root/fs/ext4/extents.c

DEFINITIONS

1. ext4_extent_block_csum
2. ext4_extent_block_csum_verify
3. ext4_extent_block_csum_set
4. ext4_ext_truncate_extend_restart
5. ext4_ext_get_access
6. __ext4_ext_dirty
7. ext4_ext_find_goal
8. ext4_ext_new_meta_block
9. ext4_ext_space_block
10. ext4_ext_space_block_idx
11. ext4_ext_space_root
12. ext4_ext_space_root_idx
13. ext4_force_split_extent_at
14. ext4_ext_calc_metadata_amount
15. ext4_ext_max_entries
16. ext4_valid_extent
17. ext4_valid_extent_idx
18. ext4_valid_extent_entries
19. __ext4_ext_check
20. ext4_ext_check_inode
21. ext4_cache_extents
22. __read_extent_tree_block
23. ext4_ext_precache
24. ext4_ext_show_path
25. ext4_ext_show_leaf
26. ext4_ext_show_move
27. ext4_ext_drop_refs
28. ext4_ext_binsearch_idx
29. ext4_ext_binsearch
30. ext4_ext_tree_init
31. ext4_find_extent
32. ext4_ext_insert_index
33. ext4_ext_split
34. ext4_ext_grow_indepth
35. ext4_ext_create_new_leaf
36. ext4_ext_search_left
37. ext4_ext_search_right
38. ext4_ext_next_allocated_block
39. ext4_ext_next_leaf_block
40. ext4_ext_correct_indexes
41. ext4_can_extents_be_merged
42. ext4_ext_try_to_merge_right
43. ext4_ext_try_to_merge_up
44. ext4_ext_try_to_merge
45. ext4_ext_check_overlap
46. ext4_ext_insert_extent
47. ext4_fill_fiemap_extents
48. ext4_fill_es_cache_info
49. ext4_ext_determine_hole
50. ext4_ext_put_gap_in_cache
51. ext4_ext_rm_idx
52. ext4_ext_calc_credits_for_single_extent
53. ext4_ext_index_trans_blocks
54. get_default_free_blocks_flags
55. ext4_rereserve_cluster
56. ext4_remove_blocks
57. ext4_ext_rm_leaf
58. ext4_ext_more_to_rm
59. ext4_ext_remove_space
60. ext4_ext_init
61. ext4_ext_release
62. ext4_zeroout_es
63. ext4_ext_zeroout
64. ext4_split_extent_at
65. ext4_split_extent
66. ext4_ext_convert_to_initialized
67. ext4_split_convert_extents
68. ext4_convert_unwritten_extents_endio
69. check_eofblocks_fl
70. convert_initialized_extent
71. ext4_ext_handle_unwritten_extents
72. get_implied_cluster_alloc
73. ext4_ext_map_blocks
74. ext4_ext_truncate
75. ext4_alloc_file_blocks
76. ext4_zero_range
77. ext4_fallocate
78. ext4_convert_unwritten_extents
79. ext4_find_delayed_extent
80. ext4_xattr_fiemap
81. _ext4_fiemap
82. ext4_fiemap
83. ext4_get_es_cache
84. ext4_access_path
85. ext4_ext_shift_path_extents
86. ext4_ext_shift_extents
87. ext4_collapse_range
88. ext4_insert_range
89. ext4_swap_extents
90. ext4_clu_mapped

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   4  * Written by Alex Tomas <alex@clusterfs.com>
   5  *
   6  * Architecture independence:
   7  *   Copyright (c) 2005, Bull S.A.
   8  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   9  */
  10 
  11 /*
  12  * Extents support for EXT4
  13  *
  14  * TODO:
  15  *   - ext4*_error() should be used in some situations
  16  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  17  *   - smart tree reduction
  18  */
  19 
  20 #include <linux/fs.h>
  21 #include <linux/time.h>
  22 #include <linux/jbd2.h>
  23 #include <linux/highuid.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/quotaops.h>
  26 #include <linux/string.h>
  27 #include <linux/slab.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/fiemap.h>
  30 #include <linux/backing-dev.h>
  31 #include "ext4_jbd2.h"
  32 #include "ext4_extents.h"
  33 #include "xattr.h"
  34 
  35 #include <trace/events/ext4.h>
  36 
  37 /*
  38  * used by extent splitting.
  39  */
  40 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  41                                         due to ENOSPC */
  42 #define EXT4_EXT_MARK_UNWRIT1   0x2  /* mark first half unwritten */
  43 #define EXT4_EXT_MARK_UNWRIT2   0x4  /* mark second half unwritten */
  44 
  45 #define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  46 #define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  47 
  48 static __le32 ext4_extent_block_csum(struct inode *inode,
  49                                      struct ext4_extent_header *eh)
  50 {
  51         struct ext4_inode_info *ei = EXT4_I(inode);
  52         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  53         __u32 csum;
  54 
  55         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  56                            EXT4_EXTENT_TAIL_OFFSET(eh));
  57         return cpu_to_le32(csum);
  58 }
  59 
  60 static int ext4_extent_block_csum_verify(struct inode *inode,
  61                                          struct ext4_extent_header *eh)
  62 {
  63         struct ext4_extent_tail *et;
  64 
  65         if (!ext4_has_metadata_csum(inode->i_sb))
  66                 return 1;
  67 
  68         et = find_ext4_extent_tail(eh);
  69         if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  70                 return 0;
  71         return 1;
  72 }
  73 
  74 static void ext4_extent_block_csum_set(struct inode *inode,
  75                                        struct ext4_extent_header *eh)
  76 {
  77         struct ext4_extent_tail *et;
  78 
  79         if (!ext4_has_metadata_csum(inode->i_sb))
  80                 return;
  81 
  82         et = find_ext4_extent_tail(eh);
  83         et->et_checksum = ext4_extent_block_csum(inode, eh);
  84 }
  85 
  86 static int ext4_split_extent(handle_t *handle,
  87                                 struct inode *inode,
  88                                 struct ext4_ext_path **ppath,
  89                                 struct ext4_map_blocks *map,
  90                                 int split_flag,
  91                                 int flags);
  92 
  93 static int ext4_split_extent_at(handle_t *handle,
  94                              struct inode *inode,
  95                              struct ext4_ext_path **ppath,
  96                              ext4_lblk_t split,
  97                              int split_flag,
  98                              int flags);
  99 
 100 static int ext4_find_delayed_extent(struct inode *inode,
 101                                     struct extent_status *newes);
 102 
 103 static int ext4_ext_truncate_extend_restart(handle_t *handle,
 104                                             struct inode *inode,
 105                                             int needed)
 106 {
 107         int err;
 108 
 109         if (!ext4_handle_valid(handle))
 110                 return 0;
 111         if (handle->h_buffer_credits >= needed)
 112                 return 0;
 113         /*
 114          * If we need to extend the journal get a few extra blocks
 115          * while we're at it for efficiency's sake.
 116          */
 117         needed += 3;
 118         err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
 119         if (err <= 0)
 120                 return err;
 121         err = ext4_truncate_restart_trans(handle, inode, needed);
 122         if (err == 0)
 123                 err = -EAGAIN;
 124 
 125         return err;
 126 }
 127 
 128 /*
 129  * could return:
 130  *  - EROFS
 131  *  - ENOMEM
 132  */
 133 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 134                                 struct ext4_ext_path *path)
 135 {
 136         if (path->p_bh) {
 137                 /* path points to block */
 138                 BUFFER_TRACE(path->p_bh, "get_write_access");
 139                 return ext4_journal_get_write_access(handle, path->p_bh);
 140         }
 141         /* path points to leaf/index in inode body */
 142         /* we use in-core data, no need to protect them */
 143         return 0;
 144 }
 145 
 146 /*
 147  * could return:
 148  *  - EROFS
 149  *  - ENOMEM
 150  *  - EIO
 151  */
 152 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
 153                      struct inode *inode, struct ext4_ext_path *path)
 154 {
 155         int err;
 156 
 157         WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 158         if (path->p_bh) {
 159                 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 160                 /* path points to block */
 161                 err = __ext4_handle_dirty_metadata(where, line, handle,
 162                                                    inode, path->p_bh);
 163         } else {
 164                 /* path points to leaf/index in inode body */
 165                 err = ext4_mark_inode_dirty(handle, inode);
 166         }
 167         return err;
 168 }
 169 
 170 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 171                               struct ext4_ext_path *path,
 172                               ext4_lblk_t block)
 173 {
 174         if (path) {
 175                 int depth = path->p_depth;
 176                 struct ext4_extent *ex;
 177 
 178                 /*
 179                  * Try to predict block placement assuming that we are
 180                  * filling in a file which will eventually be
 181                  * non-sparse --- i.e., in the case of libbfd writing
 182                  * an ELF object sections out-of-order but in a way
 183                  * the eventually results in a contiguous object or
 184                  * executable file, or some database extending a table
 185                  * space file.  However, this is actually somewhat
 186                  * non-ideal if we are writing a sparse file such as
 187                  * qemu or KVM writing a raw image file that is going
 188                  * to stay fairly sparse, since it will end up
 189                  * fragmenting the file system's free space.  Maybe we
 190                  * should have some hueristics or some way to allow
 191                  * userspace to pass a hint to file system,
 192                  * especially if the latter case turns out to be
 193                  * common.
 194                  */
 195                 ex = path[depth].p_ext;
 196                 if (ex) {
 197                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 198                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 199 
 200                         if (block > ext_block)
 201                                 return ext_pblk + (block - ext_block);
 202                         else
 203                                 return ext_pblk - (ext_block - block);
 204                 }
 205 
 206                 /* it looks like index is empty;
 207                  * try to find starting block from index itself */
 208                 if (path[depth].p_bh)
 209                         return path[depth].p_bh->b_blocknr;
 210         }
 211 
 212         /* OK. use inode's group */
 213         return ext4_inode_to_goal_block(inode);
 214 }
 215 
 216 /*
 217  * Allocation for a meta data block
 218  */
 219 static ext4_fsblk_t
 220 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 221                         struct ext4_ext_path *path,
 222                         struct ext4_extent *ex, int *err, unsigned int flags)
 223 {
 224         ext4_fsblk_t goal, newblock;
 225 
 226         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 227         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 228                                         NULL, err);
 229         return newblock;
 230 }
 231 
 232 static inline int ext4_ext_space_block(struct inode *inode, int check)
 233 {
 234         int size;
 235 
 236         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 237                         / sizeof(struct ext4_extent);
 238 #ifdef AGGRESSIVE_TEST
 239         if (!check && size > 6)
 240                 size = 6;
 241 #endif
 242         return size;
 243 }
 244 
 245 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 246 {
 247         int size;
 248 
 249         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 250                         / sizeof(struct ext4_extent_idx);
 251 #ifdef AGGRESSIVE_TEST
 252         if (!check && size > 5)
 253                 size = 5;
 254 #endif
 255         return size;
 256 }
 257 
 258 static inline int ext4_ext_space_root(struct inode *inode, int check)
 259 {
 260         int size;
 261 
 262         size = sizeof(EXT4_I(inode)->i_data);
 263         size -= sizeof(struct ext4_extent_header);
 264         size /= sizeof(struct ext4_extent);
 265 #ifdef AGGRESSIVE_TEST
 266         if (!check && size > 3)
 267                 size = 3;
 268 #endif
 269         return size;
 270 }
 271 
 272 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 273 {
 274         int size;
 275 
 276         size = sizeof(EXT4_I(inode)->i_data);
 277         size -= sizeof(struct ext4_extent_header);
 278         size /= sizeof(struct ext4_extent_idx);
 279 #ifdef AGGRESSIVE_TEST
 280         if (!check && size > 4)
 281                 size = 4;
 282 #endif
 283         return size;
 284 }
 285 
 286 static inline int
 287 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 288                            struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 289                            int nofail)
 290 {
 291         struct ext4_ext_path *path = *ppath;
 292         int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 293 
 294         return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 295                         EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 296                         EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
 297                         (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
 298 }
 299 
 300 /*
 301  * Calculate the number of metadata blocks needed
 302  * to allocate @blocks
 303  * Worse case is one block per extent
 304  */
 305 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 306 {
 307         struct ext4_inode_info *ei = EXT4_I(inode);
 308         int idxs;
 309 
 310         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 311                 / sizeof(struct ext4_extent_idx));
 312 
 313         /*
 314          * If the new delayed allocation block is contiguous with the
 315          * previous da block, it can share index blocks with the
 316          * previous block, so we only need to allocate a new index
 317          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 318          * an additional index block, and at ldxs**3 blocks, yet
 319          * another index blocks.
 320          */
 321         if (ei->i_da_metadata_calc_len &&
 322             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 323                 int num = 0;
 324 
 325                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 326                         num++;
 327                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 328                         num++;
 329                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 330                         num++;
 331                         ei->i_da_metadata_calc_len = 0;
 332                 } else
 333                         ei->i_da_metadata_calc_len++;
 334                 ei->i_da_metadata_calc_last_lblock++;
 335                 return num;
 336         }
 337 
 338         /*
 339          * In the worst case we need a new set of index blocks at
 340          * every level of the inode's extent tree.
 341          */
 342         ei->i_da_metadata_calc_len = 1;
 343         ei->i_da_metadata_calc_last_lblock = lblock;
 344         return ext_depth(inode) + 1;
 345 }
 346 
 347 static int
 348 ext4_ext_max_entries(struct inode *inode, int depth)
 349 {
 350         int max;
 351 
 352         if (depth == ext_depth(inode)) {
 353                 if (depth == 0)
 354                         max = ext4_ext_space_root(inode, 1);
 355                 else
 356                         max = ext4_ext_space_root_idx(inode, 1);
 357         } else {
 358                 if (depth == 0)
 359                         max = ext4_ext_space_block(inode, 1);
 360                 else
 361                         max = ext4_ext_space_block_idx(inode, 1);
 362         }
 363 
 364         return max;
 365 }
 366 
 367 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 368 {
 369         ext4_fsblk_t block = ext4_ext_pblock(ext);
 370         int len = ext4_ext_get_actual_len(ext);
 371         ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 372 
 373         /*
 374          * We allow neither:
 375          *  - zero length
 376          *  - overflow/wrap-around
 377          */
 378         if (lblock + len <= lblock)
 379                 return 0;
 380         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 381 }
 382 
 383 static int ext4_valid_extent_idx(struct inode *inode,
 384                                 struct ext4_extent_idx *ext_idx)
 385 {
 386         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 387 
 388         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 389 }
 390 
 391 static int ext4_valid_extent_entries(struct inode *inode,
 392                                 struct ext4_extent_header *eh,
 393                                 int depth)
 394 {
 395         unsigned short entries;
 396         if (eh->eh_entries == 0)
 397                 return 1;
 398 
 399         entries = le16_to_cpu(eh->eh_entries);
 400 
 401         if (depth == 0) {
 402                 /* leaf entries */
 403                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 404                 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 405                 ext4_fsblk_t pblock = 0;
 406                 ext4_lblk_t lblock = 0;
 407                 ext4_lblk_t prev = 0;
 408                 int len = 0;
 409                 while (entries) {
 410                         if (!ext4_valid_extent(inode, ext))
 411                                 return 0;
 412 
 413                         /* Check for overlapping extents */
 414                         lblock = le32_to_cpu(ext->ee_block);
 415                         len = ext4_ext_get_actual_len(ext);
 416                         if ((lblock <= prev) && prev) {
 417                                 pblock = ext4_ext_pblock(ext);
 418                                 es->s_last_error_block = cpu_to_le64(pblock);
 419                                 return 0;
 420                         }
 421                         ext++;
 422                         entries--;
 423                         prev = lblock + len - 1;
 424                 }
 425         } else {
 426                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 427                 while (entries) {
 428                         if (!ext4_valid_extent_idx(inode, ext_idx))
 429                                 return 0;
 430                         ext_idx++;
 431                         entries--;
 432                 }
 433         }
 434         return 1;
 435 }
 436 
 437 static int __ext4_ext_check(const char *function, unsigned int line,
 438                             struct inode *inode, struct ext4_extent_header *eh,
 439                             int depth, ext4_fsblk_t pblk)
 440 {
 441         const char *error_msg;
 442         int max = 0, err = -EFSCORRUPTED;
 443 
 444         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 445                 error_msg = "invalid magic";
 446                 goto corrupted;
 447         }
 448         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 449                 error_msg = "unexpected eh_depth";
 450                 goto corrupted;
 451         }
 452         if (unlikely(eh->eh_max == 0)) {
 453                 error_msg = "invalid eh_max";
 454                 goto corrupted;
 455         }
 456         max = ext4_ext_max_entries(inode, depth);
 457         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 458                 error_msg = "too large eh_max";
 459                 goto corrupted;
 460         }
 461         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 462                 error_msg = "invalid eh_entries";
 463                 goto corrupted;
 464         }
 465         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 466                 error_msg = "invalid extent entries";
 467                 goto corrupted;
 468         }
 469         if (unlikely(depth > 32)) {
 470                 error_msg = "too large eh_depth";
 471                 goto corrupted;
 472         }
 473         /* Verify checksum on non-root extent tree nodes */
 474         if (ext_depth(inode) != depth &&
 475             !ext4_extent_block_csum_verify(inode, eh)) {
 476                 error_msg = "extent tree corrupted";
 477                 err = -EFSBADCRC;
 478                 goto corrupted;
 479         }
 480         return 0;
 481 
 482 corrupted:
 483         ext4_error_inode(inode, function, line, 0,
 484                          "pblk %llu bad header/extent: %s - magic %x, "
 485                          "entries %u, max %u(%u), depth %u(%u)",
 486                          (unsigned long long) pblk, error_msg,
 487                          le16_to_cpu(eh->eh_magic),
 488                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 489                          max, le16_to_cpu(eh->eh_depth), depth);
 490         return err;
 491 }
 492 
 493 #define ext4_ext_check(inode, eh, depth, pblk)                  \
 494         __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
 495 
 496 int ext4_ext_check_inode(struct inode *inode)
 497 {
 498         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 499 }
 500 
 501 static void ext4_cache_extents(struct inode *inode,
 502                                struct ext4_extent_header *eh)
 503 {
 504         struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 505         ext4_lblk_t prev = 0;
 506         int i;
 507 
 508         for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 509                 unsigned int status = EXTENT_STATUS_WRITTEN;
 510                 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 511                 int len = ext4_ext_get_actual_len(ex);
 512 
 513                 if (prev && (prev != lblk))
 514                         ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
 515                                              EXTENT_STATUS_HOLE);
 516 
 517                 if (ext4_ext_is_unwritten(ex))
 518                         status = EXTENT_STATUS_UNWRITTEN;
 519                 ext4_es_cache_extent(inode, lblk, len,
 520                                      ext4_ext_pblock(ex), status);
 521                 prev = lblk + len;
 522         }
 523 }
 524 
 525 static struct buffer_head *
 526 __read_extent_tree_block(const char *function, unsigned int line,
 527                          struct inode *inode, ext4_fsblk_t pblk, int depth,
 528                          int flags)
 529 {
 530         struct buffer_head              *bh;
 531         int                             err;
 532 
 533         bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
 534         if (unlikely(!bh))
 535                 return ERR_PTR(-ENOMEM);
 536 
 537         if (!bh_uptodate_or_lock(bh)) {
 538                 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 539                 err = bh_submit_read(bh);
 540                 if (err < 0)
 541                         goto errout;
 542         }
 543         if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 544                 return bh;
 545         if (!ext4_has_feature_journal(inode->i_sb) ||
 546             (inode->i_ino !=
 547              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
 548                 err = __ext4_ext_check(function, line, inode,
 549                                        ext_block_hdr(bh), depth, pblk);
 550                 if (err)
 551                         goto errout;
 552         }
 553         set_buffer_verified(bh);
 554         /*
 555          * If this is a leaf block, cache all of its entries
 556          */
 557         if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 558                 struct ext4_extent_header *eh = ext_block_hdr(bh);
 559                 ext4_cache_extents(inode, eh);
 560         }
 561         return bh;
 562 errout:
 563         put_bh(bh);
 564         return ERR_PTR(err);
 565 
 566 }
 567 
 568 #define read_extent_tree_block(inode, pblk, depth, flags)               \
 569         __read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
 570                                  (depth), (flags))
 571 
 572 /*
 573  * This function is called to cache a file's extent information in the
 574  * extent status tree
 575  */
 576 int ext4_ext_precache(struct inode *inode)
 577 {
 578         struct ext4_inode_info *ei = EXT4_I(inode);
 579         struct ext4_ext_path *path = NULL;
 580         struct buffer_head *bh;
 581         int i = 0, depth, ret = 0;
 582 
 583         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 584                 return 0;       /* not an extent-mapped inode */
 585 
 586         down_read(&ei->i_data_sem);
 587         depth = ext_depth(inode);
 588 
 589         path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
 590                        GFP_NOFS);
 591         if (path == NULL) {
 592                 up_read(&ei->i_data_sem);
 593                 return -ENOMEM;
 594         }
 595 
 596         /* Don't cache anything if there are no external extent blocks */
 597         if (depth == 0)
 598                 goto out;
 599         path[0].p_hdr = ext_inode_hdr(inode);
 600         ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 601         if (ret)
 602                 goto out;
 603         path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 604         while (i >= 0) {
 605                 /*
 606                  * If this is a leaf block or we've reached the end of
 607                  * the index block, go up
 608                  */
 609                 if ((i == depth) ||
 610                     path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 611                         brelse(path[i].p_bh);
 612                         path[i].p_bh = NULL;
 613                         i--;
 614                         continue;
 615                 }
 616                 bh = read_extent_tree_block(inode,
 617                                             ext4_idx_pblock(path[i].p_idx++),
 618                                             depth - i - 1,
 619                                             EXT4_EX_FORCE_CACHE);
 620                 if (IS_ERR(bh)) {
 621                         ret = PTR_ERR(bh);
 622                         break;
 623                 }
 624                 i++;
 625                 path[i].p_bh = bh;
 626                 path[i].p_hdr = ext_block_hdr(bh);
 627                 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 628         }
 629         ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 630 out:
 631         up_read(&ei->i_data_sem);
 632         ext4_ext_drop_refs(path);
 633         kfree(path);
 634         return ret;
 635 }
 636 
 637 #ifdef EXT_DEBUG
 638 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 639 {
 640         int k, l = path->p_depth;
 641 
 642         ext_debug("path:");
 643         for (k = 0; k <= l; k++, path++) {
 644                 if (path->p_idx) {
 645                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 646                             ext4_idx_pblock(path->p_idx));
 647                 } else if (path->p_ext) {
 648                         ext_debug("  %d:[%d]%d:%llu ",
 649                                   le32_to_cpu(path->p_ext->ee_block),
 650                                   ext4_ext_is_unwritten(path->p_ext),
 651                                   ext4_ext_get_actual_len(path->p_ext),
 652                                   ext4_ext_pblock(path->p_ext));
 653                 } else
 654                         ext_debug("  []");
 655         }
 656         ext_debug("\n");
 657 }
 658 
 659 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 660 {
 661         int depth = ext_depth(inode);
 662         struct ext4_extent_header *eh;
 663         struct ext4_extent *ex;
 664         int i;
 665 
 666         if (!path)
 667                 return;
 668 
 669         eh = path[depth].p_hdr;
 670         ex = EXT_FIRST_EXTENT(eh);
 671 
 672         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 673 
 674         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 675                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 676                           ext4_ext_is_unwritten(ex),
 677                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 678         }
 679         ext_debug("\n");
 680 }
 681 
 682 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 683                         ext4_fsblk_t newblock, int level)
 684 {
 685         int depth = ext_depth(inode);
 686         struct ext4_extent *ex;
 687 
 688         if (depth != level) {
 689                 struct ext4_extent_idx *idx;
 690                 idx = path[level].p_idx;
 691                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 692                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
 693                                         le32_to_cpu(idx->ei_block),
 694                                         ext4_idx_pblock(idx),
 695                                         newblock);
 696                         idx++;
 697                 }
 698 
 699                 return;
 700         }
 701 
 702         ex = path[depth].p_ext;
 703         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 704                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 705                                 le32_to_cpu(ex->ee_block),
 706                                 ext4_ext_pblock(ex),
 707                                 ext4_ext_is_unwritten(ex),
 708                                 ext4_ext_get_actual_len(ex),
 709                                 newblock);
 710                 ex++;
 711         }
 712 }
 713 
 714 #else
 715 #define ext4_ext_show_path(inode, path)
 716 #define ext4_ext_show_leaf(inode, path)
 717 #define ext4_ext_show_move(inode, path, newblock, level)
 718 #endif
 719 
 720 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 721 {
 722         int depth, i;
 723 
 724         if (!path)
 725                 return;
 726         depth = path->p_depth;
 727         for (i = 0; i <= depth; i++, path++)
 728                 if (path->p_bh) {
 729                         brelse(path->p_bh);
 730                         path->p_bh = NULL;
 731                 }
 732 }
 733 
 734 /*
 735  * ext4_ext_binsearch_idx:
 736  * binary search for the closest index of the given block
 737  * the header must be checked before calling this
 738  */
 739 static void
 740 ext4_ext_binsearch_idx(struct inode *inode,
 741                         struct ext4_ext_path *path, ext4_lblk_t block)
 742 {
 743         struct ext4_extent_header *eh = path->p_hdr;
 744         struct ext4_extent_idx *r, *l, *m;
 745 
 746 
 747         ext_debug("binsearch for %u(idx):  ", block);
 748 
 749         l = EXT_FIRST_INDEX(eh) + 1;
 750         r = EXT_LAST_INDEX(eh);
 751         while (l <= r) {
 752                 m = l + (r - l) / 2;
 753                 if (block < le32_to_cpu(m->ei_block))
 754                         r = m - 1;
 755                 else
 756                         l = m + 1;
 757                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 758                                 m, le32_to_cpu(m->ei_block),
 759                                 r, le32_to_cpu(r->ei_block));
 760         }
 761 
 762         path->p_idx = l - 1;
 763         ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 764                   ext4_idx_pblock(path->p_idx));
 765 
 766 #ifdef CHECK_BINSEARCH
 767         {
 768                 struct ext4_extent_idx *chix, *ix;
 769                 int k;
 770 
 771                 chix = ix = EXT_FIRST_INDEX(eh);
 772                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 773                   if (k != 0 &&
 774                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 775                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 776                                        "first=0x%p\n", k,
 777                                        ix, EXT_FIRST_INDEX(eh));
 778                                 printk(KERN_DEBUG "%u <= %u\n",
 779                                        le32_to_cpu(ix->ei_block),
 780                                        le32_to_cpu(ix[-1].ei_block));
 781                         }
 782                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 783                                            <= le32_to_cpu(ix[-1].ei_block));
 784                         if (block < le32_to_cpu(ix->ei_block))
 785                                 break;
 786                         chix = ix;
 787                 }
 788                 BUG_ON(chix != path->p_idx);
 789         }
 790 #endif
 791 
 792 }
 793 
 794 /*
 795  * ext4_ext_binsearch:
 796  * binary search for closest extent of the given block
 797  * the header must be checked before calling this
 798  */
 799 static void
 800 ext4_ext_binsearch(struct inode *inode,
 801                 struct ext4_ext_path *path, ext4_lblk_t block)
 802 {
