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 "ext4_jbd2.h"
  32 #include "ext4_extents.h"
  33 #include "xattr.h"
  34
  35 #include <trace/events/ext4.h>
  36
  37 /*
  38  * used by extent splitting.
  39  */
  40 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  41                                         due to ENOSPC */
  42 #define EXT4_EXT_MARK_UNWRIT1   0x2  /* mark first half unwritten */
  43 #define EXT4_EXT_MARK_UNWRIT2   0x4  /* mark second half unwritten */
  44
  45 #define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  46 #define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  47
  48 static __le32 ext4_extent_block_csum(struct inode *inode,
  49                                      struct ext4_extent_header *eh)
  50 {
  51         struct ext4_inode_info *ei = EXT4_I(inode);
  52         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  53         __u32 csum;
  54
  55         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  56                            EXT4_EXTENT_TAIL_OFFSET(eh));
  57         return cpu_to_le32(csum);
  58 }
  59
  60 static int ext4_extent_block_csum_verify(struct inode *inode,
  61                                          struct ext4_extent_header *eh)
  62 {
  63         struct ext4_extent_tail *et;
  64
  65         if (!ext4_has_metadata_csum(inode->i_sb))
  66                 return 1;
  67
  68         et = find_ext4_extent_tail(eh);
  69         if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  70                 return 0;
  71         return 1;
  72 }
  73
  74 static void ext4_extent_block_csum_set(struct inode *inode,
  75                                        struct ext4_extent_header *eh)
  76 {
  77         struct ext4_extent_tail *et;
  78
  79         if (!ext4_has_metadata_csum(inode->i_sb))
  80                 return;
  81
  82         et = find_ext4_extent_tail(eh);
  83         et->et_checksum = ext4_extent_block_csum(inode, eh);
  84 }
  85
  86 static int ext4_split_extent(handle_t *handle,
  87                                 struct inode *inode,
  88                                 struct ext4_ext_path **ppath,
  89                                 struct ext4_map_blocks *map,
  90                                 int split_flag,
  91                                 int flags);
  92
  93 static int ext4_split_extent_at(handle_t *handle,
  94                              struct inode *inode,
  95                              struct ext4_ext_path **ppath,
  96                              ext4_lblk_t split,
  97                              int split_flag,
  98                              int flags);
  99
 100 static int ext4_find_delayed_extent(struct inode *inode,
 101                                     struct extent_status *newes);
 102
 103 static int ext4_ext_truncate_extend_restart(handle_t *handle,
 104                                             struct inode *inode,
 105                                             int needed)
 106 {
 107         int err;
 108
 109         if (!ext4_handle_valid(handle))
 110                 return 0;
 111         if (handle->h_buffer_credits >= needed)
 112                 return 0;
 113         /*
 114          * If we need to extend the journal get a few extra blocks
 115          * while we're at it for efficiency's sake.
 116          */
 117         needed += 3;
 118         err = ext4_journal_extend(handle, needed - handle->h_buffer_credits);
 119         if (err <= 0)
 120                 return err;
 121         err = ext4_truncate_restart_trans(handle, inode, needed);
 122         if (err == 0)
 123                 err = -EAGAIN;
 124
 125         return err;
 126 }
 127
 128 /*
 129  * could return:
 130  *  - EROFS
 131  *  - ENOMEM
 132  */
 133 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 134                                 struct ext4_ext_path *path)
 135 {
 136         if (path->p_bh) {
 137                 /* path points to block */
 138                 BUFFER_TRACE(path->p_bh, "get_write_access");
 139                 return ext4_journal_get_write_access(handle, path->p_bh);
 140         }
 141         /* path points to leaf/index in inode body */
 142         /* we use in-core data, no need to protect them */
 143         return 0;
 144 }
 145
 146 /*
 147  * could return:
 148  *  - EROFS
 149  *  - ENOMEM
 150  *  - EIO
 151  */
 152 int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle,
 153                      struct inode *inode, struct ext4_ext_path *path)
 154 {
 155         int err;
 156
 157         WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 158         if (path->p_bh) {
 159                 ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 160                 /* path points to block */
 161                 err = __ext4_handle_dirty_metadata(where, line, handle,
 162                                                    inode, path->p_bh);
 163         } else {
 164                 /* path points to leaf/index in inode body */
 165                 err = ext4_mark_inode_dirty(handle, inode);
 166         }
 167         return err;
 168 }
 169
 170 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 171                               struct ext4_ext_path *path,
 172                               ext4_lblk_t block)
 173 {
 174         if (path) {
 175                 int depth = path->p_depth;
 176                 struct ext4_extent *ex;
 177
 178                 /*
 179                  * Try to predict block placement assuming that we are
 180                  * filling in a file which will eventually be
 181                  * non-sparse --- i.e., in the case of libbfd writing
 182                  * an ELF object sections out-of-order but in a way
 183                  * the eventually results in a contiguous object or
 184                  * executable file, or some database extending a table
 185                  * space file.  However, this is actually somewhat
 186                  * non-ideal if we are writing a sparse file such as
 187                  * qemu or KVM writing a raw image file that is going
 188                  * to stay fairly sparse, since it will end up
 189                  * fragmenting the file system's free space.  Maybe we
 190                  * should have some hueristics or some way to allow
 191                  * userspace to pass a hint to file system,
 192                  * especially if the latter case turns out to be
 193                  * common.
