fs/ext4/extents.c

   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 <linux/iomap.h>
  32 #include "ext4_jbd2.h"
  33 #include "ext4_extents.h"
  34 #include "xattr.h"
  35
  36 #include <trace/events/ext4.h>
  37
  38 /*
  39  * used by extent splitting.
  40  */
  41 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  42                                         due to ENOSPC */
  43 #define EXT4_EXT_MARK_UNWRIT1   0x2  /* mark first half unwritten */
  44 #define EXT4_EXT_MARK_UNWRIT2   0x4  /* mark second half unwritten */
  45
  46 #define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  47 #define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  48
  49 static __le32 ext4_extent_block_csum(struct inode *inode,
  50                                      struct ext4_extent_header *eh)
  51 {
  52         struct ext4_inode_info *ei = EXT4_I(inode);
  53         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  54         __u32 csum;
  55
  56         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  57                            EXT4_EXTENT_TAIL_OFFSET(eh));
  58         return cpu_to_le32(csum);
  59 }
  60
  61 static int ext4_extent_block_csum_verify(struct inode *inode,
  62                                          struct ext4_extent_header *eh)
  63 {
  64         struct ext4_extent_tail *et;
  65
  66         if (!ext4_has_metadata_csum(inode->i_sb))
  67                 return 1;
  68
  69         et = find_ext4_extent_tail(eh);
  70         if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  71                 return 0;
  72         return 1;
  73 }
  74
  75 static void ext4_extent_block_csum_set(struct inode *inode,
  76                                        struct ext4_extent_header *eh)
  77 {
  78         struct ext4_extent_tail *et;
  79
  80         if (!ext4_has_metadata_csum(inode->i_sb))
  81                 return;
  82
  83         et = find_ext4_extent_tail(eh);
  84         et->et_checksum = ext4_extent_block_csum(inode, eh);
  85 }
  86
  87 static int ext4_split_extent_at(handle_t *handle,
  88                              struct inode *inode,
  89                              struct ext4_ext_path **ppath,
  90                              ext4_lblk_t split,
  91                              int split_flag,
  92                              int flags);
  93
  94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
  95 {
  96         /*
  97          * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
  98          * moment, get_block can be called only for blocks inside i_size since
  99          * page cache has been already dropped and writes are blocked by
 100          * i_mutex. So we can safely drop the i_data_sem here.
 101          */
 102         BUG_ON(EXT4_JOURNAL(inode) == NULL);
 103         ext4_discard_preallocations(inode, 0);
 104         up_write(&EXT4_I(inode)->i_data_sem);
 105         *dropped = 1;
 106         return 0;
 107 }
 108
 109 /*
 110  * Make sure 'handle' has at least 'check_cred' credits. If not, restart
 111  * transaction with 'restart_cred' credits. The function drops i_data_sem
 112  * when restarting transaction and gets it after transaction is restarted.
 113  *
 114  * The function returns 0 on success, 1 if transaction had to be restarted,
 115  * and < 0 in case of fatal error.
 116  */
 117 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
 118                                 int check_cred, int restart_cred,
 119                                 int revoke_cred)
 120 {
 121         int ret;
 122         int dropped = 0;
 123
 124         ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
 125                 revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
 126         if (dropped)
 127                 down_write(&EXT4_I(inode)->i_data_sem);
 128         return ret;
 129 }
 130
 131 /*
 132  * could return:
 133  *  - EROFS
 134  *  - ENOMEM
 135  */
 136 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 137                                 struct ext4_ext_path *path)
 138 {
 139         int err = 0;
 140
 141         if (path->p_bh) {
 142                 /* path points to block */
 143                 BUFFER_TRACE(path->p_bh, "get_write_access");
 144                 err = ext4_journal_get_write_access(handle, path->p_bh);
 145                 /*
 146                  * The extent buffer's verified bit will be set again in
 147                  * __ext4_ext_dirty(). We could leave an inconsistent
 148                  * buffer if the extents updating procudure break off du
 149                  * to some error happens, force to check it again.
 150                  */
 151                 if (!err)
 152                         clear_buffer_verified(path->p_bh);
 153         }
 154         /* path points to leaf/index in inode body */
 155         /* we use in-core data, no need to protect them */
 156         return err;
 157 }
 158
 159 /*
 160  * could return:
 161  *  - EROFS
 162  *  - ENOMEM
 163  *  - EIO
 164  */
 165 static int __ext4_ext_dirty(const char *where, unsigned int line,
 166                             handle_t *handle, struct inode *inode,
 167                             struct ext4_ext_path *path)
 168 {
 169         int err;
 170
 171         WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 172         if (path->p_bh) {
 173                 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 174                 /* path points to block */
 175                 err = __ext4_handle_dirty_metadata(where, line, handle,
 176                                                    inode, path->p_bh);
 177                 /* Extents updating done, re-set verified flag */
 178                 if (!err)
 179                         set_buffer_verified(path->p_bh);
 180         } else {
 181                 /* path points to leaf/index in inode body */
 182                 err = ext4_mark_inode_dirty(handle, inode);
 183         }
 184         return err;
 185 }
 186
 187 #define ext4_ext_dirty(handle, inode, path) \
 188                 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 189
 190 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 191                               struct ext4_ext_path *path,
 192                               ext4_lblk_t block)
 193 {
 194         if (path) {
 195                 int depth = path->p_depth;
 196                 struct ext4_extent *ex;
 197
 198                 /*
 199                  * Try to predict block placement assuming that we are
 200                  * filling in a file which will eventually be
 201                  * non-sparse --- i.e., in the case of libbfd writing
 202                  * an ELF object sections out-of-order but in a way
 203                  * the eventually results in a contiguous object or
 204                  * executable file, or some database extending a table
 205                  * space file.  However, this is actually somewhat
 206                  * non-ideal if we are writing a sparse file such as
 207                  * qemu or KVM writing a raw image file that is going
 208                  * to stay fairly sparse, since it will end up
 209                  * fragmenting the file system's free space.  Maybe we
 210                  * should have some hueristics or some way to allow
 211                  * userspace to pass a hint to file system,
 212                  * especially if the latter case turns out to be
 213                  * common.
 214                  */
 215                 ex = path[depth].p_ext;
 216                 if (ex) {
 217                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 218                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 219
 220                         if (block > ext_block)
 221                                 return ext_pblk + (block - ext_block);
 222                         else
 223                                 return ext_pblk - (ext_block - block);
 224                 }
 225
 226                 /* it looks like index is empty;
 227                  * try to find starting block from index itself */
 228                 if (path[depth].p_bh)
 229                         return path[depth].p_bh->b_blocknr;
 230         }
 231
 232         /* OK. use inode's group */
 233         return ext4_inode_to_goal_block(inode);
 234 }
 235
 236 /*
 237  * Allocation for a meta data block
 238  */
 239 static ext4_fsblk_t
 240 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 241                         struct ext4_ext_path *path,
 242                         struct ext4_extent *ex, int *err, unsigned int flags)
 243 {
 244         ext4_fsblk_t goal, newblock;
 245
 246         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 247         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 248                                         NULL, err);
 249         return newblock;
 250 }
 251
 252 static inline int ext4_ext_space_block(struct inode *inode, int check)
 253 {
 254         int size;
 255
 256         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 257                         / sizeof(struct ext4_extent);
 258 #ifdef AGGRESSIVE_TEST
 259         if (!check && size > 6)
 260                 size = 6;
 261 #endif
 262         return size;
 263 }
 264
 265 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 266 {
 267         int size;
 268
 269         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 270                         / sizeof(struct ext4_extent_idx);
 271 #ifdef AGGRESSIVE_TEST
 272         if (!check && size > 5)
 273                 size = 5;
 274 #endif
 275         return size;
 276 }
 277
 278 static inline int ext4_ext_space_root(struct inode *inode, int check)
 279 {
 280         int size;
 281
 282         size = sizeof(EXT4_I(inode)->i_data);
 283         size -= sizeof(struct ext4_extent_header);
 284         size /= sizeof(struct ext4_extent);
 285 #ifdef AGGRESSIVE_TEST
 286         if (!check && size > 3)
 287                 size = 3;
 288 #endif
 289         return size;
 290 }
 291
 292 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 293 {
 294         int size;
 295
 296         size = sizeof(EXT4_I(inode)->i_data);
 297         size -= sizeof(struct ext4_extent_header);
 298         size /= sizeof(struct ext4_extent_idx);
 299 #ifdef AGGRESSIVE_TEST
 300         if (!check && size > 4)
 301                 size = 4;
 302 #endif
 303         return size;
 304 }
 305
 306 static inline int
 307 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 308                            struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 309                            int nofail)
 310 {
 311         struct ext4_ext_path *path = *ppath;
 312         int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 313         int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
 314
 315         if (nofail)
 316                 flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
 317
 318         return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 319                         EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 320                         flags);
 321 }
 322
 323 static int
 324 ext4_ext_max_entries(struct inode *inode, int depth)
 325 {
 326         int max;
 327
 328         if (depth == ext_depth(inode)) {
 329                 if (depth == 0)
 330                         max = ext4_ext_space_root(inode, 1);
 331                 else
 332                         max = ext4_ext_space_root_idx(inode, 1);
 333         } else {
 334                 if (depth == 0)
 335                         max = ext4_ext_space_block(inode, 1);
 336                 else
 337                         max = ext4_ext_space_block_idx(inode, 1);
 338         }
 339
 340         return max;
 341 }
 342
 343 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 344 {
 345         ext4_fsblk_t block = ext4_ext_pblock(ext);
 346         int len = ext4_ext_get_actual_len(ext);
 347         ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 348
 349         /*
 350          * We allow neither:
 351          *  - zero length
 352          *  - overflow/wrap-around
 353          */
 354         if (lblock + len <= lblock)
 355                 return 0;
 356         return ext4_inode_block_valid(inode, block, len);
 357 }
 358
 359 static int ext4_valid_extent_idx(struct inode *inode,
 360                                 struct ext4_extent_idx *ext_idx)
 361 {
 362         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 363
 364         return ext4_inode_block_valid(inode, block, 1);
 365 }
 366
 367 static int ext4_valid_extent_entries(struct inode *inode,
 368                                      struct ext4_extent_header *eh,
 369                                      ext4_lblk_t lblk, ext4_fsblk_t *pblk,
 370                                      int depth)
 371 {
 372         unsigned short entries;
 373         ext4_lblk_t lblock = 0;
 374         ext4_lblk_t cur = 0;
 375
 376         if (eh->eh_entries == 0)
 377                 return 1;
 378
 379         entries = le16_to_cpu(eh->eh_entries);
 380
 381         if (depth == 0) {
 382                 /* leaf entries */
 383                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 384
 385                 /*
 386                  * The logical block in the first entry should equal to
 387                  * the number in the index block.
 388                  */
 389                 if (depth != ext_depth(inode) &&
 390                     lblk != le32_to_cpu(ext->ee_block))
 391                         return 0;
 392                 while (entries) {
 393                         if (!ext4_valid_extent(inode, ext))
 394                                 return 0;
 395
 396                         /* Check for overlapping extents */
 397                         lblock = le32_to_cpu(ext->ee_block);
 398                         if (lblock < cur) {
 399                                 *pblk = ext4_ext_pblock(ext);
 400                                 return 0;
 401                         }
 402                         cur = lblock + ext4_ext_get_actual_len(ext);
 403                         ext++;
 404                         entries--;
 405                 }
 406         } else {
 407                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 408
 409                 /*
 410                  * The logical block in the first entry should equal to
 411                  * the number in the parent index block.
 412                  */
 413                 if (depth != ext_depth(inode) &&
 414                     lblk != le32_to_cpu(ext_idx->ei_block))
 415                         return 0;
 416                 while (entries) {
 417                         if (!ext4_valid_extent_idx(inode, ext_idx))
 418                                 return 0;
 419
 420                         /* Check for overlapping index extents */
 421                         lblock = le32_to_cpu(ext_idx->ei_block);
 422                         if (lblock < cur) {
 423                                 *pblk = ext4_idx_pblock(ext_idx);
 424                                 return 0;
 425                         }
 426                         ext_idx++;
 427                         entries--;
 428                         cur = lblock + 1;
 429                 }
 430         }
 431         return 1;
 432 }
 433
 434 static int __ext4_ext_check(const char *function, unsigned int line,
 435                             struct inode *inode, struct ext4_extent_header *eh,
 436                             int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
 437 {
 438         const char *error_msg;
 439         int max = 0, err = -EFSCORRUPTED;
 440
 441         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 442                 error_msg = "invalid magic";
 443                 goto corrupted;
 444         }
 445         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 446                 error_msg = "unexpected eh_depth";
 447                 goto corrupted;
 448         }
 449         if (unlikely(eh->eh_max == 0)) {
 450                 error_msg = "invalid eh_max";
 451                 goto corrupted;
 452         }
 453         max = ext4_ext_max_entries(inode, depth);
 454         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 455                 error_msg = "too large eh_max";
 456                 goto corrupted;
 457         }
 458         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 459                 error_msg = "invalid eh_entries";
 460                 goto corrupted;
 461         }
 462         if (unlikely((eh->eh_entries == 0) && (depth > 0))) {
 463                 error_msg = "eh_entries is 0 but eh_depth is > 0";
 464                 goto corrupted;
 465         }
 466         if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
 467                 error_msg = "invalid extent entries";
 468                 goto corrupted;
 469         }
 470         if (unlikely(depth > 32)) {
 471                 error_msg = "too large eh_depth";
 472                 goto corrupted;
 473         }
 474         /* Verify checksum on non-root extent tree nodes */
 475         if (ext_depth(inode) != depth &&
 476             !ext4_extent_block_csum_verify(inode, eh)) {
 477                 error_msg = "extent tree corrupted";
 478                 err = -EFSBADCRC;
 479                 goto corrupted;
 480         }
 481         return 0;
 482
 483 corrupted:
 484         ext4_error_inode_err(inode, function, line, 0, -err,
 485                              "pblk %llu bad header/extent: %s - magic %x, "
 486                              "entries %u, max %u(%u), depth %u(%u)",
 487                              (unsigned long long) pblk, error_msg,
 488                              le16_to_cpu(eh->eh_magic),
 489                              le16_to_cpu(eh->eh_entries),
 490                              le16_to_cpu(eh->eh_max),
 491                              max, le16_to_cpu(eh->eh_depth), depth);
 492         return err;
 493 }
 494
 495 #define ext4_ext_check(inode, eh, depth, pblk)                  \
 496         __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
 497
 498 int ext4_ext_check_inode(struct inode *inode)
 499 {
 500         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 501 }
 502
 503 static void ext4_cache_extents(struct inode *inode,
 504                                struct ext4_extent_header *eh)
 505 {
 506         struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 507         ext4_lblk_t prev = 0;
 508         int i;
 509
 510         for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 511                 unsigned int status = EXTENT_STATUS_WRITTEN;
 512                 ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 513                 int len = ext4_ext_get_actual_len(ex);
 514
 515                 if (prev && (prev != lblk))
 516                         ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
 517                                              EXTENT_STATUS_HOLE);
 518
 519                 if (ext4_ext_is_unwritten(ex))
 520                         status = EXTENT_STATUS_UNWRITTEN;
 521                 ext4_es_cache_extent(inode, lblk, len,
 522                                      ext4_ext_pblock(ex), status);
 523                 prev = lblk + len;
 524         }
 525 }
 526
 527 static struct buffer_head *
 528 __read_extent_tree_block(const char *function, unsigned int line,
 529                          struct inode *inode, struct ext4_extent_idx *idx,
 530                          int depth, int flags)
 531 {
 532         struct buffer_head              *bh;
 533         int                             err;
 534         gfp_t                           gfp_flags = __GFP_MOVABLE | GFP_NOFS;
 535         ext4_fsblk_t                    pblk;
 536
 537         if (flags & EXT4_EX_NOFAIL)
 538                 gfp_flags |= __GFP_NOFAIL;
 539
 540         pblk = ext4_idx_pblock(idx);
 541         bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
 542         if (unlikely(!bh))
 543                 return ERR_PTR(-ENOMEM);
 544
 545         if (!bh_uptodate_or_lock(bh)) {
 546                 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 547                 err = ext4_read_bh(bh, 0, NULL);
 548                 if (err < 0)
 549                         goto errout;
 550         }
 551         if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 552                 return bh;
 553         err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
 554                                depth, pblk, le32_to_cpu(idx->ei_block));
 555         if (err)
 556                 goto errout;
 557         set_buffer_verified(bh);
 558         /*
 559          * If this is a leaf block, cache all of its entries
 560          */
 561         if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 562                 struct ext4_extent_header *eh = ext_block_hdr(bh);
 563                 ext4_cache_extents(inode, eh);
 564         }
 565         return bh;
 566 errout:
 567         put_bh(bh);
 568         return ERR_PTR(err);
 569
 570 }
 571
 572 #define read_extent_tree_block(inode, idx, depth, flags)                \
 573         __read_extent_tree_block(__func__, __LINE__, (inode), (idx),    \
 574                                  (depth), (flags))
 575
 576 /*
 577  * This function is called to cache a file's extent information in the
 578  * extent status tree
 579  */
 580 int ext4_ext_precache(struct inode *inode)
 581 {
 582         struct ext4_inode_info *ei = EXT4_I(inode);
 583         struct ext4_ext_path *path = NULL;
 584         struct buffer_head *bh;
 585         int i = 0, depth, ret = 0;
 586
 587         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 588                 return 0;       /* not an extent-mapped inode */
 589
 590         down_read(&ei->i_data_sem);
 591         depth = ext_depth(inode);
 592
 593         /* Don't cache anything if there are no external extent blocks */
 594         if (!depth) {
 595                 up_read(&ei->i_data_sem);
 596                 return ret;
 597         }
 598
 599         path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
 600                        GFP_NOFS);
 601         if (path == NULL) {
 602                 up_read(&ei->i_data_sem);
 603                 return -ENOMEM;
 604         }
 605
 606         path[0].p_hdr = ext_inode_hdr(inode);
 607         ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 608         if (ret)
 609                 goto out;
 610         path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 611         while (i >= 0) {
 612                 /*
 613                  * If this is a leaf block or we've reached the end of
 614                  * the index block, go up
 615                  */
 616                 if ((i == depth) ||
 617                     path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 618                         brelse(path[i].p_bh);
 619                         path[i].p_bh = NULL;
 620                         i--;
 621                         continue;
 622                 }
 623                 bh = read_extent_tree_block(inode, path[i].p_idx++,
 624                                             depth - i - 1,
 625                                             EXT4_EX_FORCE_CACHE);
 626                 if (IS_ERR(bh)) {
 627                         ret = PTR_ERR(bh);
 628                         break;
 629                 }
 630                 i++;
 631                 path[i].p_bh = bh;
 632                 path[i].p_hdr = ext_block_hdr(bh);
 633                 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 634         }
 635         ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 636 out:
 637         up_read(&ei->i_data_sem);
 638         ext4_ext_drop_refs(path);
 639         kfree(path);
 640         return ret;
 641 }
 642
 643 #ifdef EXT_DEBUG
 644 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 645 {
 646         int k, l = path->p_depth;
 647
 648         ext_debug(inode, "path:");
 649         for (k = 0; k <= l; k++, path++) {
 650                 if (path->p_idx) {
 651                         ext_debug(inode, "  %d->%llu",
 652                                   le32_to_cpu(path->p_idx->ei_block),
 653                                   ext4_idx_pblock(path->p_idx));
 654                 } else if (path->p_ext) {
 655                         ext_debug(inode, "  %d:[%d]%d:%llu ",
 656                                   le32_to_cpu(path->p_ext->ee_block),
 657                                   ext4_ext_is_unwritten(path->p_ext),
 658                                   ext4_ext_get_actual_len(path->p_ext),
 659                                   ext4_ext_pblock(path->p_ext));
 660                 } else
 661                         ext_debug(inode, "  []");
 662         }
 663         ext_debug(inode, "\n");
 664 }
 665
 666 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 667 {
 668         int depth = ext_depth(inode);
 669         struct ext4_extent_header *eh;
 670         struct ext4_extent *ex;
 671         int i;
 672
 673         if (!path)
 674                 return;
 675
 676         eh = path[depth].p_hdr;
 677         ex = EXT_FIRST_EXTENT(eh);
 678
 679         ext_debug(inode, "Displaying leaf extents\n");
 680
 681         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 682                 ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 683                           ext4_ext_is_unwritten(ex),
 684                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 685         }
 686         ext_debug(inode, "\n");
 687 }
 688
 689 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 690                         ext4_fsblk_t newblock, int level)
 691 {
 692         int depth = ext_depth(inode);
 693         struct ext4_extent *ex;
 694
 695         if (depth != level) {
 696                 struct ext4_extent_idx *idx;
 697                 idx = path[level].p_idx;
 698                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 699                         ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
 700                                   level, le32_to_cpu(idx->ei_block),
 701                                   ext4_idx_pblock(idx), newblock);
 702                         idx++;
 703                 }
 704
 705                 return;
 706         }
 707
 708         ex = path[depth].p_ext;
 709         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 710                 ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
 711                                 le32_to_cpu(ex->ee_block),
 712                                 ext4_ext_pblock(ex),
 713                                 ext4_ext_is_unwritten(ex),
 714                                 ext4_ext_get_actual_len(ex),
 715                                 newblock);
 716                 ex++;
 717         }
 718 }
 719
 720 #else
 721 #define ext4_ext_show_path(inode, path)
 722 #define ext4_ext_show_leaf(inode, path)
 723 #define ext4_ext_show_move(inode, path, newblock, level)
 724 #endif
 725
 726 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 727 {
 728         int depth, i;
 729
 730         if (!path)
 731                 return;
 732         depth = path->p_depth;
 733         for (i = 0; i <= depth; i++, path++) {
 734                 brelse(path->p_bh);
 735                 path->p_bh = NULL;
 736         }
 737 }
 738
 739 /*
 740  * ext4_ext_binsearch_idx:
 741  * binary search for the closest index of the given block
 742  * the header must be checked before calling this
 743  */
 744 static void
 745 ext4_ext_binsearch_idx(struct inode *inode,
 746                         struct ext4_ext_path *path, ext4_lblk_t block)
 747 {
 748         struct ext4_extent_header *eh = path->p_hdr;
 749         struct ext4_extent_idx *r, *l, *m;
 750
 751
 752         ext_debug(inode, "binsearch for %u(idx):  ", block);
 753
 754         l = EXT_FIRST_INDEX(eh) + 1;
 755         r = EXT_LAST_INDEX(eh);
 756         while (l <= r) {
 757                 m = l + (r - l) / 2;
 758                 if (block < le32_to_cpu(m->ei_block))
 759                         r = m - 1;
 760                 else
 761                         l = m + 1;
 762                 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 763                           le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
 764                           r, le32_to_cpu(r->ei_block));
 765         }
 766
 767         path->p_idx = l - 1;
 768         ext_debug(inode, "  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 769                   ext4_idx_pblock(path->p_idx));
 770
 771 #ifdef CHECK_BINSEARCH
 772         {
 773                 struct ext4_extent_idx *chix, *ix;
 774                 int k;
 775
 776                 chix = ix = EXT_FIRST_INDEX(eh);
 777                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 778                         if (k != 0 && le32_to_cpu(ix->ei_block) <=
 779                             le32_to_cpu(ix[-1].ei_block)) {
 780                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 781                                        "first=0x%p\n", k,
 782                                        ix, EXT_FIRST_INDEX(eh));
 783                                 printk(KERN_DEBUG "%u <= %u\n",
 784                                        le32_to_cpu(ix->ei_block),
 785                                        le32_to_cpu(ix[-1].ei_block));
 786                         }
 787                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 788                                            <= le32_to_cpu(ix[-1].ei_block));
 789                         if (block < le32_to_cpu(ix->ei_block))
 790                                 break;
 791                         chix = ix;
 792                 }
 793                 BUG_ON(chix != path->p_idx);
 794         }
 795 #endif
 796
 797 }
 798
 799 /*
 800  * ext4_ext_binsearch:
 801  * binary search for closest extent of the given block
 802  * the header must be checked before calling this
 803  */
 804 static void
 805 ext4_ext_binsearch(struct inode *inode,
 806                 struct ext4_ext_path *path, ext4_lblk_t block)
 807 {
 808         struct ext4_extent_header *eh = path->p_hdr;
 809         struct ext4_extent *r, *l, *m;
 810
 811         if (eh->eh_entries == 0) {
 812                 /*
 813                  * this leaf is empty:
 814                  * we get such a leaf in split/add case
 815                  */
 816                 return;
 817         }
 818
 819         ext_debug(inode, "binsearch for %u:  ", block);
 820
 821         l = EXT_FIRST_EXTENT(eh) + 1;
 822         r = EXT_LAST_EXTENT(eh);
 823
 824         while (l <= r) {
 825                 m = l + (r - l) / 2;
 826                 if (block < le32_to_cpu(m->ee_block))
 827                         r = m - 1;
 828                 else
 829                         l = m + 1;
 830                 ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 831                           le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
 832                           r, le32_to_cpu(r->ee_block));
 833         }
 834
 835         path->p_ext = l - 1;
 836         ext_debug(inode, "  -> %d:%llu:[%d]%d ",
 837                         le32_to_cpu(path->p_ext->ee_block),
 838                         ext4_ext_pblock(path->p_ext),
 839                         ext4_ext_is_unwritten(path->p_ext),
 840                         ext4_ext_get_actual_len(path->p_ext));
 841
 842 #ifdef CHECK_BINSEARCH
 843         {
 844                 struct ext4_extent *chex, *ex;
 845                 int k;
 846
 847                 chex = ex = EXT_FIRST_EXTENT(eh);