 803         struct ext4_extent_header *eh = path->p_hdr;
 804         struct ext4_extent *r, *l, *m;
 805 
 806         if (eh->eh_entries == 0) {
 807                 /*
 808                  * this leaf is empty:
 809                  * we get such a leaf in split/add case
 810                  */
 811                 return;
 812         }
 813 
 814         ext_debug("binsearch for %u:  ", block);
 815 
 816         l = EXT_FIRST_EXTENT(eh) + 1;
 817         r = EXT_LAST_EXTENT(eh);
 818 
 819         while (l <= r) {
 820                 m = l + (r - l) / 2;
 821                 if (block < le32_to_cpu(m->ee_block))
 822                         r = m - 1;
 823                 else
 824                         l = m + 1;
 825                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 826                                 m, le32_to_cpu(m->ee_block),
 827                                 r, le32_to_cpu(r->ee_block));
 828         }
 829 
 830         path->p_ext = l - 1;
 831         ext_debug("  -> %d:%llu:[%d]%d ",
 832                         le32_to_cpu(path->p_ext->ee_block),
 833                         ext4_ext_pblock(path->p_ext),
 834                         ext4_ext_is_unwritten(path->p_ext),
 835                         ext4_ext_get_actual_len(path->p_ext));
 836 
 837 #ifdef CHECK_BINSEARCH
 838         {
 839                 struct ext4_extent *chex, *ex;
 840                 int k;
 841 
 842                 chex = ex = EXT_FIRST_EXTENT(eh);
 843                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 844                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 845                                           <= le32_to_cpu(ex[-1].ee_block));
 846                         if (block < le32_to_cpu(ex->ee_block))
 847                                 break;
 848                         chex = ex;
 849                 }
 850                 BUG_ON(chex != path->p_ext);
 851         }
 852 #endif
 853 
 854 }
 855 
 856 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 857 {
 858         struct ext4_extent_header *eh;
 859 
 860         eh = ext_inode_hdr(inode);
 861         eh->eh_depth = 0;
 862         eh->eh_entries = 0;
 863         eh->eh_magic = EXT4_EXT_MAGIC;
 864         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 865         ext4_mark_inode_dirty(handle, inode);
 866         return 0;
 867 }
 868 
 869 struct ext4_ext_path *
 870 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 871                  struct ext4_ext_path **orig_path, int flags)
 872 {
 873         struct ext4_extent_header *eh;
 874         struct buffer_head *bh;
 875         struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 876         short int depth, i, ppos = 0;
 877         int ret;
 878 
 879         eh = ext_inode_hdr(inode);
 880         depth = ext_depth(inode);
 881         if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
 882                 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
 883                                  depth);
 884                 ret = -EFSCORRUPTED;
 885                 goto err;
 886         }
 887 
 888         if (path) {
 889                 ext4_ext_drop_refs(path);
 890                 if (depth > path[0].p_maxdepth) {
 891                         kfree(path);
 892                         *orig_path = path = NULL;
 893                 }
 894         }
 895         if (!path) {
 896                 /* account possible depth increase */
 897                 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
 898                                 GFP_NOFS);
 899                 if (unlikely(!path))
 900                         return ERR_PTR(-ENOMEM);
 901                 path[0].p_maxdepth = depth + 1;
 902         }
 903         path[0].p_hdr = eh;
 904         path[0].p_bh = NULL;
 905 
 906         i = depth;
 907         if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
 908                 ext4_cache_extents(inode, eh);
 909         /* walk through the tree */
 910         while (i) {
 911                 ext_debug("depth %d: num %d, max %d\n",
 912                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 913 
 914                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 915                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 916                 path[ppos].p_depth = i;
 917                 path[ppos].p_ext = NULL;
 918 
 919                 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
 920                                             flags);
 921                 if (IS_ERR(bh)) {
 922                         ret = PTR_ERR(bh);
 923                         goto err;
 924                 }
 925 
 926                 eh = ext_block_hdr(bh);
 927                 ppos++;
 928                 path[ppos].p_bh = bh;
 929                 path[ppos].p_hdr = eh;
 930         }
 931 
 932         path[ppos].p_depth = i;
 933         path[ppos].p_ext = NULL;
 934         path[ppos].p_idx = NULL;
 935 
 936         /* find extent */
 937         ext4_ext_binsearch(inode, path + ppos, block);
 938         /* if not an empty leaf */
 939         if (path[ppos].p_ext)
 940                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 941 
 942         ext4_ext_show_path(inode, path);
 943 
 944         return path;
 945 
 946 err:
 947         ext4_ext_drop_refs(path);
 948         kfree(path);
 949         if (orig_path)
 950                 *orig_path = NULL;
 951         return ERR_PTR(ret);
 952 }
 953 
 954 /*
 955  * ext4_ext_insert_index:
 956  * insert new index [@logical;@ptr] into the block at @curp;
 957  * check where to insert: before @curp or after @curp
 958  */
 959 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 960                                  struct ext4_ext_path *curp,
 961                                  int logical, ext4_fsblk_t ptr)
 962 {
 963         struct ext4_extent_idx *ix;
 964         int len, err;
 965 
 966         err = ext4_ext_get_access(handle, inode, curp);
 967         if (err)
 968                 return err;
 969 
 970         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 971                 EXT4_ERROR_INODE(inode,
 972                                  "logical %d == ei_block %d!",
 973                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 974                 return -EFSCORRUPTED;
 975         }
 976 
 977         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 978                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 979                 EXT4_ERROR_INODE(inode,
 980                                  "eh_entries %d >= eh_max %d!",
 981                                  le16_to_cpu(curp->p_hdr->eh_entries),
 982                                  le16_to_cpu(curp->p_hdr->eh_max));
 983                 return -EFSCORRUPTED;
 984         }
 985 
 986         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 987                 /* insert after */
 988                 ext_debug("insert new index %d after: %llu\n", logical, ptr);
 989                 ix = curp->p_idx + 1;
 990         } else {
 991                 /* insert before */
 992                 ext_debug("insert new index %d before: %llu\n", logical, ptr);
 993                 ix = curp->p_idx;
 994         }
 995 
 996         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 997         BUG_ON(len < 0);
 998         if (len > 0) {
 999                 ext_debug("insert new index %d: "
1000                                 "move %d indices from 0x%p to 0x%p\n",
1001                                 logical, len, ix, ix + 1);
1002                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1003         }
1004 
1005         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1006                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1007                 return -EFSCORRUPTED;
1008         }
1009 
1010         ix->ei_block = cpu_to_le32(logical);
1011         ext4_idx_store_pblock(ix, ptr);
1012         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1013 
1014         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1015                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1016                 return -EFSCORRUPTED;
1017         }
1018 
1019         err = ext4_ext_dirty(handle, inode, curp);
1020         ext4_std_error(inode->i_sb, err);
1021 
1022         return err;
1023 }
1024 
1025 /*
1026  * ext4_ext_split:
1027  * inserts new subtree into the path, using free index entry
1028  * at depth @at:
1029  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1030  * - makes decision where to split
1031  * - moves remaining extents and index entries (right to the split point)
1032  *   into the newly allocated blocks
1033  * - initializes subtree
1034  */
1035 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1036                           unsigned int flags,
1037                           struct ext4_ext_path *path,
1038                           struct ext4_extent *newext, int at)
1039 {
1040         struct buffer_head *bh = NULL;
1041         int depth = ext_depth(inode);
1042         struct ext4_extent_header *neh;
1043         struct ext4_extent_idx *fidx;
1044         int i = at, k, m, a;
1045         ext4_fsblk_t newblock, oldblock;
1046         __le32 border;
1047         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1048         int err = 0;
1049         size_t ext_size = 0;
1050 
1051         /* make decision: where to split? */
1052         /* FIXME: now decision is simplest: at current extent */
1053 
1054         /* if current leaf will be split, then we should use
1055          * border from split point */
1056         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1057                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1058                 return -EFSCORRUPTED;
1059         }
1060         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1061                 border = path[depth].p_ext[1].ee_block;
1062                 ext_debug("leaf will be split."
1063                                 " next leaf starts at %d\n",
1064                                   le32_to_cpu(border));
1065         } else {
1066                 border = newext->ee_block;
1067                 ext_debug("leaf will be added."
1068                                 " next leaf starts at %d\n",
1069                                 le32_to_cpu(border));
1070         }
1071 
1072         /*
1073          * If error occurs, then we break processing
1074          * and mark filesystem read-only. index won't
1075          * be inserted and tree will be in consistent
1076          * state. Next mount will repair buffers too.
1077          */
1078 
1079         /*
1080          * Get array to track all allocated blocks.
1081          * We need this to handle errors and free blocks
1082          * upon them.
1083          */
1084         ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1085         if (!ablocks)
1086                 return -ENOMEM;
1087 
1088         /* allocate all needed blocks */
1089         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1090         for (a = 0; a < depth - at; a++) {
1091                 newblock = ext4_ext_new_meta_block(handle, inode, path,
1092                                                    newext, &err, flags);
1093                 if (newblock == 0)
1094                         goto cleanup;
1095                 ablocks[a] = newblock;
1096         }
1097 
1098         /* initialize new leaf */
1099         newblock = ablocks[--a];
1100         if (unlikely(newblock == 0)) {
1101                 EXT4_ERROR_INODE(inode, "newblock == 0!");
1102                 err = -EFSCORRUPTED;
1103                 goto cleanup;
1104         }
1105         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1106         if (unlikely(!bh)) {
1107                 err = -ENOMEM;
1108                 goto cleanup;
1109         }
1110         lock_buffer(bh);
1111 
1112         err = ext4_journal_get_create_access(handle, bh);
1113         if (err)
1114                 goto cleanup;
1115 
1116         neh = ext_block_hdr(bh);
1117         neh->eh_entries = 0;
1118         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1119         neh->eh_magic = EXT4_EXT_MAGIC;
1120         neh->eh_depth = 0;
1121 
1122         /* move remainder of path[depth] to the new leaf */
1123         if (unlikely(path[depth].p_hdr->eh_entries !=
1124                      path[depth].p_hdr->eh_max)) {
1125                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1126                                  path[depth].p_hdr->eh_entries,
1127                                  path[depth].p_hdr->eh_max);
1128                 err = -EFSCORRUPTED;
1129                 goto cleanup;
1130         }
1131         /* start copy from next extent */
1132         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1133         ext4_ext_show_move(inode, path, newblock, depth);
1134         if (m) {
1135                 struct ext4_extent *ex;
1136                 ex = EXT_FIRST_EXTENT(neh);
1137                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1138                 le16_add_cpu(&neh->eh_entries, m);
1139         }
1140 
1141         /* zero out unused area in the extent block */
1142         ext_size = sizeof(struct ext4_extent_header) +
1143                 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1144         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1145         ext4_extent_block_csum_set(inode, neh);
1146         set_buffer_uptodate(bh);
1147         unlock_buffer(bh);
1148 
1149         err = ext4_handle_dirty_metadata(handle, inode, bh);
1150         if (err)
1151                 goto cleanup;
1152         brelse(bh);
1153         bh = NULL;
1154 
1155         /* correct old leaf */
1156         if (m) {
1157                 err = ext4_ext_get_access(handle, inode, path + depth);
1158                 if (err)
1159                         goto cleanup;
1160                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1161                 err = ext4_ext_dirty(handle, inode, path + depth);
1162                 if (err)
1163                         goto cleanup;
1164 
1165         }
1166 
1167         /* create intermediate indexes */
1168         k = depth - at - 1;
1169         if (unlikely(k < 0)) {
1170                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1171                 err = -EFSCORRUPTED;
1172                 goto cleanup;
1173         }
1174         if (k)
1175                 ext_debug("create %d intermediate indices\n", k);
1176         /* insert new index into current index block */
1177         /* current depth stored in i var */
1178         i = depth - 1;
1179         while (k--) {
1180                 oldblock = newblock;
1181                 newblock = ablocks[--a];
1182                 bh = sb_getblk(inode->i_sb, newblock);
1183                 if (unlikely(!bh)) {
1184                         err = -ENOMEM;
1185                         goto cleanup;
1186                 }
1187                 lock_buffer(bh);
1188 
1189                 err = ext4_journal_get_create_access(handle, bh);
1190                 if (err)
1191                         goto cleanup;
1192 
1193                 neh = ext_block_hdr(bh);
1194                 neh->eh_entries = cpu_to_le16(1);
1195                 neh->eh_magic = EXT4_EXT_MAGIC;
1196                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1197                 neh->eh_depth = cpu_to_le16(depth - i);
1198                 fidx = EXT_FIRST_INDEX(neh);
1199                 fidx->ei_block = border;
1200                 ext4_idx_store_pblock(fidx, oldblock);
1201 
1202                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1203                                 i, newblock, le32_to_cpu(border), oldblock);
1204 
1205                 /* move remainder of path[i] to the new index block */
1206                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1207                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1208                         EXT4_ERROR_INODE(inode,
1209                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1210                                          le32_to_cpu(path[i].p_ext->ee_block));
1211                         err = -EFSCORRUPTED;
1212                         goto cleanup;
1213                 }
1214                 /* start copy indexes */
1215                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1216                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1217                                 EXT_MAX_INDEX(path[i].p_hdr));
1218                 ext4_ext_show_move(inode, path, newblock, i);
1219                 if (m) {
1220                         memmove(++fidx, path[i].p_idx,
1221                                 sizeof(struct ext4_extent_idx) * m);
1222                         le16_add_cpu(&neh->eh_entries, m);
1223                 }
1224                 /* zero out unused area in the extent block */
1225                 ext_size = sizeof(struct ext4_extent_header) +
1226                    (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1227                 memset(bh->b_data + ext_size, 0,
1228                         inode->i_sb->s_blocksize - ext_size);
1229                 ext4_extent_block_csum_set(inode, neh);
1230                 set_buffer_uptodate(bh);
1231                 unlock_buffer(bh);
1232 
1233                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1234                 if (err)
1235                         goto cleanup;
1236                 brelse(bh);
1237                 bh = NULL;
1238 
1239                 /* correct old index */
1240                 if (m) {
1241                         err = ext4_ext_get_access(handle, inode, path + i);
1242                         if (err)
1243                                 goto cleanup;
1244                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1245                         err = ext4_ext_dirty(handle, inode, path + i);
1246                         if (err)
1247                                 goto cleanup;
1248                 }
1249 
1250                 i--;
1251         }
1252 
1253         /* insert new index */
1254         err = ext4_ext_insert_index(handle, inode, path + at,
1255                                     le32_to_cpu(border), newblock);
1256 
1257 cleanup:
1258         if (bh) {
1259                 if (buffer_locked(bh))
1260                         unlock_buffer(bh);
1261                 brelse(bh);
1262         }
1263 
1264         if (err) {
1265                 /* free all allocated blocks in error case */
1266                 for (i = 0; i < depth; i++) {
1267                         if (!ablocks[i])
1268                                 continue;
1269                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1270                                          EXT4_FREE_BLOCKS_METADATA);
1271                 }
1272         }
1273         kfree(ablocks);
1274 
1275         return err;
1276 }
1277 
1278 /*
1279  * ext4_ext_grow_indepth:
1280  * implements tree growing procedure:
1281  * - allocates new block
1282  * - moves top-level data (index block or leaf) into the new block
1283  * - initializes new top-level, creating index that points to the
1284  *   just created block
1285  */
1286 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1287                                  unsigned int flags)
1288 {
1289         struct ext4_extent_header *neh;
1290         struct buffer_head *bh;
1291         ext4_fsblk_t newblock, goal = 0;
1292         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1293         int err = 0;
1294         size_t ext_size = 0;
1295 
1296         /* Try to prepend new index to old one */
1297         if (ext_depth(inode))
1298                 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1299         if (goal > le32_to_cpu(es->s_first_data_block)) {
1300                 flags |= EXT4_MB_HINT_TRY_GOAL;
1301                 goal--;
1302         } else
1303                 goal = ext4_inode_to_goal_block(inode);
1304         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1305                                         NULL, &err);
1306         if (newblock == 0)
1307                 return err;
1308 
1309         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1310         if (unlikely(!bh))
1311                 return -ENOMEM;
1312         lock_buffer(bh);
1313 
1314         err = ext4_journal_get_create_access(handle, bh);
1315         if (err) {
1316                 unlock_buffer(bh);
1317                 goto out;
1318         }
1319 
1320         ext_size = sizeof(EXT4_I(inode)->i_data);
1321         /* move top-level index/leaf into new block */
1322         memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1323         /* zero out unused area in the extent block */
1324         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1325 
1326         /* set size of new block */
1327         neh = ext_block_hdr(bh);
1328         /* old root could have indexes or leaves
1329          * so calculate e_max right way */
1330         if (ext_depth(inode))
1331                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1332         else
1333                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1334         neh->eh_magic = EXT4_EXT_MAGIC;
1335         ext4_extent_block_csum_set(inode, neh);
1336         set_buffer_uptodate(bh);
1337         unlock_buffer(bh);
1338 
1339         err = ext4_handle_dirty_metadata(handle, inode, bh);
1340         if (err)
1341                 goto out;
1342 
1343         /* Update top-level index: num,max,pointer */
1344         neh = ext_inode_hdr(inode);
1345         neh->eh_entries = cpu_to_le16(1);
1346         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1347         if (neh->eh_depth == 0) {
1348                 /* Root extent block becomes index block */
1349                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1350                 EXT_FIRST_INDEX(neh)->ei_block =
1351                         EXT_FIRST_EXTENT(neh)->ee_block;
1352         }
1353         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1354                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1355                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1356                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1357 
1358         le16_add_cpu(&neh->eh_depth, 1);
1359         ext4_mark_inode_dirty(handle, inode);
1360 out:
1361         brelse(bh);
1362 
1363         return err;
1364 }
1365 
1366 /*
1367  * ext4_ext_create_new_leaf:
1368  * finds empty index and adds new leaf.
1369  * if no free index is found, then it requests in-depth growing.
1370  */
1371 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1372                                     unsigned int mb_flags,
1373                                     unsigned int gb_flags,
1374                                     struct ext4_ext_path **ppath,
1375                                     struct ext4_extent *newext)
1376 {
1377         struct ext4_ext_path *path = *ppath;
1378         struct ext4_ext_path *curp;
1379         int depth, i, err = 0;
1380 
1381 repeat:
1382         i = depth = ext_depth(inode);
1383 
1384         /* walk up to the tree and look for free index entry */
1385         curp = path + depth;
1386         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1387                 i--;
1388                 curp--;
1389         }
1390 
1391         /* we use already allocated block for index block,
1392          * so subsequent data blocks should be contiguous */
1393         if (EXT_HAS_FREE_INDEX(curp)) {
1394                 /* if we found index with free entry, then use that
1395                  * entry: create all needed subtree and add new leaf */
1396                 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1397                 if (err)
1398                         goto out;
1399 
1400                 /* refill path */
1401                 path = ext4_find_extent(inode,
1402                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1403                                     ppath, gb_flags);
1404                 if (IS_ERR(path))
1405                         err = PTR_ERR(path);
1406         } else {
1407                 /* tree is full, time to grow in depth */
1408                 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1409                 if (err)
1410                         goto out;
1411 
1412                 /* refill path */
1413                 path = ext4_find_extent(inode,
1414                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1415                                     ppath, gb_flags);
1416                 if (IS_ERR(path)) {
1417                         err = PTR_ERR(path);
1418                         goto out;
1419                 }
1420 
1421                 /*
1422                  * only first (depth 0 -> 1) produces free space;
1423                  * in all other cases we have to split the grown tree
1424                  */
1425                 depth = ext_depth(inode);
1426                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1427                         /* now we need to split */
1428                         goto repeat;
1429                 }
1430         }
1431 
1432 out:
1433         return err;
1434 }
1435 
1436 /*
1437  * search the closest allocated block to the left for *logical
1438  * and returns it at @logical + it's physical address at @phys
1439  * if *logical is the smallest allocated block, the function
1440  * returns 0 at @phys
1441  * return value contains 0 (success) or error code
1442  */
1443 static int ext4_ext_search_left(struct inode *inode,
1444                                 struct ext4_ext_path *path,
1445                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1446 {
1447         struct ext4_extent_idx *ix;
1448         struct ext4_extent *ex;
1449         int depth, ee_len;
1450 
1451         if (unlikely(path == NULL)) {
1452                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1453                 return -EFSCORRUPTED;
1454         }
1455         depth = path->p_depth;
1456         *phys = 0;
1457 
1458         if (depth == 0 && path->p_ext == NULL)
1459                 return 0;
1460 
1461         /* usually extent in the path covers blocks smaller
1462          * then *logical, but it can be that extent is the
1463          * first one in the file */
1464 
1465         ex = path[depth].p_ext;
1466         ee_len = ext4_ext_get_actual_len(ex);
1467         if (*logical < le32_to_cpu(ex->ee_block)) {
1468                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1469                         EXT4_ERROR_INODE(inode,
1470                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1471                                          *logical, le32_to_cpu(ex->ee_block));
1472                         return -EFSCORRUPTED;
1473                 }
1474                 while (--depth >= 0) {
1475                         ix = path[depth].p_idx;
1476                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1477                                 EXT4_ERROR_INODE(inode,
1478                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1479                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1480                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1481                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1482                                   depth);
1483                                 return -EFSCORRUPTED;
1484                         }
1485                 }
1486                 return 0;
1487         }
1488 
1489         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1490                 EXT4_ERROR_INODE(inode,
1491                                  "logical %d < ee_block %d + ee_len %d!",
1492                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1493                 return -EFSCORRUPTED;
1494         }
1495 
1496         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1497         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1498         return 0;
1499 }
1500 
1501 /*
1502  * search the closest allocated block to the right for *logical
1503  * and returns it at @logical + it's physical address at @phys
1504  * if *logical is the largest allocated block, the function
1505  * returns 0 at @phys
1506  * return value contains 0 (success) or error code
1507  */
1508 static int ext4_ext_search_right(struct inode *inode,
1509                                  struct ext4_ext_path *path,
1510                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1511                                  struct ext4_extent **ret_ex)
1512 {
1513         struct buffer_head *bh = NULL;
1514         struct ext4_extent_header *eh;
1515         struct ext4_extent_idx *ix;
1516         struct ext4_extent *ex;
1517         ext4_fsblk_t block;
1518         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1519         int ee_len;
1520 
1521         if (unlikely(path == NULL)) {
1522                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1523                 return -EFSCORRUPTED;
1524         }
1525         depth = path->p_depth;
1526         *phys = 0;
1527 
1528         if (depth == 0 && path->p_ext == NULL)
1529                 return 0;
1530 
1531         /* usually extent in the path covers blocks smaller
1532          * then *logical, but it can be that extent is the
1533          * first one in the file */
1534 
1535         ex = path[depth].p_ext;
1536         ee_len = ext4_ext_get_actual_len(ex);
1537         if (*logical < le32_to_cpu(ex->ee_block)) {
1538                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1539                         EXT4_ERROR_INODE(inode,
1540                                          "first_extent(path[%d].p_hdr) != ex",
1541                                          depth);
1542                         return -EFSCORRUPTED;
1543                 }
1544                 while (--depth >= 0) {
1545                         ix = path[depth].p_idx;
1546                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1547                                 EXT4_ERROR_INODE(inode,
1548                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1549                                                  *logical);
1550                                 return -EFSCORRUPTED;
1551                         }
1552                 }
1553                 goto found_extent;
1554         }
1555 
1556         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1557                 EXT4_ERROR_INODE(inode,
1558                                  "logical %d < ee_block %d + ee_len %d!",
1559                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1560                 return -EFSCORRUPTED;
1561         }
1562 
1563         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1564                 /* next allocated block in this leaf */
1565                 ex++;
1566                 goto found_extent;
1567         }
1568 
1569         /* go up and search for index to the right */
1570         while (--depth >= 0) {
1571                 ix = path[depth].p_idx;
1572                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1573                         goto got_index;
1574         }
1575 
1576         /* we've gone up to the root and found no index to the right */
1577         return 0;
1578 
1579 got_index:
1580         /* we've found index to the right, let's
1581          * follow it and find the closest allocated
1582          * block to the right */
1583         ix++;
1584         block = ext4_idx_pblock(ix);
1585         while (++depth < path->p_depth) {
1586                 /* subtract from p_depth to get proper eh_depth */
1587                 bh = read_extent_tree_block(inode, block,
1588                                             path->p_depth - depth, 0);
1589                 if (IS_ERR(bh))
1590                         return PTR_ERR(bh);
1591                 eh = ext_block_hdr(bh);
1592                 ix = EXT_FIRST_INDEX(eh);
1593                 block = ext4_idx_pblock(ix);
1594                 put_bh(bh);
1595         }
1596 
1597         bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1598         if (IS_ERR(bh))
1599                 return PTR_ERR(bh);
1600         eh = ext_block_hdr(bh);
1601         ex = EXT_FIRST_EXTENT(eh);
1602 found_extent:
1603         *logical = le32_to_cpu(ex->ee_block);
1604         *phys = ext4_ext_pblock(ex);
1605         *ret_ex = ex;
1606         if (bh)
1607                 put_bh(bh);
1608         return 0;
1609 }
1610 
1611 /*
1612  * ext4_ext_next_allocated_block:
1613  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1614  * NOTE: it considers block number from index entry as
1615  * allocated block. Thus, index entries have to be consistent
1616  * with leaves.