 194                  */
 195                 ex = path[depth].p_ext;
 196                 if (ex) {
 197                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 198                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 199
 200                         if (block > ext_block)
 201                                 return ext_pblk + (block - ext_block);
 202                         else
 203                                 return ext_pblk - (ext_block - block);
 204                 }
 205
 206                 /* it looks like index is empty;
 207                  * try to find starting block from index itself */
 208                 if (path[depth].p_bh)
 209                         return path[depth].p_bh->b_blocknr;
 210         }
 211
 212         /* OK. use inode's group */
 213         return ext4_inode_to_goal_block(inode);
 214 }
 215
 216 /*
 217  * Allocation for a meta data block
 218  */
 219 static ext4_fsblk_t
 220 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 221                         struct ext4_ext_path *path,
 222                         struct ext4_extent *ex, int *err, unsigned int flags)
 223 {
 224         ext4_fsblk_t goal, newblock;
 225
 226         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 227         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 228                                         NULL, err);
 229         return newblock;
 230 }
 231
 232 static inline int ext4_ext_space_block(struct inode *inode, int check)
 233 {
 234         int size;
 235
 236         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 237                         / sizeof(struct ext4_extent);
 238 #ifdef AGGRESSIVE_TEST
 239         if (!check && size > 6)
 240                 size = 6;
 241 #endif
 242         return size;
 243 }
 244
 245 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 246 {
 247         int size;
 248
 249         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 250                         / sizeof(struct ext4_extent_idx);
 251 #ifdef AGGRESSIVE_TEST
 252         if (!check && size > 5)
 253                 size = 5;
 254 #endif
 255         return size;
 256 }
 257
 258 static inline int ext4_ext_space_root(struct inode *inode, int check)
 259 {
 260         int size;
 261
 262         size = sizeof(EXT4_I(inode)->i_data);
 263         size -= sizeof(struct ext4_extent_header);
 264         size /= sizeof(struct ext4_extent);
 265 #ifdef AGGRESSIVE_TEST
 266         if (!check && size > 3)
 267                 size = 3;
 268 #endif
 269         return size;
 270 }
 271
 272 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 273 {
 274         int size;
 275
 276         size = sizeof(EXT4_I(inode)->i_data);
 277         size -= sizeof(struct ext4_extent_header);
 278         size /= sizeof(struct ext4_extent_idx);
 279 #ifdef AGGRESSIVE_TEST
 280         if (!check && size > 4)
 281                 size = 4;
 282 #endif
 283         return size;
 284 }
 285
 286 static inline int
 287 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 288                            struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 289                            int nofail)
 290 {
 291         struct ext4_ext_path *path = *ppath;
 292         int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 293
 294         return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 295                         EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 296                         EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
 297                         (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
 298 }
 299
 300 /*
 301  * Calculate the number of metadata blocks needed
 302  * to allocate @blocks
 303  * Worse case is one block per extent
 304  */
 305 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 306 {
 307         struct ext4_inode_info *ei = EXT4_I(inode);
 308         int idxs;
 309
 310         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 311                 / sizeof(struct ext4_extent_idx));
 312
 313         /*
 314          * If the new delayed allocation block is contiguous with the
 315          * previous da block, it can share index blocks with the
 316          * previous block, so we only need to allocate a new index
 317          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 318          * an additional index block, and at ldxs**3 blocks, yet
 319          * another index blocks.