 848                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 849                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 850                                           <= le32_to_cpu(ex[-1].ee_block));
 851                         if (block < le32_to_cpu(ex->ee_block))
 852                                 break;
 853                         chex = ex;
 854                 }
 855                 BUG_ON(chex != path->p_ext);
 856         }
 857 #endif
 858
 859 }
 860
 861 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 862 {
 863         struct ext4_extent_header *eh;
 864
 865         eh = ext_inode_hdr(inode);
 866         eh->eh_depth = 0;
 867         eh->eh_entries = 0;
 868         eh->eh_magic = EXT4_EXT_MAGIC;
 869         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 870         eh->eh_generation = 0;
 871         ext4_mark_inode_dirty(handle, inode);
 872 }
 873
 874 struct ext4_ext_path *
 875 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 876                  struct ext4_ext_path **orig_path, int flags)
 877 {
 878         struct ext4_extent_header *eh;
 879         struct buffer_head *bh;
 880         struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 881         short int depth, i, ppos = 0;
 882         int ret;
 883         gfp_t gfp_flags = GFP_NOFS;
 884
 885         if (flags & EXT4_EX_NOFAIL)
 886                 gfp_flags |= __GFP_NOFAIL;
 887
 888         eh = ext_inode_hdr(inode);
 889         depth = ext_depth(inode);
 890         if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
 891                 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
 892                                  depth);
 893                 ret = -EFSCORRUPTED;
 894                 goto err;
 895         }
 896
 897         if (path) {
 898                 ext4_ext_drop_refs(path);
 899                 if (depth > path[0].p_maxdepth) {
 900                         kfree(path);
 901                         *orig_path = path = NULL;
 902                 }
 903         }
 904         if (!path) {
 905                 /* account possible depth increase */
 906                 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
 907                                 gfp_flags);
 908                 if (unlikely(!path))
 909                         return ERR_PTR(-ENOMEM);
 910                 path[0].p_maxdepth = depth + 1;
 911         }
 912         path[0].p_hdr = eh;
 913         path[0].p_bh = NULL;
 914
 915         i = depth;
 916         if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
 917                 ext4_cache_extents(inode, eh);
 918         /* walk through the tree */
 919         while (i) {
 920                 ext_debug(inode, "depth %d: num %d, max %d\n",
 921                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 922
 923                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 924                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 925                 path[ppos].p_depth = i;
 926                 path[ppos].p_ext = NULL;
 927
 928                 bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
 929                 if (IS_ERR(bh)) {
 930                         ret = PTR_ERR(bh);
 931                         goto err;
 932                 }
 933
 934                 eh = ext_block_hdr(bh);
 935                 ppos++;
 936                 path[ppos].p_bh = bh;
 937                 path[ppos].p_hdr = eh;
 938         }
 939
 940         path[ppos].p_depth = i;
 941         path[ppos].p_ext = NULL;
 942         path[ppos].p_idx = NULL;
 943
 944         /* find extent */
 945         ext4_ext_binsearch(inode, path + ppos, block);
 946         /* if not an empty leaf */
 947         if (path[ppos].p_ext)
 948                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 949
 950         ext4_ext_show_path(inode, path);
 951
 952         return path;
 953
 954 err:
 955         ext4_ext_drop_refs(path);
 956         kfree(path);
 957         if (orig_path)
 958                 *orig_path = NULL;
 959         return ERR_PTR(ret);
 960 }
 961
 962 /*
 963  * ext4_ext_insert_index:
 964  * insert new index [@logical;@ptr] into the block at @curp;
 965  * check where to insert: before @curp or after @curp
 966  */
 967 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 968                                  struct ext4_ext_path *curp,
 969                                  int logical, ext4_fsblk_t ptr)
 970 {
 971         struct ext4_extent_idx *ix;
 972         int len, err;
 973
 974         err = ext4_ext_get_access(handle, inode, curp);
 975         if (err)
 976                 return err;
 977
 978         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 979                 EXT4_ERROR_INODE(inode,
 980                                  "logical %d == ei_block %d!",
 981                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 982                 return -EFSCORRUPTED;
 983         }
 984
 985         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 986                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 987                 EXT4_ERROR_INODE(inode,
 988                                  "eh_entries %d >= eh_max %d!",
 989                                  le16_to_cpu(curp->p_hdr->eh_entries),
 990                                  le16_to_cpu(curp->p_hdr->eh_max));
 991                 return -EFSCORRUPTED;
 992         }
 993
 994         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 995                 /* insert after */
 996                 ext_debug(inode, "insert new index %d after: %llu\n",
 997                           logical, ptr);
 998                 ix = curp->p_idx + 1;
 999         } else {
1000                 /* insert before */
1001                 ext_debug(inode, "insert new index %d before: %llu\n",
1002                           logical, ptr);
1003                 ix = curp->p_idx;
1004         }
1005
1006         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1007         BUG_ON(len < 0);
1008         if (len > 0) {
1009                 ext_debug(inode, "insert new index %d: "
1010                                 "move %d indices from 0x%p to 0x%p\n",
1011                                 logical, len, ix, ix + 1);
1012                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1013         }
1014
1015         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1016                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1017                 return -EFSCORRUPTED;
1018         }
1019
1020         ix->ei_block = cpu_to_le32(logical);
1021         ext4_idx_store_pblock(ix, ptr);
1022         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1023
1024         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1025                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1026                 return -EFSCORRUPTED;
1027         }
1028
1029         err = ext4_ext_dirty(handle, inode, curp);
1030         ext4_std_error(inode->i_sb, err);
1031
1032         return err;
1033 }
1034
1035 /*
1036  * ext4_ext_split:
1037  * inserts new subtree into the path, using free index entry
1038  * at depth @at:
1039  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1040  * - makes decision where to split
1041  * - moves remaining extents and index entries (right to the split point)
1042  *   into the newly allocated blocks
1043  * - initializes subtree
1044  */
1045 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1046                           unsigned int flags,
1047                           struct ext4_ext_path *path,
1048                           struct ext4_extent *newext, int at)
1049 {
1050         struct buffer_head *bh = NULL;
1051         int depth = ext_depth(inode);
1052         struct ext4_extent_header *neh;
1053         struct ext4_extent_idx *fidx;
1054         int i = at, k, m, a;
1055         ext4_fsblk_t newblock, oldblock;
1056         __le32 border;
1057         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1058         gfp_t gfp_flags = GFP_NOFS;
1059         int err = 0;
1060         size_t ext_size = 0;
1061
1062         if (flags & EXT4_EX_NOFAIL)
1063                 gfp_flags |= __GFP_NOFAIL;
1064
1065         /* make decision: where to split? */
1066         /* FIXME: now decision is simplest: at current extent */
1067
1068         /* if current leaf will be split, then we should use
1069          * border from split point */
1070         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1071                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1072                 return -EFSCORRUPTED;
1073         }
1074         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1075                 border = path[depth].p_ext[1].ee_block;
1076                 ext_debug(inode, "leaf will be split."
1077                                 " next leaf starts at %d\n",
1078                                   le32_to_cpu(border));
1079         } else {
1080                 border = newext->ee_block;
1081                 ext_debug(inode, "leaf will be added."
1082                                 " next leaf starts at %d\n",
1083                                 le32_to_cpu(border));
1084         }
1085
1086         /*
1087          * If error occurs, then we break processing
1088          * and mark filesystem read-only. index won't
1089          * be inserted and tree will be in consistent
1090          * state. Next mount will repair buffers too.
1091          */
1092
1093         /*
1094          * Get array to track all allocated blocks.
1095          * We need this to handle errors and free blocks
1096          * upon them.
1097          */
1098         ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1099         if (!ablocks)
1100                 return -ENOMEM;
1101
1102         /* allocate all needed blocks */
1103         ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1104         for (a = 0; a < depth - at; a++) {
1105                 newblock = ext4_ext_new_meta_block(handle, inode, path,
1106                                                    newext, &err, flags);
1107                 if (newblock == 0)
1108                         goto cleanup;
1109                 ablocks[a] = newblock;
1110         }
1111
1112         /* initialize new leaf */
1113         newblock = ablocks[--a];
1114         if (unlikely(newblock == 0)) {
1115                 EXT4_ERROR_INODE(inode, "newblock == 0!");
1116                 err = -EFSCORRUPTED;
1117                 goto cleanup;
1118         }
1119         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1120         if (unlikely(!bh)) {
1121                 err = -ENOMEM;
1122                 goto cleanup;
1123         }
1124         lock_buffer(bh);
1125
1126         err = ext4_journal_get_create_access(handle, bh);
1127         if (err)
1128                 goto cleanup;
1129
1130         neh = ext_block_hdr(bh);
1131         neh->eh_entries = 0;
1132         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1133         neh->eh_magic = EXT4_EXT_MAGIC;
1134         neh->eh_depth = 0;
1135         neh->eh_generation = 0;
1136
1137         /* move remainder of path[depth] to the new leaf */
1138         if (unlikely(path[depth].p_hdr->eh_entries !=
1139                      path[depth].p_hdr->eh_max)) {
1140                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1141                                  path[depth].p_hdr->eh_entries,
1142                                  path[depth].p_hdr->eh_max);
1143                 err = -EFSCORRUPTED;
1144                 goto cleanup;
1145         }
1146         /* start copy from next extent */
1147         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1148         ext4_ext_show_move(inode, path, newblock, depth);
1149         if (m) {
1150                 struct ext4_extent *ex;
1151                 ex = EXT_FIRST_EXTENT(neh);
1152                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1153                 le16_add_cpu(&neh->eh_entries, m);
1154         }
1155
1156         /* zero out unused area in the extent block */
1157         ext_size = sizeof(struct ext4_extent_header) +
1158                 sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1159         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1160         ext4_extent_block_csum_set(inode, neh);
1161         set_buffer_uptodate(bh);
1162         unlock_buffer(bh);
1163
1164         err = ext4_handle_dirty_metadata(handle, inode, bh);
1165         if (err)
1166                 goto cleanup;
1167         brelse(bh);
1168         bh = NULL;
1169
1170         /* correct old leaf */
1171         if (m) {
1172                 err = ext4_ext_get_access(handle, inode, path + depth);
1173                 if (err)
1174                         goto cleanup;
1175                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1176                 err = ext4_ext_dirty(handle, inode, path + depth);
1177                 if (err)
1178                         goto cleanup;
1179
1180         }
1181
1182         /* create intermediate indexes */
1183         k = depth - at - 1;
1184         if (unlikely(k < 0)) {
1185                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1186                 err = -EFSCORRUPTED;
1187                 goto cleanup;
1188         }
1189         if (k)
1190                 ext_debug(inode, "create %d intermediate indices\n", k);
1191         /* insert new index into current index block */
1192         /* current depth stored in i var */
1193         i = depth - 1;
1194         while (k--) {
1195                 oldblock = newblock;
1196                 newblock = ablocks[--a];
1197                 bh = sb_getblk(inode->i_sb, newblock);
1198                 if (unlikely(!bh)) {
1199                         err = -ENOMEM;
1200                         goto cleanup;
1201                 }
1202                 lock_buffer(bh);
1203
1204                 err = ext4_journal_get_create_access(handle, bh);
1205                 if (err)
1206                         goto cleanup;
1207
1208                 neh = ext_block_hdr(bh);
1209                 neh->eh_entries = cpu_to_le16(1);
1210                 neh->eh_magic = EXT4_EXT_MAGIC;
1211                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1212                 neh->eh_depth = cpu_to_le16(depth - i);
1213                 neh->eh_generation = 0;
1214                 fidx = EXT_FIRST_INDEX(neh);
1215                 fidx->ei_block = border;
1216                 ext4_idx_store_pblock(fidx, oldblock);
1217
1218                 ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1219                                 i, newblock, le32_to_cpu(border), oldblock);
1220
1221                 /* move remainder of path[i] to the new index block */
1222                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1223                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1224                         EXT4_ERROR_INODE(inode,
1225                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1226                                          le32_to_cpu(path[i].p_ext->ee_block));
1227                         err = -EFSCORRUPTED;
1228                         goto cleanup;
1229                 }
1230                 /* start copy indexes */
1231                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1232                 ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1233                                 EXT_MAX_INDEX(path[i].p_hdr));
1234                 ext4_ext_show_move(inode, path, newblock, i);
1235                 if (m) {
1236                         memmove(++fidx, path[i].p_idx,
1237                                 sizeof(struct ext4_extent_idx) * m);
1238                         le16_add_cpu(&neh->eh_entries, m);
1239                 }
1240                 /* zero out unused area in the extent block */
1241                 ext_size = sizeof(struct ext4_extent_header) +
1242                    (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1243                 memset(bh->b_data + ext_size, 0,
1244                         inode->i_sb->s_blocksize - ext_size);
1245                 ext4_extent_block_csum_set(inode, neh);
1246                 set_buffer_uptodate(bh);
1247                 unlock_buffer(bh);
1248
1249                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1250                 if (err)
1251                         goto cleanup;
1252                 brelse(bh);
1253                 bh = NULL;
1254
1255                 /* correct old index */
1256                 if (m) {
1257                         err = ext4_ext_get_access(handle, inode, path + i);
1258                         if (err)
1259                                 goto cleanup;
1260                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1261                         err = ext4_ext_dirty(handle, inode, path + i);
1262                         if (err)
1263                                 goto cleanup;
1264                 }
1265
1266                 i--;
1267         }
1268
1269         /* insert new index */
1270         err = ext4_ext_insert_index(handle, inode, path + at,
1271                                     le32_to_cpu(border), newblock);
1272
1273 cleanup:
1274         if (bh) {
1275                 if (buffer_locked(bh))
1276                         unlock_buffer(bh);
1277                 brelse(bh);
1278         }
1279
1280         if (err) {
1281                 /* free all allocated blocks in error case */
1282                 for (i = 0; i < depth; i++) {
1283                         if (!ablocks[i])
1284                                 continue;
1285                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1286                                          EXT4_FREE_BLOCKS_METADATA);
1287                 }
1288         }
1289         kfree(ablocks);
1290
1291         return err;
1292 }
1293
1294 /*
1295  * ext4_ext_grow_indepth:
1296  * implements tree growing procedure:
1297  * - allocates new block
1298  * - moves top-level data (index block or leaf) into the new block
1299  * - initializes new top-level, creating index that points to the
1300  *   just created block
1301  */
1302 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1303                                  unsigned int flags)
1304 {
1305         struct ext4_extent_header *neh;
1306         struct buffer_head *bh;
1307         ext4_fsblk_t newblock, goal = 0;
1308         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1309         int err = 0;
1310         size_t ext_size = 0;
1311
1312         /* Try to prepend new index to old one */
1313         if (ext_depth(inode))
1314                 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1315         if (goal > le32_to_cpu(es->s_first_data_block)) {
1316                 flags |= EXT4_MB_HINT_TRY_GOAL;
1317                 goal--;
1318         } else
1319                 goal = ext4_inode_to_goal_block(inode);
1320         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1321                                         NULL, &err);
1322         if (newblock == 0)
1323                 return err;
1324
1325         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1326         if (unlikely(!bh))
1327                 return -ENOMEM;
1328         lock_buffer(bh);
1329
1330         err = ext4_journal_get_create_access(handle, bh);
1331         if (err) {
1332                 unlock_buffer(bh);
1333                 goto out;
1334         }
1335
1336         ext_size = sizeof(EXT4_I(inode)->i_data);
1337         /* move top-level index/leaf into new block */
1338         memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1339         /* zero out unused area in the extent block */
1340         memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1341
1342         /* set size of new block */
1343         neh = ext_block_hdr(bh);
1344         /* old root could have indexes or leaves
1345          * so calculate e_max right way */
1346         if (ext_depth(inode))
1347                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1348         else
1349                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1350         neh->eh_magic = EXT4_EXT_MAGIC;
1351         ext4_extent_block_csum_set(inode, neh);
1352         set_buffer_uptodate(bh);
1353         unlock_buffer(bh);
1354
1355         err = ext4_handle_dirty_metadata(handle, inode, bh);
1356         if (err)
1357                 goto out;
1358
1359         /* Update top-level index: num,max,pointer */
1360         neh = ext_inode_hdr(inode);
1361         neh->eh_entries = cpu_to_le16(1);
1362         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1363         if (neh->eh_depth == 0) {
1364                 /* Root extent block becomes index block */
1365                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1366                 EXT_FIRST_INDEX(neh)->ei_block =
1367                         EXT_FIRST_EXTENT(neh)->ee_block;
1368         }
1369         ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1370                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1371                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1372                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1373
1374         le16_add_cpu(&neh->eh_depth, 1);
1375         err = ext4_mark_inode_dirty(handle, inode);
1376 out:
1377         brelse(bh);
1378
1379         return err;
1380 }
1381
1382 /*
1383  * ext4_ext_create_new_leaf:
1384  * finds empty index and adds new leaf.
1385  * if no free index is found, then it requests in-depth growing.
1386  */
1387 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1388                                     unsigned int mb_flags,
1389                                     unsigned int gb_flags,
1390                                     struct ext4_ext_path **ppath,
1391                                     struct ext4_extent *newext)
1392 {
1393         struct ext4_ext_path *path = *ppath;
1394         struct ext4_ext_path *curp;
1395         int depth, i, err = 0;
1396
1397 repeat:
1398         i = depth = ext_depth(inode);
1399
1400         /* walk up to the tree and look for free index entry */
1401         curp = path + depth;
1402         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1403                 i--;
1404                 curp--;
1405         }
1406
1407         /* we use already allocated block for index block,
1408          * so subsequent data blocks should be contiguous */
1409         if (EXT_HAS_FREE_INDEX(curp)) {
1410                 /* if we found index with free entry, then use that
1411                  * entry: create all needed subtree and add new leaf */
1412                 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1413                 if (err)
1414                         goto out;
1415
1416                 /* refill path */
1417                 path = ext4_find_extent(inode,
1418                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1419                                     ppath, gb_flags);
1420                 if (IS_ERR(path))
1421                         err = PTR_ERR(path);
1422         } else {
1423                 /* tree is full, time to grow in depth */
1424                 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1425                 if (err)
1426                         goto out;
1427
1428                 /* refill path */
1429                 path = ext4_find_extent(inode,
1430                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1431                                     ppath, gb_flags);
1432                 if (IS_ERR(path)) {
1433                         err = PTR_ERR(path);
1434                         goto out;
1435                 }
1436
1437                 /*
1438                  * only first (depth 0 -> 1) produces free space;
1439                  * in all other cases we have to split the grown tree
1440                  */
1441                 depth = ext_depth(inode);
1442                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1443                         /* now we need to split */
1444                         goto repeat;
1445                 }
1446         }
1447
1448 out:
1449         return err;
1450 }
1451
1452 /*
1453  * search the closest allocated block to the left for *logical
1454  * and returns it at @logical + it's physical address at @phys
1455  * if *logical is the smallest allocated block, the function
1456  * returns 0 at @phys
1457  * return value contains 0 (success) or error code
1458  */
1459 static int ext4_ext_search_left(struct inode *inode,
1460                                 struct ext4_ext_path *path,
1461                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1462 {
1463         struct ext4_extent_idx *ix;
1464         struct ext4_extent *ex;
1465         int depth, ee_len;
1466
1467         if (unlikely(path == NULL)) {
1468                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1469                 return -EFSCORRUPTED;
1470         }
1471         depth = path->p_depth;
1472         *phys = 0;
1473
1474         if (depth == 0 && path->p_ext == NULL)
1475                 return 0;
1476
1477         /* usually extent in the path covers blocks smaller
1478          * then *logical, but it can be that extent is the
1479          * first one in the file */
1480
1481         ex = path[depth].p_ext;
1482         ee_len = ext4_ext_get_actual_len(ex);
1483         if (*logical < le32_to_cpu(ex->ee_block)) {
1484                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1485                         EXT4_ERROR_INODE(inode,
1486                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1487                                          *logical, le32_to_cpu(ex->ee_block));
1488                         return -EFSCORRUPTED;
1489                 }
1490                 while (--depth >= 0) {
1491                         ix = path[depth].p_idx;
1492                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1493                                 EXT4_ERROR_INODE(inode,
1494                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1495                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1496                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1497                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1498                                   depth);
1499                                 return -EFSCORRUPTED;
1500                         }
1501                 }
1502                 return 0;
1503         }
1504
1505         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1506                 EXT4_ERROR_INODE(inode,
1507                                  "logical %d < ee_block %d + ee_len %d!",
1508                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1509                 return -EFSCORRUPTED;
1510         }
1511
1512         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1513         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1514         return 0;
1515 }
1516
1517 /*
1518  * Search the closest allocated block to the right for *logical
1519  * and returns it at @logical + it's physical address at @phys.
1520  * If not exists, return 0 and @phys is set to 0. We will return
1521  * 1 which means we found an allocated block and ret_ex is valid.
1522  * Or return a (< 0) error code.
1523  */
1524 static int ext4_ext_search_right(struct inode *inode,
1525                                  struct ext4_ext_path *path,
1526                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1527                                  struct ext4_extent *ret_ex)
1528 {
1529         struct buffer_head *bh = NULL;
1530         struct ext4_extent_header *eh;
1531         struct ext4_extent_idx *ix;
1532         struct ext4_extent *ex;
1533         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1534         int ee_len;
1535
1536         if (unlikely(path == NULL)) {
1537                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1538                 return -EFSCORRUPTED;
1539         }
1540         depth = path->p_depth;
1541         *phys = 0;
1542
1543         if (depth == 0 && path->p_ext == NULL)
1544                 return 0;
1545
1546         /* usually extent in the path covers blocks smaller
1547          * then *logical, but it can be that extent is the
1548          * first one in the file */
1549
1550         ex = path[depth].p_ext;
1551         ee_len = ext4_ext_get_actual_len(ex);
1552         if (*logical < le32_to_cpu(ex->ee_block)) {
1553                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1554                         EXT4_ERROR_INODE(inode,
1555                                          "first_extent(path[%d].p_hdr) != ex",
1556                                          depth);
1557                         return -EFSCORRUPTED;
1558                 }
1559                 while (--depth >= 0) {
1560                         ix = path[depth].p_idx;
1561                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1562                                 EXT4_ERROR_INODE(inode,
1563                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1564                                                  *logical);
1565                                 return -EFSCORRUPTED;
1566                         }
1567                 }
1568                 goto found_extent;
1569         }
1570
1571         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1572                 EXT4_ERROR_INODE(inode,
1573                                  "logical %d < ee_block %d + ee_len %d!",
1574                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1575                 return -EFSCORRUPTED;
1576         }
1577
1578         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1579                 /* next allocated block in this leaf */
1580                 ex++;
1581                 goto found_extent;
1582         }
1583
1584         /* go up and search for index to the right */
1585         while (--depth >= 0) {
1586                 ix = path[depth].p_idx;
1587                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1588                         goto got_index;
1589         }
1590
1591         /* we've gone up to the root and found no index to the right */
1592         return 0;
1593
1594 got_index:
1595         /* we've found index to the right, let's
1596          * follow it and find the closest allocated
1597          * block to the right */
1598         ix++;
1599         while (++depth < path->p_depth) {
1600                 /* subtract from p_depth to get proper eh_depth */
1601                 bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1602                 if (IS_ERR(bh))
1603                         return PTR_ERR(bh);
1604                 eh = ext_block_hdr(bh);
1605                 ix = EXT_FIRST_INDEX(eh);
1606                 put_bh(bh);
1607         }
1608
1609         bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1610         if (IS_ERR(bh))
1611                 return PTR_ERR(bh);
1612         eh = ext_block_hdr(bh);
1613         ex = EXT_FIRST_EXTENT(eh);
1614 found_extent:
1615         *logical = le32_to_cpu(ex->ee_block);
1616         *phys = ext4_ext_pblock(ex);
1617         if (ret_ex)
1618                 *ret_ex = *ex;
1619         if (bh)
1620                 put_bh(bh);
1621         return 1;
1622 }
1623
1624 /*
1625  * ext4_ext_next_allocated_block:
1626  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1627  * NOTE: it considers block number from index entry as
1628  * allocated block. Thus, index entries have to be consistent
1629  * with leaves.