1617  */
1618 ext4_lblk_t
1619 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1620 {
1621         int depth;
1622 
1623         BUG_ON(path == NULL);
1624         depth = path->p_depth;
1625 
1626         if (depth == 0 && path->p_ext == NULL)
1627                 return EXT_MAX_BLOCKS;
1628 
1629         while (depth >= 0) {
1630                 if (depth == path->p_depth) {
1631                         /* leaf */
1632                         if (path[depth].p_ext &&
1633                                 path[depth].p_ext !=
1634                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1635                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1636                 } else {
1637                         /* index */
1638                         if (path[depth].p_idx !=
1639                                         EXT_LAST_INDEX(path[depth].p_hdr))
1640                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1641                 }
1642                 depth--;
1643         }
1644 
1645         return EXT_MAX_BLOCKS;
1646 }
1647 
1648 /*
1649  * ext4_ext_next_leaf_block:
1650  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1651  */
1652 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1653 {
1654         int depth;
1655 
1656         BUG_ON(path == NULL);
1657         depth = path->p_depth;
1658 
1659         /* zero-tree has no leaf blocks at all */
1660         if (depth == 0)
1661                 return EXT_MAX_BLOCKS;
1662 
1663         /* go to index block */
1664         depth--;
1665 
1666         while (depth >= 0) {
1667                 if (path[depth].p_idx !=
1668                                 EXT_LAST_INDEX(path[depth].p_hdr))
1669                         return (ext4_lblk_t)
1670                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1671                 depth--;
1672         }
1673 
1674         return EXT_MAX_BLOCKS;
1675 }
1676 
1677 /*
1678  * ext4_ext_correct_indexes:
1679  * if leaf gets modified and modified extent is first in the leaf,
1680  * then we have to correct all indexes above.
1681  * TODO: do we need to correct tree in all cases?
1682  */
1683 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1684                                 struct ext4_ext_path *path)
1685 {
1686         struct ext4_extent_header *eh;
1687         int depth = ext_depth(inode);
1688         struct ext4_extent *ex;
1689         __le32 border;
1690         int k, err = 0;
1691 
1692         eh = path[depth].p_hdr;
1693         ex = path[depth].p_ext;
1694 
1695         if (unlikely(ex == NULL || eh == NULL)) {
1696                 EXT4_ERROR_INODE(inode,
1697                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1698                 return -EFSCORRUPTED;
1699         }
1700 
1701         if (depth == 0) {
1702                 /* there is no tree at all */
1703                 return 0;
1704         }
1705 
1706         if (ex != EXT_FIRST_EXTENT(eh)) {
1707                 /* we correct tree if first leaf got modified only */
1708                 return 0;
1709         }
1710 
1711         /*
1712          * TODO: we need correction if border is smaller than current one
1713          */
1714         k = depth - 1;
1715         border = path[depth].p_ext->ee_block;
1716         err = ext4_ext_get_access(handle, inode, path + k);
1717         if (err)
1718                 return err;
1719         path[k].p_idx->ei_block = border;
1720         err = ext4_ext_dirty(handle, inode, path + k);
1721         if (err)
1722                 return err;
1723 
1724         while (k--) {
1725                 /* change all left-side indexes */
1726                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1727                         break;
1728                 err = ext4_ext_get_access(handle, inode, path + k);
1729                 if (err)
1730                         break;
1731                 path[k].p_idx->ei_block = border;
1732                 err = ext4_ext_dirty(handle, inode, path + k);
1733                 if (err)
1734                         break;
1735         }
1736 
1737         return err;
1738 }
1739 
1740 int
1741 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1742                                 struct ext4_extent *ex2)
1743 {
1744         unsigned short ext1_ee_len, ext2_ee_len;
1745 
1746         if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1747                 return 0;
1748 
1749         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1750         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1751 
1752         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1753                         le32_to_cpu(ex2->ee_block))
1754                 return 0;
1755 
1756         /*
1757          * To allow future support for preallocated extents to be added
1758          * as an RO_COMPAT feature, refuse to merge to extents if
1759          * this can result in the top bit of ee_len being set.
1760          */
1761         if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1762                 return 0;
1763         /*
1764          * The check for IO to unwritten extent is somewhat racy as we
1765          * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1766          * dropping i_data_sem. But reserved blocks should save us in that
1767          * case.
1768          */
1769         if (ext4_ext_is_unwritten(ex1) &&
1770             (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1771              atomic_read(&EXT4_I(inode)->i_unwritten) ||
1772              (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1773                 return 0;
1774 #ifdef AGGRESSIVE_TEST
1775         if (ext1_ee_len >= 4)
1776                 return 0;
1777 #endif
1778 
1779         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1780                 return 1;
1781         return 0;
1782 }
1783 
1784 /*
1785  * This function tries to merge the "ex" extent to the next extent in the tree.
1786  * It always tries to merge towards right. If you want to merge towards
1787  * left, pass "ex - 1" as argument instead of "ex".
1788  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1789  * 1 if they got merged.
1790  */
1791 static int ext4_ext_try_to_merge_right(struct inode *inode,
1792                                  struct ext4_ext_path *path,
1793                                  struct ext4_extent *ex)
1794 {
1795         struct ext4_extent_header *eh;
1796         unsigned int depth, len;
1797         int merge_done = 0, unwritten;
1798 
1799         depth = ext_depth(inode);
1800         BUG_ON(path[depth].p_hdr == NULL);
1801         eh = path[depth].p_hdr;
1802 
1803         while (ex < EXT_LAST_EXTENT(eh)) {
1804                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1805                         break;
1806                 /* merge with next extent! */
1807                 unwritten = ext4_ext_is_unwritten(ex);
1808                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1809                                 + ext4_ext_get_actual_len(ex + 1));
1810                 if (unwritten)
1811                         ext4_ext_mark_unwritten(ex);
1812 
1813                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1814                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1815                                 * sizeof(struct ext4_extent);
1816                         memmove(ex + 1, ex + 2, len);
1817                 }
1818                 le16_add_cpu(&eh->eh_entries, -1);
1819                 merge_done = 1;
1820                 WARN_ON(eh->eh_entries == 0);
1821                 if (!eh->eh_entries)
1822                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1823         }
1824 
1825         return merge_done;
1826 }
1827 
1828 /*
1829  * This function does a very simple check to see if we can collapse
1830  * an extent tree with a single extent tree leaf block into the inode.
1831  */
1832 static void ext4_ext_try_to_merge_up(handle_t *handle,
1833                                      struct inode *inode,
1834                                      struct ext4_ext_path *path)
1835 {
1836         size_t s;
1837         unsigned max_root = ext4_ext_space_root(inode, 0);
1838         ext4_fsblk_t blk;
1839 
1840         if ((path[0].p_depth != 1) ||
1841             (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1842             (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1843                 return;
1844 
1845         /*
1846          * We need to modify the block allocation bitmap and the block
1847          * group descriptor to release the extent tree block.  If we
1848          * can't get the journal credits, give up.
1849          */
1850         if (ext4_journal_extend(handle, 2))
1851                 return;
1852 
1853         /*
1854          * Copy the extent data up to the inode
1855          */
1856         blk = ext4_idx_pblock(path[0].p_idx);
1857         s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1858                 sizeof(struct ext4_extent_idx);
1859         s += sizeof(struct ext4_extent_header);
1860 
1861         path[1].p_maxdepth = path[0].p_maxdepth;
1862         memcpy(path[0].p_hdr, path[1].p_hdr, s);
1863         path[0].p_depth = 0;
1864         path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1865                 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1866         path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1867 
1868         brelse(path[1].p_bh);
1869         ext4_free_blocks(handle, inode, NULL, blk, 1,
1870                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1871 }
1872 
1873 /*
1874  * This function tries to merge the @ex extent to neighbours in the tree.
1875  * return 1 if merge left else 0.
1876  */
1877 static void ext4_ext_try_to_merge(handle_t *handle,
1878                                   struct inode *inode,
1879                                   struct ext4_ext_path *path,
1880                                   struct ext4_extent *ex) {
1881         struct ext4_extent_header *eh;
1882         unsigned int depth;
1883         int merge_done = 0;
1884 
1885         depth = ext_depth(inode);
1886         BUG_ON(path[depth].p_hdr == NULL);
1887         eh = path[depth].p_hdr;
1888 
1889         if (ex > EXT_FIRST_EXTENT(eh))
1890                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1891 
1892         if (!merge_done)
1893                 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1894 
1895         ext4_ext_try_to_merge_up(handle, inode, path);
1896 }
1897 
1898 /*
1899  * check if a portion of the "newext" extent overlaps with an
1900  * existing extent.
1901  *
1902  * If there is an overlap discovered, it updates the length of the newext
1903  * such that there will be no overlap, and then returns 1.
1904  * If there is no overlap found, it returns 0.
1905  */
1906 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1907                                            struct inode *inode,
1908                                            struct ext4_extent *newext,
1909                                            struct ext4_ext_path *path)
1910 {
1911         ext4_lblk_t b1, b2;
1912         unsigned int depth, len1;
1913         unsigned int ret = 0;
1914 
1915         b1 = le32_to_cpu(newext->ee_block);
1916         len1 = ext4_ext_get_actual_len(newext);
1917         depth = ext_depth(inode);
1918         if (!path[depth].p_ext)
1919                 goto out;
1920         b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1921 
1922         /*
1923          * get the next allocated block if the extent in the path
1924          * is before the requested block(s)
1925          */
1926         if (b2 < b1) {
1927                 b2 = ext4_ext_next_allocated_block(path);
1928                 if (b2 == EXT_MAX_BLOCKS)
1929                         goto out;
1930                 b2 = EXT4_LBLK_CMASK(sbi, b2);
1931         }
1932 
1933         /* check for wrap through zero on extent logical start block*/
1934         if (b1 + len1 < b1) {
1935                 len1 = EXT_MAX_BLOCKS - b1;
1936                 newext->ee_len = cpu_to_le16(len1);
1937                 ret = 1;
1938         }
1939 
1940         /* check for overlap */
1941         if (b1 + len1 > b2) {
1942                 newext->ee_len = cpu_to_le16(b2 - b1);
1943                 ret = 1;
1944         }
1945 out:
1946         return ret;
1947 }
1948 
1949 /*
1950  * ext4_ext_insert_extent:
1951  * tries to merge requsted extent into the existing extent or
1952  * inserts requested extent as new one into the tree,
1953  * creating new leaf in the no-space case.
1954  */
1955 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1956                                 struct ext4_ext_path **ppath,
1957                                 struct ext4_extent *newext, int gb_flags)
1958 {
1959         struct ext4_ext_path *path = *ppath;
1960         struct ext4_extent_header *eh;
1961         struct ext4_extent *ex, *fex;
1962         struct ext4_extent *nearex; /* nearest extent */
1963         struct ext4_ext_path *npath = NULL;
1964         int depth, len, err;
1965         ext4_lblk_t next;
1966         int mb_flags = 0, unwritten;
1967 
1968         if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1969                 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1970         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1971                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1972                 return -EFSCORRUPTED;
1973         }
1974         depth = ext_depth(inode);
1975         ex = path[depth].p_ext;
1976         eh = path[depth].p_hdr;
1977         if (unlikely(path[depth].p_hdr == NULL)) {
1978                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1979                 return -EFSCORRUPTED;
1980         }
1981 
1982         /* try to insert block into found extent and return */
1983         if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1984 
1985                 /*
1986                  * Try to see whether we should rather test the extent on
1987                  * right from ex, or from the left of ex. This is because
1988                  * ext4_find_extent() can return either extent on the
1989                  * left, or on the right from the searched position. This
1990                  * will make merging more effective.
1991                  */
1992                 if (ex < EXT_LAST_EXTENT(eh) &&
1993                     (le32_to_cpu(ex->ee_block) +
1994                     ext4_ext_get_actual_len(ex) <
1995                     le32_to_cpu(newext->ee_block))) {
1996                         ex += 1;
1997                         goto prepend;
1998                 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1999                            (le32_to_cpu(newext->ee_block) +
2000                            ext4_ext_get_actual_len(newext) <
2001                            le32_to_cpu(ex->ee_block)))
2002                         ex -= 1;
2003 
2004                 /* Try to append newex to the ex */
2005                 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2006                         ext_debug("append [%d]%d block to %u:[%d]%d"
2007                                   "(from %llu)\n",
2008                                   ext4_ext_is_unwritten(newext),
2009                                   ext4_ext_get_actual_len(newext),
2010                                   le32_to_cpu(ex->ee_block),
2011                                   ext4_ext_is_unwritten(ex),
2012                                   ext4_ext_get_actual_len(ex),
2013                                   ext4_ext_pblock(ex));
2014                         err = ext4_ext_get_access(handle, inode,
2015                                                   path + depth);
2016                         if (err)
2017                                 return err;
2018                         unwritten = ext4_ext_is_unwritten(ex);
2019                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2020                                         + ext4_ext_get_actual_len(newext));
2021                         if (unwritten)
2022                                 ext4_ext_mark_unwritten(ex);
2023                         eh = path[depth].p_hdr;
2024                         nearex = ex;
2025                         goto merge;
2026                 }
2027 
2028 prepend:
2029                 /* Try to prepend newex to the ex */
2030                 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2031                         ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2032                                   "(from %llu)\n",
2033                                   le32_to_cpu(newext->ee_block),
2034                                   ext4_ext_is_unwritten(newext),
2035                                   ext4_ext_get_actual_len(newext),
2036                                   le32_to_cpu(ex->ee_block),
2037                                   ext4_ext_is_unwritten(ex),
2038                                   ext4_ext_get_actual_len(ex),
2039                                   ext4_ext_pblock(ex));
2040                         err = ext4_ext_get_access(handle, inode,
2041                                                   path + depth);
2042                         if (err)
2043                                 return err;
2044 
2045                         unwritten = ext4_ext_is_unwritten(ex);
2046                         ex->ee_block = newext->ee_block;
2047                         ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2048                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2049                                         + ext4_ext_get_actual_len(newext));
2050                         if (unwritten)
2051                                 ext4_ext_mark_unwritten(ex);
2052                         eh = path[depth].p_hdr;
2053                         nearex = ex;
2054                         goto merge;
2055                 }
2056         }
2057 
2058         depth = ext_depth(inode);
2059         eh = path[depth].p_hdr;
2060         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2061                 goto has_space;
2062 
2063         /* probably next leaf has space for us? */
2064         fex = EXT_LAST_EXTENT(eh);
2065         next = EXT_MAX_BLOCKS;
2066         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2067                 next = ext4_ext_next_leaf_block(path);
2068         if (next != EXT_MAX_BLOCKS) {
2069                 ext_debug("next leaf block - %u\n", next);
2070                 BUG_ON(npath != NULL);
2071                 npath = ext4_find_extent(inode, next, NULL, 0);
2072                 if (IS_ERR(npath))
2073                         return PTR_ERR(npath);
2074                 BUG_ON(npath->p_depth != path->p_depth);
2075                 eh = npath[depth].p_hdr;
2076                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2077                         ext_debug("next leaf isn't full(%d)\n",
2078                                   le16_to_cpu(eh->eh_entries));
2079                         path = npath;
2080                         goto has_space;
2081                 }
2082                 ext_debug("next leaf has no free space(%d,%d)\n",
2083                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2084         }
2085 
2086         /*
2087          * There is no free space in the found leaf.
2088          * We're gonna add a new leaf in the tree.
2089          */
2090         if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2091                 mb_flags |= EXT4_MB_USE_RESERVED;
2092         err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2093                                        ppath, newext);
2094         if (err)
2095                 goto cleanup;
2096         depth = ext_depth(inode);
2097         eh = path[depth].p_hdr;
2098 
2099 has_space:
2100         nearex = path[depth].p_ext;
2101 
2102         err = ext4_ext_get_access(handle, inode, path + depth);
2103         if (err)
2104                 goto cleanup;
2105 
2106         if (!nearex) {
2107                 /* there is no extent in this leaf, create first one */
2108                 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2109                                 le32_to_cpu(newext->ee_block),
2110                                 ext4_ext_pblock(newext),
2111                                 ext4_ext_is_unwritten(newext),
2112                                 ext4_ext_get_actual_len(newext));
2113                 nearex = EXT_FIRST_EXTENT(eh);
2114         } else {
2115                 if (le32_to_cpu(newext->ee_block)
2116                            > le32_to_cpu(nearex->ee_block)) {
2117                         /* Insert after */
2118                         ext_debug("insert %u:%llu:[%d]%d before: "
2119                                         "nearest %p\n",
2120                                         le32_to_cpu(newext->ee_block),
2121                                         ext4_ext_pblock(newext),
2122                                         ext4_ext_is_unwritten(newext),
2123                                         ext4_ext_get_actual_len(newext),
2124                                         nearex);
2125                         nearex++;
2126                 } else {
2127                         /* Insert before */
2128                         BUG_ON(newext->ee_block == nearex->ee_block);
2129                         ext_debug("insert %u:%llu:[%d]%d after: "
2130                                         "nearest %p\n",
2131                                         le32_to_cpu(newext->ee_block),
2132                                         ext4_ext_pblock(newext),
2133                                         ext4_ext_is_unwritten(newext),
2134                                         ext4_ext_get_actual_len(newext),
2135                                         nearex);
2136                 }
2137                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2138                 if (len > 0) {
2139                         ext_debug("insert %u:%llu:[%d]%d: "
2140                                         "move %d extents from 0x%p to 0x%p\n",
2141                                         le32_to_cpu(newext->ee_block),
2142                                         ext4_ext_pblock(newext),
2143                                         ext4_ext_is_unwritten(newext),
2144                                         ext4_ext_get_actual_len(newext),
2145                                         len, nearex, nearex + 1);
2146                         memmove(nearex + 1, nearex,
2147                                 len * sizeof(struct ext4_extent));
2148                 }
2149         }
2150 
2151         le16_add_cpu(&eh->eh_entries, 1);
2152         path[depth].p_ext = nearex;
2153         nearex->ee_block = newext->ee_block;
2154         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2155         nearex->ee_len = newext->ee_len;
2156 
2157 merge:
2158         /* try to merge extents */
2159         if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2160                 ext4_ext_try_to_merge(handle, inode, path, nearex);
2161 
2162 
2163         /* time to correct all indexes above */
2164         err = ext4_ext_correct_indexes(handle, inode, path);
2165         if (err)
2166                 goto cleanup;
2167 
2168         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2169 
2170 cleanup:
2171         ext4_ext_drop_refs(npath);
2172         kfree(npath);
2173         return err;
2174 }
2175 
2176 static int ext4_fill_fiemap_extents(struct inode *inode,
2177                                     ext4_lblk_t block, ext4_lblk_t num,
2178                                     struct fiemap_extent_info *fieinfo)
2179 {
2180         struct ext4_ext_path *path = NULL;
2181         struct ext4_extent *ex;
2182         struct extent_status es;
2183         ext4_lblk_t next, next_del, start = 0, end = 0;
2184         ext4_lblk_t last = block + num;
2185         int exists, depth = 0, err = 0;
2186         unsigned int flags = 0;
2187         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2188 
2189         while (block < last && block != EXT_MAX_BLOCKS) {
2190                 num = last - block;
2191                 /* find extent for this block */
2192                 down_read(&EXT4_I(inode)->i_data_sem);
2193 
2194                 path = ext4_find_extent(inode, block, &path, 0);
2195                 if (IS_ERR(path)) {
2196                         up_read(&EXT4_I(inode)->i_data_sem);
2197                         err = PTR_ERR(path);
2198                         path = NULL;
2199                         break;
2200                 }
2201 
2202                 depth = ext_depth(inode);
2203                 if (unlikely(path[depth].p_hdr == NULL)) {
2204                         up_read(&EXT4_I(inode)->i_data_sem);
2205                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2206                         err = -EFSCORRUPTED;
2207                         break;
2208                 }
2209                 ex = path[depth].p_ext;
2210                 next = ext4_ext_next_allocated_block(path);
2211 
2212                 flags = 0;
2213                 exists = 0;
2214                 if (!ex) {
2215                         /* there is no extent yet, so try to allocate
2216                          * all requested space */
2217                         start = block;
2218                         end = block + num;
2219                 } else if (le32_to_cpu(ex->ee_block) > block) {
2220                         /* need to allocate space before found extent */
2221                         start = block;
2222                         end = le32_to_cpu(ex->ee_block);
2223                         if (block + num < end)
2224                                 end = block + num;
2225                 } else if (block >= le32_to_cpu(ex->ee_block)
2226                                         + ext4_ext_get_actual_len(ex)) {
2227                         /* need to allocate space after found extent */
2228                         start = block;
2229                         end = block + num;
2230                         if (end >= next)
2231                                 end = next;
2232                 } else if (block >= le32_to_cpu(ex->ee_block)) {
2233                         /*
2234                          * some part of requested space is covered
2235                          * by found extent
2236                          */
2237                         start = block;
2238                         end = le32_to_cpu(ex->ee_block)
2239                                 + ext4_ext_get_actual_len(ex);
2240                         if (block + num < end)
2241                                 end = block + num;
2242                         exists = 1;
2243                 } else {
2244                         BUG();
2245                 }
2246                 BUG_ON(end <= start);
2247 
2248                 if (!exists) {
2249                         es.es_lblk = start;
2250                         es.es_len = end - start;
2251                         es.es_pblk = 0;
2252                 } else {
2253                         es.es_lblk = le32_to_cpu(ex->ee_block);
2254                         es.es_len = ext4_ext_get_actual_len(ex);
2255                         es.es_pblk = ext4_ext_pblock(ex);
2256                         if (ext4_ext_is_unwritten(ex))
2257                                 flags |= FIEMAP_EXTENT_UNWRITTEN;
2258                 }
2259 
2260                 /*
2261                  * Find delayed extent and update es accordingly. We call
2262                  * it even in !exists case to find out whether es is the
2263                  * last existing extent or not.
2264                  */
2265                 next_del = ext4_find_delayed_extent(inode, &es);
2266                 if (!exists && next_del) {
2267                         exists = 1;
2268                         flags |= (FIEMAP_EXTENT_DELALLOC |
2269                                   FIEMAP_EXTENT_UNKNOWN);
2270                 }
2271                 up_read(&EXT4_I(inode)->i_data_sem);
2272 
2273                 if (unlikely(es.es_len == 0)) {
2274                         EXT4_ERROR_INODE(inode, "es.es_len == 0");
2275                         err = -EFSCORRUPTED;
2276                         break;
2277                 }
2278 
2279                 /*
2280                  * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2281                  * we need to check next == EXT_MAX_BLOCKS because it is
2282                  * possible that an extent is with unwritten and delayed
2283                  * status due to when an extent is delayed allocated and
2284                  * is allocated by fallocate status tree will track both of
2285                  * them in a extent.
2286                  *
2287                  * So we could return a unwritten and delayed extent, and
2288                  * its block is equal to 'next'.
