_unlock(&inode->i_mutex);
4992         return err;
4993 }
4994 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4995                 __u64 start, __u64 len)
4996 {
4997         ext4_lblk_t start_blk;
4998         int error = 0;
4999
5000         /* fallback to generic here if not in extents fmt */
5001         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5002                 return generic_block_fiemap(inode, fieinfo, start, len,
5003                         ext4_get_block);
5004
5005         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5006                 return -EBADR;
5007
5008         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5009                 error = ext4_xattr_fiemap(inode, fieinfo);
5010         } else {
5011                 ext4_lblk_t len_blks;
5012                 __u64 last_blk;
5013
5014                 start_blk = start >> inode->i_sb->s_blocksize_bits;
5015                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5016                 if (last_blk >= EXT_MAX_BLOCKS)
5017                         last_blk = EXT_MAX_BLOCKS-1;
5018                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5019
5020                 /*
5021                  * Walk the extent tree gathering extent information.
5022                  * ext4_ext_fiemap_cb will push extents back to user.
5023                  */
5024                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
5025                                           ext4_ext_fiemap_cb, fieinfo);
5026         }
5027
5028         return error;
5029 }