1630  */
1631 ext4_lblk_t
1632 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1633 {
1634         int depth;
1635
1636         BUG_ON(path == NULL);
1637         depth = path->p_depth;
1638
1639         if (depth == 0 && path->p_ext == NULL)
1640                 return EXT_MAX_BLOCKS;
1641
1642         while (depth >= 0) {
1643                 struct ext4_ext_path *p = &path[depth];
1644
1645                 if (depth == path->p_depth) {
1646                         /* leaf */
1647                         if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1648                                 return le32_to_cpu(p->p_ext[1].ee_block);
1649                 } else {
1650                         /* index */
1651                         if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1652                                 return le32_to_cpu(p->p_idx[1].ei_block);
1653                 }
1654                 depth--;
1655         }
1656
1657         return EXT_MAX_BLOCKS;
1658 }
1659
1660 /*
1661  * ext4_ext_next_leaf_block:
1662  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1663  */
1664 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1665 {
1666         int depth;
1667
1668         BUG_ON(path == NULL);
1669         depth = path->p_depth;
1670
1671         /* zero-tree has no leaf blocks at all */
1672         if (depth == 0)
1673                 return EXT_MAX_BLOCKS;
1674
1675         /* go to index block */
1676         depth--;
1677
1678         while (depth >= 0) {
1679                 if (path[depth].p_idx !=
1680                                 EXT_LAST_INDEX(path[depth].p_hdr))
1681                         return (ext4_lblk_t)
1682                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1683                 depth--;
1684         }
1685
1686         return EXT_MAX_BLOCKS;
1687 }
1688
1689 /*
1690  * ext4_ext_correct_indexes:
1691  * if leaf gets modified and modified extent is first in the leaf,
1692  * then we have to correct all indexes above.
1693  * TODO: do we need to correct tree in all cases?
1694  */
1695 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1696                                 struct ext4_ext_path *path)
1697 {
1698         struct ext4_extent_header *eh;
1699         int depth = ext_depth(inode);
1700         struct ext4_extent *ex;
1701         __le32 border;
1702         int k, err = 0;
1703
1704         eh = path[depth].p_hdr;
1705         ex = path[depth].p_ext;
1706
1707         if (unlikely(ex == NULL || eh == NULL)) {
1708                 EXT4_ERROR_INODE(inode,
1709                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1710                 return -EFSCORRUPTED;
1711         }
1712
1713         if (depth == 0) {
1714                 /* there is no tree at all */
1715                 return 0;
1716         }
1717
1718         if (ex != EXT_FIRST_EXTENT(eh)) {
1719                 /* we correct tree if first leaf got modified only */
1720                 return 0;
1721         }
1722
1723         /*
1724          * TODO: we need correction if border is smaller than current one
1725          */
1726         k = depth - 1;
1727         border = path[depth].p_ext->ee_block;
1728         err = ext4_ext_get_access(handle, inode, path + k);
1729         if (err)
1730                 return err;
1731         path[k].p_idx->ei_block = border;
1732         err = ext4_ext_dirty(handle, inode, path + k);
1733         if (err)
1734                 return err;
1735
1736         while (k--) {
1737                 /* change all left-side indexes */
1738                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1739                         break;
1740                 err = ext4_ext_get_access(handle, inode, path + k);
1741                 if (err)
1742                         break;
1743                 path[k].p_idx->ei_block = border;
1744                 err = ext4_ext_dirty(handle, inode, path + k);
1745                 if (err)
1746                         break;
1747         }
1748
1749         return err;
1750 }
1751
1752 static int ext4_can_extents_be_merged(struct inode *inode,
1753                                       struct ext4_extent *ex1,
1754                                       struct ext4_extent *ex2)
1755 {
1756         unsigned short ext1_ee_len, ext2_ee_len;
1757
1758         if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1759                 return 0;
1760
1761         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1762         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1763
1764         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1765                         le32_to_cpu(ex2->ee_block))
1766                 return 0;
1767
1768         if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1769                 return 0;
1770
1771         if (ext4_ext_is_unwritten(ex1) &&
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                         ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1852                 return;
1853
1854         /*
1855          * Copy the extent data up to the inode
1856          */
1857         blk = ext4_idx_pblock(path[0].p_idx);
1858         s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1859                 sizeof(struct ext4_extent_idx);
1860         s += sizeof(struct ext4_extent_header);
1861
1862         path[1].p_maxdepth = path[0].p_maxdepth;
1863         memcpy(path[0].p_hdr, path[1].p_hdr, s);
1864         path[0].p_depth = 0;
1865         path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1866                 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1867         path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1868
1869         brelse(path[1].p_bh);
1870         ext4_free_blocks(handle, inode, NULL, blk, 1,
1871                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1872 }
1873
1874 /*
1875  * This function tries to merge the @ex extent to neighbours in the tree, then
1876  * tries to collapse the extent tree into the inode.
1877  */
1878 static void ext4_ext_try_to_merge(handle_t *handle,
1879                                   struct inode *inode,
1880                                   struct ext4_ext_path *path,
1881                                   struct ext4_extent *ex)
1882 {
1883         struct ext4_extent_header *eh;
1884         unsigned int depth;
1885         int merge_done = 0;
1886
1887         depth = ext_depth(inode);
1888         BUG_ON(path[depth].p_hdr == NULL);
1889         eh = path[depth].p_hdr;
1890
1891         if (ex > EXT_FIRST_EXTENT(eh))
1892                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1893
1894         if (!merge_done)
1895                 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1896
1897         ext4_ext_try_to_merge_up(handle, inode, path);
1898 }
1899
1900 /*
1901  * check if a portion of the "newext" extent overlaps with an
1902  * existing extent.
1903  *
1904  * If there is an overlap discovered, it updates the length of the newext
1905  * such that there will be no overlap, and then returns 1.
1906  * If there is no overlap found, it returns 0.
1907  */
1908 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1909                                            struct inode *inode,
1910                                            struct ext4_extent *newext,
1911                                            struct ext4_ext_path *path)
1912 {
1913         ext4_lblk_t b1, b2;
1914         unsigned int depth, len1;
1915         unsigned int ret = 0;
1916
1917         b1 = le32_to_cpu(newext->ee_block);
1918         len1 = ext4_ext_get_actual_len(newext);
1919         depth = ext_depth(inode);
1920         if (!path[depth].p_ext)
1921                 goto out;
1922         b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1923
1924         /*
1925          * get the next allocated block if the extent in the path
1926          * is before the requested block(s)
1927          */
1928         if (b2 < b1) {
1929                 b2 = ext4_ext_next_allocated_block(path);
1930                 if (b2 == EXT_MAX_BLOCKS)
1931                         goto out;
1932                 b2 = EXT4_LBLK_CMASK(sbi, b2);
1933         }
1934
1935         /* check for wrap through zero on extent logical start block*/
1936         if (b1 + len1 < b1) {
1937                 len1 = EXT_MAX_BLOCKS - b1;
1938                 newext->ee_len = cpu_to_le16(len1);
1939                 ret = 1;
1940         }
1941
1942         /* check for overlap */
1943         if (b1 + len1 > b2) {
1944                 newext->ee_len = cpu_to_le16(b2 - b1);
1945                 ret = 1;
1946         }
1947 out:
1948         return ret;
1949 }
1950
1951 /*
1952  * ext4_ext_insert_extent:
1953  * tries to merge requested extent into the existing extent or
1954  * inserts requested extent as new one into the tree,
1955  * creating new leaf in the no-space case.
1956  */
1957 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1958                                 struct ext4_ext_path **ppath,
1959                                 struct ext4_extent *newext, int gb_flags)
1960 {
1961         struct ext4_ext_path *path = *ppath;
1962         struct ext4_extent_header *eh;
1963         struct ext4_extent *ex, *fex;
1964         struct ext4_extent *nearex; /* nearest extent */
1965         struct ext4_ext_path *npath = NULL;
1966         int depth, len, err;
1967         ext4_lblk_t next;
1968         int mb_flags = 0, unwritten;
1969
1970         if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1971                 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1972         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1973                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1974                 return -EFSCORRUPTED;
1975         }
1976         depth = ext_depth(inode);
1977         ex = path[depth].p_ext;
1978         eh = path[depth].p_hdr;
1979         if (unlikely(path[depth].p_hdr == NULL)) {
1980                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1981                 return -EFSCORRUPTED;
1982         }
1983
1984         /* try to insert block into found extent and return */
1985         if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1986
1987                 /*
1988                  * Try to see whether we should rather test the extent on
1989                  * right from ex, or from the left of ex. This is because
1990                  * ext4_find_extent() can return either extent on the
1991                  * left, or on the right from the searched position. This
1992                  * will make merging more effective.
1993                  */
1994                 if (ex < EXT_LAST_EXTENT(eh) &&
1995                     (le32_to_cpu(ex->ee_block) +
1996                     ext4_ext_get_actual_len(ex) <
1997                     le32_to_cpu(newext->ee_block))) {
1998                         ex += 1;
1999                         goto prepend;
2000                 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2001                            (le32_to_cpu(newext->ee_block) +
2002                            ext4_ext_get_actual_len(newext) <
2003                            le32_to_cpu(ex->ee_block)))
2004                         ex -= 1;
2005
2006                 /* Try to append newex to the ex */
2007                 if (ext4_can_extents_be_merged(inode, ex, newext)) {
2008                         ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2009                                   "(from %llu)\n",
2010                                   ext4_ext_is_unwritten(newext),
2011                                   ext4_ext_get_actual_len(newext),
2012                                   le32_to_cpu(ex->ee_block),
2013                                   ext4_ext_is_unwritten(ex),
2014                                   ext4_ext_get_actual_len(ex),
2015                                   ext4_ext_pblock(ex));
2016                         err = ext4_ext_get_access(handle, inode,
2017                                                   path + depth);
2018                         if (err)
2019                                 return err;
2020                         unwritten = ext4_ext_is_unwritten(ex);
2021                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2022                                         + ext4_ext_get_actual_len(newext));
2023                         if (unwritten)
2024                                 ext4_ext_mark_unwritten(ex);
2025                         eh = path[depth].p_hdr;
2026                         nearex = ex;
2027                         goto merge;
2028                 }
2029
2030 prepend:
2031                 /* Try to prepend newex to the ex */
2032                 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2033                         ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2034                                   "(from %llu)\n",
2035                                   le32_to_cpu(newext->ee_block),
2036                                   ext4_ext_is_unwritten(newext),
2037                                   ext4_ext_get_actual_len(newext),
2038                                   le32_to_cpu(ex->ee_block),
2039                                   ext4_ext_is_unwritten(ex),
2040                                   ext4_ext_get_actual_len(ex),
2041                                   ext4_ext_pblock(ex));
2042                         err = ext4_ext_get_access(handle, inode,
2043                                                   path + depth);
2044                         if (err)
2045                                 return err;
2046
2047                         unwritten = ext4_ext_is_unwritten(ex);
2048                         ex->ee_block = newext->ee_block;
2049                         ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2050                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2051                                         + ext4_ext_get_actual_len(newext));
2052                         if (unwritten)
2053                                 ext4_ext_mark_unwritten(ex);
2054                         eh = path[depth].p_hdr;
2055                         nearex = ex;
2056                         goto merge;
2057                 }
2058         }
2059
2060         depth = ext_depth(inode);
2061         eh = path[depth].p_hdr;
2062         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2063                 goto has_space;
2064
2065         /* probably next leaf has space for us? */
2066         fex = EXT_LAST_EXTENT(eh);
2067         next = EXT_MAX_BLOCKS;
2068         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2069                 next = ext4_ext_next_leaf_block(path);
2070         if (next != EXT_MAX_BLOCKS) {
2071                 ext_debug(inode, "next leaf block - %u\n", next);
2072                 BUG_ON(npath != NULL);
2073                 npath = ext4_find_extent(inode, next, NULL, gb_flags);
2074                 if (IS_ERR(npath))
2075                         return PTR_ERR(npath);
2076                 BUG_ON(npath->p_depth != path->p_depth);
2077                 eh = npath[depth].p_hdr;
2078                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2079                         ext_debug(inode, "next leaf isn't full(%d)\n",
2080                                   le16_to_cpu(eh->eh_entries));
2081                         path = npath;
2082                         goto has_space;
2083                 }
2084                 ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2085                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2086         }
2087
2088         /*
2089          * There is no free space in the found leaf.
2090          * We're gonna add a new leaf in the tree.
2091          */
2092         if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2093                 mb_flags |= EXT4_MB_USE_RESERVED;
2094         err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2095                                        ppath, newext);
2096         if (err)
2097                 goto cleanup;
2098         depth = ext_depth(inode);
2099         eh = path[depth].p_hdr;
2100
2101 has_space:
2102         nearex = path[depth].p_ext;
2103
2104         err = ext4_ext_get_access(handle, inode, path + depth);
2105         if (err)
2106                 goto cleanup;
2107
2108         if (!nearex) {
2109                 /* there is no extent in this leaf, create first one */
2110                 ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2111                                 le32_to_cpu(newext->ee_block),
2112                                 ext4_ext_pblock(newext),
2113                                 ext4_ext_is_unwritten(newext),
2114                                 ext4_ext_get_actual_len(newext));
2115                 nearex = EXT_FIRST_EXTENT(eh);
2116         } else {
2117                 if (le32_to_cpu(newext->ee_block)
2118                            > le32_to_cpu(nearex->ee_block)) {
2119                         /* Insert after */
2120                         ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2121                                         "nearest %p\n",
2122                                         le32_to_cpu(newext->ee_block),
2123                                         ext4_ext_pblock(newext),
2124                                         ext4_ext_is_unwritten(newext),
2125                                         ext4_ext_get_actual_len(newext),
2126                                         nearex);
2127                         nearex++;
2128                 } else {
2129                         /* Insert before */
2130                         BUG_ON(newext->ee_block == nearex->ee_block);
2131                         ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2132                                         "nearest %p\n",
2133                                         le32_to_cpu(newext->ee_block),
2134                                         ext4_ext_pblock(newext),
2135                                         ext4_ext_is_unwritten(newext),
2136                                         ext4_ext_get_actual_len(newext),
2137                                         nearex);
2138                 }
2139                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2140                 if (len > 0) {
2141                         ext_debug(inode, "insert %u:%llu:[%d]%d: "
2142                                         "move %d extents from 0x%p to 0x%p\n",
2143                                         le32_to_cpu(newext->ee_block),
2144                                         ext4_ext_pblock(newext),
2145                                         ext4_ext_is_unwritten(newext),
2146                                         ext4_ext_get_actual_len(newext),
2147                                         len, nearex, nearex + 1);
2148                         memmove(nearex + 1, nearex,
2149                                 len * sizeof(struct ext4_extent));
2150                 }
2151         }
2152
2153         le16_add_cpu(&eh->eh_entries, 1);
2154         path[depth].p_ext = nearex;
2155         nearex->ee_block = newext->ee_block;
2156         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2157         nearex->ee_len = newext->ee_len;
2158
2159 merge:
2160         /* try to merge extents */
2161         if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2162                 ext4_ext_try_to_merge(handle, inode, path, nearex);
2163
2164
2165         /* time to correct all indexes above */
2166         err = ext4_ext_correct_indexes(handle, inode, path);
2167         if (err)
2168                 goto cleanup;
2169
2170         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2171
2172 cleanup:
2173         ext4_ext_drop_refs(npath);
2174         kfree(npath);
2175         return err;
2176 }
2177
2178 static int ext4_fill_es_cache_info(struct inode *inode,
2179                                    ext4_lblk_t block, ext4_lblk_t num,
2180                                    struct fiemap_extent_info *fieinfo)
2181 {
2182         ext4_lblk_t next, end = block + num - 1;
2183         struct extent_status es;
2184         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2185         unsigned int flags;
2186         int err;
2187
2188         while (block <= end) {
2189                 next = 0;
2190                 flags = 0;
2191                 if (!ext4_es_lookup_extent(inode, block, &next, &es))
2192                         break;
2193                 if (ext4_es_is_unwritten(&es))
2194                         flags |= FIEMAP_EXTENT_UNWRITTEN;
2195                 if (ext4_es_is_delayed(&es))
2196                         flags |= (FIEMAP_EXTENT_DELALLOC |
2197                                   FIEMAP_EXTENT_UNKNOWN);
2198                 if (ext4_es_is_hole(&es))
2199                         flags |= EXT4_FIEMAP_EXTENT_HOLE;
2200                 if (next == 0)
2201                         flags |= FIEMAP_EXTENT_LAST;
2202                 if (flags & (FIEMAP_EXTENT_DELALLOC|
2203                              EXT4_FIEMAP_EXTENT_HOLE))
2204                         es.es_pblk = 0;
2205                 else
2206                         es.es_pblk = ext4_es_pblock(&es);
2207                 err = fiemap_fill_next_extent(fieinfo,
2208                                 (__u64)es.es_lblk << blksize_bits,
2209                                 (__u64)es.es_pblk << blksize_bits,
2210                                 (__u64)es.es_len << blksize_bits,
2211                                 flags);
2212                 if (next == 0)
2213                         break;
2214                 block = next;
2215                 if (err < 0)
2216                         return err;
2217                 if (err == 1)
2218                         return 0;
2219         }
2220         return 0;
2221 }
2222
2223
2224 /*
2225  * ext4_ext_determine_hole - determine hole around given block
2226  * @inode:      inode we lookup in
2227  * @path:       path in extent tree to @lblk
2228  * @lblk:       pointer to logical block around which we want to determine hole
2229  *
2230  * Determine hole length (and start if easily possible) around given logical
2231  * block. We don't try too hard to find the beginning of the hole but @path
2232  * actually points to extent before @lblk, we provide it.
2233  *
2234  * The function returns the length of a hole starting at @lblk. We update @lblk
2235  * to the beginning of the hole if we managed to find it.
2236  */
2237 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2238                                            struct ext4_ext_path *path,
2239                                            ext4_lblk_t *lblk)
2240 {
2241         int depth = ext_depth(inode);
2242         struct ext4_extent *ex;
2243         ext4_lblk_t len;
2244
2245         ex = path[depth].p_ext;
2246         if (ex == NULL) {
2247                 /* there is no extent yet, so gap is [0;-] */
2248                 *lblk = 0;
2249                 len = EXT_MAX_BLOCKS;
2250         } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2251                 len = le32_to_cpu(ex->ee_block) - *lblk;
2252         } else if (*lblk >= le32_to_cpu(ex->ee_block)
2253                         + ext4_ext_get_actual_len(ex)) {
2254                 ext4_lblk_t next;
2255
2256                 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2257                 next = ext4_ext_next_allocated_block(path);
2258                 BUG_ON(next == *lblk);
2259                 len = next - *lblk;
2260         } else {
2261                 BUG();
2262         }
2263         return len;
2264 }
2265
2266 /*
2267  * ext4_ext_put_gap_in_cache:
2268  * calculate boundaries of the gap that the requested block fits into
2269  * and cache this gap
2270  */
2271 static void
2272 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2273                           ext4_lblk_t hole_len)
2274 {
2275         struct extent_status es;
2276
2277         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2278                                   hole_start + hole_len - 1, &es);
2279         if (es.es_len) {
2280                 /* There's delayed extent containing lblock? */
2281                 if (es.es_lblk <= hole_start)
2282                         return;
2283                 hole_len = min(es.es_lblk - hole_start, hole_len);
2284         }
2285         ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2286         ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2287                               EXTENT_STATUS_HOLE);
2288 }
2289
2290 /*
2291  * ext4_ext_rm_idx:
2292  * removes index from the index block.
2293  */
2294 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2295                         struct ext4_ext_path *path, int depth)
2296 {
2297         int err;
2298         ext4_fsblk_t leaf;
2299
2300         /* free index block */
2301         depth--;
2302         path = path + depth;
2303         leaf = ext4_idx_pblock(path->p_idx);
2304         if (unlikely(path->p_hdr->eh_entries == 0)) {
2305                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2306                 return -EFSCORRUPTED;
2307         }
2308         err = ext4_ext_get_access(handle, inode, path);
2309         if (err)
2310                 return err;
2311
2312         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2313                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2314                 len *= sizeof(struct ext4_extent_idx);
2315                 memmove(path->p_idx, path->p_idx + 1, len);
2316         }
2317
2318         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2319         err = ext4_ext_dirty(handle, inode, path);
2320         if (err)
2321                 return err;
2322         ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2323         trace_ext4_ext_rm_idx(inode, leaf);
2324
2325         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2326                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2327
2328         while (--depth >= 0) {
2329                 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2330                         break;
2331                 path--;
2332                 err = ext4_ext_get_access(handle, inode, path);
2333                 if (err)
2334                         break;
2335                 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2336                 err = ext4_ext_dirty(handle, inode, path);
2337                 if (err)
2338                         break;
2339         }
2340         return err;
2341 }
2342
2343 /*
2344  * ext4_ext_calc_credits_for_single_extent:
2345  * This routine returns max. credits that needed to insert an extent
2346  * to the extent tree.
2347  * When pass the actual path, the caller should calculate credits
2348  * under i_data_sem.
2349  */
2350 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2351                                                 struct ext4_ext_path *path)
2352 {
2353         if (path) {
2354                 int depth = ext_depth(inode);
2355                 int ret = 0;
2356
2357                 /* probably there is space in leaf? */
2358                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2359                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2360
2361                         /*
2362                          *  There are some space in the leaf tree, no
2363                          *  need to account for leaf block credit
2364                          *
2365                          *  bitmaps and block group descriptor blocks
2366                          *  and other metadata blocks still need to be
2367                          *  accounted.
2368                          */
2369                         /* 1 bitmap, 1 block group descriptor */
2370                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2371                         return ret;
2372                 }
2373         }
2374
2375         return ext4_chunk_trans_blocks(inode, nrblocks);
2376 }
2377
2378 /*
2379  * How many index/leaf blocks need to change/allocate to add @extents extents?
2380  *
2381  * If we add a single extent, then in the worse case, each tree level
2382  * index/leaf need to be changed in case of the tree split.
2383  *
2384  * If more extents are inserted, they could cause the whole tree split more
2385  * than once, but this is really rare.
2386  */
2387 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2388 {
2389         int index;
2390         int depth;
2391
2392         /* If we are converting the inline data, only one is needed here. */
2393         if (ext4_has_inline_data(inode))
2394                 return 1;
2395
2396         depth = ext_depth(inode);
2397
2398         if (extents <= 1)
2399                 index = depth * 2;
2400         else
2401                 index = depth * 3;
2402
2403         return index;
2404 }
2405
2406 static inline int get_default_free_blocks_flags(struct inode *inode)
2407 {
2408         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2409             ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2410                 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2411         else if (ext4_should_journal_data(inode))
2412                 return EXT4_FREE_BLOCKS_FORGET;
2413         return 0;
2414 }
2415
2416 /*
2417  * ext4_rereserve_cluster - increment the reserved cluster count when
2418  *                          freeing a cluster with a pending reservation
2419  *
2420  * @inode - file containing the cluster
2421  * @lblk - logical block in cluster to be reserved
2422  *
2423  * Increments the reserved cluster count and adjusts quota in a bigalloc
2424  * file system when freeing a partial cluster containing at least one
2425  * delayed and unwritten block.  A partial cluster meeting that
2426  * requirement will have a pending reservation.  If so, the
2427  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2428  * defer reserved and allocated space accounting to a subsequent call
2429  * to this function.