2289                  */
2290                 if (next == next_del && next == EXT_MAX_BLOCKS) {
2291                         flags |= FIEMAP_EXTENT_LAST;
2292                         if (unlikely(next_del != EXT_MAX_BLOCKS ||
2293                                      next != EXT_MAX_BLOCKS)) {
2294                                 EXT4_ERROR_INODE(inode,
2295                                                  "next extent == %u, next "
2296                                                  "delalloc extent = %u",
2297                                                  next, next_del);
2298                                 err = -EFSCORRUPTED;
2299                                 break;
2300                         }
2301                 }
2302 
2303                 if (exists) {
2304                         err = fiemap_fill_next_extent(fieinfo,
2305                                 (__u64)es.es_lblk << blksize_bits,
2306                                 (__u64)es.es_pblk << blksize_bits,
2307                                 (__u64)es.es_len << blksize_bits,
2308                                 flags);
2309                         if (err < 0)
2310                                 break;
2311                         if (err == 1) {
2312                                 err = 0;
2313                                 break;
2314                         }
2315                 }
2316 
2317                 block = es.es_lblk + es.es_len;
2318         }
2319 
2320         ext4_ext_drop_refs(path);
2321         kfree(path);
2322         return err;
2323 }
2324 
2325 static int ext4_fill_es_cache_info(struct inode *inode,
2326                                    ext4_lblk_t block, ext4_lblk_t num,
2327                                    struct fiemap_extent_info *fieinfo)
2328 {
2329         ext4_lblk_t next, end = block + num - 1;
2330         struct extent_status es;
2331         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2332         unsigned int flags;
2333         int err;
2334 
2335         while (block <= end) {
2336                 next = 0;
2337                 flags = 0;
2338                 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2339                         break;
2340                 if (ext4_es_is_unwritten(&es))
2341                         flags |= FIEMAP_EXTENT_UNWRITTEN;
2342                 if (ext4_es_is_delayed(&es))
2343                         flags |= (FIEMAP_EXTENT_DELALLOC |
2344                                   FIEMAP_EXTENT_UNKNOWN);
2345                 if (ext4_es_is_hole(&es))
2346                         flags |= EXT4_FIEMAP_EXTENT_HOLE;
2347                 if (next == 0)
2348                         flags |= FIEMAP_EXTENT_LAST;
2349                 if (flags & (FIEMAP_EXTENT_DELALLOC|
2350                              EXT4_FIEMAP_EXTENT_HOLE))
2351                         es.es_pblk = 0;
2352                 else
2353                         es.es_pblk = ext4_es_pblock(&es);
2354                 err = fiemap_fill_next_extent(fieinfo,
2355                                 (__u64)es.es_lblk << blksize_bits,
2356                                 (__u64)es.es_pblk << blksize_bits,
2357                                 (__u64)es.es_len << blksize_bits,
2358                                 flags);
2359                 if (next == 0)
2360                         break;
2361                 block = next;
2362                 if (err < 0)
2363                         return err;
2364                 if (err == 1)
2365                         return 0;
2366         }
2367         return 0;
2368 }
2369 
2370 
2371 /*
2372  * ext4_ext_determine_hole - determine hole around given block
2373  * @inode:      inode we lookup in
2374  * @path:       path in extent tree to @lblk
2375  * @lblk:       pointer to logical block around which we want to determine hole
2376  *
2377  * Determine hole length (and start if easily possible) around given logical
2378  * block. We don't try too hard to find the beginning of the hole but @path
2379  * actually points to extent before @lblk, we provide it.
2380  *
2381  * The function returns the length of a hole starting at @lblk. We update @lblk
2382  * to the beginning of the hole if we managed to find it.
2383  */
2384 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2385                                            struct ext4_ext_path *path,
2386                                            ext4_lblk_t *lblk)
2387 {
2388         int depth = ext_depth(inode);
2389         struct ext4_extent *ex;
2390         ext4_lblk_t len;
2391 
2392         ex = path[depth].p_ext;
2393         if (ex == NULL) {
2394                 /* there is no extent yet, so gap is [0;-] */
2395                 *lblk = 0;
2396                 len = EXT_MAX_BLOCKS;
2397         } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2398                 len = le32_to_cpu(ex->ee_block) - *lblk;
2399         } else if (*lblk >= le32_to_cpu(ex->ee_block)
2400                         + ext4_ext_get_actual_len(ex)) {
2401                 ext4_lblk_t next;
2402 
2403                 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2404                 next = ext4_ext_next_allocated_block(path);
2405                 BUG_ON(next == *lblk);
2406                 len = next - *lblk;
2407         } else {
2408                 BUG();
2409         }
2410         return len;
2411 }
2412 
2413 /*
2414  * ext4_ext_put_gap_in_cache:
2415  * calculate boundaries of the gap that the requested block fits into
2416  * and cache this gap
2417  */
2418 static void
2419 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2420                           ext4_lblk_t hole_len)
2421 {
2422         struct extent_status es;
2423 
2424         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2425                                   hole_start + hole_len - 1, &es);
2426         if (es.es_len) {
2427                 /* There's delayed extent containing lblock? */
2428                 if (es.es_lblk <= hole_start)
2429                         return;
2430                 hole_len = min(es.es_lblk - hole_start, hole_len);
2431         }
2432         ext_debug(" -> %u:%u\n", hole_start, hole_len);
2433         ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2434                               EXTENT_STATUS_HOLE);
2435 }
2436 
2437 /*
2438  * ext4_ext_rm_idx:
2439  * removes index from the index block.
2440  */
2441 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2442                         struct ext4_ext_path *path, int depth)
2443 {
2444         int err;
2445         ext4_fsblk_t leaf;
2446 
2447         /* free index block */
2448         depth--;
2449         path = path + depth;
2450         leaf = ext4_idx_pblock(path->p_idx);
2451         if (unlikely(path->p_hdr->eh_entries == 0)) {
2452                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2453                 return -EFSCORRUPTED;
2454         }
2455         err = ext4_ext_get_access(handle, inode, path);
2456         if (err)
2457                 return err;
2458 
2459         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2460                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2461                 len *= sizeof(struct ext4_extent_idx);
2462                 memmove(path->p_idx, path->p_idx + 1, len);
2463         }
2464 
2465         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2466         err = ext4_ext_dirty(handle, inode, path);
2467         if (err)
2468                 return err;
2469         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2470         trace_ext4_ext_rm_idx(inode, leaf);
2471 
2472         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2473                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2474 
2475         while (--depth >= 0) {
2476                 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2477                         break;
2478                 path--;
2479                 err = ext4_ext_get_access(handle, inode, path);
2480                 if (err)
2481                         break;
2482                 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2483                 err = ext4_ext_dirty(handle, inode, path);
2484                 if (err)
2485                         break;
2486         }
2487         return err;
2488 }
2489 
2490 /*
2491  * ext4_ext_calc_credits_for_single_extent:
2492  * This routine returns max. credits that needed to insert an extent
2493  * to the extent tree.
2494  * When pass the actual path, the caller should calculate credits
2495  * under i_data_sem.
2496  */
2497 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2498                                                 struct ext4_ext_path *path)
2499 {
2500         if (path) {
2501                 int depth = ext_depth(inode);
2502                 int ret = 0;
2503 
2504                 /* probably there is space in leaf? */
2505                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2506                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2507 
2508                         /*
2509                          *  There are some space in the leaf tree, no
2510                          *  need to account for leaf block credit
2511                          *
2512                          *  bitmaps and block group descriptor blocks
2513                          *  and other metadata blocks still need to be
2514                          *  accounted.
2515                          */
2516                         /* 1 bitmap, 1 block group descriptor */
2517                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2518                         return ret;
2519                 }
2520         }
2521 
2522         return ext4_chunk_trans_blocks(inode, nrblocks);
2523 }
2524 
2525 /*
2526  * How many index/leaf blocks need to change/allocate to add @extents extents?
2527  *
2528  * If we add a single extent, then in the worse case, each tree level
2529  * index/leaf need to be changed in case of the tree split.
2530  *
2531  * If more extents are inserted, they could cause the whole tree split more
2532  * than once, but this is really rare.
2533  */
2534 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2535 {
2536         int index;
2537         int depth;
2538 
2539         /* If we are converting the inline data, only one is needed here. */
2540         if (ext4_has_inline_data(inode))
2541                 return 1;
2542 
2543         depth = ext_depth(inode);
2544 
2545         if (extents <= 1)
2546                 index = depth * 2;
2547         else
2548                 index = depth * 3;
2549 
2550         return index;
2551 }
2552 
2553 static inline int get_default_free_blocks_flags(struct inode *inode)
2554 {
2555         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2556             ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2557                 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2558         else if (ext4_should_journal_data(inode))
2559                 return EXT4_FREE_BLOCKS_FORGET;
2560         return 0;
2561 }
2562 
2563 /*
2564  * ext4_rereserve_cluster - increment the reserved cluster count when
2565  *                          freeing a cluster with a pending reservation
2566  *
2567  * @inode - file containing the cluster
2568  * @lblk - logical block in cluster to be reserved
2569  *
2570  * Increments the reserved cluster count and adjusts quota in a bigalloc
2571  * file system when freeing a partial cluster containing at least one
2572  * delayed and unwritten block.  A partial cluster meeting that
2573  * requirement will have a pending reservation.  If so, the
2574  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2575  * defer reserved and allocated space accounting to a subsequent call
2576  * to this function.
2577  */
2578 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2579 {
2580         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2581         struct ext4_inode_info *ei = EXT4_I(inode);
2582 
2583         dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2584 
2585         spin_lock(&ei->i_block_reservation_lock);
2586         ei->i_reserved_data_blocks++;
2587         percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2588         spin_unlock(&ei->i_block_reservation_lock);
2589 
2590         percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2591         ext4_remove_pending(inode, lblk);
2592 }
2593 
2594 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2595                               struct ext4_extent *ex,
2596                               struct partial_cluster *partial,
2597                               ext4_lblk_t from, ext4_lblk_t to)
2598 {
2599         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2600         unsigned short ee_len = ext4_ext_get_actual_len(ex);
2601         ext4_fsblk_t last_pblk, pblk;
2602         ext4_lblk_t num;
2603         int flags;
2604 
2605         /* only extent tail removal is allowed */
2606         if (from < le32_to_cpu(ex->ee_block) ||
2607             to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2608                 ext4_error(sbi->s_sb,
2609                            "strange request: removal(2) %u-%u from %u:%u",
2610                            from, to, le32_to_cpu(ex->ee_block), ee_len);
2611                 return 0;
2612         }
2613 
2614 #ifdef EXTENTS_STATS
2615         spin_lock(&sbi->s_ext_stats_lock);
2616         sbi->s_ext_blocks += ee_len;
2617         sbi->s_ext_extents++;
2618         if (ee_len < sbi->s_ext_min)
2619                 sbi->s_ext_min = ee_len;
2620         if (ee_len > sbi->s_ext_max)
2621                 sbi->s_ext_max = ee_len;
2622         if (ext_depth(inode) > sbi->s_depth_max)
2623                 sbi->s_depth_max = ext_depth(inode);
2624         spin_unlock(&sbi->s_ext_stats_lock);
2625 #endif
2626 
2627         trace_ext4_remove_blocks(inode, ex, from, to, partial);
2628 
2629         /*
2630          * if we have a partial cluster, and it's different from the
2631          * cluster of the last block in the extent, we free it
2632          */
2633         last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2634 
2635         if (partial->state != initial &&
2636             partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2637                 if (partial->state == tofree) {
2638                         flags = get_default_free_blocks_flags(inode);
2639                         if (ext4_is_pending(inode, partial->lblk))
2640                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2641                         ext4_free_blocks(handle, inode, NULL,
2642                                          EXT4_C2B(sbi, partial->pclu),
2643                                          sbi->s_cluster_ratio, flags);
2644                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2645                                 ext4_rereserve_cluster(inode, partial->lblk);
2646                 }
2647                 partial->state = initial;
2648         }
2649 
2650         num = le32_to_cpu(ex->ee_block) + ee_len - from;
2651         pblk = ext4_ext_pblock(ex) + ee_len - num;
2652 
2653         /*
2654          * We free the partial cluster at the end of the extent (if any),
2655          * unless the cluster is used by another extent (partial_cluster
2656          * state is nofree).  If a partial cluster exists here, it must be
2657          * shared with the last block in the extent.
2658          */
2659         flags = get_default_free_blocks_flags(inode);
2660 
2661         /* partial, left end cluster aligned, right end unaligned */
2662         if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2663             (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2664             (partial->state != nofree)) {
2665                 if (ext4_is_pending(inode, to))
2666                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2667                 ext4_free_blocks(handle, inode, NULL,
2668                                  EXT4_PBLK_CMASK(sbi, last_pblk),
2669                                  sbi->s_cluster_ratio, flags);
2670                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2671                         ext4_rereserve_cluster(inode, to);
2672                 partial->state = initial;
2673                 flags = get_default_free_blocks_flags(inode);
2674         }
2675 
2676         flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2677 
2678         /*
2679          * For bigalloc file systems, we never free a partial cluster
2680          * at the beginning of the extent.  Instead, we check to see if we
2681          * need to free it on a subsequent call to ext4_remove_blocks,
2682          * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2683          */
2684         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2685         ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2686 
2687         /* reset the partial cluster if we've freed past it */
2688         if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2689                 partial->state = initial;
2690 
2691         /*
2692          * If we've freed the entire extent but the beginning is not left
2693          * cluster aligned and is not marked as ineligible for freeing we
2694          * record the partial cluster at the beginning of the extent.  It
2695          * wasn't freed by the preceding ext4_free_blocks() call, and we
2696          * need to look farther to the left to determine if it's to be freed
2697          * (not shared with another extent). Else, reset the partial
2698          * cluster - we're either  done freeing or the beginning of the
2699          * extent is left cluster aligned.
2700          */
2701         if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2702                 if (partial->state == initial) {
2703                         partial->pclu = EXT4_B2C(sbi, pblk);
2704                         partial->lblk = from;
2705                         partial->state = tofree;
2706                 }
2707         } else {
2708                 partial->state = initial;
2709         }
2710 
2711         return 0;
2712 }
2713 
2714 /*
2715  * ext4_ext_rm_leaf() Removes the extents associated with the
2716  * blocks appearing between "start" and "end".  Both "start"
2717  * and "end" must appear in the same extent or EIO is returned.
2718  *
2719  * @handle: The journal handle
2720  * @inode:  The files inode
2721  * @path:   The path to the leaf
2722  * @partial_cluster: The cluster which we'll have to free if all extents
2723  *                   has been released from it.  However, if this value is
2724  *                   negative, it's a cluster just to the right of the
2725  *                   punched region and it must not be freed.
2726  * @start:  The first block to remove
2727  * @end:   The last block to remove
2728  */
2729 static int
2730 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2731                  struct ext4_ext_path *path,
2732                  struct partial_cluster *partial,
2733                  ext4_lblk_t start, ext4_lblk_t end)
2734 {
2735         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2736         int err = 0, correct_index = 0;
2737         int depth = ext_depth(inode), credits;
2738         struct ext4_extent_header *eh;
2739         ext4_lblk_t a, b;
2740         unsigned num;
2741         ext4_lblk_t ex_ee_block;
2742         unsigned short ex_ee_len;
2743         unsigned unwritten = 0;
2744         struct ext4_extent *ex;
2745         ext4_fsblk_t pblk;
2746 
2747         /* the header must be checked already in ext4_ext_remove_space() */
2748         ext_debug("truncate since %u in leaf to %u\n", start, end);
2749         if (!path[depth].p_hdr)
2750                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2751         eh = path[depth].p_hdr;
2752         if (unlikely(path[depth].p_hdr == NULL)) {
2753                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2754                 return -EFSCORRUPTED;
2755         }
2756         /* find where to start removing */
2757         ex = path[depth].p_ext;
2758         if (!ex)
2759                 ex = EXT_LAST_EXTENT(eh);
2760 
2761         ex_ee_block = le32_to_cpu(ex->ee_block);
2762         ex_ee_len = ext4_ext_get_actual_len(ex);
2763 
2764         trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2765 
2766         while (ex >= EXT_FIRST_EXTENT(eh) &&
2767                         ex_ee_block + ex_ee_len > start) {
2768 
2769                 if (ext4_ext_is_unwritten(ex))
2770                         unwritten = 1;
2771                 else
2772                         unwritten = 0;
2773 
2774                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2775                           unwritten, ex_ee_len);
2776                 path[depth].p_ext = ex;
2777 
2778                 a = ex_ee_block > start ? ex_ee_block : start;
2779                 b = ex_ee_block+ex_ee_len - 1 < end ?
2780                         ex_ee_block+ex_ee_len - 1 : end;
2781 
2782                 ext_debug("  border %u:%u\n", a, b);
2783 
2784                 /* If this extent is beyond the end of the hole, skip it */
2785                 if (end < ex_ee_block) {
2786                         /*
2787                          * We're going to skip this extent and move to another,
2788                          * so note that its first cluster is in use to avoid
2789                          * freeing it when removing blocks.  Eventually, the
2790                          * right edge of the truncated/punched region will
2791                          * be just to the left.
2792                          */
2793                         if (sbi->s_cluster_ratio > 1) {
2794                                 pblk = ext4_ext_pblock(ex);
2795                                 partial->pclu = EXT4_B2C(sbi, pblk);
2796                                 partial->state = nofree;
2797                         }
2798                         ex--;
2799                         ex_ee_block = le32_to_cpu(ex->ee_block);
2800                         ex_ee_len = ext4_ext_get_actual_len(ex);
2801                         continue;
2802                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2803                         EXT4_ERROR_INODE(inode,
2804                                          "can not handle truncate %u:%u "
2805                                          "on extent %u:%u",
2806                                          start, end, ex_ee_block,
2807                                          ex_ee_block + ex_ee_len - 1);
2808                         err = -EFSCORRUPTED;
2809                         goto out;
2810                 } else if (a != ex_ee_block) {
2811                         /* remove tail of the extent */
2812                         num = a - ex_ee_block;
2813                 } else {
2814                         /* remove whole extent: excellent! */
2815                         num = 0;
2816                 }
2817                 /*
2818                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2819                  * descriptor) for each block group; assume two block
2820                  * groups plus ex_ee_len/blocks_per_block_group for
2821                  * the worst case
2822                  */
2823                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2824                 if (ex == EXT_FIRST_EXTENT(eh)) {
2825                         correct_index = 1;
2826                         credits += (ext_depth(inode)) + 1;
2827                 }
2828                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2829 
2830                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2831                 if (err)
2832                         goto out;
2833 
2834                 err = ext4_ext_get_access(handle, inode, path + depth);
2835                 if (err)
2836                         goto out;
2837 
2838                 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2839                 if (err)
2840                         goto out;
2841 
2842                 if (num == 0)
2843                         /* this extent is removed; mark slot entirely unused */
2844                         ext4_ext_store_pblock(ex, 0);
2845 
2846                 ex->ee_len = cpu_to_le16(num);
2847                 /*
2848                  * Do not mark unwritten if all the blocks in the
2849                  * extent have been removed.
2850                  */
2851                 if (unwritten && num)
2852                         ext4_ext_mark_unwritten(ex);
2853                 /*
2854                  * If the extent was completely released,
2855                  * we need to remove it from the leaf
2856                  */
2857                 if (num == 0) {
2858                         if (end != EXT_MAX_BLOCKS - 1) {
2859                                 /*
2860                                  * For hole punching, we need to scoot all the
2861                                  * extents up when an extent is removed so that
2862                                  * we dont have blank extents in the middle
2863                                  */
2864                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2865                                         sizeof(struct ext4_extent));
2866 
2867                                 /* Now get rid of the one at the end */
2868                                 memset(EXT_LAST_EXTENT(eh), 0,
2869                                         sizeof(struct ext4_extent));
2870                         }
2871                         le16_add_cpu(&eh->eh_entries, -1);
2872                 }
2873 
2874                 err = ext4_ext_dirty(handle, inode, path + depth);
2875                 if (err)
2876                         goto out;
2877 
2878                 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2879                                 ext4_ext_pblock(ex));
2880                 ex--;
2881                 ex_ee_block = le32_to_cpu(ex->ee_block);
2882                 ex_ee_len = ext4_ext_get_actual_len(ex);
2883         }
2884 
2885         if (correct_index && eh->eh_entries)
2886                 err = ext4_ext_correct_indexes(handle, inode, path);
2887 
2888         /*
2889          * If there's a partial cluster and at least one extent remains in
2890          * the leaf, free the partial cluster if it isn't shared with the
2891          * current extent.  If it is shared with the current extent
2892          * we reset the partial cluster because we've reached the start of the
2893          * truncated/punched region and we're done removing blocks.
2894          */
2895         if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2896                 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2897                 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2898                         int flags = get_default_free_blocks_flags(inode);
2899 
2900                         if (ext4_is_pending(inode, partial->lblk))
2901                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2902                         ext4_free_blocks(handle, inode, NULL,
2903                                          EXT4_C2B(sbi, partial->pclu),
2904                                          sbi->s_cluster_ratio, flags);
2905                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2906                                 ext4_rereserve_cluster(inode, partial->lblk);
2907                 }
2908                 partial->state = initial;
2909         }
2910 
2911         /* if this leaf is free, then we should
2912          * remove it from index block above */
2913         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2914                 err = ext4_ext_rm_idx(handle, inode, path, depth);
2915 
2916 out:
2917         return err;
2918 }
2919 
2920 /*
2921  * ext4_ext_more_to_rm:
2922  * returns 1 if current index has to be freed (even partial)
2923  */
2924 static int
2925 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2926 {
2927         BUG_ON(path->p_idx == NULL);
2928 
2929         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2930                 return 0;
2931 
2932         /*
2933          * if truncate on deeper level happened, it wasn't partial,
2934          * so we have to consider current index for truncation
2935          */
2936         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2937                 return 0;
2938         return 1;
2939 }
2940 
2941 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2942                           ext4_lblk_t end)
2943 {
2944         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2945         int depth = ext_depth(inode);
2946         struct ext4_ext_path *path = NULL;
2947         struct partial_cluster partial;
2948         handle_t *handle;
2949         int i = 0, err = 0;
2950 
2951         partial.pclu = 0;
2952         partial.lblk = 0;
2953         partial.state = initial;
2954 
2955         ext_debug("truncate since %u to %u\n", start, end);
2956 
2957         /* probably first extent we're gonna free will be last in block */
2958         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2959         if (IS_ERR(handle))
2960                 return PTR_ERR(handle);
2961 
2962 again:
2963         trace_ext4_ext_remove_space(inode, start, end, depth);
2964 
2965         /*
2966          * Check if we are removing extents inside the extent tree. If that
2967          * is the case, we are going to punch a hole inside the extent tree
2968          * so we have to check whether we need to split the extent covering
2969          * the last block to remove so we can easily remove the part of it
2970          * in ext4_ext_rm_leaf().
2971          */
2972         if (end < EXT_MAX_BLOCKS - 1) {
2973                 struct ext4_extent *ex;
2974                 ext4_lblk_t ee_block, ex_end, lblk;
2975                 ext4_fsblk_t pblk;
2976 
2977                 /* find extent for or closest extent to this block */
2978                 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2979                 if (IS_ERR(path)) {
2980                         ext4_journal_stop(handle);
2981                         return PTR_ERR(path);
2982                 }
2983                 depth = ext_depth(inode);
2984                 /* Leaf not may not exist only if inode has no blocks at all */
2985                 ex = path[depth].p_ext;
2986                 if (!ex) {
2987                         if (depth) {
2988                                 EXT4_ERROR_INODE(inode,
2989                                                  "path[%d].p_hdr == NULL",
2990                                                  depth);
2991                                 err = -EFSCORRUPTED;
2992                         }
2993                         goto out;
2994                 }
2995 
2996                 ee_block = le32_to_cpu(ex->ee_block);
2997                 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2998 
2999                 /*
3000                  * See if the last block is inside the extent, if so split
3001                  * the extent at 'end' block so we can easily remove the
3002                  * tail of the first part of the split extent in
3003                  * ext4_ext_rm_leaf().
3004                  */
3005                 if (end >= ee_block && end < ex_end) {
3006 
3007                         /*
3008                          * If we're going to split the extent, note that
3009                          * the cluster containing the block after 'end' is
3010                          * in use to avoid freeing it when removing blocks.
3011                          */
3012                         if (sbi->s_cluster_ratio > 1) {
3013                                 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
3014                                 partial.pclu = EXT4_B2C(sbi, pblk);
3015                                 partial.state = nofree;
3016                         }
3017 
3018                         /*
3019                          * Split the extent in two so that 'end' is the last
3020                          * block in the first new extent. Also we should not
3021                          * fail removing space due to ENOSPC so try to use
3022                          * reserved block if that happens.
3023                          */
3024                         err = ext4_force_split_extent_at(handle, inode, &path,
3025                                                          end + 1, 1);
3026                         if (err < 0)
3027                                 goto out;
3028 
3029                 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3030                            partial.state == initial) {
3031                         /*
3032                          * If we're punching, there's an extent to the right.
3033                          * If the partial cluster hasn't been set, set it to
3034                          * that extent's first cluster and its state to nofree
3035                          * so it won't be freed should it contain blocks to be
3036                          * removed. If it's already set (tofree/nofree), we're
3037                          * retrying and keep the original partial cluster info
3038                          * so a cluster marked tofree as a result of earlier
3039                          * extent removal is not lost.
3040                          */
3041                         lblk = ex_end + 1;
3042                         err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3043                                                     &ex);
3044                         if (err)
3045                                 goto out;
3046                         if (pblk) {
3047                                 partial.pclu = EXT4_B2C(sbi, pblk);
3048                                 partial.state = nofree;
3049                         }
3050                 }
3051         }
3052         /*
3053          * We start scanning from right side, freeing all the blocks
3054          * after i_size and walking into the tree depth-wise.