 320          */
 321         if (ei->i_da_metadata_calc_len &&
 322             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 323                 int num = 0;
 324
 325                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 326                         num++;
 327                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 328                         num++;
 329                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 330                         num++;
 331                         ei->i_da_metadata_calc_len = 0;
 332                 } else
 333                         ei->i_da_metadata_calc_len++;
 334                 ei->i_da_metadata_calc_last_lblock++;
 335                 return num;
 336         }
 337
 338         /*
 339          * In the worst case we need a new set of index blocks at
 340          * every level of the inode's extent tree.
 341          */
 342         ei->i_da_metadata_calc_len = 1;
 343         ei->i_da_metadata_calc_last_lblock = lblock;
 344         return ext_depth(inode) + 1;
 345 }
 346
 347 static int
 348 ext4_ext_max_entries(struct inode *inode, int depth)
 349 {
 350         int max;
 351
 352         if (depth == ext_depth(inode)) {
 353                 if (depth == 0)
 354                         max = ext4_ext_space_root(inode, 1);
 355                 else
 356                         max = ext4_ext_space_root_idx(inode, 1);
 357         } else {
 358                 if (depth == 0)
 359                         max = ext4_ext_space_block(inode, 1);
 360                 else
 361                         max = ext4_ext_space_block_idx(inode, 1);
 362         }
 363
 364         return max;
 365 }
 366
 367 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 368 {
 369         ext4_fsblk_t block = ext4_ext_pblock(ext);
 370         int len = ext4_ext_get_actual_len(ext);
 371         ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 372
 373         /*
 374          * We allow neither:
 375          *  - zero length
 376          *  - overflow/wrap-around
 377          */
 378         if (lblock + len <= lblock)
 379                 return 0;
 380         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 381 }
 382
 383 static int ext4_valid_extent_idx(struct inode *inode,
 384                                 struct ext4_extent_idx *ext_idx)
 385 {
 386         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 387
 388         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 389 }
 390
 391 static int ext4_valid_extent_entries(struct inode *inode,
 392                                 struct ext4_extent_header *eh,
 393                                 int depth)
 394 {
 395         unsigned short entries;
 396         if (eh->eh_entries == 0)
 397                 return 1;
 398
 399         entries = le16_to_cpu(eh->eh_entries);
 400
 401         if (depth == 0) {
 402                 /* leaf entries */
 403                 struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 404                 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 405                 ext4_fsblk_t pblock = 0;
 406                 ext4_lblk_t lblock = 0;
 407                 ext4_lblk_t prev = 0;
 408                 int len = 0;
 409                 while (entries) {
 410                         if (!ext4_valid_extent(inode, ext))
 411                                 return 0;
 412
 413                         /* Check for overlapping extents */
 414                         lblock = le32_to_cpu(ext->ee_block);
 415                         len = ext4_ext_get_actual_len(ext);
 416                         if ((lblock <= prev) && prev) {
 417                                 pblock = ext4_ext_pblock(ext);
 418                                 es->s_last_error_block = cpu_to_le64(pblock);
 419                                 return 0;
 420                         }
 421                         ext++;
 422                         entries--;
 423                         prev = lblock + len - 1;
 424                 }
 425         } else {
 426                 struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 427                 while (entries) {
 428                         if (!ext4_valid_extent_idx(inode, ext_idx))
 429                                 return 0;
 430                         ext_idx++;
 431                         entries--;
 432                 }
 433         }
 434         return 1;
 435 }
 436
 437 static int __ext4_ext_check(const char *function, unsigned int line,
 438                             struct inode *inode, struct ext4_extent_header *eh,
 439                             int depth, ext4_fsblk_t pblk)
 440 {
 441         const char *error_msg;
 442         int max = 0, err = -EFSCORRUPTED;
 443
 444         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 445                 error_msg = "invalid magic";
 446                 goto corrupted;
 447         }
 448         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 449                 error_msg = "unexpected eh_depth";
 450                 goto corrupted;
 451         }
 452         if (unlikely(eh->eh_max == 0)) {