2430  */
2431 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2432 {
2433         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2434         struct ext4_inode_info *ei = EXT4_I(inode);
2435
2436         dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2437
2438         spin_lock(&ei->i_block_reservation_lock);
2439         ei->i_reserved_data_blocks++;
2440         percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2441         spin_unlock(&ei->i_block_reservation_lock);
2442
2443         percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2444         ext4_remove_pending(inode, lblk);
2445 }
2446
2447 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2448                               struct ext4_extent *ex,
2449                               struct partial_cluster *partial,
2450                               ext4_lblk_t from, ext4_lblk_t to)
2451 {
2452         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2453         unsigned short ee_len = ext4_ext_get_actual_len(ex);
2454         ext4_fsblk_t last_pblk, pblk;
2455         ext4_lblk_t num;
2456         int flags;
2457
2458         /* only extent tail removal is allowed */
2459         if (from < le32_to_cpu(ex->ee_block) ||
2460             to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2461                 ext4_error(sbi->s_sb,
2462                            "strange request: removal(2) %u-%u from %u:%u",
2463                            from, to, le32_to_cpu(ex->ee_block), ee_len);
2464                 return 0;
2465         }
2466
2467 #ifdef EXTENTS_STATS
2468         spin_lock(&sbi->s_ext_stats_lock);
2469         sbi->s_ext_blocks += ee_len;
2470         sbi->s_ext_extents++;
2471         if (ee_len < sbi->s_ext_min)
2472                 sbi->s_ext_min = ee_len;
2473         if (ee_len > sbi->s_ext_max)
2474                 sbi->s_ext_max = ee_len;
2475         if (ext_depth(inode) > sbi->s_depth_max)
2476                 sbi->s_depth_max = ext_depth(inode);
2477         spin_unlock(&sbi->s_ext_stats_lock);
2478 #endif
2479
2480         trace_ext4_remove_blocks(inode, ex, from, to, partial);
2481
2482         /*
2483          * if we have a partial cluster, and it's different from the
2484          * cluster of the last block in the extent, we free it
2485          */
2486         last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2487
2488         if (partial->state != initial &&
2489             partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2490                 if (partial->state == tofree) {
2491                         flags = get_default_free_blocks_flags(inode);
2492                         if (ext4_is_pending(inode, partial->lblk))
2493                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2494                         ext4_free_blocks(handle, inode, NULL,
2495                                          EXT4_C2B(sbi, partial->pclu),
2496                                          sbi->s_cluster_ratio, flags);
2497                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2498                                 ext4_rereserve_cluster(inode, partial->lblk);
2499                 }
2500                 partial->state = initial;
2501         }
2502
2503         num = le32_to_cpu(ex->ee_block) + ee_len - from;
2504         pblk = ext4_ext_pblock(ex) + ee_len - num;
2505
2506         /*
2507          * We free the partial cluster at the end of the extent (if any),
2508          * unless the cluster is used by another extent (partial_cluster
2509          * state is nofree).  If a partial cluster exists here, it must be
2510          * shared with the last block in the extent.
2511          */
2512         flags = get_default_free_blocks_flags(inode);
2513
2514         /* partial, left end cluster aligned, right end unaligned */
2515         if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2516             (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2517             (partial->state != nofree)) {
2518                 if (ext4_is_pending(inode, to))
2519                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2520                 ext4_free_blocks(handle, inode, NULL,
2521                                  EXT4_PBLK_CMASK(sbi, last_pblk),
2522                                  sbi->s_cluster_ratio, flags);
2523                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2524                         ext4_rereserve_cluster(inode, to);
2525                 partial->state = initial;
2526                 flags = get_default_free_blocks_flags(inode);
2527         }
2528
2529         flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2530
2531         /*
2532          * For bigalloc file systems, we never free a partial cluster
2533          * at the beginning of the extent.  Instead, we check to see if we
2534          * need to free it on a subsequent call to ext4_remove_blocks,
2535          * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2536          */
2537         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2538         ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2539
2540         /* reset the partial cluster if we've freed past it */
2541         if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2542                 partial->state = initial;
2543
2544         /*
2545          * If we've freed the entire extent but the beginning is not left
2546          * cluster aligned and is not marked as ineligible for freeing we
2547          * record the partial cluster at the beginning of the extent.  It
2548          * wasn't freed by the preceding ext4_free_blocks() call, and we
2549          * need to look farther to the left to determine if it's to be freed
2550          * (not shared with another extent). Else, reset the partial
2551          * cluster - we're either  done freeing or the beginning of the
2552          * extent is left cluster aligned.
2553          */
2554         if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2555                 if (partial->state == initial) {
2556                         partial->pclu = EXT4_B2C(sbi, pblk);
2557                         partial->lblk = from;
2558                         partial->state = tofree;
2559                 }
2560         } else {
2561                 partial->state = initial;
2562         }
2563
2564         return 0;
2565 }
2566
2567 /*
2568  * ext4_ext_rm_leaf() Removes the extents associated with the
2569  * blocks appearing between "start" and "end".  Both "start"
2570  * and "end" must appear in the same extent or EIO is returned.
2571  *
2572  * @handle: The journal handle
2573  * @inode:  The files inode
2574  * @path:   The path to the leaf
2575  * @partial_cluster: The cluster which we'll have to free if all extents
2576  *                   has been released from it.  However, if this value is
2577  *                   negative, it's a cluster just to the right of the
2578  *                   punched region and it must not be freed.
2579  * @start:  The first block to remove
2580  * @end:   The last block to remove
2581  */
2582 static int
2583 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2584                  struct ext4_ext_path *path,
2585                  struct partial_cluster *partial,
2586                  ext4_lblk_t start, ext4_lblk_t end)
2587 {
2588         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2589         int err = 0, correct_index = 0;
2590         int depth = ext_depth(inode), credits, revoke_credits;
2591         struct ext4_extent_header *eh;
2592         ext4_lblk_t a, b;
2593         unsigned num;
2594         ext4_lblk_t ex_ee_block;
2595         unsigned short ex_ee_len;
2596         unsigned unwritten = 0;
2597         struct ext4_extent *ex;
2598         ext4_fsblk_t pblk;
2599
2600         /* the header must be checked already in ext4_ext_remove_space() */
2601         ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2602         if (!path[depth].p_hdr)
2603                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2604         eh = path[depth].p_hdr;
2605         if (unlikely(path[depth].p_hdr == NULL)) {
2606                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2607                 return -EFSCORRUPTED;
2608         }
2609         /* find where to start removing */
2610         ex = path[depth].p_ext;
2611         if (!ex)
2612                 ex = EXT_LAST_EXTENT(eh);
2613
2614         ex_ee_block = le32_to_cpu(ex->ee_block);
2615         ex_ee_len = ext4_ext_get_actual_len(ex);
2616
2617         trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2618
2619         while (ex >= EXT_FIRST_EXTENT(eh) &&
2620                         ex_ee_block + ex_ee_len > start) {
2621
2622                 if (ext4_ext_is_unwritten(ex))
2623                         unwritten = 1;
2624                 else
2625                         unwritten = 0;
2626
2627                 ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2628                           unwritten, ex_ee_len);
2629                 path[depth].p_ext = ex;
2630
2631                 a = ex_ee_block > start ? ex_ee_block : start;
2632                 b = ex_ee_block+ex_ee_len - 1 < end ?
2633                         ex_ee_block+ex_ee_len - 1 : end;
2634
2635                 ext_debug(inode, "  border %u:%u\n", a, b);
2636
2637                 /* If this extent is beyond the end of the hole, skip it */
2638                 if (end < ex_ee_block) {
2639                         /*
2640                          * We're going to skip this extent and move to another,
2641                          * so note that its first cluster is in use to avoid
2642                          * freeing it when removing blocks.  Eventually, the
2643                          * right edge of the truncated/punched region will
2644                          * be just to the left.
2645                          */
2646                         if (sbi->s_cluster_ratio > 1) {
2647                                 pblk = ext4_ext_pblock(ex);
2648                                 partial->pclu = EXT4_B2C(sbi, pblk);
2649                                 partial->state = nofree;
2650                         }
2651                         ex--;
2652                         ex_ee_block = le32_to_cpu(ex->ee_block);
2653                         ex_ee_len = ext4_ext_get_actual_len(ex);
2654                         continue;
2655                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2656                         EXT4_ERROR_INODE(inode,
2657                                          "can not handle truncate %u:%u "
2658                                          "on extent %u:%u",
2659                                          start, end, ex_ee_block,
2660                                          ex_ee_block + ex_ee_len - 1);
2661                         err = -EFSCORRUPTED;
2662                         goto out;
2663                 } else if (a != ex_ee_block) {
2664                         /* remove tail of the extent */
2665                         num = a - ex_ee_block;
2666                 } else {
2667                         /* remove whole extent: excellent! */
2668                         num = 0;
2669                 }
2670                 /*
2671                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2672                  * descriptor) for each block group; assume two block
2673                  * groups plus ex_ee_len/blocks_per_block_group for
2674                  * the worst case
2675                  */
2676                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2677                 if (ex == EXT_FIRST_EXTENT(eh)) {
2678                         correct_index = 1;
2679                         credits += (ext_depth(inode)) + 1;
2680                 }
2681                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2682                 /*
2683                  * We may end up freeing some index blocks and data from the
2684                  * punched range. Note that partial clusters are accounted for
2685                  * by ext4_free_data_revoke_credits().
2686                  */
2687                 revoke_credits =
2688                         ext4_free_metadata_revoke_credits(inode->i_sb,
2689                                                           ext_depth(inode)) +
2690                         ext4_free_data_revoke_credits(inode, b - a + 1);
2691
2692                 err = ext4_datasem_ensure_credits(handle, inode, credits,
2693                                                   credits, revoke_credits);
2694                 if (err) {
2695                         if (err > 0)
2696                                 err = -EAGAIN;
2697                         goto out;
2698                 }
2699
2700                 err = ext4_ext_get_access(handle, inode, path + depth);
2701                 if (err)
2702                         goto out;
2703
2704                 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2705                 if (err)
2706                         goto out;
2707
2708                 if (num == 0)
2709                         /* this extent is removed; mark slot entirely unused */
2710                         ext4_ext_store_pblock(ex, 0);
2711
2712                 ex->ee_len = cpu_to_le16(num);
2713                 /*
2714                  * Do not mark unwritten if all the blocks in the
2715                  * extent have been removed.
2716                  */
2717                 if (unwritten && num)
2718                         ext4_ext_mark_unwritten(ex);
2719                 /*
2720                  * If the extent was completely released,
2721                  * we need to remove it from the leaf
2722                  */
2723                 if (num == 0) {
2724                         if (end != EXT_MAX_BLOCKS - 1) {
2725                                 /*
2726                                  * For hole punching, we need to scoot all the
2727                                  * extents up when an extent is removed so that
2728                                  * we dont have blank extents in the middle
2729                                  */
2730                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2731                                         sizeof(struct ext4_extent));
2732
2733                                 /* Now get rid of the one at the end */
2734                                 memset(EXT_LAST_EXTENT(eh), 0,
2735                                         sizeof(struct ext4_extent));
2736                         }
2737                         le16_add_cpu(&eh->eh_entries, -1);
2738                 }
2739
2740                 err = ext4_ext_dirty(handle, inode, path + depth);
2741                 if (err)
2742                         goto out;
2743
2744                 ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2745                                 ext4_ext_pblock(ex));
2746                 ex--;
2747                 ex_ee_block = le32_to_cpu(ex->ee_block);
2748                 ex_ee_len = ext4_ext_get_actual_len(ex);
2749         }
2750
2751         if (correct_index && eh->eh_entries)
2752                 err = ext4_ext_correct_indexes(handle, inode, path);
2753
2754         /*
2755          * If there's a partial cluster and at least one extent remains in
2756          * the leaf, free the partial cluster if it isn't shared with the
2757          * current extent.  If it is shared with the current extent
2758          * we reset the partial cluster because we've reached the start of the
2759          * truncated/punched region and we're done removing blocks.
2760          */
2761         if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2762                 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2763                 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2764                         int flags = get_default_free_blocks_flags(inode);
2765
2766                         if (ext4_is_pending(inode, partial->lblk))
2767                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2768                         ext4_free_blocks(handle, inode, NULL,
2769                                          EXT4_C2B(sbi, partial->pclu),
2770                                          sbi->s_cluster_ratio, flags);
2771                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2772                                 ext4_rereserve_cluster(inode, partial->lblk);
2773                 }
2774                 partial->state = initial;
2775         }
2776
2777         /* if this leaf is free, then we should
2778          * remove it from index block above */
2779         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2780                 err = ext4_ext_rm_idx(handle, inode, path, depth);
2781
2782 out:
2783         return err;
2784 }
2785
2786 /*
2787  * ext4_ext_more_to_rm:
2788  * returns 1 if current index has to be freed (even partial)
2789  */
2790 static int
2791 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2792 {
2793         BUG_ON(path->p_idx == NULL);
2794
2795         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2796                 return 0;
2797
2798         /*
2799          * if truncate on deeper level happened, it wasn't partial,
2800          * so we have to consider current index for truncation
2801          */
2802         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2803                 return 0;
2804         return 1;
2805 }
2806
2807 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2808                           ext4_lblk_t end)
2809 {
2810         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2811         int depth = ext_depth(inode);
2812         struct ext4_ext_path *path = NULL;
2813         struct partial_cluster partial;
2814         handle_t *handle;
2815         int i = 0, err = 0;
2816
2817         partial.pclu = 0;
2818         partial.lblk = 0;
2819         partial.state = initial;
2820
2821         ext_debug(inode, "truncate since %u to %u\n", start, end);
2822
2823         /* probably first extent we're gonna free will be last in block */
2824         handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2825                         depth + 1,
2826                         ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2827         if (IS_ERR(handle))
2828                 return PTR_ERR(handle);
2829
2830 again:
2831         trace_ext4_ext_remove_space(inode, start, end, depth);
2832
2833         /*
2834          * Check if we are removing extents inside the extent tree. If that
2835          * is the case, we are going to punch a hole inside the extent tree
2836          * so we have to check whether we need to split the extent covering
2837          * the last block to remove so we can easily remove the part of it
2838          * in ext4_ext_rm_leaf().
2839          */
2840         if (end < EXT_MAX_BLOCKS - 1) {
2841                 struct ext4_extent *ex;
2842                 ext4_lblk_t ee_block, ex_end, lblk;
2843                 ext4_fsblk_t pblk;
2844
2845                 /* find extent for or closest extent to this block */
2846                 path = ext4_find_extent(inode, end, NULL,
2847                                         EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2848                 if (IS_ERR(path)) {
2849                         ext4_journal_stop(handle);
2850                         return PTR_ERR(path);
2851                 }
2852                 depth = ext_depth(inode);
2853                 /* Leaf not may not exist only if inode has no blocks at all */
2854                 ex = path[depth].p_ext;
2855                 if (!ex) {
2856                         if (depth) {
2857                                 EXT4_ERROR_INODE(inode,
2858                                                  "path[%d].p_hdr == NULL",
2859                                                  depth);
2860                                 err = -EFSCORRUPTED;
2861                         }
2862                         goto out;
2863                 }
2864
2865                 ee_block = le32_to_cpu(ex->ee_block);
2866                 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2867
2868                 /*
2869                  * See if the last block is inside the extent, if so split
2870                  * the extent at 'end' block so we can easily remove the
2871                  * tail of the first part of the split extent in
2872                  * ext4_ext_rm_leaf().
2873                  */
2874                 if (end >= ee_block && end < ex_end) {
2875
2876                         /*
2877                          * If we're going to split the extent, note that
2878                          * the cluster containing the block after 'end' is
2879                          * in use to avoid freeing it when removing blocks.
2880                          */
2881                         if (sbi->s_cluster_ratio > 1) {
2882                                 pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2883                                 partial.pclu = EXT4_B2C(sbi, pblk);
2884                                 partial.state = nofree;
2885                         }
2886
2887                         /*
2888                          * Split the extent in two so that 'end' is the last
2889                          * block in the first new extent. Also we should not
2890                          * fail removing space due to ENOSPC so try to use
2891                          * reserved block if that happens.
2892                          */
2893                         err = ext4_force_split_extent_at(handle, inode, &path,
2894                                                          end + 1, 1);
2895                         if (err < 0)
2896                                 goto out;
2897
2898                 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2899                            partial.state == initial) {
2900                         /*
2901                          * If we're punching, there's an extent to the right.
2902                          * If the partial cluster hasn't been set, set it to
2903                          * that extent's first cluster and its state to nofree
2904                          * so it won't be freed should it contain blocks to be
2905                          * removed. If it's already set (tofree/nofree), we're
2906                          * retrying and keep the original partial cluster info
2907                          * so a cluster marked tofree as a result of earlier
2908                          * extent removal is not lost.
2909                          */
2910                         lblk = ex_end + 1;
2911                         err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2912                                                     NULL);
2913                         if (err < 0)
2914                                 goto out;
2915                         if (pblk) {
2916                                 partial.pclu = EXT4_B2C(sbi, pblk);
2917                                 partial.state = nofree;
2918                         }
2919                 }
2920         }
2921         /*
2922          * We start scanning from right side, freeing all the blocks
2923          * after i_size and walking into the tree depth-wise.
2924          */
2925         depth = ext_depth(inode);
2926         if (path) {
2927                 int k = i = depth;
2928                 while (--k > 0)
2929                         path[k].p_block =
2930                                 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2931         } else {
2932                 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2933                                GFP_NOFS | __GFP_NOFAIL);
2934                 if (path == NULL) {
2935                         ext4_journal_stop(handle);
2936                         return -ENOMEM;
2937                 }
2938                 path[0].p_maxdepth = path[0].p_depth = depth;
2939                 path[0].p_hdr = ext_inode_hdr(inode);
2940                 i = 0;
2941
2942                 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2943                         err = -EFSCORRUPTED;
2944                         goto out;
2945                 }
2946         }
2947         err = 0;
2948
2949         while (i >= 0 && err == 0) {
2950                 if (i == depth) {
2951                         /* this is leaf block */
2952                         err = ext4_ext_rm_leaf(handle, inode, path,
2953                                                &partial, start, end);
2954                         /* root level has p_bh == NULL, brelse() eats this */
2955                         brelse(path[i].p_bh);
2956                         path[i].p_bh = NULL;
2957                         i--;
2958                         continue;
2959                 }
2960
2961                 /* this is index block */
2962                 if (!path[i].p_hdr) {
2963                         ext_debug(inode, "initialize header\n");
2964                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2965                 }
2966
2967                 if (!path[i].p_idx) {
2968                         /* this level hasn't been touched yet */
2969                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2970                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2971                         ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2972                                   path[i].p_hdr,
2973                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2974                 } else {
2975                         /* we were already here, see at next index */
2976                         path[i].p_idx--;
2977                 }
2978
2979                 ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2980                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2981                                 path[i].p_idx);
2982                 if (ext4_ext_more_to_rm(path + i)) {
2983                         struct buffer_head *bh;
2984                         /* go to the next level */
2985                         ext_debug(inode, "move to level %d (block %llu)\n",
2986                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2987                         memset(path + i + 1, 0, sizeof(*path));
2988                         bh = read_extent_tree_block(inode, path[i].p_idx,
2989                                                     depth - i - 1,
2990                                                     EXT4_EX_NOCACHE);
2991                         if (IS_ERR(bh)) {
2992                                 /* should we reset i_size? */
2993                                 err = PTR_ERR(bh);
2994                                 break;
2995                         }
2996                         /* Yield here to deal with large extent trees.
2997                          * Should be a no-op if we did IO above. */
2998                         cond_resched();
2999                         if (WARN_ON(i + 1 > depth)) {
3000                                 err = -EFSCORRUPTED;
3001                                 break;
3002                         }
3003                         path[i + 1].p_bh = bh;
3004
3005                         /* save actual number of indexes since this
3006                          * number is changed at the next iteration */
3007                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3008                         i++;
3009                 } else {
3010                         /* we finished processing this index, go up */
3011                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3012                                 /* index is empty, remove it;
3013                                  * handle must be already prepared by the
3014                                  * truncatei_leaf() */
3015                                 err = ext4_ext_rm_idx(handle, inode, path, i);
3016                         }
3017                         /* root level has p_bh == NULL, brelse() eats this */
3018                         brelse(path[i].p_bh);
3019                         path[i].p_bh = NULL;
3020                         i--;
3021                         ext_debug(inode, "return to level %d\n", i);
3022                 }
3023         }
3024
3025         trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3026                                          path->p_hdr->eh_entries);
3027
3028         /*
3029          * if there's a partial cluster and we have removed the first extent
3030          * in the file, then we also free the partial cluster, if any
3031          */
3032         if (partial.state == tofree && err == 0) {
3033                 int flags = get_default_free_blocks_flags(inode);
3034
3035                 if (ext4_is_pending(inode, partial.lblk))
3036                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3037                 ext4_free_blocks(handle, inode, NULL,
3038                                  EXT4_C2B(sbi, partial.pclu),
3039                                  sbi->s_cluster_ratio, flags);
3040                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3041                         ext4_rereserve_cluster(inode, partial.lblk);
3042                 partial.state = initial;
3043         }
3044
3045         /* TODO: flexible tree reduction should be here */
3046         if (path->p_hdr->eh_entries == 0) {
3047                 /*
3048                  * truncate to zero freed all the tree,
3049                  * so we need to correct eh_depth
3050                  */
3051                 err = ext4_ext_get_access(handle, inode, path);
3052                 if (err == 0) {
3053                         ext_inode_hdr(inode)->eh_depth = 0;
3054                         ext_inode_hdr(inode)->eh_max =
3055                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
3056                         err = ext4_ext_dirty(handle, inode, path);
3057                 }
3058         }
3059 out:
3060         ext4_ext_drop_refs(path);
3061         kfree(path);
3062         path = NULL;
3063         if (err == -EAGAIN)
3064                 goto again;
3065         ext4_journal_stop(handle);
3066
3067         return err;
3068 }
3069
3070 /*
3071  * called at mount time
3072  */
3073 void ext4_ext_init(struct super_block *sb)
3074 {
3075         /*
3076          * possible initialization would be here
3077          */
3078
3079         if (ext4_has_feature_extents(sb)) {
3080 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3081                 printk(KERN_INFO "EXT4-fs: file extents enabled"
3082 #ifdef AGGRESSIVE_TEST
3083                        ", aggressive tests"
3084 #endif
3085 #ifdef CHECK_BINSEARCH
3086                        ", check binsearch"
3087 #endif
3088 #ifdef EXTENTS_STATS
3089                        ", stats"
3090 #endif
3091                        "\n");
3092 #endif
3093 #ifdef EXTENTS_STATS
3094                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3095                 EXT4_SB(sb)->s_ext_min = 1 << 30;
3096                 EXT4_SB(sb)->s_ext_max = 0;
3097 #endif
3098         }
3099 }
3100
3101 /*
3102  * called at umount time
3103  */
3104 void ext4_ext_release(struct super_block *sb)
3105 {
3106         if (!ext4_has_feature_extents(sb))
3107                 return;
3108
3109 #ifdef EXTENTS_STATS
3110         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3111                 struct ext4_sb_info *sbi = EXT4_SB(sb);
3112                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3113                         sbi->s_ext_blocks, sbi->s_ext_extents,
3114                         sbi->s_ext_blocks / sbi->s_ext_extents);
3115                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3116                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3117         }
3118 #endif
3119 }
3120
3121 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3122 {
3123         ext4_lblk_t  ee_block;
3124         ext4_fsblk_t ee_pblock;
3125         unsigned int ee_len;
3126
3127         ee_block  = le32_to_cpu(ex->ee_block);
3128         ee_len    = ext4_ext_get_actual_len(ex);
3129         ee_pblock = ext4_ext_pblock(ex);
3130
3131         if (ee_len == 0)
3132                 return 0;
3133
3134         return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3135                                      EXTENT_STATUS_WRITTEN);
3136 }
3137
3138 /* FIXME!! we need to try to merge to left or right after zero-out  */
3139 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3140 {
3141         ext4_fsblk_t ee_pblock;
3142         unsigned int ee_len;
3143
3144         ee_len    = ext4_ext_get_actual_len(ex);
3145         ee_pblock = ext4_ext_pblock(ex);
3146         return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3147                                   ee_len);
3148 }
3149
3150 /*
3151  * ext4_split_extent_at() splits an extent at given block.
3152  *
3153  * @handle: the journal handle
3154  * @inode: the file inode
3155  * @path: the path to the extent
3156  * @split: the logical block where the extent is splitted.
3157  * @split_flags: indicates if the extent could be zeroout if split fails, and
3158  *               the states(init or unwritten) of new extents.
3159  * @flags: flags used to insert new extent to extent tree.
3160  *
3161  *
3162  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3163  * of which are determined by split_flag.
3164  *
3165  * There are two cases:
3166  *  a> the extent are splitted into two extent.
3167  *  b> split is not needed, and just mark the extent.
3168  *
3169  * return 0 on success.
3170  */
3171 static int ext4_split_extent_at(handle_t *handle,
3172                              struct inode *inode,
3173                              struct ext4_ext_path **ppath,
3174                              ext4_lblk_t split,
3175                              int split_flag,
3176                              int flags)
3177 {
3178         struct ext4_ext_path *path = *ppath;
3179         ext4_fsblk_t newblock;
3180         ext4_lblk_t ee_block;
3181         struct ext4_extent *ex, newex, orig_ex, zero_ex;
3182         struct ext4_extent *ex2 = NULL;
3183         unsigned int ee_len, depth;
3184         int err = 0;
3185
3186         BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3187                (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3188
3189         ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3190
3191         ext4_ext_show_leaf(inode, path);
3192
3193         depth = ext_depth(inode);
3194         ex = path[depth].p_ext;
3195         ee_block = le32_to_cpu(ex->ee_block);
3196         ee_len = ext4_ext_get_actual_len(ex);
3197         newblock = split - ee_block + ext4_ext_pblock(ex);
3198
3199         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3200         BUG_ON(!ext4_ext_is_unwritten(ex) &&
3201                split_flag & (EXT4_EXT_MAY_ZEROOUT |
3202                              EXT4_EXT_MARK_UNWRIT1 |
3203                              EXT4_EXT_MARK_UNWRIT2));
3204
3205         err = ext4_ext_get_access(handle, inode, path + depth);
3206         if (err)
3207                 goto out;
3208
3209         if (split == ee_block) {
3210                 /*
3211                  * case b: block @split is the block that the extent begins with
3212                  * then we just change the state of the extent, and splitting
3213                  * is not needed.