3055          */
3056         depth = ext_depth(inode);
3057         if (path) {
3058                 int k = i = depth;
3059                 while (--k > 0)
3060                         path[k].p_block =
3061                                 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3062         } else {
3063                 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3064                                GFP_NOFS);
3065                 if (path == NULL) {
3066                         ext4_journal_stop(handle);
3067                         return -ENOMEM;
3068                 }
3069                 path[0].p_maxdepth = path[0].p_depth = depth;
3070                 path[0].p_hdr = ext_inode_hdr(inode);
3071                 i = 0;
3072 
3073                 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3074                         err = -EFSCORRUPTED;
3075                         goto out;
3076                 }
3077         }
3078         err = 0;
3079 
3080         while (i >= 0 && err == 0) {
3081                 if (i == depth) {
3082                         /* this is leaf block */
3083                         err = ext4_ext_rm_leaf(handle, inode, path,
3084                                                &partial, start, end);
3085                         /* root level has p_bh == NULL, brelse() eats this */
3086                         brelse(path[i].p_bh);
3087                         path[i].p_bh = NULL;
3088                         i--;
3089                         continue;
3090                 }
3091 
3092                 /* this is index block */
3093                 if (!path[i].p_hdr) {
3094                         ext_debug("initialize header\n");
3095                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3096                 }
3097 
3098                 if (!path[i].p_idx) {
3099                         /* this level hasn't been touched yet */
3100                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3101                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3102                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
3103                                   path[i].p_hdr,
3104                                   le16_to_cpu(path[i].p_hdr->eh_entries));
3105                 } else {
3106                         /* we were already here, see at next index */
3107                         path[i].p_idx--;
3108                 }
3109 
3110                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3111                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
3112                                 path[i].p_idx);
3113                 if (ext4_ext_more_to_rm(path + i)) {
3114                         struct buffer_head *bh;
3115                         /* go to the next level */
3116                         ext_debug("move to level %d (block %llu)\n",
3117                                   i + 1, ext4_idx_pblock(path[i].p_idx));
3118                         memset(path + i + 1, 0, sizeof(*path));
3119                         bh = read_extent_tree_block(inode,
3120                                 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3121                                 EXT4_EX_NOCACHE);
3122                         if (IS_ERR(bh)) {
3123                                 /* should we reset i_size? */
3124                                 err = PTR_ERR(bh);
3125                                 break;
3126                         }
3127                         /* Yield here to deal with large extent trees.
3128                          * Should be a no-op if we did IO above. */
3129                         cond_resched();
3130                         if (WARN_ON(i + 1 > depth)) {
3131                                 err = -EFSCORRUPTED;
3132                                 break;
3133                         }
3134                         path[i + 1].p_bh = bh;
3135 
3136                         /* save actual number of indexes since this
3137                          * number is changed at the next iteration */
3138                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3139                         i++;
3140                 } else {
3141                         /* we finished processing this index, go up */
3142                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3143                                 /* index is empty, remove it;
3144                                  * handle must be already prepared by the
3145                                  * truncatei_leaf() */
3146                                 err = ext4_ext_rm_idx(handle, inode, path, i);
3147                         }
3148                         /* root level has p_bh == NULL, brelse() eats this */
3149                         brelse(path[i].p_bh);
3150                         path[i].p_bh = NULL;
3151                         i--;
3152                         ext_debug("return to level %d\n", i);
3153                 }
3154         }
3155 
3156         trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3157                                          path->p_hdr->eh_entries);
3158 
3159         /*
3160          * if there's a partial cluster and we have removed the first extent
3161          * in the file, then we also free the partial cluster, if any
3162          */
3163         if (partial.state == tofree && err == 0) {
3164                 int flags = get_default_free_blocks_flags(inode);
3165 
3166                 if (ext4_is_pending(inode, partial.lblk))
3167                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3168                 ext4_free_blocks(handle, inode, NULL,
3169                                  EXT4_C2B(sbi, partial.pclu),
3170                                  sbi->s_cluster_ratio, flags);
3171                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3172                         ext4_rereserve_cluster(inode, partial.lblk);
3173                 partial.state = initial;
3174         }
3175 
3176         /* TODO: flexible tree reduction should be here */
3177         if (path->p_hdr->eh_entries == 0) {
3178                 /*
3179                  * truncate to zero freed all the tree,
3180                  * so we need to correct eh_depth
3181                  */
3182                 err = ext4_ext_get_access(handle, inode, path);
3183                 if (err == 0) {
3184                         ext_inode_hdr(inode)->eh_depth = 0;
3185                         ext_inode_hdr(inode)->eh_max =
3186                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
3187                         err = ext4_ext_dirty(handle, inode, path);
3188                 }
3189         }
3190 out:
3191         ext4_ext_drop_refs(path);
3192         kfree(path);
3193         path = NULL;
3194         if (err == -EAGAIN)
3195                 goto again;
3196         ext4_journal_stop(handle);
3197 
3198         return err;
3199 }
3200 
3201 /*
3202  * called at mount time
3203  */
3204 void ext4_ext_init(struct super_block *sb)
3205 {
3206         /*
3207          * possible initialization would be here
3208          */
3209 
3210         if (ext4_has_feature_extents(sb)) {
3211 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3212                 printk(KERN_INFO "EXT4-fs: file extents enabled"
3213 #ifdef AGGRESSIVE_TEST
3214                        ", aggressive tests"
3215 #endif
3216 #ifdef CHECK_BINSEARCH
3217                        ", check binsearch"
3218 #endif
3219 #ifdef EXTENTS_STATS
3220                        ", stats"
3221 #endif
3222                        "\n");
3223 #endif
3224 #ifdef EXTENTS_STATS
3225                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3226                 EXT4_SB(sb)->s_ext_min = 1 << 30;
3227                 EXT4_SB(sb)->s_ext_max = 0;
3228 #endif
3229         }
3230 }
3231 
3232 /*
3233  * called at umount time
3234  */
3235 void ext4_ext_release(struct super_block *sb)
3236 {
3237         if (!ext4_has_feature_extents(sb))
3238                 return;
3239 
3240 #ifdef EXTENTS_STATS
3241         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3242                 struct ext4_sb_info *sbi = EXT4_SB(sb);
3243                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3244                         sbi->s_ext_blocks, sbi->s_ext_extents,
3245                         sbi->s_ext_blocks / sbi->s_ext_extents);
3246                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3247                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3248         }
3249 #endif
3250 }
3251 
3252 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3253 {
3254         ext4_lblk_t  ee_block;
3255         ext4_fsblk_t ee_pblock;
3256         unsigned int ee_len;
3257 
3258         ee_block  = le32_to_cpu(ex->ee_block);
3259         ee_len    = ext4_ext_get_actual_len(ex);
3260         ee_pblock = ext4_ext_pblock(ex);
3261 
3262         if (ee_len == 0)
3263                 return 0;
3264 
3265         return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3266                                      EXTENT_STATUS_WRITTEN);
3267 }
3268 
3269 /* FIXME!! we need to try to merge to left or right after zero-out  */
3270 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3271 {
3272         ext4_fsblk_t ee_pblock;
3273         unsigned int ee_len;
3274 
3275         ee_len    = ext4_ext_get_actual_len(ex);
3276         ee_pblock = ext4_ext_pblock(ex);
3277         return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3278                                   ee_len);
3279 }
3280 
3281 /*
3282  * ext4_split_extent_at() splits an extent at given block.
3283  *
3284  * @handle: the journal handle
3285  * @inode: the file inode
3286  * @path: the path to the extent
3287  * @split: the logical block where the extent is splitted.
3288  * @split_flags: indicates if the extent could be zeroout if split fails, and
3289  *               the states(init or unwritten) of new extents.
3290  * @flags: flags used to insert new extent to extent tree.
3291  *
3292  *
3293  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3294  * of which are deterimined by split_flag.
3295  *
3296  * There are two cases:
3297  *  a> the extent are splitted into two extent.
3298  *  b> split is not needed, and just mark the extent.
3299  *
3300  * return 0 on success.
3301  */
3302 static int ext4_split_extent_at(handle_t *handle,
3303                              struct inode *inode,
3304                              struct ext4_ext_path **ppath,
3305                              ext4_lblk_t split,
3306                              int split_flag,
3307                              int flags)
3308 {
3309         struct ext4_ext_path *path = *ppath;
3310         ext4_fsblk_t newblock;
3311         ext4_lblk_t ee_block;
3312         struct ext4_extent *ex, newex, orig_ex, zero_ex;
3313         struct ext4_extent *ex2 = NULL;
3314         unsigned int ee_len, depth;
3315         int err = 0;
3316 
3317         BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3318                (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3319 
3320         ext_debug("ext4_split_extents_at: inode %lu, logical"
3321                 "block %llu\n", inode->i_ino, (unsigned long long)split);
3322 
3323         ext4_ext_show_leaf(inode, path);
3324 
3325         depth = ext_depth(inode);
3326         ex = path[depth].p_ext;
3327         ee_block = le32_to_cpu(ex->ee_block);
3328         ee_len = ext4_ext_get_actual_len(ex);
3329         newblock = split - ee_block + ext4_ext_pblock(ex);
3330 
3331         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3332         BUG_ON(!ext4_ext_is_unwritten(ex) &&
3333                split_flag & (EXT4_EXT_MAY_ZEROOUT |
3334                              EXT4_EXT_MARK_UNWRIT1 |
3335                              EXT4_EXT_MARK_UNWRIT2));
3336 
3337         err = ext4_ext_get_access(handle, inode, path + depth);
3338         if (err)
3339                 goto out;
3340 
3341         if (split == ee_block) {
3342                 /*
3343                  * case b: block @split is the block that the extent begins with
3344                  * then we just change the state of the extent, and splitting
3345                  * is not needed.
3346                  */
3347                 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3348                         ext4_ext_mark_unwritten(ex);
3349                 else
3350                         ext4_ext_mark_initialized(ex);
3351 
3352                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3353                         ext4_ext_try_to_merge(handle, inode, path, ex);
3354 
3355                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3356                 goto out;
3357         }
3358 
3359         /* case a */
3360         memcpy(&orig_ex, ex, sizeof(orig_ex));
3361         ex->ee_len = cpu_to_le16(split - ee_block);
3362         if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3363                 ext4_ext_mark_unwritten(ex);
3364 
3365         /*
3366          * path may lead to new leaf, not to original leaf any more
3367          * after ext4_ext_insert_extent() returns,
3368          */
3369         err = ext4_ext_dirty(handle, inode, path + depth);
3370         if (err)
3371                 goto fix_extent_len;
3372 
3373         ex2 = &newex;
3374         ex2->ee_block = cpu_to_le32(split);
3375         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3376         ext4_ext_store_pblock(ex2, newblock);
3377         if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3378                 ext4_ext_mark_unwritten(ex2);
3379 
3380         err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3381         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3382                 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3383                         if (split_flag & EXT4_EXT_DATA_VALID1) {
3384                                 err = ext4_ext_zeroout(inode, ex2);
3385                                 zero_ex.ee_block = ex2->ee_block;
3386                                 zero_ex.ee_len = cpu_to_le16(
3387                                                 ext4_ext_get_actual_len(ex2));
3388                                 ext4_ext_store_pblock(&zero_ex,
3389                                                       ext4_ext_pblock(ex2));
3390                         } else {
3391                                 err = ext4_ext_zeroout(inode, ex);
3392                                 zero_ex.ee_block = ex->ee_block;
3393                                 zero_ex.ee_len = cpu_to_le16(
3394                                                 ext4_ext_get_actual_len(ex));
3395                                 ext4_ext_store_pblock(&zero_ex,
3396                                                       ext4_ext_pblock(ex));
3397                         }
3398                 } else {
3399                         err = ext4_ext_zeroout(inode, &orig_ex);
3400                         zero_ex.ee_block = orig_ex.ee_block;
3401                         zero_ex.ee_len = cpu_to_le16(
3402                                                 ext4_ext_get_actual_len(&orig_ex));
3403                         ext4_ext_store_pblock(&zero_ex,
3404                                               ext4_ext_pblock(&orig_ex));
3405                 }
3406 
3407                 if (err)
3408                         goto fix_extent_len;
3409                 /* update the extent length and mark as initialized */
3410                 ex->ee_len = cpu_to_le16(ee_len);
3411                 ext4_ext_try_to_merge(handle, inode, path, ex);
3412                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3413                 if (err)
3414                         goto fix_extent_len;
3415 
3416                 /* update extent status tree */
3417                 err = ext4_zeroout_es(inode, &zero_ex);
3418 
3419                 goto out;
3420         } else if (err)
3421                 goto fix_extent_len;
3422 
3423 out:
3424         ext4_ext_show_leaf(inode, path);
3425         return err;
3426 
3427 fix_extent_len:
3428         ex->ee_len = orig_ex.ee_len;
3429         ext4_ext_dirty(handle, inode, path + path->p_depth);
3430         return err;
3431 }
3432 
3433 /*
3434  * ext4_split_extents() splits an extent and mark extent which is covered
3435  * by @map as split_flags indicates
3436  *
3437  * It may result in splitting the extent into multiple extents (up to three)
3438  * There are three possibilities:
3439  *   a> There is no split required
3440  *   b> Splits in two extents: Split is happening at either end of the extent
3441  *   c> Splits in three extents: Somone is splitting in middle of the extent
3442  *
3443  */
3444 static int ext4_split_extent(handle_t *handle,
3445                               struct inode *inode,
3446                               struct ext4_ext_path **ppath,
3447                               struct ext4_map_blocks *map,
3448                               int split_flag,
3449                               int flags)
3450 {
3451         struct ext4_ext_path *path = *ppath;
3452         ext4_lblk_t ee_block;
3453         struct ext4_extent *ex;
3454         unsigned int ee_len, depth;
3455         int err = 0;
3456         int unwritten;
3457         int split_flag1, flags1;
3458         int allocated = map->m_len;
3459 
3460         depth = ext_depth(inode);
3461         ex = path[depth].p_ext;
3462         ee_block = le32_to_cpu(ex->ee_block);
3463         ee_len = ext4_ext_get_actual_len(ex);
3464         unwritten = ext4_ext_is_unwritten(ex);
3465 
3466         if (map->m_lblk + map->m_len < ee_block + ee_len) {
3467                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3468                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3469                 if (unwritten)
3470                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3471                                        EXT4_EXT_MARK_UNWRIT2;
3472                 if (split_flag & EXT4_EXT_DATA_VALID2)
3473                         split_flag1 |= EXT4_EXT_DATA_VALID1;
3474                 err = ext4_split_extent_at(handle, inode, ppath,
3475                                 map->m_lblk + map->m_len, split_flag1, flags1);
3476                 if (err)
3477                         goto out;
3478         } else {
3479                 allocated = ee_len - (map->m_lblk - ee_block);
3480         }
3481         /*
3482          * Update path is required because previous ext4_split_extent_at() may
3483          * result in split of original leaf or extent zeroout.
3484          */
3485         path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3486         if (IS_ERR(path))
3487                 return PTR_ERR(path);
3488         depth = ext_depth(inode);
3489         ex = path[depth].p_ext;
3490         if (!ex) {
3491                 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3492                                  (unsigned long) map->m_lblk);
3493                 return -EFSCORRUPTED;
3494         }
3495         unwritten = ext4_ext_is_unwritten(ex);
3496         split_flag1 = 0;
3497 
3498         if (map->m_lblk >= ee_block) {
3499                 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3500                 if (unwritten) {
3501                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3502                         split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3503                                                      EXT4_EXT_MARK_UNWRIT2);
3504                 }
3505                 err = ext4_split_extent_at(handle, inode, ppath,
3506                                 map->m_lblk, split_flag1, flags);
3507                 if (err)
3508                         goto out;
3509         }
3510 
3511         ext4_ext_show_leaf(inode, path);
3512 out:
3513         return err ? err : allocated;
3514 }
3515 
3516 /*
3517  * This function is called by ext4_ext_map_blocks() if someone tries to write
3518  * to an unwritten extent. It may result in splitting the unwritten
3519  * extent into multiple extents (up to three - one initialized and two
3520  * unwritten).
3521  * There are three possibilities:
3522  *   a> There is no split required: Entire extent should be initialized
3523  *   b> Splits in two extents: Write is happening at either end of the extent
3524  *   c> Splits in three extents: Somone is writing in middle of the extent
3525  *
3526  * Pre-conditions:
3527  *  - The extent pointed to by 'path' is unwritten.
3528  *  - The extent pointed to by 'path' contains a superset
3529  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3530  *
3531  * Post-conditions on success:
3532  *  - the returned value is the number of blocks beyond map->l_lblk
3533  *    that are allocated and initialized.
3534  *    It is guaranteed to be >= map->m_len.
3535  */
3536 static int ext4_ext_convert_to_initialized(handle_t *handle,
3537                                            struct inode *inode,
3538                                            struct ext4_map_blocks *map,
3539                                            struct ext4_ext_path **ppath,
3540                                            int flags)
3541 {
3542         struct ext4_ext_path *path = *ppath;
3543         struct ext4_sb_info *sbi;
3544         struct ext4_extent_header *eh;
3545         struct ext4_map_blocks split_map;
3546         struct ext4_extent zero_ex1, zero_ex2;
3547         struct ext4_extent *ex, *abut_ex;
3548         ext4_lblk_t ee_block, eof_block;
3549         unsigned int ee_len, depth, map_len = map->m_len;
3550         int allocated = 0, max_zeroout = 0;
3551         int err = 0;
3552         int split_flag = EXT4_EXT_DATA_VALID2;
3553 
3554         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3555                 "block %llu, max_blocks %u\n", inode->i_ino,
3556                 (unsigned long long)map->m_lblk, map_len);
3557 
3558         sbi = EXT4_SB(inode->i_sb);
3559         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3560                         >> inode->i_sb->s_blocksize_bits;
3561         if (eof_block < map->m_lblk + map_len)
3562                 eof_block = map->m_lblk + map_len;
3563 
3564         depth = ext_depth(inode);
3565         eh = path[depth].p_hdr;
3566         ex = path[depth].p_ext;
3567         ee_block = le32_to_cpu(ex->ee_block);
3568         ee_len = ext4_ext_get_actual_len(ex);
3569         zero_ex1.ee_len = 0;
3570         zero_ex2.ee_len = 0;
3571 
3572         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3573 
3574         /* Pre-conditions */
3575         BUG_ON(!ext4_ext_is_unwritten(ex));
3576         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3577 
3578         /*
3579          * Attempt to transfer newly initialized blocks from the currently
3580          * unwritten extent to its neighbor. This is much cheaper
3581          * than an insertion followed by a merge as those involve costly
3582          * memmove() calls. Transferring to the left is the common case in
3583          * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3584          * followed by append writes.
3585          *
3586          * Limitations of the current logic:
3587          *  - L1: we do not deal with writes covering the whole extent.
3588          *    This would require removing the extent if the transfer
3589          *    is possible.
3590          *  - L2: we only attempt to merge with an extent stored in the
3591          *    same extent tree node.
3592          */
3593         if ((map->m_lblk == ee_block) &&
3594                 /* See if we can merge left */
3595                 (map_len < ee_len) &&           /*L1*/
3596                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3597                 ext4_lblk_t prev_lblk;
3598                 ext4_fsblk_t prev_pblk, ee_pblk;
3599                 unsigned int prev_len;
3600 
3601                 abut_ex = ex - 1;
3602                 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3603                 prev_len = ext4_ext_get_actual_len(abut_ex);
3604                 prev_pblk = ext4_ext_pblock(abut_ex);
3605                 ee_pblk = ext4_ext_pblock(ex);
3606 
3607                 /*
3608                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3609                  * upon those conditions:
3610                  * - C1: abut_ex is initialized,
3611                  * - C2: abut_ex is logically abutting ex,
3612                  * - C3: abut_ex is physically abutting ex,
3613                  * - C4: abut_ex can receive the additional blocks without
3614                  *   overflowing the (initialized) length limit.
3615                  */
3616                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3617                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3618                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3619                         (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3620                         err = ext4_ext_get_access(handle, inode, path + depth);
3621                         if (err)
3622                                 goto out;
3623 
3624                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3625                                 map, ex, abut_ex);
3626 
3627                         /* Shift the start of ex by 'map_len' blocks */
3628                         ex->ee_block = cpu_to_le32(ee_block + map_len);
3629                         ext4_ext_store_pblock(ex, ee_pblk + map_len);
3630                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3631                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3632 
3633                         /* Extend abut_ex by 'map_len' blocks */
3634                         abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3635 
3636                         /* Result: number of initialized blocks past m_lblk */
3637                         allocated = map_len;
3638                 }
3639         } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3640                    (map_len < ee_len) &&        /*L1*/
3641                    ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3642                 /* See if we can merge right */
3643                 ext4_lblk_t next_lblk;
3644                 ext4_fsblk_t next_pblk, ee_pblk;
3645                 unsigned int next_len;
3646 
3647                 abut_ex = ex + 1;
3648                 next_lblk = le32_to_cpu(abut_ex->ee_block);
3649                 next_len = ext4_ext_get_actual_len(abut_ex);
3650                 next_pblk = ext4_ext_pblock(abut_ex);
3651                 ee_pblk = ext4_ext_pblock(ex);
3652 
3653                 /*
3654                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3655                  * upon those conditions:
3656                  * - C1: abut_ex is initialized,
3657                  * - C2: abut_ex is logically abutting ex,
3658                  * - C3: abut_ex is physically abutting ex,
3659                  * - C4: abut_ex can receive the additional blocks without
3660                  *   overflowing the (initialized) length limit.
3661                  */
3662                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3663                     ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3664                     ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3665                     (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3666                         err = ext4_ext_get_access(handle, inode, path + depth);
3667                         if (err)
3668                                 goto out;
3669 
3670                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3671                                 map, ex, abut_ex);
3672 
3673                         /* Shift the start of abut_ex by 'map_len' blocks */
3674                         abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3675                         ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3676                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3677                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3678 
3679                         /* Extend abut_ex by 'map_len' blocks */
3680                         abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3681 
3682                         /* Result: number of initialized blocks past m_lblk */
3683                         allocated = map_len;
3684                 }
3685         }
3686         if (allocated) {
3687                 /* Mark the block containing both extents as dirty */
3688                 ext4_ext_dirty(handle, inode, path + depth);
3689 
3690                 /* Update path to point to the right extent */
3691                 path[depth].p_ext = abut_ex;
3692                 goto out;
3693         } else
3694                 allocated = ee_len - (map->m_lblk - ee_block);
3695 
3696         WARN_ON(map->m_lblk < ee_block);
3697         /*
3698          * It is safe to convert extent to initialized via explicit
3699          * zeroout only if extent is fully inside i_size or new_size.
3700          */
3701         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3702 
3703         if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3704                 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3705                         (inode->i_sb->s_blocksize_bits - 10);
3706 
3707         if (IS_ENCRYPTED(inode))
3708                 max_zeroout = 0;
3709 
3710         /*
3711          * five cases:
3712          * 1. split the extent into three extents.
3713          * 2. split the extent into two extents, zeroout the head of the first
3714          *    extent.
3715          * 3. split the extent into two extents, zeroout the tail of the second
3716          *    extent.
3717          * 4. split the extent into two extents with out zeroout.
3718          * 5. no splitting needed, just possibly zeroout the head and / or the
3719          *    tail of the extent.
3720          */
3721         split_map.m_lblk = map->m_lblk;
3722         split_map.m_len = map->m_len;
3723 
3724         if (max_zeroout && (allocated > split_map.m_len)) {
3725                 if (allocated <= max_zeroout) {
3726                         /* case 3 or 5 */
3727                         zero_ex1.ee_block =
3728                                  cpu_to_le32(split_map.m_lblk +
3729                                              split_map.m_len);
3730                         zero_ex1.ee_len =
3731                                 cpu_to_le16(allocated - split_map.m_len);
3732                         ext4_ext_store_pblock(&zero_ex1,
3733                                 ext4_ext_pblock(ex) + split_map.m_lblk +
3734                                 split_map.m_len - ee_block);
3735                         err = ext4_ext_zeroout(inode, &zero_ex1);
3736                         if (err)
3737                                 goto out;
3738                         split_map.m_len = allocated;
3739                 }
3740                 if (split_map.m_lblk - ee_block + split_map.m_len <
3741                                                                 max_zeroout) {
3742                         /* case 2 or 5 */
3743                         if (split_map.m_lblk != ee_block) {
3744                                 zero_ex2.ee_block = ex->ee_block;
3745                                 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3746                                                         ee_block);
3747                                 ext4_ext_store_pblock(&zero_ex2,
3748                                                       ext4_ext_pblock(ex));
3749                                 err = ext4_ext_zeroout(inode, &zero_ex2);
3750                                 if (err)
3751                                         goto out;
3752                         }
3753 
3754                         split_map.m_len += split_map.m_lblk - ee_block;
3755                         split_map.m_lblk = ee_block;
3756                         allocated = map->m_len;
3757                 }
3758         }
3759 
3760         err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3761                                 flags);
3762         if (err > 0)
3763                 err = 0;
3764 out:
3765         /* If we have gotten a failure, don't zero out status tree */
3766         if (!err) {
3767                 err = ext4_zeroout_es(inode, &zero_ex1);
3768                 if (!err)
3769                         err = ext4_zeroout_es(inode, &zero_ex2);
3770         }
3771         return err ? err : allocated;
3772 }
3773 
3774 /*
3775  * This function is called by ext4_ext_map_blocks() from
3776  * ext4_get_blocks_dio_write() when DIO to write
3777  * to an unwritten extent.