 453                 error_msg = "invalid eh_max";
 454                 goto corrupted;
 455         }
 456         max = ext4_ext_max_entries(inode, depth);
 457         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 458                 error_msg = "too large eh_max";
 459                 goto corrupted;
 460         }
 461         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 462                 error_msg = "invalid eh_entries";
 463                 goto corrupted;
 464         }
 465         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 466                 error_msg = "invalid extent entries";
 467                 goto corrupted;
 468         }
 469         if (unlikely(depth > 32)) {
 470                 error_msg = "too large eh_depth";
 471                 goto corrupted;
 472         }
 473         /* Verify checksum on non-root extent tree nodes */
 474         if (ext_depth(inode) != depth &&
 475             !ext4_extent_block_csum_verify(inode, eh)) {
 476                 error_msg = "extent tree corrupted";
 477                 err = -EFSBADCRC;
 478                 goto corrupted;
 479         }
 480         return 0;
 481
 482 corrupted:
 483         ext4_error_inode(inode, function, line, 0,
 484                          "pblk %llu bad header/extent: %s - magic %x, "
 485                          "entries %u, max %u(%u), depth %u(%u)",
 486                          (unsigned long long) pblk, error_msg,
 487                          le16_to_cpu(eh->eh_magic),
 488                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 489                          max, le16_to_cpu(eh->eh_depth), depth);
 490         return err;
 491 }
 492
 493 #define ext4_ext_check(inode, eh, depth, pblk)                  \
 494         __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
 495
 496 int ext4_ext_check_inode(struct inode *inode)
 497 {
 498         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 499 }
 500
 501 static struct buffer_head *
 502 __read_extent_tree_block(const char *function, unsigned int line,
 503                          struct inode *inode, ext4_fsblk_t pblk, int depth,
 504                          int flags)
 505 {
 506         struct buffer_head              *bh;
 507         int                             err;
 508
 509         bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
 510         if (unlikely(!bh))
 511                 return ERR_PTR(-ENOMEM);
 512
 513         if (!bh_uptodate_or_lock(bh)) {
 514                 trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 515                 err = bh_submit_read(bh);
 516                 if (err < 0)
 517                         goto errout;
 518         }
 519         if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 520                 return bh;
 521         err = __ext4_ext_check(function, line, inode,
 522                                ext_block_hdr(bh), depth, pblk);
 523         if (err)
 524                 goto errout;
 525         set_buffer_verified(bh);
 526         /*
 527          * If this is a leaf block, cache all of its entries
 528          */
 529         if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 530                 struct ext4_extent_header *eh = ext_block_hdr(bh);
 531                 struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 532                 ext4_lblk_t prev = 0;
 533                 int i;
 534
 535                 for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 536                         unsigned int status = EXTENT_STATUS_WRITTEN;
 537                         ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 538                         int len = ext4_ext_get_actual_len(ex);
 539
 540                         if (prev && (prev != lblk))
 541                                 ext4_es_cache_extent(inode, prev,
 542                                                      lblk - prev, ~0,
 543                                                      EXTENT_STATUS_HOLE);
 544
 545                         if (ext4_ext_is_unwritten(ex))
 546                                 status = EXTENT_STATUS_UNWRITTEN;
 547                         ext4_es_cache_extent(inode, lblk, len,
 548                                              ext4_ext_pblock(ex), status);
 549                         prev = lblk + len;
 550                 }
 551         }
 552         return bh;
 553 errout:
 554         put_bh(bh);
 555         return ERR_PTR(err);
 556
 557 }
 558
 559 #define read_extent_tree_block(inode, pblk, depth, flags)               \
 560         __read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
 561                                  (depth), (flags))
 562
 563 /*
 564  * This function is called to cache a file's extent information in the
 565  * extent status tree
 566  */
 567 int ext4_ext_precache(struct inode *inode)
 568 {
 569         struct ext4_inode_info *ei = EXT4_I(inode);
 570         struct ext4_ext_path *path = NULL;
 571         struct buffer_head *bh;
 572         int i = 0, depth, ret = 0;
 573
 574         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 575                 return 0;       /* not an extent-mapped inode */