3214                  */
3215                 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3216                         ext4_ext_mark_unwritten(ex);
3217                 else
3218                         ext4_ext_mark_initialized(ex);
3219
3220                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3221                         ext4_ext_try_to_merge(handle, inode, path, ex);
3222
3223                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3224                 goto out;
3225         }
3226
3227         /* case a */
3228         memcpy(&orig_ex, ex, sizeof(orig_ex));
3229         ex->ee_len = cpu_to_le16(split - ee_block);
3230         if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3231                 ext4_ext_mark_unwritten(ex);
3232
3233         /*
3234          * path may lead to new leaf, not to original leaf any more
3235          * after ext4_ext_insert_extent() returns,
3236          */
3237         err = ext4_ext_dirty(handle, inode, path + depth);
3238         if (err)
3239                 goto fix_extent_len;
3240
3241         ex2 = &newex;
3242         ex2->ee_block = cpu_to_le32(split);
3243         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3244         ext4_ext_store_pblock(ex2, newblock);
3245         if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3246                 ext4_ext_mark_unwritten(ex2);
3247
3248         err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3249         if (err != -ENOSPC && err != -EDQUOT)
3250                 goto out;
3251
3252         if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3253                 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3254                         if (split_flag & EXT4_EXT_DATA_VALID1) {
3255                                 err = ext4_ext_zeroout(inode, ex2);
3256                                 zero_ex.ee_block = ex2->ee_block;
3257                                 zero_ex.ee_len = cpu_to_le16(
3258                                                 ext4_ext_get_actual_len(ex2));
3259                                 ext4_ext_store_pblock(&zero_ex,
3260                                                       ext4_ext_pblock(ex2));
3261                         } else {
3262                                 err = ext4_ext_zeroout(inode, ex);
3263                                 zero_ex.ee_block = ex->ee_block;
3264                                 zero_ex.ee_len = cpu_to_le16(
3265                                                 ext4_ext_get_actual_len(ex));
3266                                 ext4_ext_store_pblock(&zero_ex,
3267                                                       ext4_ext_pblock(ex));
3268                         }
3269                 } else {
3270                         err = ext4_ext_zeroout(inode, &orig_ex);
3271                         zero_ex.ee_block = orig_ex.ee_block;
3272                         zero_ex.ee_len = cpu_to_le16(
3273                                                 ext4_ext_get_actual_len(&orig_ex));
3274                         ext4_ext_store_pblock(&zero_ex,
3275                                               ext4_ext_pblock(&orig_ex));
3276                 }
3277
3278                 if (!err) {
3279                         /* update the extent length and mark as initialized */
3280                         ex->ee_len = cpu_to_le16(ee_len);
3281                         ext4_ext_try_to_merge(handle, inode, path, ex);
3282                         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3283                         if (!err)
3284                                 /* update extent status tree */
3285                                 err = ext4_zeroout_es(inode, &zero_ex);
3286                         /* If we failed at this point, we don't know in which
3287                          * state the extent tree exactly is so don't try to fix
3288                          * length of the original extent as it may do even more
3289                          * damage.
3290                          */
3291                         goto out;
3292                 }
3293         }
3294
3295 fix_extent_len:
3296         ex->ee_len = orig_ex.ee_len;
3297         /*
3298          * Ignore ext4_ext_dirty return value since we are already in error path
3299          * and err is a non-zero error code.
3300          */
3301         ext4_ext_dirty(handle, inode, path + path->p_depth);
3302         return err;
3303 out:
3304         ext4_ext_show_leaf(inode, path);
3305         return err;
3306 }
3307
3308 /*
3309  * ext4_split_extents() splits an extent and mark extent which is covered
3310  * by @map as split_flags indicates
3311  *
3312  * It may result in splitting the extent into multiple extents (up to three)
3313  * There are three possibilities:
3314  *   a> There is no split required
3315  *   b> Splits in two extents: Split is happening at either end of the extent
3316  *   c> Splits in three extents: Somone is splitting in middle of the extent
3317  *
3318  */
3319 static int ext4_split_extent(handle_t *handle,
3320                               struct inode *inode,
3321                               struct ext4_ext_path **ppath,
3322                               struct ext4_map_blocks *map,
3323                               int split_flag,
3324                               int flags)
3325 {
3326         struct ext4_ext_path *path = *ppath;
3327         ext4_lblk_t ee_block;
3328         struct ext4_extent *ex;
3329         unsigned int ee_len, depth;
3330         int err = 0;
3331         int unwritten;
3332         int split_flag1, flags1;
3333         int allocated = map->m_len;
3334
3335         depth = ext_depth(inode);
3336         ex = path[depth].p_ext;
3337         ee_block = le32_to_cpu(ex->ee_block);
3338         ee_len = ext4_ext_get_actual_len(ex);
3339         unwritten = ext4_ext_is_unwritten(ex);
3340
3341         if (map->m_lblk + map->m_len < ee_block + ee_len) {
3342                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3343                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3344                 if (unwritten)
3345                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3346                                        EXT4_EXT_MARK_UNWRIT2;
3347                 if (split_flag & EXT4_EXT_DATA_VALID2)
3348                         split_flag1 |= EXT4_EXT_DATA_VALID1;
3349                 err = ext4_split_extent_at(handle, inode, ppath,
3350                                 map->m_lblk + map->m_len, split_flag1, flags1);
3351                 if (err)
3352                         goto out;
3353         } else {
3354                 allocated = ee_len - (map->m_lblk - ee_block);
3355         }
3356         /*
3357          * Update path is required because previous ext4_split_extent_at() may
3358          * result in split of original leaf or extent zeroout.
3359          */
3360         path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3361         if (IS_ERR(path))
3362                 return PTR_ERR(path);
3363         depth = ext_depth(inode);
3364         ex = path[depth].p_ext;
3365         if (!ex) {
3366                 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3367                                  (unsigned long) map->m_lblk);
3368                 return -EFSCORRUPTED;
3369         }
3370         unwritten = ext4_ext_is_unwritten(ex);
3371         split_flag1 = 0;
3372
3373         if (map->m_lblk >= ee_block) {
3374                 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3375                 if (unwritten) {
3376                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3377                         split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3378                                                      EXT4_EXT_MARK_UNWRIT2);
3379                 }
3380                 err = ext4_split_extent_at(handle, inode, ppath,
3381                                 map->m_lblk, split_flag1, flags);
3382                 if (err)
3383                         goto out;
3384         }
3385
3386         ext4_ext_show_leaf(inode, path);
3387 out:
3388         return err ? err : allocated;
3389 }
3390
3391 /*
3392  * This function is called by ext4_ext_map_blocks() if someone tries to write
3393  * to an unwritten extent. It may result in splitting the unwritten
3394  * extent into multiple extents (up to three - one initialized and two
3395  * unwritten).
3396  * There are three possibilities:
3397  *   a> There is no split required: Entire extent should be initialized
3398  *   b> Splits in two extents: Write is happening at either end of the extent
3399  *   c> Splits in three extents: Somone is writing in middle of the extent
3400  *
3401  * Pre-conditions:
3402  *  - The extent pointed to by 'path' is unwritten.
3403  *  - The extent pointed to by 'path' contains a superset
3404  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3405  *
3406  * Post-conditions on success:
3407  *  - the returned value is the number of blocks beyond map->l_lblk
3408  *    that are allocated and initialized.
3409  *    It is guaranteed to be >= map->m_len.
3410  */
3411 static int ext4_ext_convert_to_initialized(handle_t *handle,
3412                                            struct inode *inode,
3413                                            struct ext4_map_blocks *map,
3414                                            struct ext4_ext_path **ppath,
3415                                            int flags)
3416 {
3417         struct ext4_ext_path *path = *ppath;
3418         struct ext4_sb_info *sbi;
3419         struct ext4_extent_header *eh;
3420         struct ext4_map_blocks split_map;
3421         struct ext4_extent zero_ex1, zero_ex2;
3422         struct ext4_extent *ex, *abut_ex;
3423         ext4_lblk_t ee_block, eof_block;
3424         unsigned int ee_len, depth, map_len = map->m_len;
3425         int allocated = 0, max_zeroout = 0;
3426         int err = 0;
3427         int split_flag = EXT4_EXT_DATA_VALID2;
3428
3429         ext_debug(inode, "logical block %llu, max_blocks %u\n",
3430                   (unsigned long long)map->m_lblk, map_len);
3431
3432         sbi = EXT4_SB(inode->i_sb);
3433         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3434                         >> inode->i_sb->s_blocksize_bits;
3435         if (eof_block < map->m_lblk + map_len)
3436                 eof_block = map->m_lblk + map_len;
3437
3438         depth = ext_depth(inode);
3439         eh = path[depth].p_hdr;
3440         ex = path[depth].p_ext;
3441         ee_block = le32_to_cpu(ex->ee_block);
3442         ee_len = ext4_ext_get_actual_len(ex);
3443         zero_ex1.ee_len = 0;
3444         zero_ex2.ee_len = 0;
3445
3446         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3447
3448         /* Pre-conditions */
3449         BUG_ON(!ext4_ext_is_unwritten(ex));
3450         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3451
3452         /*
3453          * Attempt to transfer newly initialized blocks from the currently
3454          * unwritten extent to its neighbor. This is much cheaper
3455          * than an insertion followed by a merge as those involve costly
3456          * memmove() calls. Transferring to the left is the common case in
3457          * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3458          * followed by append writes.
3459          *
3460          * Limitations of the current logic:
3461          *  - L1: we do not deal with writes covering the whole extent.
3462          *    This would require removing the extent if the transfer
3463          *    is possible.
3464          *  - L2: we only attempt to merge with an extent stored in the
3465          *    same extent tree node.
3466          */
3467         if ((map->m_lblk == ee_block) &&
3468                 /* See if we can merge left */
3469                 (map_len < ee_len) &&           /*L1*/
3470                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3471                 ext4_lblk_t prev_lblk;
3472                 ext4_fsblk_t prev_pblk, ee_pblk;
3473                 unsigned int prev_len;
3474
3475                 abut_ex = ex - 1;
3476                 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3477                 prev_len = ext4_ext_get_actual_len(abut_ex);
3478                 prev_pblk = ext4_ext_pblock(abut_ex);
3479                 ee_pblk = ext4_ext_pblock(ex);
3480
3481                 /*
3482                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3483                  * upon those conditions:
3484                  * - C1: abut_ex is initialized,
3485                  * - C2: abut_ex is logically abutting ex,
3486                  * - C3: abut_ex is physically abutting ex,
3487                  * - C4: abut_ex can receive the additional blocks without
3488                  *   overflowing the (initialized) length limit.
3489                  */
3490                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3491                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3492                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3493                         (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3494                         err = ext4_ext_get_access(handle, inode, path + depth);
3495                         if (err)
3496                                 goto out;
3497
3498                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3499                                 map, ex, abut_ex);
3500
3501                         /* Shift the start of ex by 'map_len' blocks */
3502                         ex->ee_block = cpu_to_le32(ee_block + map_len);
3503                         ext4_ext_store_pblock(ex, ee_pblk + map_len);
3504                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3505                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3506
3507                         /* Extend abut_ex by 'map_len' blocks */
3508                         abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3509
3510                         /* Result: number of initialized blocks past m_lblk */
3511                         allocated = map_len;
3512                 }
3513         } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3514                    (map_len < ee_len) &&        /*L1*/
3515                    ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3516                 /* See if we can merge right */
3517                 ext4_lblk_t next_lblk;
3518                 ext4_fsblk_t next_pblk, ee_pblk;
3519                 unsigned int next_len;
3520
3521                 abut_ex = ex + 1;
3522                 next_lblk = le32_to_cpu(abut_ex->ee_block);
3523                 next_len = ext4_ext_get_actual_len(abut_ex);
3524                 next_pblk = ext4_ext_pblock(abut_ex);
3525                 ee_pblk = ext4_ext_pblock(ex);
3526
3527                 /*
3528                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3529                  * upon those conditions:
3530                  * - C1: abut_ex is initialized,
3531                  * - C2: abut_ex is logically abutting ex,
3532                  * - C3: abut_ex is physically abutting ex,
3533                  * - C4: abut_ex can receive the additional blocks without
3534                  *   overflowing the (initialized) length limit.
3535                  */
3536                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3537                     ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3538                     ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3539                     (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3540                         err = ext4_ext_get_access(handle, inode, path + depth);
3541                         if (err)
3542                                 goto out;
3543
3544                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3545                                 map, ex, abut_ex);
3546
3547                         /* Shift the start of abut_ex by 'map_len' blocks */
3548                         abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3549                         ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3550                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3551                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3552
3553                         /* Extend abut_ex by 'map_len' blocks */
3554                         abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3555
3556                         /* Result: number of initialized blocks past m_lblk */
3557                         allocated = map_len;
3558                 }
3559         }
3560         if (allocated) {
3561                 /* Mark the block containing both extents as dirty */
3562                 err = ext4_ext_dirty(handle, inode, path + depth);
3563
3564                 /* Update path to point to the right extent */
3565                 path[depth].p_ext = abut_ex;
3566                 goto out;
3567         } else
3568                 allocated = ee_len - (map->m_lblk - ee_block);
3569
3570         WARN_ON(map->m_lblk < ee_block);
3571         /*
3572          * It is safe to convert extent to initialized via explicit
3573          * zeroout only if extent is fully inside i_size or new_size.
3574          */
3575         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3576
3577         if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3578                 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3579                         (inode->i_sb->s_blocksize_bits - 10);
3580
3581         /*
3582          * five cases:
3583          * 1. split the extent into three extents.
3584          * 2. split the extent into two extents, zeroout the head of the first
3585          *    extent.
3586          * 3. split the extent into two extents, zeroout the tail of the second
3587          *    extent.
3588          * 4. split the extent into two extents with out zeroout.
3589          * 5. no splitting needed, just possibly zeroout the head and / or the
3590          *    tail of the extent.
3591          */
3592         split_map.m_lblk = map->m_lblk;
3593         split_map.m_len = map->m_len;
3594
3595         if (max_zeroout && (allocated > split_map.m_len)) {
3596                 if (allocated <= max_zeroout) {
3597                         /* case 3 or 5 */
3598                         zero_ex1.ee_block =
3599                                  cpu_to_le32(split_map.m_lblk +
3600                                              split_map.m_len);
3601                         zero_ex1.ee_len =
3602                                 cpu_to_le16(allocated - split_map.m_len);
3603                         ext4_ext_store_pblock(&zero_ex1,
3604                                 ext4_ext_pblock(ex) + split_map.m_lblk +
3605                                 split_map.m_len - ee_block);
3606                         err = ext4_ext_zeroout(inode, &zero_ex1);
3607                         if (err)
3608                                 goto out;
3609                         split_map.m_len = allocated;
3610                 }
3611                 if (split_map.m_lblk - ee_block + split_map.m_len <
3612                                                                 max_zeroout) {
3613                         /* case 2 or 5 */
3614                         if (split_map.m_lblk != ee_block) {
3615                                 zero_ex2.ee_block = ex->ee_block;
3616                                 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3617                                                         ee_block);
3618                                 ext4_ext_store_pblock(&zero_ex2,
3619                                                       ext4_ext_pblock(ex));
3620                                 err = ext4_ext_zeroout(inode, &zero_ex2);
3621                                 if (err)
3622                                         goto out;
3623                         }
3624
3625                         split_map.m_len += split_map.m_lblk - ee_block;
3626                         split_map.m_lblk = ee_block;
3627                         allocated = map->m_len;
3628                 }
3629         }
3630
3631         err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3632                                 flags);
3633         if (err > 0)
3634                 err = 0;
3635 out:
3636         /* If we have gotten a failure, don't zero out status tree */
3637         if (!err) {
3638                 err = ext4_zeroout_es(inode, &zero_ex1);
3639                 if (!err)
3640                         err = ext4_zeroout_es(inode, &zero_ex2);
3641         }
3642         return err ? err : allocated;
3643 }
3644
3645 /*
3646  * This function is called by ext4_ext_map_blocks() from
3647  * ext4_get_blocks_dio_write() when DIO to write
3648  * to an unwritten extent.
3649  *
3650  * Writing to an unwritten extent may result in splitting the unwritten
3651  * extent into multiple initialized/unwritten extents (up to three)
3652  * There are three possibilities:
3653  *   a> There is no split required: Entire extent should be unwritten
3654  *   b> Splits in two extents: Write is happening at either end of the extent
3655  *   c> Splits in three extents: Somone is writing in middle of the extent
3656  *
3657  * This works the same way in the case of initialized -> unwritten conversion.
3658  *
3659  * One of more index blocks maybe needed if the extent tree grow after
3660  * the unwritten extent split. To prevent ENOSPC occur at the IO
3661  * complete, we need to split the unwritten extent before DIO submit
3662  * the IO. The unwritten extent called at this time will be split
3663  * into three unwritten extent(at most). After IO complete, the part
3664  * being filled will be convert to initialized by the end_io callback function
3665  * via ext4_convert_unwritten_extents().
3666  *
3667  * Returns the size of unwritten extent to be written on success.
3668  */
3669 static int ext4_split_convert_extents(handle_t *handle,
3670                                         struct inode *inode,
3671                                         struct ext4_map_blocks *map,
3672                                         struct ext4_ext_path **ppath,
3673                                         int flags)
3674 {
3675         struct ext4_ext_path *path = *ppath;
3676         ext4_lblk_t eof_block;
3677         ext4_lblk_t ee_block;
3678         struct ext4_extent *ex;
3679         unsigned int ee_len;
3680         int split_flag = 0, depth;
3681
3682         ext_debug(inode, "logical block %llu, max_blocks %u\n",
3683                   (unsigned long long)map->m_lblk, map->m_len);
3684
3685         eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3686                         >> inode->i_sb->s_blocksize_bits;
3687         if (eof_block < map->m_lblk + map->m_len)
3688                 eof_block = map->m_lblk + map->m_len;
3689         /*
3690          * It is safe to convert extent to initialized via explicit
3691          * zeroout only if extent is fully inside i_size or new_size.
3692          */
3693         depth = ext_depth(inode);
3694         ex = path[depth].p_ext;
3695         ee_block = le32_to_cpu(ex->ee_block);
3696         ee_len = ext4_ext_get_actual_len(ex);
3697
3698         /* Convert to unwritten */
3699         if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3700                 split_flag |= EXT4_EXT_DATA_VALID1;
3701         /* Convert to initialized */
3702         } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3703                 split_flag |= ee_block + ee_len <= eof_block ?
3704                               EXT4_EXT_MAY_ZEROOUT : 0;
3705                 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3706         }
3707         flags |= EXT4_GET_BLOCKS_PRE_IO;
3708         return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3709 }
3710
3711 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3712                                                 struct inode *inode,
3713                                                 struct ext4_map_blocks *map,
3714                                                 struct ext4_ext_path **ppath)
3715 {
3716         struct ext4_ext_path *path = *ppath;
3717         struct ext4_extent *ex;
3718         ext4_lblk_t ee_block;
3719         unsigned int ee_len;
3720         int depth;
3721         int err = 0;
3722
3723         depth = ext_depth(inode);
3724         ex = path[depth].p_ext;
3725         ee_block = le32_to_cpu(ex->ee_block);
3726         ee_len = ext4_ext_get_actual_len(ex);
3727
3728         ext_debug(inode, "logical block %llu, max_blocks %u\n",
3729                   (unsigned long long)ee_block, ee_len);
3730
3731         /* If extent is larger than requested it is a clear sign that we still
3732          * have some extent state machine issues left. So extent_split is still
3733          * required.
3734          * TODO: Once all related issues will be fixed this situation should be
3735          * illegal.
3736          */
3737         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3738 #ifdef CONFIG_EXT4_DEBUG
3739                 ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3740                              " len %u; IO logical block %llu, len %u",
3741                              inode->i_ino, (unsigned long long)ee_block, ee_len,
3742                              (unsigned long long)map->m_lblk, map->m_len);
3743 #endif
3744                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3745                                                  EXT4_GET_BLOCKS_CONVERT);
3746                 if (err < 0)
3747                         return err;
3748                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3749                 if (IS_ERR(path))
3750                         return PTR_ERR(path);
3751                 depth = ext_depth(inode);
3752                 ex = path[depth].p_ext;
3753         }
3754
3755         err = ext4_ext_get_access(handle, inode, path + depth);
3756         if (err)
3757                 goto out;
3758         /* first mark the extent as initialized */
3759         ext4_ext_mark_initialized(ex);
3760
3761         /* note: ext4_ext_correct_indexes() isn't needed here because
3762          * borders are not changed
3763          */
3764         ext4_ext_try_to_merge(handle, inode, path, ex);
3765
3766         /* Mark modified extent as dirty */
3767         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3768 out:
3769         ext4_ext_show_leaf(inode, path);
3770         return err;
3771 }
3772
3773 static int
3774 convert_initialized_extent(handle_t *handle, struct inode *inode,
3775                            struct ext4_map_blocks *map,
3776                            struct ext4_ext_path **ppath,
3777                            unsigned int *allocated)
3778 {
3779         struct ext4_ext_path *path = *ppath;
3780         struct ext4_extent *ex;
3781         ext4_lblk_t ee_block;
3782         unsigned int ee_len;
3783         int depth;
3784         int err = 0;
3785
3786         /*
3787          * Make sure that the extent is no bigger than we support with
3788          * unwritten extent
3789          */
3790         if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3791                 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3792
3793         depth = ext_depth(inode);
3794         ex = path[depth].p_ext;
3795         ee_block = le32_to_cpu(ex->ee_block);
3796         ee_len = ext4_ext_get_actual_len(ex);
3797
3798         ext_debug(inode, "logical block %llu, max_blocks %u\n",
3799                   (unsigned long long)ee_block, ee_len);
3800
3801         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3802                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3803                                 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3804                 if (err < 0)
3805                         return err;
3806                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3807                 if (IS_ERR(path))
3808                         return PTR_ERR(path);
3809                 depth = ext_depth(inode);
3810                 ex = path[depth].p_ext;
3811                 if (!ex) {
3812                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3813                                          (unsigned long) map->m_lblk);
3814                         return -EFSCORRUPTED;
3815                 }
3816         }
3817
3818         err = ext4_ext_get_access(handle, inode, path + depth);
3819         if (err)
3820                 return err;
3821         /* first mark the extent as unwritten */
3822         ext4_ext_mark_unwritten(ex);
3823
3824         /* note: ext4_ext_correct_indexes() isn't needed here because
3825          * borders are not changed
3826          */
3827         ext4_ext_try_to_merge(handle, inode, path, ex);
3828
3829         /* Mark modified extent as dirty */
3830         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3831         if (err)
3832                 return err;
3833         ext4_ext_show_leaf(inode, path);
3834
3835         ext4_update_inode_fsync_trans(handle, inode, 1);
3836
3837         map->m_flags |= EXT4_MAP_UNWRITTEN;
3838         if (*allocated > map->m_len)
3839                 *allocated = map->m_len;
3840         map->m_len = *allocated;
3841         return 0;
3842 }
3843
3844 static int
3845 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3846                         struct ext4_map_blocks *map,
3847                         struct ext4_ext_path **ppath, int flags,
3848                         unsigned int allocated, ext4_fsblk_t newblock)
3849 {
3850         struct ext4_ext_path __maybe_unused *path = *ppath;
3851         int ret = 0;
3852         int err = 0;
3853
3854         ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3855                   (unsigned long long)map->m_lblk, map->m_len, flags,
3856                   allocated);
3857         ext4_ext_show_leaf(inode, path);
3858
3859         /*
3860          * When writing into unwritten space, we should not fail to
3861          * allocate metadata blocks for the new extent block if needed.