3778  *
3779  * Writing to an unwritten extent may result in splitting the unwritten
3780  * extent into multiple initialized/unwritten extents (up to three)
3781  * There are three possibilities:
3782  *   a> There is no split required: Entire extent should be unwritten
3783  *   b> Splits in two extents: Write is happening at either end of the extent
3784  *   c> Splits in three extents: Somone is writing in middle of the extent
3785  *
3786  * This works the same way in the case of initialized -> unwritten conversion.
3787  *
3788  * One of more index blocks maybe needed if the extent tree grow after
3789  * the unwritten extent split. To prevent ENOSPC occur at the IO
3790  * complete, we need to split the unwritten extent before DIO submit
3791  * the IO. The unwritten extent called at this time will be split
3792  * into three unwritten extent(at most). After IO complete, the part
3793  * being filled will be convert to initialized by the end_io callback function
3794  * via ext4_convert_unwritten_extents().
3795  *
3796  * Returns the size of unwritten extent to be written on success.
3797  */
3798 static int ext4_split_convert_extents(handle_t *handle,
3799                                         struct inode *inode,
3800                                         struct ext4_map_blocks *map,
3801                                         struct ext4_ext_path **ppath,
3802                                         int flags)
3803 {
3804         struct ext4_ext_path *path = *ppath;
3805         ext4_lblk_t eof_block;
3806         ext4_lblk_t ee_block;
3807         struct ext4_extent *ex;
3808         unsigned int ee_len;
3809         int split_flag = 0, depth;
3810 
3811         ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3812                   __func__, inode->i_ino,
3813                   (unsigned long long)map->m_lblk, map->m_len);
3814 
3815         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3816                         >> inode->i_sb->s_blocksize_bits;
3817         if (eof_block < map->m_lblk + map->m_len)
3818                 eof_block = map->m_lblk + map->m_len;
3819         /*
3820          * It is safe to convert extent to initialized via explicit
3821          * zeroout only if extent is fully insde i_size or new_size.
3822          */
3823         depth = ext_depth(inode);
3824         ex = path[depth].p_ext;
3825         ee_block = le32_to_cpu(ex->ee_block);
3826         ee_len = ext4_ext_get_actual_len(ex);
3827 
3828         /* Convert to unwritten */
3829         if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3830                 split_flag |= EXT4_EXT_DATA_VALID1;
3831         /* Convert to initialized */
3832         } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3833                 split_flag |= ee_block + ee_len <= eof_block ?
3834                               EXT4_EXT_MAY_ZEROOUT : 0;
3835                 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3836         }
3837         flags |= EXT4_GET_BLOCKS_PRE_IO;
3838         return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3839 }
3840 
3841 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3842                                                 struct inode *inode,
3843                                                 struct ext4_map_blocks *map,
3844                                                 struct ext4_ext_path **ppath)
3845 {
3846         struct ext4_ext_path *path = *ppath;
3847         struct ext4_extent *ex;
3848         ext4_lblk_t ee_block;
3849         unsigned int ee_len;
3850         int depth;
3851         int err = 0;
3852 
3853         depth = ext_depth(inode);
3854         ex = path[depth].p_ext;
3855         ee_block = le32_to_cpu(ex->ee_block);
3856         ee_len = ext4_ext_get_actual_len(ex);
3857 
3858         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3859                 "block %llu, max_blocks %u\n", inode->i_ino,
3860                   (unsigned long long)ee_block, ee_len);
3861 
3862         /* If extent is larger than requested it is a clear sign that we still
3863          * have some extent state machine issues left. So extent_split is still
3864          * required.
3865          * TODO: Once all related issues will be fixed this situation should be
3866          * illegal.
3867          */
3868         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3869 #ifdef CONFIG_EXT4_DEBUG
3870                 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3871                              " len %u; IO logical block %llu, len %u",
3872                              inode->i_ino, (unsigned long long)ee_block, ee_len,
3873                              (unsigned long long)map->m_lblk, map->m_len);
3874 #endif
3875                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3876                                                  EXT4_GET_BLOCKS_CONVERT);
3877                 if (err < 0)
3878                         return err;
3879                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3880                 if (IS_ERR(path))
3881                         return PTR_ERR(path);
3882                 depth = ext_depth(inode);
3883                 ex = path[depth].p_ext;
3884         }
3885 
3886         err = ext4_ext_get_access(handle, inode, path + depth);
3887         if (err)
3888                 goto out;
3889         /* first mark the extent as initialized */
3890         ext4_ext_mark_initialized(ex);
3891 
3892         /* note: ext4_ext_correct_indexes() isn't needed here because
3893          * borders are not changed
3894          */
3895         ext4_ext_try_to_merge(handle, inode, path, ex);
3896 
3897         /* Mark modified extent as dirty */
3898         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3899 out:
3900         ext4_ext_show_leaf(inode, path);
3901         return err;
3902 }
3903 
3904 /*
3905  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3906  */
3907 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3908                               ext4_lblk_t lblk,
3909                               struct ext4_ext_path *path,
3910                               unsigned int len)
3911 {
3912         int i, depth;
3913         struct ext4_extent_header *eh;
3914         struct ext4_extent *last_ex;
3915 
3916         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3917                 return 0;
3918 
3919         depth = ext_depth(inode);
3920         eh = path[depth].p_hdr;
3921 
3922         /*
3923          * We're going to remove EOFBLOCKS_FL entirely in future so we
3924          * do not care for this case anymore. Simply remove the flag
3925          * if there are no extents.
3926          */
3927         if (unlikely(!eh->eh_entries))
3928                 goto out;
3929         last_ex = EXT_LAST_EXTENT(eh);
3930         /*
3931          * We should clear the EOFBLOCKS_FL flag if we are writing the
3932          * last block in the last extent in the file.  We test this by
3933          * first checking to see if the caller to
3934          * ext4_ext_get_blocks() was interested in the last block (or
3935          * a block beyond the last block) in the current extent.  If
3936          * this turns out to be false, we can bail out from this
3937          * function immediately.
3938          */
3939         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3940             ext4_ext_get_actual_len(last_ex))
3941                 return 0;
3942         /*
3943          * If the caller does appear to be planning to write at or
3944          * beyond the end of the current extent, we then test to see
3945          * if the current extent is the last extent in the file, by
3946          * checking to make sure it was reached via the rightmost node
3947          * at each level of the tree.
3948          */
3949         for (i = depth-1; i >= 0; i--)
3950                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3951                         return 0;
3952 out:
3953         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3954         return ext4_mark_inode_dirty(handle, inode);
3955 }
3956 
3957 static int
3958 convert_initialized_extent(handle_t *handle, struct inode *inode,
3959                            struct ext4_map_blocks *map,
3960                            struct ext4_ext_path **ppath,
3961                            unsigned int allocated)
3962 {
3963         struct ext4_ext_path *path = *ppath;
3964         struct ext4_extent *ex;
3965         ext4_lblk_t ee_block;
3966         unsigned int ee_len;
3967         int depth;
3968         int err = 0;
3969 
3970         /*
3971          * Make sure that the extent is no bigger than we support with
3972          * unwritten extent
3973          */
3974         if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3975                 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3976 
3977         depth = ext_depth(inode);
3978         ex = path[depth].p_ext;
3979         ee_block = le32_to_cpu(ex->ee_block);
3980         ee_len = ext4_ext_get_actual_len(ex);
3981 
3982         ext_debug("%s: inode %lu, logical"
3983                 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3984                   (unsigned long long)ee_block, ee_len);
3985 
3986         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3987                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3988                                 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3989                 if (err < 0)
3990                         return err;
3991                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3992                 if (IS_ERR(path))
3993                         return PTR_ERR(path);
3994                 depth = ext_depth(inode);
3995                 ex = path[depth].p_ext;
3996                 if (!ex) {
3997                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3998                                          (unsigned long) map->m_lblk);
3999                         return -EFSCORRUPTED;
4000                 }
4001         }
4002 
4003         err = ext4_ext_get_access(handle, inode, path + depth);
4004         if (err)
4005                 return err;
4006         /* first mark the extent as unwritten */
4007         ext4_ext_mark_unwritten(ex);
4008 
4009         /* note: ext4_ext_correct_indexes() isn't needed here because
4010          * borders are not changed
4011          */
4012         ext4_ext_try_to_merge(handle, inode, path, ex);
4013 
4014         /* Mark modified extent as dirty */
4015         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
4016         if (err)
4017                 return err;
4018         ext4_ext_show_leaf(inode, path);
4019 
4020         ext4_update_inode_fsync_trans(handle, inode, 1);
4021         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
4022         if (err)
4023                 return err;
4024         map->m_flags |= EXT4_MAP_UNWRITTEN;
4025         if (allocated > map->m_len)
4026                 allocated = map->m_len;
4027         map->m_len = allocated;
4028         return allocated;
4029 }
4030 
4031 static int
4032 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4033                         struct ext4_map_blocks *map,
4034                         struct ext4_ext_path **ppath, int flags,
4035                         unsigned int allocated, ext4_fsblk_t newblock)
4036 {
4037         struct ext4_ext_path *path = *ppath;
4038         int ret = 0;
4039         int err = 0;
4040 
4041         ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4042                   "block %llu, max_blocks %u, flags %x, allocated %u\n",
4043                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4044                   flags, allocated);
4045         ext4_ext_show_leaf(inode, path);
4046 
4047         /*
4048          * When writing into unwritten space, we should not fail to
4049          * allocate metadata blocks for the new extent block if needed.
4050          */
4051         flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4052 
4053         trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4054                                                     allocated, newblock);
4055 
4056         /* get_block() before submit the IO, split the extent */
4057         if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4058                 ret = ext4_split_convert_extents(handle, inode, map, ppath,
4059                                          flags | EXT4_GET_BLOCKS_CONVERT);
4060                 if (ret <= 0)
4061                         goto out;
4062                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4063                 goto out;
4064         }
4065         /* IO end_io complete, convert the filled extent to written */
4066         if (flags & EXT4_GET_BLOCKS_CONVERT) {
4067                 if (flags & EXT4_GET_BLOCKS_ZERO) {
4068                         if (allocated > map->m_len)
4069                                 allocated = map->m_len;
4070                         err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4071                                                  allocated);
4072                         if (err < 0)
4073                                 goto out2;
4074                 }
4075                 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4076                                                            ppath);
4077                 if (ret >= 0) {
4078                         ext4_update_inode_fsync_trans(handle, inode, 1);
4079                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
4080                                                  path, map->m_len);
4081                 } else
4082                         err = ret;
4083                 map->m_flags |= EXT4_MAP_MAPPED;
4084                 map->m_pblk = newblock;
4085                 if (allocated > map->m_len)
4086                         allocated = map->m_len;
4087                 map->m_len = allocated;
4088                 goto out2;
4089         }
4090         /* buffered IO case */
4091         /*
4092          * repeat fallocate creation request
4093          * we already have an unwritten extent
4094          */
4095         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4096                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4097                 goto map_out;
4098         }
4099 
4100         /* buffered READ or buffered write_begin() lookup */
4101         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4102                 /*
4103                  * We have blocks reserved already.  We
4104                  * return allocated blocks so that delalloc
4105                  * won't do block reservation for us.  But
4106                  * the buffer head will be unmapped so that
4107                  * a read from the block returns 0s.
4108                  */
4109                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4110                 goto out1;
4111         }
4112 
4113         /* buffered write, writepage time, convert*/
4114         ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4115         if (ret >= 0)
4116                 ext4_update_inode_fsync_trans(handle, inode, 1);
4117 out:
4118         if (ret <= 0) {
4119                 err = ret;
4120                 goto out2;
4121         } else
4122                 allocated = ret;
4123         map->m_flags |= EXT4_MAP_NEW;
4124         if (allocated > map->m_len)
4125                 allocated = map->m_len;
4126         map->m_len = allocated;
4127 
4128 map_out:
4129         map->m_flags |= EXT4_MAP_MAPPED;
4130         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4131                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4132                                          map->m_len);
4133                 if (err < 0)
4134                         goto out2;
4135         }
4136 out1:
4137         if (allocated > map->m_len)
4138                 allocated = map->m_len;
4139         ext4_ext_show_leaf(inode, path);
4140         map->m_pblk = newblock;
4141         map->m_len = allocated;
4142 out2:
4143         return err ? err : allocated;
4144 }
4145 
4146 /*
4147  * get_implied_cluster_alloc - check to see if the requested
4148  * allocation (in the map structure) overlaps with a cluster already
4149  * allocated in an extent.
4150  *      @sb     The filesystem superblock structure
4151  *      @map    The requested lblk->pblk mapping
4152  *      @ex     The extent structure which might contain an implied
4153  *                      cluster allocation
4154  *
4155  * This function is called by ext4_ext_map_blocks() after we failed to
4156  * find blocks that were already in the inode's extent tree.  Hence,
4157  * we know that the beginning of the requested region cannot overlap
4158  * the extent from the inode's extent tree.  There are three cases we
4159  * want to catch.  The first is this case:
4160  *
4161  *               |--- cluster # N--|
4162  *    |--- extent ---|  |---- requested region ---|
4163  *                      |==========|
4164  *
4165  * The second case that we need to test for is this one:
4166  *
4167  *   |--------- cluster # N ----------------|
4168  *         |--- requested region --|   |------- extent ----|
4169  *         |=======================|
4170  *
4171  * The third case is when the requested region lies between two extents
4172  * within the same cluster:
4173  *          |------------- cluster # N-------------|
4174  * |----- ex -----|                  |---- ex_right ----|
4175  *                  |------ requested region ------|
4176  *                  |================|
4177  *
4178  * In each of the above cases, we need to set the map->m_pblk and
4179  * map->m_len so it corresponds to the return the extent labelled as
4180  * "|====|" from cluster #N, since it is already in use for data in
4181  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
4182  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4183  * as a new "allocated" block region.  Otherwise, we will return 0 and
4184  * ext4_ext_map_blocks() will then allocate one or more new clusters
4185  * by calling ext4_mb_new_blocks().
4186  */
4187 static int get_implied_cluster_alloc(struct super_block *sb,
4188                                      struct ext4_map_blocks *map,
4189                                      struct ext4_extent *ex,
4190                                      struct ext4_ext_path *path)
4191 {
4192         struct ext4_sb_info *sbi = EXT4_SB(sb);
4193         ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4194         ext4_lblk_t ex_cluster_start, ex_cluster_end;
4195         ext4_lblk_t rr_cluster_start;
4196         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4197         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4198         unsigned short ee_len = ext4_ext_get_actual_len(ex);
4199 
4200         /* The extent passed in that we are trying to match */
4201         ex_cluster_start = EXT4_B2C(sbi, ee_block);
4202         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4203 
4204         /* The requested region passed into ext4_map_blocks() */
4205         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4206 
4207         if ((rr_cluster_start == ex_cluster_end) ||
4208             (rr_cluster_start == ex_cluster_start)) {
4209                 if (rr_cluster_start == ex_cluster_end)
4210                         ee_start += ee_len - 1;
4211                 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4212                 map->m_len = min(map->m_len,
4213                                  (unsigned) sbi->s_cluster_ratio - c_offset);
4214                 /*
4215                  * Check for and handle this case:
4216                  *
4217                  *   |--------- cluster # N-------------|
4218                  *                     |------- extent ----|
4219                  *         |--- requested region ---|
4220                  *         |===========|
4221                  */
4222 
4223                 if (map->m_lblk < ee_block)
4224                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
4225 
4226                 /*
4227                  * Check for the case where there is already another allocated
4228                  * block to the right of 'ex' but before the end of the cluster.
4229                  *
4230                  *          |------------- cluster # N-------------|
4231                  * |----- ex -----|                  |---- ex_right ----|
4232                  *                  |------ requested region ------|
4233                  *                  |================|
4234                  */
4235                 if (map->m_lblk > ee_block) {
4236                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4237                         map->m_len = min(map->m_len, next - map->m_lblk);
4238                 }
4239 
4240                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4241                 return 1;
4242         }
4243 
4244         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4245         return 0;
4246 }
4247 
4248 
4249 /*
4250  * Block allocation/map/preallocation routine for extents based files
4251  *
4252  *
4253  * Need to be called with
4254  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4255  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4256  *
4257  * return > 0, number of of blocks already mapped/allocated
4258  *          if create == 0 and these are pre-allocated blocks
4259  *              buffer head is unmapped
4260  *          otherwise blocks are mapped
4261  *
4262  * return = 0, if plain look up failed (blocks have not been allocated)
4263  *          buffer head is unmapped
4264  *
4265  * return < 0, error case.
4266  */
4267 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4268                         struct ext4_map_blocks *map, int flags)
4269 {
4270         struct ext4_ext_path *path = NULL;
4271         struct ext4_extent newex, *ex, *ex2;
4272         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4273         ext4_fsblk_t newblock = 0;
4274         int free_on_err = 0, err = 0, depth, ret;
4275         unsigned int allocated = 0, offset = 0;
4276         unsigned int allocated_clusters = 0;
4277         struct ext4_allocation_request ar;
4278         ext4_lblk_t cluster_offset;
4279         bool map_from_cluster = false;
4280 
4281         ext_debug("blocks %u/%u requested for inode %lu\n",
4282                   map->m_lblk, map->m_len, inode->i_ino);
4283         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4284 
4285         /* find extent for this block */
4286         path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4287         if (IS_ERR(path)) {
4288                 err = PTR_ERR(path);
4289                 path = NULL;
4290                 goto out2;
4291         }
4292 
4293         depth = ext_depth(inode);
4294 
4295         /*
4296          * consistent leaf must not be empty;
4297          * this situation is possible, though, _during_ tree modification;
4298          * this is why assert can't be put in ext4_find_extent()
4299          */
4300         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4301                 EXT4_ERROR_INODE(inode, "bad extent address "
4302                                  "lblock: %lu, depth: %d pblock %lld",
4303                                  (unsigned long) map->m_lblk, depth,
4304                                  path[depth].p_block);
4305                 err = -EFSCORRUPTED;
4306                 goto out2;
4307         }
4308 
4309         ex = path[depth].p_ext;
4310         if (ex) {
4311                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4312                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4313                 unsigned short ee_len;
4314 
4315 
4316                 /*
4317                  * unwritten extents are treated as holes, except that
4318                  * we split out initialized portions during a write.
4319                  */
4320                 ee_len = ext4_ext_get_actual_len(ex);
4321 
4322                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4323 
4324                 /* if found extent covers block, simply return it */
4325                 if (in_range(map->m_lblk, ee_block, ee_len)) {
4326                         newblock = map->m_lblk - ee_block + ee_start;
4327                         /* number of remaining blocks in the extent */
4328                         allocated = ee_len - (map->m_lblk - ee_block);
4329                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4330                                   ee_block, ee_len, newblock);
4331 
4332                         /*
4333                          * If the extent is initialized check whether the
4334                          * caller wants to convert it to unwritten.
4335                          */
4336                         if ((!ext4_ext_is_unwritten(ex)) &&
4337                             (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4338                                 allocated = convert_initialized_extent(
4339                                                 handle, inode, map, &path,
4340                                                 allocated);
4341                                 goto out2;
4342                         } else if (!ext4_ext_is_unwritten(ex))
4343                                 goto out;
4344 
4345                         ret = ext4_ext_handle_unwritten_extents(
4346                                 handle, inode, map, &path, flags,
4347                                 allocated, newblock);
4348                         if (ret < 0)
4349                                 err = ret;
4350                         else
4351                                 allocated = ret;
4352                         goto out2;
4353                 }
4354         }
4355 
4356         /*
4357          * requested block isn't allocated yet;
4358          * we couldn't try to create block if create flag is zero
4359          */
4360         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4361                 ext4_lblk_t hole_start, hole_len;
4362 
4363                 hole_start = map->m_lblk;
4364                 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4365                 /*
4366                  * put just found gap into cache to speed up
4367                  * subsequent requests
4368                  */
4369                 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4370 
4371                 /* Update hole_len to reflect hole size after map->m_lblk */
4372                 if (hole_start != map->m_lblk)
4373                         hole_len -= map->m_lblk - hole_start;
4374                 map->m_pblk = 0;
4375                 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4376 
4377                 goto out2;
4378         }
4379 
4380         /*
4381          * Okay, we need to do block allocation.
4382          */
4383         newex.ee_block = cpu_to_le32(map->m_lblk);
4384         cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4385 
4386         /*
4387          * If we are doing bigalloc, check to see if the extent returned
4388          * by ext4_find_extent() implies a cluster we can use.
4389          */
4390         if (cluster_offset && ex &&
4391             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4392                 ar.len = allocated = map->m_len;
4393                 newblock = map->m_pblk;
4394                 map_from_cluster = true;
4395                 goto got_allocated_blocks;
4396         }
4397 
4398         /* find neighbour allocated blocks */
4399         ar.lleft = map->m_lblk;
4400         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4401         if (err)
4402                 goto out2;
4403         ar.lright = map->m_lblk;
4404         ex2 = NULL;
4405         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4406         if (err)
4407                 goto out2;
4408 
4409         /* Check if the extent after searching to the right implies a
4410          * cluster we can use. */
4411         if ((sbi->s_cluster_ratio > 1) && ex2 &&
4412             get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4413                 ar.len = allocated = map->m_len;
4414                 newblock = map->m_pblk;
4415                 map_from_cluster = true;
4416                 goto got_allocated_blocks;
4417         }
4418 
4419         /*
4420          * See if request is beyond maximum number of blocks we can have in
4421          * a single extent. For an initialized extent this limit is
4422          * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4423          * EXT_UNWRITTEN_MAX_LEN.
4424          */
4425         if (map->m_len > EXT_INIT_MAX_LEN &&
4426             !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4427                 map->m_len = EXT_INIT_MAX_LEN;
4428         else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4429                  (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4430                 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4431 
4432         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4433         newex.ee_len = cpu_to_le16(map->m_len);
4434         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4435         if (err)
4436                 allocated = ext4_ext_get_actual_len(&newex);
4437         else
4438                 allocated = map->m_len;
4439 
4440         /* allocate new block */
4441         ar.inode = inode;
4442         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4443         ar.logical = map->m_lblk;
4444         /*
4445          * We calculate the offset from the beginning of the cluster
4446          * for the logical block number, since when we allocate a
4447          * physical cluster, the physical block should start at the
4448          * same offset from the beginning of the cluster.  This is
4449          * needed so that future calls to get_implied_cluster_alloc()
4450          * work correctly.
4451          */
4452         offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4453         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4454         ar.goal -= offset;
4455         ar.logical -= offset;
4456         if (S_ISREG(inode->i_mode))
4457                 ar.flags = EXT4_MB_HINT_DATA;
4458         else
4459                 /* disable in-core preallocation for non-regular files */
4460                 ar.flags = 0;
4461         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4462                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4463         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4464                 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4465         if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4466                 ar.flags |= EXT4_MB_USE_RESERVED;
4467         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4468         if (!newblock)
4469                 goto out2;
4470         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4471                   ar.goal, newblock, allocated);
4472         free_on_err = 1;
4473         allocated_clusters = ar.len;
4474         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4475         if (ar.len > allocated)
4476                 ar.len = allocated;
4477 
4478 got_allocated_blocks:
4479         /* try to insert new extent into found leaf and return */
4480         ext4_ext_store_pblock(&newex, newblock + offset);
4481         newex.ee_len = cpu_to_le16(ar.len);
4482         /* Mark unwritten */
4483         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4484                 ext4_ext_mark_unwritten(&newex);
4485                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4486         }
4487 
4488         err = 0;
4489         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4490                 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4491                                          path, ar.len);
4492         if (!err)
4493                 err = ext4_ext_insert_extent(handle, inode, &path,
4494                                              &newex, flags);
4495 
4496         if (err && free_on_err) {
4497                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4498                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4499                 /* free data blocks we just allocated */
4500                 /* not a good idea to call discard here directly,
4501                  * but otherwise we'd need to call it every free() */
4502                 ext4_discard_preallocations(inode);
4503                 ext4_free_blocks(handle, inode, NULL, newblock,
4504                                  EXT4_C2B(sbi, allocated_clusters), fb_flags);
4505                 goto out2;
4506         }
4507 
4508         /* previous routine could use block we allocated */
4509         newblock = ext4_ext_pblock(&newex);
4510         allocated = ext4_ext_get_actual_len(&newex);
4511         if (allocated > map->m_len)
4512                 allocated = map->m_len;
4513         map->m_flags |= EXT4_MAP_NEW;
4514 
4515         /*
4516          * Reduce the reserved cluster count to reflect successful deferred
4517          * allocation of delayed allocated clusters or direct allocation of
4518          * clusters discovered to be delayed allocated.  Once allocated, a
4519          * cluster is not included in the reserved count.
4520          */
4521         if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4522                 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4523                         /*
4524                          * When allocating delayed allocated clusters, simply
4525                          * reduce the reserved cluster count and claim quota
4526                          */
4527                         ext4_da_update_reserve_space(inode, allocated_clusters,
4528                                                         1);
4529                 } else {
4530                         ext4_lblk_t lblk, len;
4531                         unsigned int n;
4532 
4533                         /*
4534                          * When allocating non-delayed allocated clusters
4535                          * (from fallocate, filemap, DIO, or clusters
4536                          * allocated when delalloc has been disabled by
4537                          * ext4_nonda_switch), reduce the reserved cluster
4538                          * count by the number of allocated clusters that
4539                          * have previously been delayed allocated.  Quota
4540                          * has been claimed by ext4_mb_new_blocks() above,
4541                          * so release the quota reservations made for any
4542                          * previously delayed allocated clusters.