3862          */
3863         flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3864
3865         trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3866                                                     allocated, newblock);
3867
3868         /* get_block() before submitting IO, split the extent */
3869         if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3870                 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3871                                          flags | EXT4_GET_BLOCKS_CONVERT);
3872                 if (ret < 0) {
3873                         err = ret;
3874                         goto out2;
3875                 }
3876                 /*
3877                  * shouldn't get a 0 return when splitting an extent unless
3878                  * m_len is 0 (bug) or extent has been corrupted
3879                  */
3880                 if (unlikely(ret == 0)) {
3881                         EXT4_ERROR_INODE(inode,
3882                                          "unexpected ret == 0, m_len = %u",
3883                                          map->m_len);
3884                         err = -EFSCORRUPTED;
3885                         goto out2;
3886                 }
3887                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3888                 goto out;
3889         }
3890         /* IO end_io complete, convert the filled extent to written */
3891         if (flags & EXT4_GET_BLOCKS_CONVERT) {
3892                 err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3893                                                            ppath);
3894                 if (err < 0)
3895                         goto out2;
3896                 ext4_update_inode_fsync_trans(handle, inode, 1);
3897                 goto map_out;
3898         }
3899         /* buffered IO cases */
3900         /*
3901          * repeat fallocate creation request
3902          * we already have an unwritten extent
3903          */
3904         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3905                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3906                 goto map_out;
3907         }
3908
3909         /* buffered READ or buffered write_begin() lookup */
3910         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3911                 /*
3912                  * We have blocks reserved already.  We
3913                  * return allocated blocks so that delalloc
3914                  * won't do block reservation for us.  But
3915                  * the buffer head will be unmapped so that
3916                  * a read from the block returns 0s.
3917                  */
3918                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3919                 goto out1;
3920         }
3921
3922         /*
3923          * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3924          * For buffered writes, at writepage time, etc.  Convert a
3925          * discovered unwritten extent to written.
3926          */
3927         ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3928         if (ret < 0) {
3929                 err = ret;
3930                 goto out2;
3931         }
3932         ext4_update_inode_fsync_trans(handle, inode, 1);
3933         /*
3934          * shouldn't get a 0 return when converting an unwritten extent
3935          * unless m_len is 0 (bug) or extent has been corrupted
3936          */
3937         if (unlikely(ret == 0)) {
3938                 EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3939                                  map->m_len);
3940                 err = -EFSCORRUPTED;
3941                 goto out2;
3942         }
3943
3944 out:
3945         allocated = ret;
3946         map->m_flags |= EXT4_MAP_NEW;
3947 map_out:
3948         map->m_flags |= EXT4_MAP_MAPPED;
3949 out1:
3950         map->m_pblk = newblock;
3951         if (allocated > map->m_len)
3952                 allocated = map->m_len;
3953         map->m_len = allocated;
3954         ext4_ext_show_leaf(inode, path);
3955 out2:
3956         return err ? err : allocated;
3957 }
3958
3959 /*
3960  * get_implied_cluster_alloc - check to see if the requested
3961  * allocation (in the map structure) overlaps with a cluster already
3962  * allocated in an extent.
3963  *      @sb     The filesystem superblock structure
3964  *      @map    The requested lblk->pblk mapping
3965  *      @ex     The extent structure which might contain an implied
3966  *                      cluster allocation
3967  *
3968  * This function is called by ext4_ext_map_blocks() after we failed to
3969  * find blocks that were already in the inode's extent tree.  Hence,
3970  * we know that the beginning of the requested region cannot overlap
3971  * the extent from the inode's extent tree.  There are three cases we
3972  * want to catch.  The first is this case:
3973  *
3974  *               |--- cluster # N--|
3975  *    |--- extent ---|  |---- requested region ---|
3976  *                      |==========|
3977  *
3978  * The second case that we need to test for is this one:
3979  *
3980  *   |--------- cluster # N ----------------|
3981  *         |--- requested region --|   |------- extent ----|
3982  *         |=======================|
3983  *
3984  * The third case is when the requested region lies between two extents
3985  * within the same cluster:
3986  *          |------------- cluster # N-------------|
3987  * |----- ex -----|                  |---- ex_right ----|
3988  *                  |------ requested region ------|
3989  *                  |================|
3990  *
3991  * In each of the above cases, we need to set the map->m_pblk and
3992  * map->m_len so it corresponds to the return the extent labelled as
3993  * "|====|" from cluster #N, since it is already in use for data in
3994  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3995  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3996  * as a new "allocated" block region.  Otherwise, we will return 0 and
3997  * ext4_ext_map_blocks() will then allocate one or more new clusters
3998  * by calling ext4_mb_new_blocks().
3999  */
4000 static int get_implied_cluster_alloc(struct super_block *sb,
4001                                      struct ext4_map_blocks *map,
4002                                      struct ext4_extent *ex,
4003                                      struct ext4_ext_path *path)
4004 {
4005         struct ext4_sb_info *sbi = EXT4_SB(sb);
4006         ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4007         ext4_lblk_t ex_cluster_start, ex_cluster_end;
4008         ext4_lblk_t rr_cluster_start;
4009         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4010         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4011         unsigned short ee_len = ext4_ext_get_actual_len(ex);
4012
4013         /* The extent passed in that we are trying to match */
4014         ex_cluster_start = EXT4_B2C(sbi, ee_block);
4015         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4016
4017         /* The requested region passed into ext4_map_blocks() */
4018         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4019
4020         if ((rr_cluster_start == ex_cluster_end) ||
4021             (rr_cluster_start == ex_cluster_start)) {
4022                 if (rr_cluster_start == ex_cluster_end)
4023                         ee_start += ee_len - 1;
4024                 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4025                 map->m_len = min(map->m_len,
4026                                  (unsigned) sbi->s_cluster_ratio - c_offset);
4027                 /*
4028                  * Check for and handle this case:
4029                  *
4030                  *   |--------- cluster # N-------------|
4031                  *                     |------- extent ----|
4032                  *         |--- requested region ---|
4033                  *         |===========|
4034                  */
4035
4036                 if (map->m_lblk < ee_block)
4037                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
4038
4039                 /*
4040                  * Check for the case where there is already another allocated
4041                  * block to the right of 'ex' but before the end of the cluster.
4042                  *
4043                  *          |------------- cluster # N-------------|
4044                  * |----- ex -----|                  |---- ex_right ----|
4045                  *                  |------ requested region ------|
4046                  *                  |================|
4047                  */
4048                 if (map->m_lblk > ee_block) {
4049                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4050                         map->m_len = min(map->m_len, next - map->m_lblk);
4051                 }
4052
4053                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4054                 return 1;
4055         }
4056
4057         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4058         return 0;
4059 }
4060
4061
4062 /*
4063  * Block allocation/map/preallocation routine for extents based files
4064  *
4065  *
4066  * Need to be called with
4067  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4068  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4069  *
4070  * return > 0, number of blocks already mapped/allocated
4071  *          if create == 0 and these are pre-allocated blocks
4072  *              buffer head is unmapped
4073  *          otherwise blocks are mapped
4074  *
4075  * return = 0, if plain look up failed (blocks have not been allocated)
4076  *          buffer head is unmapped
4077  *
4078  * return < 0, error case.
4079  */
4080 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4081                         struct ext4_map_blocks *map, int flags)
4082 {
4083         struct ext4_ext_path *path = NULL;
4084         struct ext4_extent newex, *ex, ex2;
4085         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4086         ext4_fsblk_t newblock = 0, pblk;
4087         int err = 0, depth, ret;
4088         unsigned int allocated = 0, offset = 0;
4089         unsigned int allocated_clusters = 0;
4090         struct ext4_allocation_request ar;
4091         ext4_lblk_t cluster_offset;
4092
4093         ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4094         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4095
4096         /* find extent for this block */
4097         path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4098         if (IS_ERR(path)) {
4099                 err = PTR_ERR(path);
4100                 path = NULL;
4101                 goto out;
4102         }
4103
4104         depth = ext_depth(inode);
4105
4106         /*
4107          * consistent leaf must not be empty;
4108          * this situation is possible, though, _during_ tree modification;
4109          * this is why assert can't be put in ext4_find_extent()
4110          */
4111         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4112                 EXT4_ERROR_INODE(inode, "bad extent address "
4113                                  "lblock: %lu, depth: %d pblock %lld",
4114                                  (unsigned long) map->m_lblk, depth,
4115                                  path[depth].p_block);
4116                 err = -EFSCORRUPTED;
4117                 goto out;
4118         }
4119
4120         ex = path[depth].p_ext;
4121         if (ex) {
4122                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4123                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4124                 unsigned short ee_len;
4125
4126
4127                 /*
4128                  * unwritten extents are treated as holes, except that
4129                  * we split out initialized portions during a write.
4130                  */
4131                 ee_len = ext4_ext_get_actual_len(ex);
4132
4133                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4134
4135                 /* if found extent covers block, simply return it */
4136                 if (in_range(map->m_lblk, ee_block, ee_len)) {
4137                         newblock = map->m_lblk - ee_block + ee_start;
4138                         /* number of remaining blocks in the extent */
4139                         allocated = ee_len - (map->m_lblk - ee_block);
4140                         ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4141                                   map->m_lblk, ee_block, ee_len, newblock);
4142
4143                         /*
4144                          * If the extent is initialized check whether the
4145                          * caller wants to convert it to unwritten.
4146                          */
4147                         if ((!ext4_ext_is_unwritten(ex)) &&
4148                             (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4149                                 err = convert_initialized_extent(handle,
4150                                         inode, map, &path, &allocated);
4151                                 goto out;
4152                         } else if (!ext4_ext_is_unwritten(ex)) {
4153                                 map->m_flags |= EXT4_MAP_MAPPED;
4154                                 map->m_pblk = newblock;
4155                                 if (allocated > map->m_len)
4156                                         allocated = map->m_len;
4157                                 map->m_len = allocated;
4158                                 ext4_ext_show_leaf(inode, path);
4159                                 goto out;
4160                         }
4161
4162                         ret = ext4_ext_handle_unwritten_extents(
4163                                 handle, inode, map, &path, flags,
4164                                 allocated, newblock);
4165                         if (ret < 0)
4166                                 err = ret;
4167                         else
4168                                 allocated = ret;
4169                         goto out;
4170                 }
4171         }
4172
4173         /*
4174          * requested block isn't allocated yet;
4175          * we couldn't try to create block if create flag is zero
4176          */
4177         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4178                 ext4_lblk_t hole_start, hole_len;
4179
4180                 hole_start = map->m_lblk;
4181                 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4182                 /*
4183                  * put just found gap into cache to speed up
4184                  * subsequent requests
4185                  */
4186                 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4187
4188                 /* Update hole_len to reflect hole size after map->m_lblk */
4189                 if (hole_start != map->m_lblk)
4190                         hole_len -= map->m_lblk - hole_start;
4191                 map->m_pblk = 0;
4192                 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4193
4194                 goto out;
4195         }
4196
4197         /*
4198          * Okay, we need to do block allocation.
4199          */
4200         newex.ee_block = cpu_to_le32(map->m_lblk);
4201         cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4202
4203         /*
4204          * If we are doing bigalloc, check to see if the extent returned
4205          * by ext4_find_extent() implies a cluster we can use.
4206          */
4207         if (cluster_offset && ex &&
4208             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4209                 ar.len = allocated = map->m_len;
4210                 newblock = map->m_pblk;
4211                 goto got_allocated_blocks;
4212         }
4213
4214         /* find neighbour allocated blocks */
4215         ar.lleft = map->m_lblk;
4216         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4217         if (err)
4218                 goto out;
4219         ar.lright = map->m_lblk;
4220         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4221         if (err < 0)
4222                 goto out;
4223
4224         /* Check if the extent after searching to the right implies a
4225          * cluster we can use. */
4226         if ((sbi->s_cluster_ratio > 1) && err &&
4227             get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4228                 ar.len = allocated = map->m_len;
4229                 newblock = map->m_pblk;
4230                 goto got_allocated_blocks;
4231         }
4232
4233         /*
4234          * See if request is beyond maximum number of blocks we can have in
4235          * a single extent. For an initialized extent this limit is
4236          * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4237          * EXT_UNWRITTEN_MAX_LEN.
4238          */
4239         if (map->m_len > EXT_INIT_MAX_LEN &&
4240             !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4241                 map->m_len = EXT_INIT_MAX_LEN;
4242         else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4243                  (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4244                 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4245
4246         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4247         newex.ee_len = cpu_to_le16(map->m_len);
4248         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4249         if (err)
4250                 allocated = ext4_ext_get_actual_len(&newex);
4251         else
4252                 allocated = map->m_len;
4253
4254         /* allocate new block */
4255         ar.inode = inode;
4256         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4257         ar.logical = map->m_lblk;
4258         /*
4259          * We calculate the offset from the beginning of the cluster
4260          * for the logical block number, since when we allocate a
4261          * physical cluster, the physical block should start at the
4262          * same offset from the beginning of the cluster.  This is
4263          * needed so that future calls to get_implied_cluster_alloc()
4264          * work correctly.
4265          */
4266         offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4267         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4268         ar.goal -= offset;
4269         ar.logical -= offset;
4270         if (S_ISREG(inode->i_mode))
4271                 ar.flags = EXT4_MB_HINT_DATA;
4272         else
4273                 /* disable in-core preallocation for non-regular files */
4274                 ar.flags = 0;
4275         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4276                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4277         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4278                 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4279         if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4280                 ar.flags |= EXT4_MB_USE_RESERVED;
4281         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4282         if (!newblock)
4283                 goto out;
4284         allocated_clusters = ar.len;
4285         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4286         ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4287                   ar.goal, newblock, ar.len, allocated);
4288         if (ar.len > allocated)
4289                 ar.len = allocated;
4290
4291 got_allocated_blocks:
4292         /* try to insert new extent into found leaf and return */
4293         pblk = newblock + offset;
4294         ext4_ext_store_pblock(&newex, pblk);
4295         newex.ee_len = cpu_to_le16(ar.len);
4296         /* Mark unwritten */
4297         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4298                 ext4_ext_mark_unwritten(&newex);
4299                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4300         }
4301
4302         err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4303         if (err) {
4304                 if (allocated_clusters) {
4305                         int fb_flags = 0;
4306
4307                         /*
4308                          * free data blocks we just allocated.
4309                          * not a good idea to call discard here directly,
4310                          * but otherwise we'd need to call it every free().
4311                          */
4312                         ext4_discard_preallocations(inode, 0);
4313                         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4314                                 fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4315                         ext4_free_blocks(handle, inode, NULL, newblock,
4316                                          EXT4_C2B(sbi, allocated_clusters),
4317                                          fb_flags);
4318                 }
4319                 goto out;
4320         }
4321
4322         /*
4323          * Reduce the reserved cluster count to reflect successful deferred
4324          * allocation of delayed allocated clusters or direct allocation of
4325          * clusters discovered to be delayed allocated.  Once allocated, a
4326          * cluster is not included in the reserved count.
4327          */
4328         if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4329                 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4330                         /*
4331                          * When allocating delayed allocated clusters, simply
4332                          * reduce the reserved cluster count and claim quota
4333                          */
4334                         ext4_da_update_reserve_space(inode, allocated_clusters,
4335                                                         1);
4336                 } else {
4337                         ext4_lblk_t lblk, len;
4338                         unsigned int n;
4339
4340                         /*
4341                          * When allocating non-delayed allocated clusters
4342                          * (from fallocate, filemap, DIO, or clusters
4343                          * allocated when delalloc has been disabled by
4344                          * ext4_nonda_switch), reduce the reserved cluster
4345                          * count by the number of allocated clusters that
4346                          * have previously been delayed allocated.  Quota
4347                          * has been claimed by ext4_mb_new_blocks() above,
4348                          * so release the quota reservations made for any
4349                          * previously delayed allocated clusters.
4350                          */
4351                         lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4352                         len = allocated_clusters << sbi->s_cluster_bits;
4353                         n = ext4_es_delayed_clu(inode, lblk, len);
4354                         if (n > 0)
4355                                 ext4_da_update_reserve_space(inode, (int) n, 0);
4356                 }
4357         }
4358
4359         /*
4360          * Cache the extent and update transaction to commit on fdatasync only
4361          * when it is _not_ an unwritten extent.
4362          */
4363         if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4364                 ext4_update_inode_fsync_trans(handle, inode, 1);
4365         else
4366                 ext4_update_inode_fsync_trans(handle, inode, 0);
4367
4368         map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4369         map->m_pblk = pblk;
4370         map->m_len = ar.len;
4371         allocated = map->m_len;
4372         ext4_ext_show_leaf(inode, path);
4373 out:
4374         ext4_ext_drop_refs(path);
4375         kfree(path);
4376
4377         trace_ext4_ext_map_blocks_exit(inode, flags, map,
4378                                        err ? err : allocated);
4379         return err ? err : allocated;
4380 }
4381
4382 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4383 {
4384         struct super_block *sb = inode->i_sb;
4385         ext4_lblk_t last_block;
4386         int err = 0;
4387
4388         /*
4389          * TODO: optimization is possible here.
4390          * Probably we need not scan at all,
4391          * because page truncation is enough.
4392          */
4393
4394         /* we have to know where to truncate from in crash case */
4395         EXT4_I(inode)->i_disksize = inode->i_size;
4396         err = ext4_mark_inode_dirty(handle, inode);
4397         if (err)
4398                 return err;
4399
4400         last_block = (inode->i_size + sb->s_blocksize - 1)
4401                         >> EXT4_BLOCK_SIZE_BITS(sb);
4402 retry:
4403         err = ext4_es_remove_extent(inode, last_block,
4404                                     EXT_MAX_BLOCKS - last_block);
4405         if (err == -ENOMEM) {
4406                 cond_resched();
4407                 congestion_wait(BLK_RW_ASYNC, HZ/50);
4408                 goto retry;
4409         }
4410         if (err)
4411                 return err;
4412 retry_remove_space:
4413         err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4414         if (err == -ENOMEM) {
4415                 cond_resched();
4416                 congestion_wait(BLK_RW_ASYNC, HZ/50);
4417                 goto retry_remove_space;
4418         }
4419         return err;
4420 }
4421
4422 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4423                                   ext4_lblk_t len, loff_t new_size,
4424                                   int flags)
4425 {
4426         struct inode *inode = file_inode(file);
4427         handle_t *handle;
4428         int ret = 0;
4429         int ret2 = 0, ret3 = 0;
4430         int retries = 0;
4431         int depth = 0;
4432         struct ext4_map_blocks map;
4433         unsigned int credits;
4434         loff_t epos;
4435
4436         BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4437         map.m_lblk = offset;
4438         map.m_len = len;
4439         /*
4440          * Don't normalize the request if it can fit in one extent so
4441          * that it doesn't get unnecessarily split into multiple
4442          * extents.
4443          */
4444         if (len <= EXT_UNWRITTEN_MAX_LEN)
4445                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4446
4447         /*
4448          * credits to insert 1 extent into extent tree
4449          */
4450         credits = ext4_chunk_trans_blocks(inode, len);
4451         depth = ext_depth(inode);
4452
4453 retry:
4454         while (ret >= 0 && len) {
4455                 /*
4456                  * Recalculate credits when extent tree depth changes.
4457                  */
4458                 if (depth != ext_depth(inode)) {
4459                         credits = ext4_chunk_trans_blocks(inode, len);
4460                         depth = ext_depth(inode);
4461                 }
4462
4463                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4464                                             credits);
4465                 if (IS_ERR(handle)) {
4466                         ret = PTR_ERR(handle);
4467                         break;
4468                 }
4469                 ret = ext4_map_blocks(handle, inode, &map, flags);
4470                 if (ret <= 0) {
4471                         ext4_debug("inode #%lu: block %u: len %u: "
4472                                    "ext4_ext_map_blocks returned %d",
4473                                    inode->i_ino, map.m_lblk,
4474                                    map.m_len, ret);
4475                         ext4_mark_inode_dirty(handle, inode);
4476                         ret2 = ext4_journal_stop(handle);
4477                         break;
4478                 }
4479                 map.m_lblk += ret;
4480                 map.m_len = len = len - ret;
4481                 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4482                 inode->i_ctime = current_time(inode);
4483                 if (new_size) {
4484                         if (epos > new_size)
4485                                 epos = new_size;
4486                         if (ext4_update_inode_size(inode, epos) & 0x1)
4487                                 inode->i_mtime = inode->i_ctime;
4488                 }
4489                 ret2 = ext4_mark_inode_dirty(handle, inode);
4490                 ext4_update_inode_fsync_trans(handle, inode, 1);
4491                 ret3 = ext4_journal_stop(handle);
4492                 ret2 = ret3 ? ret3 : ret2;
4493                 if (unlikely(ret2))
4494                         break;
4495         }
4496         if (ret == -ENOSPC &&
4497                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4498                 ret = 0;
4499                 goto retry;
4500         }
4501
4502         return ret > 0 ? ret2 : ret;
4503 }
4504
4505 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len);
4506
4507 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len);
4508
4509 static long ext4_zero_range(struct file *file, loff_t offset,
4510                             loff_t len, int mode)
4511 {
4512         struct inode *inode = file_inode(file);
4513         handle_t *handle = NULL;
4514         unsigned int max_blocks;
4515         loff_t new_size = 0;
4516         int ret = 0;
4517         int flags;
4518         int credits;
4519         int partial_begin, partial_end;
4520         loff_t start, end;
4521         ext4_lblk_t lblk;
4522         unsigned int blkbits = inode->i_blkbits;
4523
4524         trace_ext4_zero_range(inode, offset, len, mode);
4525
4526         /* Call ext4_force_commit to flush all data in case of data=journal. */
4527         if (ext4_should_journal_data(inode)) {
4528                 ret = ext4_force_commit(inode->i_sb);
4529                 if (ret)
4530                         return ret;
4531         }
4532
4533         /*
4534          * Round up offset. This is not fallocate, we need to zero out
4535          * blocks, so convert interior block aligned part of the range to
4536          * unwritten and possibly manually zero out unaligned parts of the
4537          * range.
4538          */
4539         start = round_up(offset, 1 << blkbits);
4540         end = round_down((offset + len), 1 << blkbits);
4541
4542         if (start < offset || end > offset + len)
4543                 return -EINVAL;
4544         partial_begin = offset & ((1 << blkbits) - 1);
4545         partial_end = (offset + len) & ((1 << blkbits) - 1);
4546
4547         lblk = start >> blkbits;
4548         max_blocks = (end >> blkbits);
4549         if (max_blocks < lblk)
4550                 max_blocks = 0;
4551         else
4552                 max_blocks -= lblk;
4553
4554         inode_lock(inode);
4555
4556         /*
4557          * Indirect files do not support unwritten extents
4558          */
4559         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4560                 ret = -EOPNOTSUPP;
4561                 goto out_mutex;
4562         }
4563
4564         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4565             (offset + len > inode->i_size ||
4566              offset + len > EXT4_I(inode)->i_disksize)) {
4567                 new_size = offset + len;
4568                 ret = inode_newsize_ok(inode, new_size);
4569                 if (ret)
4570                         goto out_mutex;
4571         }
4572
4573         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4574
4575         /* Wait all existing dio workers, newcomers will block on i_mutex */
4576         inode_dio_wait(inode);
4577
4578         ret = file_modified(file);
4579         if (ret)
4580                 goto out_mutex;
4581
4582         /* Preallocate the range including the unaligned edges */
4583         if (partial_begin || partial_end) {
4584                 ret = ext4_alloc_file_blocks(file,
4585                                 round_down(offset, 1 << blkbits) >> blkbits,
4586                                 (round_up((offset + len), 1 << blkbits) -
4587                                  round_down(offset, 1 << blkbits)) >> blkbits,
4588                                 new_size, flags);
4589                 if (ret)
4590                         goto out_mutex;
4591
4592         }
4593
4594         /* Zero range excluding the unaligned edges */
4595         if (max_blocks > 0) {
4596                 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4597                           EXT4_EX_NOCACHE);
4598
4599                 /*
4600                  * Prevent page faults from reinstantiating pages we have
4601                  * released from page cache.
4602                  */
4603                 down_write(&EXT4_I(inode)->i_mmap_sem);
4604
4605                 ret = ext4_break_layouts(inode);
4606                 if (ret) {
4607                         up_write(&EXT4_I(inode)->i_mmap_sem);
4608                         goto out_mutex;
4609                 }
4610
4611                 ret = ext4_update_disksize_before_punch(inode, offset, len);
4612                 if (ret) {
4613                         up_write(&EXT4_I(inode)->i_mmap_sem);
4614                         goto out_mutex;
4615                 }
4616                 /* Now release the pages and zero block aligned part of pages */
4617                 truncate_pagecache_range(inode, start, end - 1);
4618                 inode->i_mtime = inode->i_ctime = current_time(inode);
4619
4620                 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4621                                              flags);
4622                 up_write(&EXT4_I(inode)->i_mmap_sem);
4623                 if (ret)
4624                         goto out_mutex;
4625         }
4626         if (!partial_begin && !partial_end)
4627                 goto out_mutex;
4628
4629         /*
4630          * In worst case we have to writeout two nonadjacent unwritten
4631          * blocks and update the inode
4632          */
4633         credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4634         if (ext4_should_journal_data(inode))
4635                 credits += 2;
4636         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4637         if (IS_ERR(handle)) {
4638                 ret = PTR_ERR(handle);
4639                 ext4_std_error(inode->i_sb, ret);
4640                 goto out_mutex;
4641         }
4642
4643         inode->i_mtime = inode->i_ctime = current_time(inode);
4644         if (new_size)
4645                 ext4_update_inode_size(inode, new_size);
4646         ret = ext4_mark_inode_dirty(handle, inode);
4647         if (unlikely(ret))
4648                 goto out_handle;
4649         /* Zero out partial block at the edges of the range */
4650         ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4651         if (ret >= 0)
4652                 ext4_update_inode_fsync_trans(handle, inode, 1);
4653
4654         if (file->f_flags & O_SYNC)
4655                 ext4_handle_sync(handle);
4656
4657 out_handle:
4658         ext4_journal_stop(handle);
4659 out_mutex:
4660         inode_unlock(inode);
4661         return ret;
4662 }
4663
4664 /*
4665  * preallocate space for a file. This implements ext4's fallocate file
4666  * operation, which gets called from sys_fallocate system call.