4543                          */
4544                         lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4545                         len = allocated_clusters << sbi->s_cluster_bits;
4546                         n = ext4_es_delayed_clu(inode, lblk, len);
4547                         if (n > 0)
4548                                 ext4_da_update_reserve_space(inode, (int) n, 0);
4549                 }
4550         }
4551 
4552         /*
4553          * Cache the extent and update transaction to commit on fdatasync only
4554          * when it is _not_ an unwritten extent.
4555          */
4556         if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4557                 ext4_update_inode_fsync_trans(handle, inode, 1);
4558         else
4559                 ext4_update_inode_fsync_trans(handle, inode, 0);
4560 out:
4561         if (allocated > map->m_len)
4562                 allocated = map->m_len;
4563         ext4_ext_show_leaf(inode, path);
4564         map->m_flags |= EXT4_MAP_MAPPED;
4565         map->m_pblk = newblock;
4566         map->m_len = allocated;
4567 out2:
4568         ext4_ext_drop_refs(path);
4569         kfree(path);
4570 
4571         trace_ext4_ext_map_blocks_exit(inode, flags, map,
4572                                        err ? err : allocated);
4573         return err ? err : allocated;
4574 }
4575 
4576 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4577 {
4578         struct super_block *sb = inode->i_sb;
4579         ext4_lblk_t last_block;
4580         int err = 0;
4581 
4582         /*
4583          * TODO: optimization is possible here.
4584          * Probably we need not scan at all,
4585          * because page truncation is enough.
4586          */
4587 
4588         /* we have to know where to truncate from in crash case */
4589         EXT4_I(inode)->i_disksize = inode->i_size;
4590         err = ext4_mark_inode_dirty(handle, inode);
4591         if (err)
4592                 return err;
4593 
4594         last_block = (inode->i_size + sb->s_blocksize - 1)
4595                         >> EXT4_BLOCK_SIZE_BITS(sb);
4596 retry:
4597         err = ext4_es_remove_extent(inode, last_block,
4598                                     EXT_MAX_BLOCKS - last_block);
4599         if (err == -ENOMEM) {
4600                 cond_resched();
4601                 congestion_wait(BLK_RW_ASYNC, HZ/50);
4602                 goto retry;
4603         }
4604         if (err)
4605                 return err;
4606         return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4607 }
4608 
4609 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4610                                   ext4_lblk_t len, loff_t new_size,
4611                                   int flags)
4612 {
4613         struct inode *inode = file_inode(file);
4614         handle_t *handle;
4615         int ret = 0;
4616         int ret2 = 0;
4617         int retries = 0;
4618         int depth = 0;
4619         struct ext4_map_blocks map;
4620         unsigned int credits;
4621         loff_t epos;
4622 
4623         BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4624         map.m_lblk = offset;
4625         map.m_len = len;
4626         /*
4627          * Don't normalize the request if it can fit in one extent so
4628          * that it doesn't get unnecessarily split into multiple
4629          * extents.
4630          */
4631         if (len <= EXT_UNWRITTEN_MAX_LEN)
4632                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4633 
4634         /*
4635          * credits to insert 1 extent into extent tree
4636          */
4637         credits = ext4_chunk_trans_blocks(inode, len);
4638         depth = ext_depth(inode);
4639 
4640 retry:
4641         while (ret >= 0 && len) {
4642                 /*
4643                  * Recalculate credits when extent tree depth changes.
4644                  */
4645                 if (depth != ext_depth(inode)) {
4646                         credits = ext4_chunk_trans_blocks(inode, len);
4647                         depth = ext_depth(inode);
4648                 }
4649 
4650                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4651                                             credits);
4652                 if (IS_ERR(handle)) {
4653                         ret = PTR_ERR(handle);
4654                         break;
4655                 }
4656                 ret = ext4_map_blocks(handle, inode, &map, flags);
4657                 if (ret <= 0) {
4658                         ext4_debug("inode #%lu: block %u: len %u: "
4659                                    "ext4_ext_map_blocks returned %d",
4660                                    inode->i_ino, map.m_lblk,
4661                                    map.m_len, ret);
4662                         ext4_mark_inode_dirty(handle, inode);
4663                         ret2 = ext4_journal_stop(handle);
4664                         break;
4665                 }
4666                 map.m_lblk += ret;
4667                 map.m_len = len = len - ret;
4668                 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4669                 inode->i_ctime = current_time(inode);
4670                 if (new_size) {
4671                         if (epos > new_size)
4672                                 epos = new_size;
4673                         if (ext4_update_inode_size(inode, epos) & 0x1)
4674                                 inode->i_mtime = inode->i_ctime;
4675                 } else {
4676                         if (epos > inode->i_size)
4677                                 ext4_set_inode_flag(inode,
4678                                                     EXT4_INODE_EOFBLOCKS);
4679                 }
4680                 ext4_mark_inode_dirty(handle, inode);
4681                 ext4_update_inode_fsync_trans(handle, inode, 1);
4682                 ret2 = ext4_journal_stop(handle);
4683                 if (ret2)
4684                         break;
4685         }
4686         if (ret == -ENOSPC &&
4687                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4688                 ret = 0;
4689                 goto retry;
4690         }
4691 
4692         return ret > 0 ? ret2 : ret;
4693 }
4694 
4695 static long ext4_zero_range(struct file *file, loff_t offset,
4696                             loff_t len, int mode)
4697 {
4698         struct inode *inode = file_inode(file);
4699         handle_t *handle = NULL;
4700         unsigned int max_blocks;
4701         loff_t new_size = 0;
4702         int ret = 0;
4703         int flags;
4704         int credits;
4705         int partial_begin, partial_end;
4706         loff_t start, end;
4707         ext4_lblk_t lblk;
4708         unsigned int blkbits = inode->i_blkbits;
4709 
4710         trace_ext4_zero_range(inode, offset, len, mode);
4711 
4712         if (!S_ISREG(inode->i_mode))
4713                 return -EINVAL;
4714 
4715         /* Call ext4_force_commit to flush all data in case of data=journal. */
4716         if (ext4_should_journal_data(inode)) {
4717                 ret = ext4_force_commit(inode->i_sb);
4718                 if (ret)
4719                         return ret;
4720         }
4721 
4722         /*
4723          * Round up offset. This is not fallocate, we neet to zero out
4724          * blocks, so convert interior block aligned part of the range to
4725          * unwritten and possibly manually zero out unaligned parts of the
4726          * range.
4727          */
4728         start = round_up(offset, 1 << blkbits);
4729         end = round_down((offset + len), 1 << blkbits);
4730 
4731         if (start < offset || end > offset + len)
4732                 return -EINVAL;
4733         partial_begin = offset & ((1 << blkbits) - 1);
4734         partial_end = (offset + len) & ((1 << blkbits) - 1);
4735 
4736         lblk = start >> blkbits;
4737         max_blocks = (end >> blkbits);
4738         if (max_blocks < lblk)
4739                 max_blocks = 0;
4740         else
4741                 max_blocks -= lblk;
4742 
4743         inode_lock(inode);
4744 
4745         /*
4746          * Indirect files do not support unwritten extnets
4747          */
4748         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4749                 ret = -EOPNOTSUPP;
4750                 goto out_mutex;
4751         }
4752 
4753         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4754             (offset + len > i_size_read(inode) ||
4755              offset + len > EXT4_I(inode)->i_disksize)) {
4756                 new_size = offset + len;
4757                 ret = inode_newsize_ok(inode, new_size);
4758                 if (ret)
4759                         goto out_mutex;
4760         }
4761 
4762         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4763         if (mode & FALLOC_FL_KEEP_SIZE)
4764                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4765 
4766         /* Wait all existing dio workers, newcomers will block on i_mutex */
4767         inode_dio_wait(inode);
4768 
4769         /* Preallocate the range including the unaligned edges */
4770         if (partial_begin || partial_end) {
4771                 ret = ext4_alloc_file_blocks(file,
4772                                 round_down(offset, 1 << blkbits) >> blkbits,
4773                                 (round_up((offset + len), 1 << blkbits) -
4774                                  round_down(offset, 1 << blkbits)) >> blkbits,
4775                                 new_size, flags);
4776                 if (ret)
4777                         goto out_mutex;
4778 
4779         }
4780 
4781         /* Zero range excluding the unaligned edges */
4782         if (max_blocks > 0) {
4783                 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4784                           EXT4_EX_NOCACHE);
4785 
4786                 /*
4787                  * Prevent page faults from reinstantiating pages we have
4788                  * released from page cache.
4789                  */
4790                 down_write(&EXT4_I(inode)->i_mmap_sem);
4791 
4792                 ret = ext4_break_layouts(inode);
4793                 if (ret) {
4794                         up_write(&EXT4_I(inode)->i_mmap_sem);
4795                         goto out_mutex;
4796                 }
4797 
4798                 ret = ext4_update_disksize_before_punch(inode, offset, len);
4799                 if (ret) {
4800                         up_write(&EXT4_I(inode)->i_mmap_sem);
4801                         goto out_mutex;
4802                 }
4803                 /* Now release the pages and zero block aligned part of pages */
4804                 truncate_pagecache_range(inode, start, end - 1);
4805                 inode->i_mtime = inode->i_ctime = current_time(inode);
4806 
4807                 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4808                                              flags);
4809                 up_write(&EXT4_I(inode)->i_mmap_sem);
4810                 if (ret)
4811                         goto out_mutex;
4812         }
4813         if (!partial_begin && !partial_end)
4814                 goto out_mutex;
4815 
4816         /*
4817          * In worst case we have to writeout two nonadjacent unwritten
4818          * blocks and update the inode
4819          */
4820         credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4821         if (ext4_should_journal_data(inode))
4822                 credits += 2;
4823         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4824         if (IS_ERR(handle)) {
4825                 ret = PTR_ERR(handle);
4826                 ext4_std_error(inode->i_sb, ret);
4827                 goto out_mutex;
4828         }
4829 
4830         inode->i_mtime = inode->i_ctime = current_time(inode);
4831         if (new_size) {
4832                 ext4_update_inode_size(inode, new_size);
4833         } else {
4834                 /*
4835                 * Mark that we allocate beyond EOF so the subsequent truncate
4836                 * can proceed even if the new size is the same as i_size.
4837                 */
4838                 if ((offset + len) > i_size_read(inode))
4839                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4840         }
4841         ext4_mark_inode_dirty(handle, inode);
4842 
4843         /* Zero out partial block at the edges of the range */
4844         ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4845         if (ret >= 0)
4846                 ext4_update_inode_fsync_trans(handle, inode, 1);
4847 
4848         if (file->f_flags & O_SYNC)
4849                 ext4_handle_sync(handle);
4850 
4851         ext4_journal_stop(handle);
4852 out_mutex:
4853         inode_unlock(inode);
4854         return ret;
4855 }
4856 
4857 /*
4858  * preallocate space for a file. This implements ext4's fallocate file
4859  * operation, which gets called from sys_fallocate system call.
4860  * For block-mapped files, posix_fallocate should fall back to the method
4861  * of writing zeroes to the required new blocks (the same behavior which is
4862  * expected for file systems which do not support fallocate() system call).
4863  */
4864 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4865 {
4866         struct inode *inode = file_inode(file);
4867         loff_t new_size = 0;
4868         unsigned int max_blocks;
4869         int ret = 0;
4870         int flags;
4871         ext4_lblk_t lblk;
4872         unsigned int blkbits = inode->i_blkbits;
4873 
4874         /*
4875          * Encrypted inodes can't handle collapse range or insert
4876          * range since we would need to re-encrypt blocks with a
4877          * different IV or XTS tweak (which are based on the logical
4878          * block number).
4879          *
4880          * XXX It's not clear why zero range isn't working, but we'll
4881          * leave it disabled for encrypted inodes for now.  This is a
4882          * bug we should fix....
4883          */
4884         if (IS_ENCRYPTED(inode) &&
4885             (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4886                      FALLOC_FL_ZERO_RANGE)))
4887                 return -EOPNOTSUPP;
4888 
4889         /* Return error if mode is not supported */
4890         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4891                      FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4892                      FALLOC_FL_INSERT_RANGE))
4893                 return -EOPNOTSUPP;
4894 
4895         if (mode & FALLOC_FL_PUNCH_HOLE)
4896                 return ext4_punch_hole(inode, offset, len);
4897 
4898         ret = ext4_convert_inline_data(inode);
4899         if (ret)
4900                 return ret;
4901 
4902         if (mode & FALLOC_FL_COLLAPSE_RANGE)
4903                 return ext4_collapse_range(inode, offset, len);
4904 
4905         if (mode & FALLOC_FL_INSERT_RANGE)
4906                 return ext4_insert_range(inode, offset, len);
4907 
4908         if (mode & FALLOC_FL_ZERO_RANGE)
4909                 return ext4_zero_range(file, offset, len, mode);
4910 
4911         trace_ext4_fallocate_enter(inode, offset, len, mode);
4912         lblk = offset >> blkbits;
4913 
4914         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4915         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4916         if (mode & FALLOC_FL_KEEP_SIZE)
4917                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4918 
4919         inode_lock(inode);
4920 
4921         /*
4922          * We only support preallocation for extent-based files only
4923          */
4924         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4925                 ret = -EOPNOTSUPP;
4926                 goto out;
4927         }
4928 
4929         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4930             (offset + len > i_size_read(inode) ||
4931              offset + len > EXT4_I(inode)->i_disksize)) {
4932                 new_size = offset + len;
4933                 ret = inode_newsize_ok(inode, new_size);
4934                 if (ret)
4935                         goto out;
4936         }
4937 
4938         /* Wait all existing dio workers, newcomers will block on i_mutex */
4939         inode_dio_wait(inode);
4940 
4941         ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4942         if (ret)
4943                 goto out;
4944 
4945         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4946                 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4947                                                 EXT4_I(inode)->i_sync_tid);
4948         }
4949 out:
4950         inode_unlock(inode);
4951         trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4952         return ret;
4953 }
4954 
4955 /*
4956  * This function convert a range of blocks to written extents
4957  * The caller of this function will pass the start offset and the size.
4958  * all unwritten extents within this range will be converted to
4959  * written extents.
4960  *
4961  * This function is called from the direct IO end io call back
4962  * function, to convert the fallocated extents after IO is completed.
4963  * Returns 0 on success.
4964  */
4965 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4966                                    loff_t offset, ssize_t len)
4967 {
4968         unsigned int max_blocks;
4969         int ret = 0;
4970         int ret2 = 0;
4971         struct ext4_map_blocks map;
4972         unsigned int credits, blkbits = inode->i_blkbits;
4973 
4974         map.m_lblk = offset >> blkbits;
4975         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4976 
4977         /*
4978          * This is somewhat ugly but the idea is clear: When transaction is
4979          * reserved, everything goes into it. Otherwise we rather start several
4980          * smaller transactions for conversion of each extent separately.
4981          */
4982         if (handle) {
4983                 handle = ext4_journal_start_reserved(handle,
4984                                                      EXT4_HT_EXT_CONVERT);
4985                 if (IS_ERR(handle))
4986                         return PTR_ERR(handle);
4987                 credits = 0;
4988         } else {
4989                 /*
4990                  * credits to insert 1 extent into extent tree
4991                  */
4992                 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4993         }
4994         while (ret >= 0 && ret < max_blocks) {
4995                 map.m_lblk += ret;
4996                 map.m_len = (max_blocks -= ret);
4997                 if (credits) {
4998                         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4999                                                     credits);
5000                         if (IS_ERR(handle)) {
5001                                 ret = PTR_ERR(handle);
5002                                 break;
5003                         }
5004                 }
5005                 ret = ext4_map_blocks(handle, inode, &map,
5006                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
5007                 if (ret <= 0)
5008                         ext4_warning(inode->i_sb,
5009                                      "inode #%lu: block %u: len %u: "
5010                                      "ext4_ext_map_blocks returned %d",
5011                                      inode->i_ino, map.m_lblk,
5012                                      map.m_len, ret);
5013                 ext4_mark_inode_dirty(handle, inode);
5014                 if (credits)
5015                         ret2 = ext4_journal_stop(handle);
5016                 if (ret <= 0 || ret2)
5017                         break;
5018         }
5019         if (!credits)
5020                 ret2 = ext4_journal_stop(handle);
5021         return ret > 0 ? ret2 : ret;
5022 }
5023 
5024 /*
5025  * If newes is not existing extent (newes->ec_pblk equals zero) find
5026  * delayed extent at start of newes and update newes accordingly and
5027  * return start of the next delayed extent.
5028  *
5029  * If newes is existing extent (newes->ec_pblk is not equal zero)
5030  * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5031  * extent found. Leave newes unmodified.
5032  */
5033 static int ext4_find_delayed_extent(struct inode *inode,
5034                                     struct extent_status *newes)
5035 {
5036         struct extent_status es;
5037         ext4_lblk_t block, next_del;
5038 
5039         if (newes->es_pblk == 0) {
5040                 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5041                                           newes->es_lblk,
5042                                           newes->es_lblk + newes->es_len - 1,
5043                                           &es);
5044 
5045                 /*
5046                  * No extent in extent-tree contains block @newes->es_pblk,
5047                  * then the block may stay in 1)a hole or 2)delayed-extent.
5048                  */
5049                 if (es.es_len == 0)
5050                         /* A hole found. */
5051                         return 0;
5052 
5053                 if (es.es_lblk > newes->es_lblk) {
5054                         /* A hole found. */
5055                         newes->es_len = min(es.es_lblk - newes->es_lblk,
5056                                             newes->es_len);
5057                         return 0;
5058                 }
5059 
5060                 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5061         }
5062 
5063         block = newes->es_lblk + newes->es_len;
5064         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5065                                   EXT_MAX_BLOCKS, &es);
5066         if (es.es_len == 0)
5067                 next_del = EXT_MAX_BLOCKS;
5068         else
5069                 next_del = es.es_lblk;
5070 
5071         return next_del;
5072 }
5073 
5074 static int ext4_xattr_fiemap(struct inode *inode,
5075                                 struct fiemap_extent_info *fieinfo)
5076 {
5077         __u64 physical = 0;
5078         __u64 length;
5079         __u32 flags = FIEMAP_EXTENT_LAST;
5080         int blockbits = inode->i_sb->s_blocksize_bits;
5081         int error = 0;
5082 
5083         /* in-inode? */
5084         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5085                 struct ext4_iloc iloc;
5086                 int offset;     /* offset of xattr in inode */
5087 
5088                 error = ext4_get_inode_loc(inode, &iloc);
5089                 if (error)
5090                         return error;
5091                 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5092                 offset = EXT4_GOOD_OLD_INODE_SIZE +
5093                                 EXT4_I(inode)->i_extra_isize;
5094                 physical += offset;
5095                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5096                 flags |= FIEMAP_EXTENT_DATA_INLINE;
5097                 brelse(iloc.bh);
5098         } else { /* external block */
5099                 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5100                 length = inode->i_sb->s_blocksize;
5101         }
5102 
5103         if (physical)
5104                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5105                                                 length, flags);
5106         return (error < 0 ? error : 0);
5107 }
5108 
5109 static int _ext4_fiemap(struct inode *inode,
5110                         struct fiemap_extent_info *fieinfo,
5111                         __u64 start, __u64 len,
5112                         int (*fill)(struct inode *, ext4_lblk_t,
5113                                     ext4_lblk_t,
5114                                     struct fiemap_extent_info *))
5115 {
5116         ext4_lblk_t start_blk;
5117         u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5118 
5119         int error = 0;
5120 
5121         if (ext4_has_inline_data(inode)) {
5122                 int has_inline = 1;
5123 
5124                 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5125                                                 start, len);
5126 
5127                 if (has_inline)
5128                         return error;
5129         }
5130 
5131         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5132                 error = ext4_ext_precache(inode);
5133                 if (error)
5134                         return error;
5135                 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5136         }
5137 
5138         /* fallback to generic here if not in extents fmt */
5139         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5140             fill == ext4_fill_fiemap_extents)
5141                 return generic_block_fiemap(inode, fieinfo, start, len,
5142                         ext4_get_block);
5143 
5144         if (fill == ext4_fill_es_cache_info)
5145                 ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5146         if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5147                 return -EBADR;
5148 
5149         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5150                 error = ext4_xattr_fiemap(inode, fieinfo);
5151         } else {
5152                 ext4_lblk_t len_blks;
5153                 __u64 last_blk;
5154 
5155                 start_blk = start >> inode->i_sb->s_blocksize_bits;
5156                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5157                 if (last_blk >= EXT_MAX_BLOCKS)
5158                         last_blk = EXT_MAX_BLOCKS-1;
5159                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5160 
5161                 /*
5162                  * Walk the extent tree gathering extent information
5163                  * and pushing extents back to the user.
5164                  */
5165                 error = fill(inode, start_blk, len_blks, fieinfo);
5166         }
5167         return error;
5168 }
5169 
5170 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5171                 __u64 start, __u64 len)
5172 {
5173         return _ext4_fiemap(inode, fieinfo, start, len,
5174                             ext4_fill_fiemap_extents);
5175 }
5176 
5177 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5178                       __u64 start, __u64 len)
5179 {
5180         if (ext4_has_inline_data(inode)) {
5181                 int has_inline;
5182 
5183                 down_read(&EXT4_I(inode)->xattr_sem);
5184                 has_inline = ext4_has_inline_data(inode);
5185                 up_read(&EXT4_I(inode)->xattr_sem);
5186                 if (has_inline)
5187                         return 0;
5188         }
5189 
5190         return _ext4_fiemap(inode, fieinfo, start, len,
5191                             ext4_fill_es_cache_info);
5192 }
5193 
5194 
5195 /*
5196  * ext4_access_path:
5197  * Function to access the path buffer for marking it dirty.
5198  * It also checks if there are sufficient credits left in the journal handle
5199  * to update path.
5200  */
5201 static int
5202 ext4_access_path(handle_t *handle, struct inode *inode,
5203                 struct ext4_ext_path *path)
5204 {
5205         int credits, err;
5206 
5207         if (!ext4_handle_valid(handle))
5208                 return 0;
5209 
5210         /*
5211          * Check if need to extend journal credits
5212          * 3 for leaf, sb, and inode plus 2 (bmap and group
5213          * descriptor) for each block group; assume two block
5214          * groups
5215          */
5216         if (handle->h_buffer_credits < 7) {
5217                 credits = ext4_writepage_trans_blocks(inode);
5218                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5219                 /* EAGAIN is success */
5220                 if (err && err != -EAGAIN)
5221                         return err;
5222         }
5223 
5224         err = ext4_ext_get_access(handle, inode, path);
5225         return err;
5226 }
5227 
5228 /*
5229  * ext4_ext_shift_path_extents:
5230  * Shift the extents of a path structure lying between path[depth].p_ext
5231  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5232  * if it is right shift or left shift operation.
5233  */
5234 static int
5235 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5236                             struct inode *inode, handle_t *handle,
5237                             enum SHIFT_DIRECTION SHIFT)
5238 {
5239         int depth, err = 0;
5240         struct ext4_extent *ex_start, *ex_last;
5241         bool update = 0;
5242         depth = path->p_depth;
5243 
5244         while (depth >= 0) {
5245                 if (depth == path->p_depth) {
5246                         ex_start = path[depth].p_ext;
5247                         if (!ex_start)
5248                                 return -EFSCORRUPTED;
5249 
5250                         ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5251 
5252                         err = ext4_access_path(handle, inode, path + depth);
5253                         if (err)
5254                                 goto out;
5255 
5256                         if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5257                                 update = 1;
5258 
5259                         while (ex_start <= ex_last) {
5260                                 if (SHIFT == SHIFT_LEFT) {
5261                                         le32_add_cpu(&ex_start->ee_block,
5262                                                 -shift);
5263                                         /* Try to merge to the left. */
5264                                         if ((ex_start >
5265                                             EXT_FIRST_EXTENT(path[depth].p_hdr))
5266                                             &&
5267                                             ext4_ext_try_to_merge_right(inode,
5268                                             path, ex_start - 1))
5269                                                 ex_last--;
5270                                         else
5271                                                 ex_start++;
5272                                 } else {
5273                                         le32_add_cpu(&ex_last->ee_block, shift);
5274                                         ext4_ext_try_to_merge_right(inode, path,
5275                                                 ex_last);
5276                                         ex_last--;
5277                                 }
5278                         }
5279                         err = ext4_ext_dirty(handle, inode, path + depth);
5280                         if (err)
5281                                 goto out;
5282 
5283                         if (--depth < 0 || !update)
5284                                 break;
5285                 }
5286 
5287                 /* Update index too */
5288                 err = ext4_access_path(handle, inode, path + depth);
5289                 if (err)
5290                         goto out;
5291 
5292                 if (SHIFT == SHIFT_LEFT)
5293                         le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5294                 else
5295                         le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5296                 err = ext4_ext_dirty(handle, inode, path + depth);
5297                 if (err)
5298                         goto out;
5299 
5300                 /* we are done if current index is not a starting index */
5301                 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5302                         break;
5303 
5304                 depth--;
5305         }
5306 
5307 out:
5308         return err;
5309 }
5310 
5311 /*
5312  * ext4_ext_shift_extents:
5313  * All the extents which lies in the range from @start to the last allocated
5314  * block for the @inode are shifted either towards left or right (depending
5315  * upon @SHIFT) by @shift blocks.