4667  * For block-mapped files, posix_fallocate should fall back to the method
4668  * of writing zeroes to the required new blocks (the same behavior which is
4669  * expected for file systems which do not support fallocate() system call).
4670  */
4671 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4672 {
4673         struct inode *inode = file_inode(file);
4674         loff_t new_size = 0;
4675         unsigned int max_blocks;
4676         int ret = 0;
4677         int flags;
4678         ext4_lblk_t lblk;
4679         unsigned int blkbits = inode->i_blkbits;
4680
4681         /*
4682          * Encrypted inodes can't handle collapse range or insert
4683          * range since we would need to re-encrypt blocks with a
4684          * different IV or XTS tweak (which are based on the logical
4685          * block number).
4686          */
4687         if (IS_ENCRYPTED(inode) &&
4688             (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4689                 return -EOPNOTSUPP;
4690
4691         /* Return error if mode is not supported */
4692         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4693                      FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4694                      FALLOC_FL_INSERT_RANGE))
4695                 return -EOPNOTSUPP;
4696
4697         ext4_fc_start_update(inode);
4698         inode_lock(inode);
4699         ret = ext4_convert_inline_data(inode);
4700         inode_unlock(inode);
4701         if (ret)
4702                 goto exit;
4703
4704         if (mode & FALLOC_FL_PUNCH_HOLE) {
4705                 ret = ext4_punch_hole(file, offset, len);
4706                 goto exit;
4707         }
4708
4709         if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4710                 ret = ext4_collapse_range(file, offset, len);
4711                 goto exit;
4712         }
4713
4714         if (mode & FALLOC_FL_INSERT_RANGE) {
4715                 ret = ext4_insert_range(file, offset, len);
4716                 goto exit;
4717         }
4718
4719         if (mode & FALLOC_FL_ZERO_RANGE) {
4720                 ret = ext4_zero_range(file, offset, len, mode);
4721                 goto exit;
4722         }
4723         trace_ext4_fallocate_enter(inode, offset, len, mode);
4724         lblk = offset >> blkbits;
4725
4726         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4727         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4728
4729         inode_lock(inode);
4730
4731         /*
4732          * We only support preallocation for extent-based files only
4733          */
4734         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4735                 ret = -EOPNOTSUPP;
4736                 goto out;
4737         }
4738
4739         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4740             (offset + len > inode->i_size ||
4741              offset + len > EXT4_I(inode)->i_disksize)) {
4742                 new_size = offset + len;
4743                 ret = inode_newsize_ok(inode, new_size);
4744                 if (ret)
4745                         goto out;
4746         }
4747
4748         /* Wait all existing dio workers, newcomers will block on i_mutex */
4749         inode_dio_wait(inode);
4750
4751         ret = file_modified(file);
4752         if (ret)
4753                 goto out;
4754
4755         ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4756         if (ret)
4757                 goto out;
4758
4759         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4760                 ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4761                                         EXT4_I(inode)->i_sync_tid);
4762         }
4763 out:
4764         inode_unlock(inode);
4765         trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4766 exit:
4767         ext4_fc_stop_update(inode);
4768         return ret;
4769 }
4770
4771 /*
4772  * This function convert a range of blocks to written extents
4773  * The caller of this function will pass the start offset and the size.
4774  * all unwritten extents within this range will be converted to
4775  * written extents.
4776  *
4777  * This function is called from the direct IO end io call back
4778  * function, to convert the fallocated extents after IO is completed.
4779  * Returns 0 on success.
4780  */
4781 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4782                                    loff_t offset, ssize_t len)
4783 {
4784         unsigned int max_blocks;
4785         int ret = 0, ret2 = 0, ret3 = 0;
4786         struct ext4_map_blocks map;
4787         unsigned int blkbits = inode->i_blkbits;
4788         unsigned int credits = 0;
4789
4790         map.m_lblk = offset >> blkbits;
4791         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4792
4793         if (!handle) {
4794                 /*
4795                  * credits to insert 1 extent into extent tree
4796                  */
4797                 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4798         }
4799         while (ret >= 0 && ret < max_blocks) {
4800                 map.m_lblk += ret;
4801                 map.m_len = (max_blocks -= ret);
4802                 if (credits) {
4803                         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4804                                                     credits);
4805                         if (IS_ERR(handle)) {
4806                                 ret = PTR_ERR(handle);
4807                                 break;
4808                         }
4809                 }
4810                 ret = ext4_map_blocks(handle, inode, &map,
4811                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4812                 if (ret <= 0)
4813                         ext4_warning(inode->i_sb,
4814                                      "inode #%lu: block %u: len %u: "
4815                                      "ext4_ext_map_blocks returned %d",
4816                                      inode->i_ino, map.m_lblk,
4817                                      map.m_len, ret);
4818                 ret2 = ext4_mark_inode_dirty(handle, inode);
4819                 if (credits) {
4820                         ret3 = ext4_journal_stop(handle);
4821                         if (unlikely(ret3))
4822                                 ret2 = ret3;
4823                 }
4824
4825                 if (ret <= 0 || ret2)
4826                         break;
4827         }
4828         return ret > 0 ? ret2 : ret;
4829 }
4830
4831 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4832 {
4833         int ret = 0, err = 0;
4834         struct ext4_io_end_vec *io_end_vec;
4835
4836         /*
4837          * This is somewhat ugly but the idea is clear: When transaction is
4838          * reserved, everything goes into it. Otherwise we rather start several
4839          * smaller transactions for conversion of each extent separately.
4840          */
4841         if (handle) {
4842                 handle = ext4_journal_start_reserved(handle,
4843                                                      EXT4_HT_EXT_CONVERT);
4844                 if (IS_ERR(handle))
4845                         return PTR_ERR(handle);
4846         }
4847
4848         list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4849                 ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4850                                                      io_end_vec->offset,
4851                                                      io_end_vec->size);
4852                 if (ret)
4853                         break;
4854         }
4855
4856         if (handle)
4857                 err = ext4_journal_stop(handle);
4858
4859         return ret < 0 ? ret : err;
4860 }
4861
4862 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4863 {
4864         __u64 physical = 0;
4865         __u64 length = 0;
4866         int blockbits = inode->i_sb->s_blocksize_bits;
4867         int error = 0;
4868         u16 iomap_type;
4869
4870         /* in-inode? */
4871         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4872                 struct ext4_iloc iloc;
4873                 int offset;     /* offset of xattr in inode */
4874
4875                 error = ext4_get_inode_loc(inode, &iloc);
4876                 if (error)
4877                         return error;
4878                 physical = (__u64)iloc.bh->b_blocknr << blockbits;
4879                 offset = EXT4_GOOD_OLD_INODE_SIZE +
4880                                 EXT4_I(inode)->i_extra_isize;
4881                 physical += offset;
4882                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4883                 brelse(iloc.bh);
4884                 iomap_type = IOMAP_INLINE;
4885         } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4886                 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4887                 length = inode->i_sb->s_blocksize;
4888                 iomap_type = IOMAP_MAPPED;
4889         } else {
4890                 /* no in-inode or external block for xattr, so return -ENOENT */
4891                 error = -ENOENT;
4892                 goto out;
4893         }
4894
4895         iomap->addr = physical;
4896         iomap->offset = 0;
4897         iomap->length = length;
4898         iomap->type = iomap_type;
4899         iomap->flags = 0;
4900 out:
4901         return error;
4902 }
4903
4904 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4905                                   loff_t length, unsigned flags,
4906                                   struct iomap *iomap, struct iomap *srcmap)
4907 {
4908         int error;
4909
4910         error = ext4_iomap_xattr_fiemap(inode, iomap);
4911         if (error == 0 && (offset >= iomap->length))
4912                 error = -ENOENT;
4913         return error;
4914 }
4915
4916 static const struct iomap_ops ext4_iomap_xattr_ops = {
4917         .iomap_begin            = ext4_iomap_xattr_begin,
4918 };
4919
4920 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4921 {
4922         u64 maxbytes;
4923
4924         if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4925                 maxbytes = inode->i_sb->s_maxbytes;
4926         else
4927                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4928
4929         if (*len == 0)
4930                 return -EINVAL;
4931         if (start > maxbytes)
4932                 return -EFBIG;
4933
4934         /*
4935          * Shrink request scope to what the fs can actually handle.
4936          */
4937         if (*len > maxbytes || (maxbytes - *len) < start)
4938                 *len = maxbytes - start;
4939         return 0;
4940 }
4941
4942 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4943                 u64 start, u64 len)
4944 {
4945         int error = 0;
4946
4947         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4948                 error = ext4_ext_precache(inode);
4949                 if (error)
4950                         return error;
4951                 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4952         }
4953
4954         /*
4955          * For bitmap files the maximum size limit could be smaller than
4956          * s_maxbytes, so check len here manually instead of just relying on the
4957          * generic check.
4958          */
4959         error = ext4_fiemap_check_ranges(inode, start, &len);
4960         if (error)
4961                 return error;
4962
4963         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4964                 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4965                 return iomap_fiemap(inode, fieinfo, start, len,
4966                                     &ext4_iomap_xattr_ops);
4967         }
4968
4969         return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4970 }
4971
4972 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4973                       __u64 start, __u64 len)
4974 {
4975         ext4_lblk_t start_blk, len_blks;
4976         __u64 last_blk;
4977         int error = 0;
4978
4979         if (ext4_has_inline_data(inode)) {
4980                 int has_inline;
4981
4982                 down_read(&EXT4_I(inode)->xattr_sem);
4983                 has_inline = ext4_has_inline_data(inode);
4984                 up_read(&EXT4_I(inode)->xattr_sem);
4985                 if (has_inline)
4986                         return 0;
4987         }
4988
4989         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4990                 error = ext4_ext_precache(inode);
4991                 if (error)
4992                         return error;
4993                 fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4994         }
4995
4996         error = fiemap_prep(inode, fieinfo, start, &len, 0);
4997         if (error)
4998                 return error;
4999
5000         error = ext4_fiemap_check_ranges(inode, start, &len);
5001         if (error)
5002                 return error;
5003
5004         start_blk = start >> inode->i_sb->s_blocksize_bits;
5005         last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5006         if (last_blk >= EXT_MAX_BLOCKS)
5007                 last_blk = EXT_MAX_BLOCKS-1;
5008         len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5009
5010         /*
5011          * Walk the extent tree gathering extent information
5012          * and pushing extents back to the user.
5013          */
5014         return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5015 }
5016
5017 /*
5018  * ext4_ext_shift_path_extents:
5019  * Shift the extents of a path structure lying between path[depth].p_ext
5020  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5021  * if it is right shift or left shift operation.
5022  */
5023 static int
5024 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5025                             struct inode *inode, handle_t *handle,
5026                             enum SHIFT_DIRECTION SHIFT)
5027 {
5028         int depth, err = 0;
5029         struct ext4_extent *ex_start, *ex_last;
5030         bool update = false;
5031         int credits, restart_credits;
5032         depth = path->p_depth;
5033
5034         while (depth >= 0) {
5035                 if (depth == path->p_depth) {
5036                         ex_start = path[depth].p_ext;
5037                         if (!ex_start)
5038                                 return -EFSCORRUPTED;
5039
5040                         ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5041                         /* leaf + sb + inode */
5042                         credits = 3;
5043                         if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5044                                 update = true;
5045                                 /* extent tree + sb + inode */
5046                                 credits = depth + 2;
5047                         }
5048
5049                         restart_credits = ext4_writepage_trans_blocks(inode);
5050                         err = ext4_datasem_ensure_credits(handle, inode, credits,
5051                                         restart_credits, 0);
5052                         if (err) {
5053                                 if (err > 0)
5054                                         err = -EAGAIN;
5055                                 goto out;
5056                         }
5057
5058                         err = ext4_ext_get_access(handle, inode, path + depth);
5059                         if (err)
5060                                 goto out;
5061
5062                         while (ex_start <= ex_last) {
5063                                 if (SHIFT == SHIFT_LEFT) {
5064                                         le32_add_cpu(&ex_start->ee_block,
5065                                                 -shift);
5066                                         /* Try to merge to the left. */
5067                                         if ((ex_start >
5068                                             EXT_FIRST_EXTENT(path[depth].p_hdr))
5069                                             &&
5070                                             ext4_ext_try_to_merge_right(inode,
5071                                             path, ex_start - 1))
5072                                                 ex_last--;
5073                                         else
5074                                                 ex_start++;
5075                                 } else {
5076                                         le32_add_cpu(&ex_last->ee_block, shift);
5077                                         ext4_ext_try_to_merge_right(inode, path,
5078                                                 ex_last);
5079                                         ex_last--;
5080                                 }
5081                         }
5082                         err = ext4_ext_dirty(handle, inode, path + depth);
5083                         if (err)
5084                                 goto out;
5085
5086                         if (--depth < 0 || !update)
5087                                 break;
5088                 }
5089
5090                 /* Update index too */
5091                 err = ext4_ext_get_access(handle, inode, path + depth);
5092                 if (err)
5093                         goto out;
5094
5095                 if (SHIFT == SHIFT_LEFT)
5096                         le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5097                 else
5098                         le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5099                 err = ext4_ext_dirty(handle, inode, path + depth);
5100                 if (err)
5101                         goto out;
5102
5103                 /* we are done if current index is not a starting index */
5104                 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5105                         break;
5106
5107                 depth--;
5108         }
5109
5110 out:
5111         return err;
5112 }
5113
5114 /*
5115  * ext4_ext_shift_extents:
5116  * All the extents which lies in the range from @start to the last allocated
5117  * block for the @inode are shifted either towards left or right (depending
5118  * upon @SHIFT) by @shift blocks.
5119  * On success, 0 is returned, error otherwise.
5120  */
5121 static int
5122 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5123                        ext4_lblk_t start, ext4_lblk_t shift,
5124                        enum SHIFT_DIRECTION SHIFT)
5125 {
5126         struct ext4_ext_path *path;
5127         int ret = 0, depth;
5128         struct ext4_extent *extent;
5129         ext4_lblk_t stop, *iterator, ex_start, ex_end;
5130         ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5131
5132         /* Let path point to the last extent */
5133         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5134                                 EXT4_EX_NOCACHE);
5135         if (IS_ERR(path))
5136                 return PTR_ERR(path);
5137
5138         depth = path->p_depth;
5139         extent = path[depth].p_ext;
5140         if (!extent)
5141                 goto out;
5142
5143         stop = le32_to_cpu(extent->ee_block);
5144
5145        /*
5146         * For left shifts, make sure the hole on the left is big enough to
5147         * accommodate the shift.  For right shifts, make sure the last extent
5148         * won't be shifted beyond EXT_MAX_BLOCKS.
5149         */
5150         if (SHIFT == SHIFT_LEFT) {
5151                 path = ext4_find_extent(inode, start - 1, &path,
5152                                         EXT4_EX_NOCACHE);
5153                 if (IS_ERR(path))
5154                         return PTR_ERR(path);
5155                 depth = path->p_depth;
5156                 extent =  path[depth].p_ext;
5157                 if (extent) {
5158                         ex_start = le32_to_cpu(extent->ee_block);
5159                         ex_end = le32_to_cpu(extent->ee_block) +
5160                                 ext4_ext_get_actual_len(extent);
5161                 } else {
5162                         ex_start = 0;
5163                         ex_end = 0;
5164                 }
5165
5166                 if ((start == ex_start && shift > ex_start) ||
5167                     (shift > start - ex_end)) {
5168                         ret = -EINVAL;
5169                         goto out;
5170                 }
5171         } else {
5172                 if (shift > EXT_MAX_BLOCKS -
5173                     (stop + ext4_ext_get_actual_len(extent))) {
5174                         ret = -EINVAL;
5175                         goto out;
5176                 }
5177         }
5178
5179         /*
5180          * In case of left shift, iterator points to start and it is increased
5181          * till we reach stop. In case of right shift, iterator points to stop
5182          * and it is decreased till we reach start.
5183          */
5184 again:
5185         if (SHIFT == SHIFT_LEFT)
5186                 iterator = &start;
5187         else
5188                 iterator = &stop;
5189
5190         if (tmp != EXT_MAX_BLOCKS)
5191                 *iterator = tmp;
5192
5193         /*
5194          * Its safe to start updating extents.  Start and stop are unsigned, so
5195          * in case of right shift if extent with 0 block is reached, iterator
5196          * becomes NULL to indicate the end of the loop.
5197          */
5198         while (iterator && start <= stop) {
5199                 path = ext4_find_extent(inode, *iterator, &path,
5200                                         EXT4_EX_NOCACHE);
5201                 if (IS_ERR(path))
5202                         return PTR_ERR(path);
5203                 depth = path->p_depth;
5204                 extent = path[depth].p_ext;
5205                 if (!extent) {
5206                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5207                                          (unsigned long) *iterator);
5208                         return -EFSCORRUPTED;
5209                 }
5210                 if (SHIFT == SHIFT_LEFT && *iterator >
5211                     le32_to_cpu(extent->ee_block)) {
5212                         /* Hole, move to the next extent */
5213                         if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5214                                 path[depth].p_ext++;
5215                         } else {
5216                                 *iterator = ext4_ext_next_allocated_block(path);
5217                                 continue;
5218                         }
5219                 }
5220
5221                 tmp = *iterator;
5222                 if (SHIFT == SHIFT_LEFT) {
5223                         extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5224                         *iterator = le32_to_cpu(extent->ee_block) +
5225                                         ext4_ext_get_actual_len(extent);
5226                 } else {
5227                         extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5228                         if (le32_to_cpu(extent->ee_block) > 0)
5229                                 *iterator = le32_to_cpu(extent->ee_block) - 1;
5230                         else
5231                                 /* Beginning is reached, end of the loop */
5232                                 iterator = NULL;
5233                         /* Update path extent in case we need to stop */
5234                         while (le32_to_cpu(extent->ee_block) < start)
5235                                 extent++;
5236                         path[depth].p_ext = extent;
5237                 }
5238                 ret = ext4_ext_shift_path_extents(path, shift, inode,
5239                                 handle, SHIFT);
5240                 /* iterator can be NULL which means we should break */
5241                 if (ret == -EAGAIN)
5242                         goto again;
5243                 if (ret)
5244                         break;
5245         }
5246 out:
5247         ext4_ext_drop_refs(path);
5248         kfree(path);
5249         return ret;
5250 }
5251
5252 /*
5253  * ext4_collapse_range:
5254  * This implements the fallocate's collapse range functionality for ext4
5255  * Returns: 0 and non-zero on error.
5256  */
5257 static int ext4_collapse_range(struct file *file, loff_t offset, loff_t len)
5258 {
5259         struct inode *inode = file_inode(file);
5260         struct super_block *sb = inode->i_sb;
5261         ext4_lblk_t punch_start, punch_stop;
5262         handle_t *handle;
5263         unsigned int credits;
5264         loff_t new_size, ioffset;
5265         int ret;
5266
5267         /*
5268          * We need to test this early because xfstests assumes that a
5269          * collapse range of (0, 1) will return EOPNOTSUPP if the file
5270          * system does not support collapse range.
5271          */
5272         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5273                 return -EOPNOTSUPP;
5274
5275         /* Collapse range works only on fs cluster size aligned regions. */
5276         if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5277                 return -EINVAL;
5278
5279         trace_ext4_collapse_range(inode, offset, len);
5280
5281         punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5282         punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5283
5284         /* Call ext4_force_commit to flush all data in case of data=journal. */
5285         if (ext4_should_journal_data(inode)) {
5286                 ret = ext4_force_commit(inode->i_sb);
5287                 if (ret)
5288                         return ret;
5289         }
5290
5291         inode_lock(inode);
5292         /*
5293          * There is no need to overlap collapse range with EOF, in which case
5294          * it is effectively a truncate operation
5295          */
5296         if (offset + len >= inode->i_size) {
5297                 ret = -EINVAL;
5298                 goto out_mutex;
5299         }
5300
5301         /* Currently just for extent based files */
5302         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5303                 ret = -EOPNOTSUPP;
5304                 goto out_mutex;
5305         }
5306
5307         /* Wait for existing dio to complete */
5308         inode_dio_wait(inode);
5309
5310         ret = file_modified(file);
5311         if (ret)
5312                 goto out_mutex;
5313
5314         /*
5315          * Prevent page faults from reinstantiating pages we have released from
5316          * page cache.
5317          */
5318         down_write(&EXT4_I(inode)->i_mmap_sem);
5319
5320         ret = ext4_break_layouts(inode);
5321         if (ret)
5322                 goto out_mmap;
5323
5324         /*
5325          * Need to round down offset to be aligned with page size boundary
5326          * for page size > block size.
5327          */
5328         ioffset = round_down(offset, PAGE_SIZE);
5329         /*
5330          * Write tail of the last page before removed range since it will get
5331          * removed from the page cache below.
5332          */
5333         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5334         if (ret)
5335                 goto out_mmap;
5336         /*
5337          * Write data that will be shifted to preserve them when discarding
5338          * page cache below. We are also protected from pages becoming dirty
5339          * by i_mmap_sem.
5340          */
5341         ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5342                                            LLONG_MAX);
5343         if (ret)
5344                 goto out_mmap;
5345         truncate_pagecache(inode, ioffset);
5346
5347         credits = ext4_writepage_trans_blocks(inode);
5348         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5349         if (IS_ERR(handle)) {
5350                 ret = PTR_ERR(handle);
5351                 goto out_mmap;
5352         }
5353         ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5354
5355         down_write(&EXT4_I(inode)->i_data_sem);
5356         ext4_discard_preallocations(inode, 0);
5357
5358         ret = ext4_es_remove_extent(inode, punch_start,
5359                                     EXT_MAX_BLOCKS - punch_start);
5360         if (ret) {
5361                 up_write(&EXT4_I(inode)->i_data_sem);
5362                 goto out_stop;
5363         }
5364
5365         ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5366         if (ret) {
5367                 up_write(&EXT4_I(inode)->i_data_sem);
5368                 goto out_stop;
5369         }
5370         ext4_discard_preallocations(inode, 0);
5371
5372         ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5373                                      punch_stop - punch_start, SHIFT_LEFT);
5374         if (ret) {
5375                 up_write(&EXT4_I(inode)->i_data_sem);
5376                 goto out_stop;
5377         }
5378
5379         new_size = inode->i_size - len;
5380         i_size_write(inode, new_size);
5381         EXT4_I(inode)->i_disksize = new_size;
5382
5383         up_write(&EXT4_I(inode)->i_data_sem);
5384         if (IS_SYNC(inode))
5385                 ext4_handle_sync(handle);
5386         inode->i_mtime = inode->i_ctime = current_time(inode);
5387         ret = ext4_mark_inode_dirty(handle, inode);
5388         ext4_update_inode_fsync_trans(handle, inode, 1);
5389
5390 out_stop:
5391         ext4_journal_stop(handle);
5392         ext4_fc_stop_ineligible(sb);
5393 out_mmap:
5394         up_write(&EXT4_I(inode)->i_mmap_sem);
5395 out_mutex:
5396         inode_unlock(inode);
5397         return ret;
5398 }
5399
5400 /*
5401  * ext4_insert_range:
5402  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5403  * The data blocks starting from @offset to the EOF are shifted by @len
5404  * towards right to create a hole in the @inode. Inode size is increased
5405  * by len bytes.
5406  * Returns 0 on success, error otherwise.
5407  */
5408 static int ext4_insert_range(struct file *file, loff_t offset, loff_t len)
5409 {
5410         struct inode *inode = file_inode(file);
5411         struct super_block *sb = inode->i_sb;
5412         handle_t *handle;
5413         struct ext4_ext_path *path;
5414         struct ext4_extent *extent;
5415         ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5416         unsigned int credits, ee_len;
5417         int ret = 0, depth, split_flag = 0;
5418         loff_t ioffset;
5419
5420         /*
5421          * We need to test this early because xfstests assumes that an
5422          * insert range of (0, 1) will return EOPNOTSUPP if the file
5423          * system does not support insert range.
5424          */
5425         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5426                 return -EOPNOTSUPP;
5427
5428         /* Insert range works only on fs cluster size aligned regions. */
5429         if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5430                 return -EINVAL;
5431
5432         trace_ext4_insert_range(inode, offset, len);
5433
5434         offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5435         len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5436
5437         /* Call ext4_force_commit to flush all data in case of data=journal */
5438         if (ext4_should_journal_data(inode)) {
5439                 ret = ext4_force_commit(inode->i_sb);
5440                 if (ret)
5441                         return ret;
5442         }
5443
5444         inode_lock(inode);
5445         /* Currently just for extent based files */
5446         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5447                 ret = -EOPNOTSUPP;
5448                 goto out_mutex;
5449         }
5450
5451         /* Check whether the maximum file size would be exceeded */
5452         if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5453                 ret = -EFBIG;
5454                 goto out_mutex;
5455         }
5456
5457         /* Offset must be less than i_size */
5458         if (offset >= inode->i_size) {
5459                 ret = -EINVAL;
5460                 goto out_mutex;
5461         }
5462
5463         /* Wait for existing dio to complete */
5464         inode_dio_wait(inode);
5465
5466         ret = file_modified(file);
5467         if (ret)
5468                 goto out_mutex;
5469
5470         /*
5471          * Prevent page faults from reinstantiating pages we have released from
5472          * page cache.
5473          */
5474         down_write(&EXT4_I(inode)->i_mmap_sem);
5475
5476         ret = ext4_break_layouts(inode);
5477         if (ret)
5478                 goto out_mmap;
5479
5480         /*
5481          * Need to round down to align start offset to page size boundary
5482          * for page size > block size.
5483          */
5484         ioffset = round_down(offset, PAGE_SIZE);
5485         /* Write out all dirty pages */
5486         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5487                         LLONG_MAX);
5488         if (ret)
5489                 goto out_mmap;
5490         truncate_pagecache(inode, ioffset);
5491
5492         credits = ext4_writepage_trans_blocks(inode);
5493         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5494         if (IS_ERR(handle)) {
5495                 ret = PTR_ERR(handle);
5496                 goto out_mmap;
5497         }
5498         ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5499
5500         /* Expand file to avoid data loss if there is error while shifting */
5501         inode->i_size += len;
5502         EXT4_I(inode)->i_disksize += len;
5503         inode->i_mtime = inode->i_ctime = current_time(inode);
5504         ret = ext4_mark_inode_dirty(handle, inode);
5505         if (ret)
5506                 goto out_stop;
5507
5508         down_write(&EXT4_I(inode)->i_data_sem);
5509         ext4_discard_preallocations(inode, 0);
5510
5511         path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5512         if (IS_ERR(path)) {
5513                 up_write(&EXT4_I(inode)->i_data_sem);
5514                 goto out_stop;
5515         }
5516
5517         depth = ext_depth(inode);
5518         extent = path[depth].p_ext;
5519         if (extent) {
5520                 ee_start_lblk = le32_to_cpu(extent->ee_block);
5521                 ee_len = ext4_ext_get_actual_len(extent);
5522
5523                 /*
5524                  * If offset_lblk is not the starting block of extent, split
5525                  * the extent @offset_lblk
5526                  */
5527                 if ((offset_lblk > ee_start_lblk) &&
5528                                 (offset_lblk < (ee_start_lblk + ee_len))) {
5529                         if (ext4_ext_is_unwritten(extent))
5530                                 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5531                                         EXT4_EXT_MARK_UNWRIT2;
5532                         ret = ext4_split_extent_at(handle, inode, &path,
5533                                         offset_lblk, split_flag,
5534                                         EXT4_EX_NOCACHE |
5535                                         EXT4_GET_BLOCKS_PRE_IO |
5536                                         EXT4_GET_BLOCKS_METADATA_NOFAIL);
5537                 }
5538
5539                 ext4_ext_drop_refs(path);
5540                 kfree(path);
5541                 if (ret < 0) {
5542                         up_write(&EXT4_I(inode)->i_data_sem);
5543                         goto out_stop;
5544                 }
5545         } else {
5546                 ext4_ext_drop_refs(path);
5547                 kfree(path);
5548         }
5549
5550         ret = ext4_es_remove_extent(inode, offset_lblk,
5551                         EXT_MAX_BLOCKS - offset_lblk);
5552         if (ret) {
5553                 up_write(&EXT4_I(inode)->i_data_sem);
5554                 goto out_stop;
5555         }
5556
5557         /*
5558          * if offset_lblk lies in a hole which is at start of file, use
5559          * ee_start_lblk to shift extents
5560          */
5561         ret = ext4_ext_shift_extents(inode, handle,
5562                 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5563                 len_lblk, SHIFT_RIGHT);
5564
5565         up_write(&EXT4_I(inode)->i_data_sem);
5566         if (IS_SYNC(inode))
5567                 ext4_handle_sync(handle);
5568         if (ret >= 0)
5569                 ext4_update_inode_fsync_trans(handle, inode, 1);
5570
5571 out_stop:
5572         ext4_journal_stop(handle);
5573         ext4_fc_stop_ineligible(sb);
5574 out_mmap:
5575         up_write(&EXT4_I(inode)->i_mmap_sem);
5576 out_mutex:
5577         inode_unlock(inode);
5578         return ret;
5579 }
5580
5581 /**
5582  * ext4_swap_extents() - Swap extents between two inodes
5583  * @handle: handle for this transaction
5584  * @inode1:     First inode
5585  * @inode2:     Second inode
5586  * @lblk1:      Start block for first inode
5587  * @lblk2:      Start block for second inode
5588  * @count:      Number of blocks to swap
5589  * @unwritten: Mark second inode's extents as unwritten after swap
5590  * @erp:        Pointer to save error value
5591  *
5592  * This helper routine does exactly what is promise "swap extents". All other
5593  * stuff such as page-cache locking consistency, bh mapping consistency or
5594  * extent's data copying must be performed by caller.
5595  * Locking:
5596  *              i_mutex is held for both inodes
5597  *              i_data_sem is locked for write for both inodes
5598  * Assumptions:
5599  *              All pages from requested range are locked for both inodes
5600  */
5601 int
5602 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5603                   struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5604                   ext4_lblk_t count, int unwritten, int *erp)
5605 {
5606         struct ext4_ext_path *path1 = NULL;
5607         struct ext4_ext_path *path2 = NULL;
5608         int replaced_count = 0;
5609
5610         BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5611         BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5612         BUG_ON(!inode_is_locked(inode1));
5613         BUG_ON(!inode_is_locked(inode2));
5614
5615         *erp = ext4_es_remove_extent(inode1, lblk1, count);
5616         if (unlikely(*erp))
5617                 return 0;
5618         *erp = ext4_es_remove_extent(inode2, lblk2, count);
5619         if (unlikely(*erp))
5620                 return 0;
5621
5622         while (count) {
5623                 struct ext4_extent *ex1, *ex2, tmp_ex;
5624                 ext4_lblk_t e1_blk, e2_blk;
5625                 int e1_len, e2_len, len;
5626                 int split = 0;
5627
5628                 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5629                 if (IS_ERR(path1)) {
5630                         *erp = PTR_ERR(path1);
5631                         path1 = NULL;
5632                 finish:
5633                         count = 0;
5634                         goto repeat;
5635                 }
5636                 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5637                 if (IS_ERR(path2)) {
5638                         *erp = PTR_ERR(path2);
5639                         path2 = NULL;
5640                         goto finish;
5641                 }
5642                 ex1 = path1[path1->p_depth].p_ext;
5643                 ex2 = path2[path2->p_depth].p_ext;
5644                 /* Do we have something to swap ? */
5645                 if (unlikely(!ex2 || !ex1))
5646                         goto finish;
5647
5648                 e1_blk = le32_to_cpu(ex1->ee_block);
5649                 e2_blk = le32_to_cpu(ex2->ee_block);
5650                 e1_len = ext4_ext_get_actual_len(ex1);
5651                 e2_len = ext4_ext_get_actual_len(ex2);
5652
5653                 /* Hole handling */
5654                 if (!in_range(lblk1, e1_blk, e1_len) ||
5655                     !in_range(lblk2, e2_blk, e2_len)) {
5656                         ext4_lblk_t next1, next2;
5657
5658                         /* if hole after extent, then go to next extent */
5659                         next1 = ext4_ext_next_allocated_block(path1);
5660                         next2 = ext4_ext_next_allocated_block(path2);
5661                         /* If hole before extent, then shift to that extent */
5662                         if (e1_blk > lblk1)
5663                                 next1 = e1_blk;
5664                         if (e2_blk > lblk2)
5665                                 next2 = e2_blk;
5666                         /* Do we have something to swap */
5667                         if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5668                                 goto finish;
5669                         /* Move to the rightest boundary */
5670                         len = next1 - lblk1;
5671                         if (len < next2 - lblk2)
5672                                 len = next2 - lblk2;
5673                         if (len > count)
5674                                 len = count;
5675                         lblk1 += len;
5676                         lblk2 += len;
5677                         count -= len;
5678                         goto repeat;
5679                 }
5680
5681                 /* Prepare left boundary */
5682                 if (e1_blk < lblk1) {
5683                         split = 1;
5684                         *erp = ext4_force_split_extent_at(handle, inode1,
5685                                                 &path1, lblk1, 0);
5686                         if (unlikely(*erp))
5687                                 goto finish;
5688                 }
5689                 if (e2_blk < lblk2) {
5690                         split = 1;
5691                         *erp = ext4_force_split_extent_at(handle, inode2,
5692                                                 &path2,  lblk2, 0);
5693                         if (unlikely(*erp))
5694                                 goto finish;
5695                 }
5696                 /* ext4_split_extent_at() may result in leaf extent split,
5697                  * path must to be revalidated. */
5698                 if (split)
5699                         goto repeat;
5700
5701                 /* Prepare right boundary */
5702                 len = count;
5703                 if (len > e1_blk + e1_len - lblk1)
5704                         len = e1_blk + e1_len - lblk1;
5705                 if (len > e2_blk + e2_len - lblk2)
5706                         len = e2_blk + e2_len - lblk2;
5707
5708                 if (len != e1_len) {
5709                         split = 1;
5710                         *erp = ext4_force_split_extent_at(handle, inode1,
5711                                                 &path1, lblk1 + len, 0);
5712                         if (unlikely(*erp))
5713                                 goto finish;
5714                 }
5715                 if (len != e2_len) {
5716                         split = 1;
5717                         *erp = ext4_force_split_extent_at(handle, inode2,
5718                                                 &path2, lblk2 + len, 0);
5719                         if (*erp)
5720                                 goto finish;
5721                 }
5722                 /* ext4_split_extent_at() may result in leaf extent split,
5723                  * path must to be revalidated. */
5724                 if (split)
5725                         goto repeat;
5726
5727                 BUG_ON(e2_len != e1_len);
5728                 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5729                 if (unlikely(*erp))
5730                         goto finish;
5731                 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5732                 if (unlikely(*erp))
5733                         goto finish;
5734
5735                 /* Both extents are fully inside boundaries. Swap it now */
5736                 tmp_ex = *ex1;
5737                 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5738                 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5739                 ex1->ee_len = cpu_to_le16(e2_len);
5740                 ex2->ee_len = cpu_to_le16(e1_len);
5741                 if (unwritten)
5742                         ext4_ext_mark_unwritten(ex2);
5743                 if (ext4_ext_is_unwritten(&tmp_ex))
5744                         ext4_ext_mark_unwritten(ex1);
5745
5746                 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5747                 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5748                 *erp = ext4_ext_dirty(handle, inode2, path2 +
5749                                       path2->p_depth);
5750                 if (unlikely(*erp))
5751                         goto finish;
5752                 *erp = ext4_ext_dirty(handle, inode1, path1 +
5753                                       path1->p_depth);
5754                 /*
5755                  * Looks scarry ah..? second inode already points to new blocks,
5756                  * and it was successfully dirtied. But luckily error may happen
5757                  * only due to journal error, so full transaction will be
5758                  * aborted anyway.
5759                  */
5760                 if (unlikely(*erp))
5761                         goto finish;
5762                 lblk1 += len;
5763                 lblk2 += len;
5764                 replaced_count += len;
5765                 count -= len;
5766
5767         repeat:
5768                 ext4_ext_drop_refs(path1);
5769                 kfree(path1);
5770                 ext4_ext_drop_refs(path2);
5771                 kfree(path2);
5772                 path1 = path2 = NULL;
5773         }
5774         return replaced_count;
5775 }
5776
5777 /*
5778  * ext4_clu_mapped - determine whether any block in a logical cluster has
5779  *                   been mapped to a physical cluster
5780  *
5781  * @inode - file containing the logical cluster
5782  * @lclu - logical cluster of interest
5783  *
5784  * Returns 1 if any block in the logical cluster is mapped, signifying
5785  * that a physical cluster has been allocated for it.  Otherwise,
5786  * returns 0.  Can also return negative error codes.  Derived from
5787  * ext4_ext_map_blocks().
5788  */
5789 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5790 {
5791         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5792         struct ext4_ext_path *path;
5793         int depth, mapped = 0, err = 0;
5794         struct ext4_extent *extent;
5795         ext4_lblk_t first_lblk, first_lclu, last_lclu;
5796
5797         /* search for the extent closest to the first block in the cluster */
5798         path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5799         if (IS_ERR(path)) {
5800                 err = PTR_ERR(path);
5801                 path = NULL;
5802                 goto out;
5803         }
5804
5805         depth = ext_depth(inode);
5806
5807         /*
5808          * A consistent leaf must not be empty.  This situation is possible,
5809          * though, _during_ tree modification, and it's why an assert can't
5810          * be put in ext4_find_extent().
5811          */
5812         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5813                 EXT4_ERROR_INODE(inode,
5814                     "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5815                                  (unsigned long) EXT4_C2B(sbi, lclu),
5816                                  depth, path[depth].p_block);
5817                 err = -EFSCORRUPTED;
5818                 goto out;
5819         }
5820
5821         extent = path[depth].p_ext;
5822
5823         /* can't be mapped if the extent tree is empty */
5824         if (extent == NULL)
5825                 goto out;
5826
5827         first_lblk = le32_to_cpu(extent->ee_block);
5828         first_lclu = EXT4_B2C(sbi, first_lblk);
5829
5830         /*
5831          * Three possible outcomes at this point - found extent spanning
5832          * the target cluster, to the left of the target cluster, or to the
5833          * right of the target cluster.  The first two cases are handled here.
5834          * The last case indicates the target cluster is not mapped.
5835          */
5836         if (lclu >= first_lclu) {
5837                 last_lclu = EXT4_B2C(sbi, first_lblk +
5838                                      ext4_ext_get_actual_len(extent) - 1);
5839                 if (lclu <= last_lclu) {
5840                         mapped = 1;
5841                 } else {
5842                         first_lblk = ext4_ext_next_allocated_block(path);
5843                         first_lclu = EXT4_B2C(sbi, first_lblk);
5844                         if (lclu == first_lclu)
5845                                 mapped = 1;
5846                 }
5847         }
5848
5849 out:
5850         ext4_ext_drop_refs(path);
5851         kfree(path);
5852
5853         return err ? err : mapped;
5854 }
5855
5856 /*
5857  * Updates physical block address and unwritten status of extent
5858  * starting at lblk start and of len. If such an extent doesn't exist,
5859  * this function splits the extent tree appropriately to create an
5860  * extent like this.  This function is called in the fast commit
5861  * replay path.  Returns 0 on success and error on failure.
5862  */
5863 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5864                               int len, int unwritten, ext4_fsblk_t pblk)
5865 {
5866         struct ext4_ext_path *path = NULL, *ppath;
5867         struct ext4_extent *ex;
5868         int ret;
5869
5870         path = ext4_find_extent(inode, start, NULL, 0);
5871         if (IS_ERR(path))
5872                 return PTR_ERR(path);
5873         ex = path[path->p_depth].p_ext;
5874         if (!ex) {
5875                 ret = -EFSCORRUPTED;
5876                 goto out;
5877         }
5878
5879         if (le32_to_cpu(ex->ee_block) != start ||
5880                 ext4_ext_get_actual_len(ex) != len) {
5881                 /* We need to split this extent to match our extent first */
5882                 ppath = path;
5883                 down_write(&EXT4_I(inode)->i_data_sem);
5884                 ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5885                 up_write(&EXT4_I(inode)->i_data_sem);
5886                 if (ret)
5887                         goto out;
5888                 kfree(path);
5889                 path = ext4_find_extent(inode, start, NULL, 0);
5890                 if (IS_ERR(path))
5891                         return -1;
5892                 ppath = path;
5893                 ex = path[path->p_depth].p_ext;
5894                 WARN_ON(le32_to_cpu(ex->ee_block) != start);
5895                 if (ext4_ext_get_actual_len(ex) != len) {
5896                         down_write(&EXT4_I(inode)->i_data_sem);
5897                         ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5898                                                          start + len, 1);
5899                         up_write(&EXT4_I(inode)->i_data_sem);
5900                         if (ret)
5901                                 goto out;
5902                         kfree(path);
5903                         path = ext4_find_extent(inode, start, NULL, 0);
5904                         if (IS_ERR(path))
5905                                 return -EINVAL;
5906                         ex = path[path->p_depth].p_ext;
5907                 }
5908         }
5909         if (unwritten)
5910                 ext4_ext_mark_unwritten(ex);
5911         else
5912                 ext4_ext_mark_initialized(ex);
5913         ext4_ext_store_pblock(ex, pblk);
5914         down_write(&EXT4_I(inode)->i_data_sem);
5915         ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5916         up_write(&EXT4_I(inode)->i_data_sem);
5917 out:
5918         ext4_ext_drop_refs(path);
5919         kfree(path);
5920         ext4_mark_inode_dirty(NULL, inode);
5921         return ret;
5922 }
5923
5924 /* Try to shrink the extent tree */
5925 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5926 {
5927         struct ext4_ext_path *path = NULL;
5928         struct ext4_extent *ex;
5929         ext4_lblk_t old_cur, cur = 0;
5930
5931         while (cur < end) {
5932                 path = ext4_find_extent(inode, cur, NULL, 0);
5933                 if (IS_ERR(path))
5934                         return;
5935                 ex = path[path->p_depth].p_ext;
5936                 if (!ex) {
5937                         ext4_ext_drop_refs(path);
5938                         kfree(path);
5939                         ext4_mark_inode_dirty(NULL, inode);
5940                         return;
5941                 }
5942                 old_cur = cur;
5943                 cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5944                 if (cur <= old_cur)
5945                         cur = old_cur + 1;
5946                 ext4_ext_try_to_merge(NULL, inode, path, ex);
5947                 down_write(&EXT4_I(inode)->i_data_sem);
5948                 ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5949                 up_write(&EXT4_I(inode)->i_data_sem);
5950                 ext4_mark_inode_dirty(NULL, inode);
5951                 ext4_ext_drop_refs(path);
5952                 kfree(path);
5953         }
5954 }
5955
5956 /* Check if *cur is a hole and if it is, skip it */
5957 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5958 {
5959         int ret;
5960         struct ext4_map_blocks map;
5961
5962         map.m_lblk = *cur;
5963         map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5964
5965         ret = ext4_map_blocks(NULL, inode, &map, 0);
5966         if (ret < 0)
5967                 return ret;
5968         if (ret != 0)
5969                 return 0;
5970         *cur = *cur + map.m_len;
5971         return 0;
5972 }
5973
5974 /* Count number of blocks used by this inode and update i_blocks */
5975 int ext4_ext_replay_set_iblocks(struct inode *inode)
5976 {
5977         struct ext4_ext_path *path = NULL, *path2 = NULL;
5978         struct ext4_extent *ex;
5979         ext4_lblk_t cur = 0, end;
5980         int numblks = 0, i, ret = 0;
5981         ext4_fsblk_t cmp1, cmp2;
5982         struct ext4_map_blocks map;
5983
5984         /* Determin the size of the file first */
5985         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5986                                         EXT4_EX_NOCACHE);
5987         if (IS_ERR(path))
5988                 return PTR_ERR(path);
5989         ex = path[path->p_depth].p_ext;
5990         if (!ex) {
5991                 ext4_ext_drop_refs(path);
5992                 kfree(path);
5993                 goto out;
5994         }
5995         end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5996         ext4_ext_drop_refs(path);
5997         kfree(path);
5998
5999         /* Count the number of data blocks */
6000         cur = 0;
6001         while (cur < end) {
6002                 map.m_lblk = cur;
6003                 map.m_len = end - cur;
6004                 ret = ext4_map_blocks(NULL, inode, &map, 0);
6005                 if (ret < 0)
6006                         break;
6007                 if (ret > 0)
6008                         numblks += ret;
6009                 cur = cur + map.m_len;
6010         }
6011
6012         /*
6013          * Count the number of extent tree blocks. We do it by looking up
6014          * two successive extents and determining the difference between
6015          * their paths. When path is different for 2 successive extents
6016          * we compare the blocks in the path at each level and increment
6017          * iblocks by total number of differences found.
6018          */
6019         cur = 0;
6020         ret = skip_hole(inode, &cur);
6021         if (ret < 0)
6022                 goto out;
6023         path = ext4_find_extent(inode, cur, NULL, 0);
6024         if (IS_ERR(path))
6025                 goto out;
6026         numblks += path->p_depth;
6027         ext4_ext_drop_refs(path);
6028         kfree(path);
6029         while (cur < end) {
6030                 path = ext4_find_extent(inode, cur, NULL, 0);
6031                 if (IS_ERR(path))
6032                         break;
6033                 ex = path[path->p_depth].p_ext;
6034                 if (!ex) {
6035                         ext4_ext_drop_refs(path);
6036                         kfree(path);
6037                         return 0;
6038                 }
6039                 cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6040                                         ext4_ext_get_actual_len(ex));
6041                 ret = skip_hole(inode, &cur);
6042                 if (ret < 0) {
6043                         ext4_ext_drop_refs(path);
6044                         kfree(path);
6045                         break;
6046                 }
6047                 path2 = ext4_find_extent(inode, cur, NULL, 0);
6048                 if (IS_ERR(path2)) {
6049                         ext4_ext_drop_refs(path);
6050                         kfree(path);
6051                         break;
6052                 }
6053                 ex = path2[path2->p_depth].p_ext;
6054                 for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6055                         cmp1 = cmp2 = 0;
6056                         if (i <= path->p_depth)
6057                                 cmp1 = path[i].p_bh ?
6058                                         path[i].p_bh->b_blocknr : 0;
6059                         if (i <= path2->p_depth)
6060                                 cmp2 = path2[i].p_bh ?
6061                                         path2[i].p_bh->b_blocknr : 0;
6062                         if (cmp1 != cmp2 && cmp2 != 0)
6063                                 numblks++;
6064                 }
6065                 ext4_ext_drop_refs(path);
6066                 ext4_ext_drop_refs(path2);
6067                 kfree(path);
6068                 kfree(path2);
6069         }
6070
6071 out:
6072         inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6073         ext4_mark_inode_dirty(NULL, inode);
6074         return 0;
6075 }
6076
6077 int ext4_ext_clear_bb(struct inode *inode)
6078 {
6079         struct ext4_ext_path *path = NULL;
6080         struct ext4_extent *ex;
6081         ext4_lblk_t cur = 0, end;
6082         int j, ret = 0;
6083         struct ext4_map_blocks map;
6084
6085         /* Determin the size of the file first */
6086         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6087                                         EXT4_EX_NOCACHE);
6088         if (IS_ERR(path))
6089                 return PTR_ERR(path);
6090         ex = path[path->p_depth].p_ext;
6091         if (!ex) {
6092                 ext4_ext_drop_refs(path);
6093                 kfree(path);
6094                 return 0;
6095         }
6096         end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6097         ext4_ext_drop_refs(path);
6098         kfree(path);
6099
6100         cur = 0;
6101         while (cur < end) {
6102                 map.m_lblk = cur;
6103                 map.m_len = end - cur;
6104                 ret = ext4_map_blocks(NULL, inode, &map, 0);
6105                 if (ret < 0)
6106                         break;
6107                 if (ret > 0) {
6108                         path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6109                         if (!IS_ERR_OR_NULL(path)) {
6110                                 for (j = 0; j < path->p_depth; j++) {
6111
6112                                         ext4_mb_mark_bb(inode->i_sb,
6113                                                         path[j].p_block, 1, 0);
6114                                         ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6115                                                         0, path[j].p_block, 1, 1);
6116                                 }
6117                                 ext4_ext_drop_refs(path);
6118                                 kfree(path);
6119                         }
6120                         ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6121                         ext4_fc_record_regions(inode->i_sb, inode->i_ino,
6122                                         map.m_lblk, map.m_pblk, map.m_len, 1);
6123                 }
6124                 cur = cur + map.m_len;
6125         }
6126
6127         return 0;
6128 }