5316  * On success, 0 is returned, error otherwise.
5317  */
5318 static int
5319 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5320                        ext4_lblk_t start, ext4_lblk_t shift,
5321                        enum SHIFT_DIRECTION SHIFT)
5322 {
5323         struct ext4_ext_path *path;
5324         int ret = 0, depth;
5325         struct ext4_extent *extent;
5326         ext4_lblk_t stop, *iterator, ex_start, ex_end;
5327 
5328         /* Let path point to the last extent */
5329         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5330                                 EXT4_EX_NOCACHE);
5331         if (IS_ERR(path))
5332                 return PTR_ERR(path);
5333 
5334         depth = path->p_depth;
5335         extent = path[depth].p_ext;
5336         if (!extent)
5337                 goto out;
5338 
5339         stop = le32_to_cpu(extent->ee_block);
5340 
5341        /*
5342         * For left shifts, make sure the hole on the left is big enough to
5343         * accommodate the shift.  For right shifts, make sure the last extent
5344         * won't be shifted beyond EXT_MAX_BLOCKS.
5345         */
5346         if (SHIFT == SHIFT_LEFT) {
5347                 path = ext4_find_extent(inode, start - 1, &path,
5348                                         EXT4_EX_NOCACHE);
5349                 if (IS_ERR(path))
5350                         return PTR_ERR(path);
5351                 depth = path->p_depth;
5352                 extent =  path[depth].p_ext;
5353                 if (extent) {
5354                         ex_start = le32_to_cpu(extent->ee_block);
5355                         ex_end = le32_to_cpu(extent->ee_block) +
5356                                 ext4_ext_get_actual_len(extent);
5357                 } else {
5358                         ex_start = 0;
5359                         ex_end = 0;
5360                 }
5361 
5362                 if ((start == ex_start && shift > ex_start) ||
5363                     (shift > start - ex_end)) {
5364                         ret = -EINVAL;
5365                         goto out;
5366                 }
5367         } else {
5368                 if (shift > EXT_MAX_BLOCKS -
5369                     (stop + ext4_ext_get_actual_len(extent))) {
5370                         ret = -EINVAL;
5371                         goto out;
5372                 }
5373         }
5374 
5375         /*
5376          * In case of left shift, iterator points to start and it is increased
5377          * till we reach stop. In case of right shift, iterator points to stop
5378          * and it is decreased till we reach start.
5379          */
5380         if (SHIFT == SHIFT_LEFT)
5381                 iterator = &start;
5382         else
5383                 iterator = &stop;
5384 
5385         /*
5386          * Its safe to start updating extents.  Start and stop are unsigned, so
5387          * in case of right shift if extent with 0 block is reached, iterator
5388          * becomes NULL to indicate the end of the loop.
5389          */
5390         while (iterator && start <= stop) {
5391                 path = ext4_find_extent(inode, *iterator, &path,
5392                                         EXT4_EX_NOCACHE);
5393                 if (IS_ERR(path))
5394                         return PTR_ERR(path);
5395                 depth = path->p_depth;
5396                 extent = path[depth].p_ext;
5397                 if (!extent) {
5398                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5399                                          (unsigned long) *iterator);
5400                         return -EFSCORRUPTED;
5401                 }
5402                 if (SHIFT == SHIFT_LEFT && *iterator >
5403                     le32_to_cpu(extent->ee_block)) {
5404                         /* Hole, move to the next extent */
5405                         if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5406                                 path[depth].p_ext++;
5407                         } else {
5408                                 *iterator = ext4_ext_next_allocated_block(path);
5409                                 continue;
5410                         }
5411                 }
5412 
5413                 if (SHIFT == SHIFT_LEFT) {
5414                         extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5415                         *iterator = le32_to_cpu(extent->ee_block) +
5416                                         ext4_ext_get_actual_len(extent);
5417                 } else {
5418                         extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5419                         if (le32_to_cpu(extent->ee_block) > 0)
5420                                 *iterator = le32_to_cpu(extent->ee_block) - 1;
5421                         else
5422                                 /* Beginning is reached, end of the loop */
5423                                 iterator = NULL;
5424                         /* Update path extent in case we need to stop */
5425                         while (le32_to_cpu(extent->ee_block) < start)
5426                                 extent++;
5427                         path[depth].p_ext = extent;
5428                 }
5429                 ret = ext4_ext_shift_path_extents(path, shift, inode,
5430                                 handle, SHIFT);
5431                 if (ret)
5432                         break;
5433         }
5434 out:
5435         ext4_ext_drop_refs(path);
5436         kfree(path);
5437         return ret;
5438 }
5439 
5440 /*
5441  * ext4_collapse_range:
5442  * This implements the fallocate's collapse range functionality for ext4
5443  * Returns: 0 and non-zero on error.
5444  */
5445 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5446 {
5447         struct super_block *sb = inode->i_sb;
5448         ext4_lblk_t punch_start, punch_stop;
5449         handle_t *handle;
5450         unsigned int credits;
5451         loff_t new_size, ioffset;
5452         int ret;
5453 
5454         /*
5455          * We need to test this early because xfstests assumes that a
5456          * collapse range of (0, 1) will return EOPNOTSUPP if the file
5457          * system does not support collapse range.
5458          */
5459         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5460                 return -EOPNOTSUPP;
5461 
5462         /* Collapse range works only on fs block size aligned offsets. */
5463         if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5464             len & (EXT4_CLUSTER_SIZE(sb) - 1))
5465                 return -EINVAL;
5466 
5467         if (!S_ISREG(inode->i_mode))
5468                 return -EINVAL;
5469 
5470         trace_ext4_collapse_range(inode, offset, len);
5471 
5472         punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5473         punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5474 
5475         /* Call ext4_force_commit to flush all data in case of data=journal. */
5476         if (ext4_should_journal_data(inode)) {
5477                 ret = ext4_force_commit(inode->i_sb);
5478                 if (ret)
5479                         return ret;
5480         }
5481 
5482         inode_lock(inode);
5483         /*
5484          * There is no need to overlap collapse range with EOF, in which case
5485          * it is effectively a truncate operation
5486          */
5487         if (offset + len >= i_size_read(inode)) {
5488                 ret = -EINVAL;
5489                 goto out_mutex;
5490         }
5491 
5492         /* Currently just for extent based files */
5493         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5494                 ret = -EOPNOTSUPP;
5495                 goto out_mutex;
5496         }
5497 
5498         /* Wait for existing dio to complete */
5499         inode_dio_wait(inode);
5500 
5501         /*
5502          * Prevent page faults from reinstantiating pages we have released from
5503          * page cache.
5504          */
5505         down_write(&EXT4_I(inode)->i_mmap_sem);
5506 
5507         ret = ext4_break_layouts(inode);
5508         if (ret)
5509                 goto out_mmap;
5510 
5511         /*
5512          * Need to round down offset to be aligned with page size boundary
5513          * for page size > block size.
5514          */
5515         ioffset = round_down(offset, PAGE_SIZE);
5516         /*
5517          * Write tail of the last page before removed range since it will get
5518          * removed from the page cache below.
5519          */
5520         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5521         if (ret)
5522                 goto out_mmap;
5523         /*
5524          * Write data that will be shifted to preserve them when discarding
5525          * page cache below. We are also protected from pages becoming dirty
5526          * by i_mmap_sem.
5527          */
5528         ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5529                                            LLONG_MAX);
5530         if (ret)
5531                 goto out_mmap;
5532         truncate_pagecache(inode, ioffset);
5533 
5534         credits = ext4_writepage_trans_blocks(inode);
5535         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5536         if (IS_ERR(handle)) {
5537                 ret = PTR_ERR(handle);
5538                 goto out_mmap;
5539         }
5540 
5541         down_write(&EXT4_I(inode)->i_data_sem);
5542         ext4_discard_preallocations(inode);
5543 
5544         ret = ext4_es_remove_extent(inode, punch_start,
5545                                     EXT_MAX_BLOCKS - punch_start);
5546         if (ret) {
5547                 up_write(&EXT4_I(inode)->i_data_sem);
5548                 goto out_stop;
5549         }
5550 
5551         ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5552         if (ret) {
5553                 up_write(&EXT4_I(inode)->i_data_sem);
5554                 goto out_stop;
5555         }
5556         ext4_discard_preallocations(inode);
5557 
5558         ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5559                                      punch_stop - punch_start, SHIFT_LEFT);
5560         if (ret) {
5561                 up_write(&EXT4_I(inode)->i_data_sem);
5562                 goto out_stop;
5563         }
5564 
5565         new_size = i_size_read(inode) - len;
5566         i_size_write(inode, new_size);
5567         EXT4_I(inode)->i_disksize = new_size;
5568 
5569         up_write(&EXT4_I(inode)->i_data_sem);
5570         if (IS_SYNC(inode))
5571                 ext4_handle_sync(handle);
5572         inode->i_mtime = inode->i_ctime = current_time(inode);
5573         ext4_mark_inode_dirty(handle, inode);
5574         ext4_update_inode_fsync_trans(handle, inode, 1);
5575 
5576 out_stop:
5577         ext4_journal_stop(handle);
5578 out_mmap:
5579         up_write(&EXT4_I(inode)->i_mmap_sem);
5580 out_mutex:
5581         inode_unlock(inode);
5582         return ret;
5583 }
5584 
5585 /*
5586  * ext4_insert_range:
5587  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5588  * The data blocks starting from @offset to the EOF are shifted by @len
5589  * towards right to create a hole in the @inode. Inode size is increased
5590  * by len bytes.
5591  * Returns 0 on success, error otherwise.
5592  */
5593 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5594 {
5595         struct super_block *sb = inode->i_sb;
5596         handle_t *handle;
5597         struct ext4_ext_path *path;
5598         struct ext4_extent *extent;
5599         ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5600         unsigned int credits, ee_len;
5601         int ret = 0, depth, split_flag = 0;
5602         loff_t ioffset;
5603 
5604         /*
5605          * We need to test this early because xfstests assumes that an
5606          * insert range of (0, 1) will return EOPNOTSUPP if the file
5607          * system does not support insert range.
5608          */
5609         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5610                 return -EOPNOTSUPP;
5611 
5612         /* Insert range works only on fs block size aligned offsets. */
5613         if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5614                         len & (EXT4_CLUSTER_SIZE(sb) - 1))
5615                 return -EINVAL;
5616 
5617         if (!S_ISREG(inode->i_mode))
5618                 return -EOPNOTSUPP;
5619 
5620         trace_ext4_insert_range(inode, offset, len);
5621 
5622         offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5623         len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5624 
5625         /* Call ext4_force_commit to flush all data in case of data=journal */
5626         if (ext4_should_journal_data(inode)) {
5627                 ret = ext4_force_commit(inode->i_sb);
5628                 if (ret)
5629                         return ret;
5630         }
5631 
5632         inode_lock(inode);
5633         /* Currently just for extent based files */
5634         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5635                 ret = -EOPNOTSUPP;
5636                 goto out_mutex;
5637         }
5638 
5639         /* Check for wrap through zero */
5640         if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5641                 ret = -EFBIG;
5642                 goto out_mutex;
5643         }
5644 
5645         /* Offset should be less than i_size */
5646         if (offset >= i_size_read(inode)) {
5647                 ret = -EINVAL;
5648                 goto out_mutex;
5649         }
5650 
5651         /* Wait for existing dio to complete */
5652         inode_dio_wait(inode);
5653 
5654         /*
5655          * Prevent page faults from reinstantiating pages we have released from
5656          * page cache.
5657          */
5658         down_write(&EXT4_I(inode)->i_mmap_sem);
5659 
5660         ret = ext4_break_layouts(inode);
5661         if (ret)
5662                 goto out_mmap;
5663 
5664         /*
5665          * Need to round down to align start offset to page size boundary
5666          * for page size > block size.
5667          */
5668         ioffset = round_down(offset, PAGE_SIZE);
5669         /* Write out all dirty pages */
5670         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5671                         LLONG_MAX);
5672         if (ret)
5673                 goto out_mmap;
5674         truncate_pagecache(inode, ioffset);
5675 
5676         credits = ext4_writepage_trans_blocks(inode);
5677         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5678         if (IS_ERR(handle)) {
5679                 ret = PTR_ERR(handle);
5680                 goto out_mmap;
5681         }
5682 
5683         /* Expand file to avoid data loss if there is error while shifting */
5684         inode->i_size += len;
5685         EXT4_I(inode)->i_disksize += len;
5686         inode->i_mtime = inode->i_ctime = current_time(inode);
5687         ret = ext4_mark_inode_dirty(handle, inode);
5688         if (ret)
5689                 goto out_stop;
5690 
5691         down_write(&EXT4_I(inode)->i_data_sem);
5692         ext4_discard_preallocations(inode);
5693 
5694         path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5695         if (IS_ERR(path)) {
5696                 up_write(&EXT4_I(inode)->i_data_sem);
5697                 goto out_stop;
5698         }
5699 
5700         depth = ext_depth(inode);
5701         extent = path[depth].p_ext;
5702         if (extent) {
5703                 ee_start_lblk = le32_to_cpu(extent->ee_block);
5704                 ee_len = ext4_ext_get_actual_len(extent);
5705 
5706                 /*
5707                  * If offset_lblk is not the starting block of extent, split
5708                  * the extent @offset_lblk
5709                  */
5710                 if ((offset_lblk > ee_start_lblk) &&
5711                                 (offset_lblk < (ee_start_lblk + ee_len))) {
5712                         if (ext4_ext_is_unwritten(extent))
5713                                 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5714                                         EXT4_EXT_MARK_UNWRIT2;
5715                         ret = ext4_split_extent_at(handle, inode, &path,
5716                                         offset_lblk, split_flag,
5717                                         EXT4_EX_NOCACHE |
5718                                         EXT4_GET_BLOCKS_PRE_IO |
5719                                         EXT4_GET_BLOCKS_METADATA_NOFAIL);
5720                 }
5721 
5722                 ext4_ext_drop_refs(path);
5723                 kfree(path);
5724                 if (ret < 0) {
5725                         up_write(&EXT4_I(inode)->i_data_sem);
5726                         goto out_stop;
5727                 }
5728         } else {
5729                 ext4_ext_drop_refs(path);
5730                 kfree(path);
5731         }
5732 
5733         ret = ext4_es_remove_extent(inode, offset_lblk,
5734                         EXT_MAX_BLOCKS - offset_lblk);
5735         if (ret) {
5736                 up_write(&EXT4_I(inode)->i_data_sem);
5737                 goto out_stop;
5738         }
5739 
5740         /*
5741          * if offset_lblk lies in a hole which is at start of file, use
5742          * ee_start_lblk to shift extents
5743          */
5744         ret = ext4_ext_shift_extents(inode, handle,
5745                 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5746                 len_lblk, SHIFT_RIGHT);
5747 
5748         up_write(&EXT4_I(inode)->i_data_sem);
5749         if (IS_SYNC(inode))
5750                 ext4_handle_sync(handle);
5751         if (ret >= 0)
5752                 ext4_update_inode_fsync_trans(handle, inode, 1);
5753 
5754 out_stop:
5755         ext4_journal_stop(handle);
5756 out_mmap:
5757         up_write(&EXT4_I(inode)->i_mmap_sem);
5758 out_mutex:
5759         inode_unlock(inode);
5760         return ret;
5761 }
5762 
5763 /**
5764  * ext4_swap_extents() - Swap extents between two inodes
5765  * @handle: handle for this transaction
5766  * @inode1:     First inode
5767  * @inode2:     Second inode
5768  * @lblk1:      Start block for first inode
5769  * @lblk2:      Start block for second inode
5770  * @count:      Number of blocks to swap
5771  * @unwritten: Mark second inode's extents as unwritten after swap
5772  * @erp:        Pointer to save error value
5773  *
5774  * This helper routine does exactly what is promise "swap extents". All other
5775  * stuff such as page-cache locking consistency, bh mapping consistency or
5776  * extent's data copying must be performed by caller.
5777  * Locking:
5778  *              i_mutex is held for both inodes
5779  *              i_data_sem is locked for write for both inodes
5780  * Assumptions:
5781  *              All pages from requested range are locked for both inodes
5782  */
5783 int
5784 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5785                   struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5786                   ext4_lblk_t count, int unwritten, int *erp)
5787 {
5788         struct ext4_ext_path *path1 = NULL;
5789         struct ext4_ext_path *path2 = NULL;
5790         int replaced_count = 0;
5791 
5792         BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5793         BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5794         BUG_ON(!inode_is_locked(inode1));
5795         BUG_ON(!inode_is_locked(inode2));
5796 
5797         *erp = ext4_es_remove_extent(inode1, lblk1, count);
5798         if (unlikely(*erp))
5799                 return 0;
5800         *erp = ext4_es_remove_extent(inode2, lblk2, count);
5801         if (unlikely(*erp))
5802                 return 0;
5803 
5804         while (count) {
5805                 struct ext4_extent *ex1, *ex2, tmp_ex;
5806                 ext4_lblk_t e1_blk, e2_blk;
5807                 int e1_len, e2_len, len;
5808                 int split = 0;
5809 
5810                 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5811                 if (IS_ERR(path1)) {
5812                         *erp = PTR_ERR(path1);
5813                         path1 = NULL;
5814                 finish:
5815                         count = 0;
5816                         goto repeat;
5817                 }
5818                 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5819                 if (IS_ERR(path2)) {
5820                         *erp = PTR_ERR(path2);
5821                         path2 = NULL;
5822                         goto finish;
5823                 }
5824                 ex1 = path1[path1->p_depth].p_ext;
5825                 ex2 = path2[path2->p_depth].p_ext;
5826                 /* Do we have somthing to swap ? */
5827                 if (unlikely(!ex2 || !ex1))
5828                         goto finish;
5829 
5830                 e1_blk = le32_to_cpu(ex1->ee_block);
5831                 e2_blk = le32_to_cpu(ex2->ee_block);
5832                 e1_len = ext4_ext_get_actual_len(ex1);
5833                 e2_len = ext4_ext_get_actual_len(ex2);
5834 
5835                 /* Hole handling */
5836                 if (!in_range(lblk1, e1_blk, e1_len) ||
5837                     !in_range(lblk2, e2_blk, e2_len)) {
5838                         ext4_lblk_t next1, next2;
5839 
5840                         /* if hole after extent, then go to next extent */
5841                         next1 = ext4_ext_next_allocated_block(path1);
5842                         next2 = ext4_ext_next_allocated_block(path2);
5843                         /* If hole before extent, then shift to that extent */
5844                         if (e1_blk > lblk1)
5845                                 next1 = e1_blk;
5846                         if (e2_blk > lblk2)
5847                                 next2 = e2_blk;
5848                         /* Do we have something to swap */
5849                         if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5850                                 goto finish;
5851                         /* Move to the rightest boundary */
5852                         len = next1 - lblk1;
5853                         if (len < next2 - lblk2)
5854                                 len = next2 - lblk2;
5855                         if (len > count)
5856                                 len = count;
5857                         lblk1 += len;
5858                         lblk2 += len;
5859                         count -= len;
5860                         goto repeat;
5861                 }
5862 
5863                 /* Prepare left boundary */
5864                 if (e1_blk < lblk1) {
5865                         split = 1;
5866                         *erp = ext4_force_split_extent_at(handle, inode1,
5867                                                 &path1, lblk1, 0);
5868                         if (unlikely(*erp))
5869                                 goto finish;
5870                 }
5871                 if (e2_blk < lblk2) {
5872                         split = 1;
5873                         *erp = ext4_force_split_extent_at(handle, inode2,
5874                                                 &path2,  lblk2, 0);
5875                         if (unlikely(*erp))
5876                                 goto finish;
5877                 }
5878                 /* ext4_split_extent_at() may result in leaf extent split,
5879                  * path must to be revalidated. */
5880                 if (split)
5881                         goto repeat;
5882 
5883                 /* Prepare right boundary */
5884                 len = count;
5885                 if (len > e1_blk + e1_len - lblk1)
5886                         len = e1_blk + e1_len - lblk1;
5887                 if (len > e2_blk + e2_len - lblk2)
5888                         len = e2_blk + e2_len - lblk2;
5889 
5890                 if (len != e1_len) {
5891                         split = 1;
5892                         *erp = ext4_force_split_extent_at(handle, inode1,
5893                                                 &path1, lblk1 + len, 0);
5894                         if (unlikely(*erp))
5895                                 goto finish;
5896                 }
5897                 if (len != e2_len) {
5898                         split = 1;
5899                         *erp = ext4_force_split_extent_at(handle, inode2,
5900                                                 &path2, lblk2 + len, 0);
5901                         if (*erp)
5902                                 goto finish;
5903                 }
5904                 /* ext4_split_extent_at() may result in leaf extent split,
5905                  * path must to be revalidated. */
5906                 if (split)
5907                         goto repeat;
5908 
5909                 BUG_ON(e2_len != e1_len);
5910                 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5911                 if (unlikely(*erp))
5912                         goto finish;
5913                 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5914                 if (unlikely(*erp))
5915                         goto finish;
5916 
5917                 /* Both extents are fully inside boundaries. Swap it now */
5918                 tmp_ex = *ex1;
5919                 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5920                 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5921                 ex1->ee_len = cpu_to_le16(e2_len);
5922                 ex2->ee_len = cpu_to_le16(e1_len);
5923                 if (unwritten)
5924                         ext4_ext_mark_unwritten(ex2);
5925                 if (ext4_ext_is_unwritten(&tmp_ex))
5926                         ext4_ext_mark_unwritten(ex1);
5927 
5928                 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5929                 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5930                 *erp = ext4_ext_dirty(handle, inode2, path2 +
5931                                       path2->p_depth);
5932                 if (unlikely(*erp))
5933                         goto finish;
5934                 *erp = ext4_ext_dirty(handle, inode1, path1 +
5935                                       path1->p_depth);
5936                 /*
5937                  * Looks scarry ah..? second inode already points to new blocks,
5938                  * and it was successfully dirtied. But luckily error may happen
5939                  * only due to journal error, so full transaction will be
5940                  * aborted anyway.
5941                  */
5942                 if (unlikely(*erp))
5943                         goto finish;
5944                 lblk1 += len;
5945                 lblk2 += len;
5946                 replaced_count += len;
5947                 count -= len;
5948 
5949         repeat:
5950                 ext4_ext_drop_refs(path1);
5951                 kfree(path1);
5952                 ext4_ext_drop_refs(path2);
5953                 kfree(path2);
5954                 path1 = path2 = NULL;
5955         }
5956         return replaced_count;
5957 }
5958 
5959 /*
5960  * ext4_clu_mapped - determine whether any block in a logical cluster has
5961  *                   been mapped to a physical cluster
5962  *
5963  * @inode - file containing the logical cluster
5964  * @lclu - logical cluster of interest
5965  *
5966  * Returns 1 if any block in the logical cluster is mapped, signifying
5967  * that a physical cluster has been allocated for it.  Otherwise,
5968  * returns 0.  Can also return negative error codes.  Derived from
5969  * ext4_ext_map_blocks().
5970  */
5971 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5972 {
5973         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5974         struct ext4_ext_path *path;
5975         int depth, mapped = 0, err = 0;
5976         struct ext4_extent *extent;
5977         ext4_lblk_t first_lblk, first_lclu, last_lclu;
5978 
5979         /* search for the extent closest to the first block in the cluster */
5980         path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5981         if (IS_ERR(path)) {
5982                 err = PTR_ERR(path);
5983                 path = NULL;
5984                 goto out;
5985         }
5986 
5987         depth = ext_depth(inode);
5988 
5989         /*
5990          * A consistent leaf must not be empty.  This situation is possible,
5991          * though, _during_ tree modification, and it's why an assert can't
5992          * be put in ext4_find_extent().
5993          */
5994         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5995                 EXT4_ERROR_INODE(inode,
5996                     "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5997                                  (unsigned long) EXT4_C2B(sbi, lclu),
5998                                  depth, path[depth].p_block);
5999                 err = -EFSCORRUPTED;
6000                 goto out;
6001         }
6002 
6003         extent = path[depth].p_ext;
6004 
6005         /* can't be mapped if the extent tree is empty */
6006         if (extent == NULL)
6007                 goto out;
6008 
6009         first_lblk = le32_to_cpu(extent->ee_block);
6010         first_lclu = EXT4_B2C(sbi, first_lblk);
6011 
6012         /*
6013          * Three possible outcomes at this point - found extent spanning
6014          * the target cluster, to the left of the target cluster, or to the
6015          * right of the target cluster.  The first two cases are handled here.
6016          * The last case indicates the target cluster is not mapped.
6017          */
6018         if (lclu >= first_lclu) {
6019                 last_lclu = EXT4_B2C(sbi, first_lblk +
6020                                      ext4_ext_get_actual_len(extent) - 1);
6021                 if (lclu <= last_lclu) {
6022                         mapped = 1;
6023                 } else {
6024                         first_lblk = ext4_ext_next_allocated_block(path);
6025                         first_lclu = EXT4_B2C(sbi, first_lblk);
6026                         if (lclu == first_lclu)
6027                                 mapped = 1;
6028                 }
6029         }
6030 
6031 out:
6032         ext4_ext_drop_refs(path);
6033         kfree(path);
6034 
6035         return err ? err : mapped;
6036 }