 576
 577         down_read(&ei->i_data_sem);
 578         depth = ext_depth(inode);
 579
 580         path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
 581                        GFP_NOFS);
 582         if (path == NULL) {
 583                 up_read(&ei->i_data_sem);
 584                 return -ENOMEM;
 585         }
 586
 587         /* Don't cache anything if there are no external extent blocks */
 588         if (depth == 0)
 589                 goto out;
 590         path[0].p_hdr = ext_inode_hdr(inode);
 591         ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 592         if (ret)
 593                 goto out;
 594         path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 595         while (i >= 0) {
 596                 /*
 597                  * If this is a leaf block or we've reached the end of
 598                  * the index block, go up
 599                  */
 600                 if ((i == depth) ||
 601                     path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 602                         brelse(path[i].p_bh);
 603                         path[i].p_bh = NULL;
 604                         i--;
 605                         continue;
 606                 }
 607                 bh = read_extent_tree_block(inode,
 608                                             ext4_idx_pblock(path[i].p_idx++),
 609                                             depth - i - 1,
 610                                             EXT4_EX_FORCE_CACHE);
 611                 if (IS_ERR(bh)) {
 612                         ret = PTR_ERR(bh);
 613                         break;
 614                 }
 615                 i++;
 616                 path[i].p_bh = bh;
 617                 path[i].p_hdr = ext_block_hdr(bh);
 618                 path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 619         }
 620         ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 621 out:
 622         up_read(&ei->i_data_sem);
 623         ext4_ext_drop_refs(path);
 624         kfree(path);
 625         return ret;
 626 }
 627
 628 #ifdef EXT_DEBUG
 629 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 630 {
 631         int k, l = path->p_depth;
 632
 633         ext_debug("path:");
 634         for (k = 0; k <= l; k++, path++) {
 635                 if (path->p_idx) {
 636                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 637                             ext4_idx_pblock(path->p_idx));
 638                 } else if (path->p_ext) {
 639                         ext_debug("  %d:[%d]%d:%llu ",
 640                                   le32_to_cpu(path->p_ext->ee_block),
 641                                   ext4_ext_is_unwritten(path->p_ext),
 642                                   ext4_ext_get_actual_len(path->p_ext),
 643                                   ext4_ext_pblock(path->p_ext));
 644                 } else
 645                         ext_debug("  []");
 646         }
 647         ext_debug("\n");
 648 }
 649
 650 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 651 {
 652         int depth = ext_depth(inode);
 653         struct ext4_extent_header *eh;
 654         struct ext4_extent *ex;
 655         int i;
 656
 657         if (!path)
 658                 return;
 659
 660         eh = path[depth].p_hdr;
 661         ex = EXT_FIRST_EXTENT(eh);
 662
 663         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 664
 665         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 666                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 667                           ext4_ext_is_unwritten(ex),
 668                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 669         }
 670         ext_debug("\n");
 671 }
 672
 673 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 674                         ext4_fsblk_t newblock, int level)
 675 {
 676         int depth = ext_depth(inode);
 677         struct ext4_extent *ex;
 678
 679         if (depth != level) {
 680                 struct ext4_extent_idx *idx;
 681                 idx = path[level].p_idx;
 682                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 683                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
 684                                         le32_to_cpu(idx->ei_block),
 685                                         ext4_idx_pblock(idx),
 686                                         newblock);
 687                         idx++;
 688                 }
 689
 690                 return;
 691         }
 692
 693         ex = path[depth].p_ext;
 694         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 695                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 696                                 le32_to_cpu(ex->ee_block),
 697                                 ext4_ext_pblock(ex),
 698                                 ext4_ext_is_unwritten(ex),
 699                                 ext4_ext_get_actual_len(ex),
 700                                 newblock);
 701                 ex++;
 702         }
 703 }
 704
 705 #else
 706 #define ext4_ext_show_path(inode, path)
 707 #define ext4_ext_show_leaf(inode, path)
 708 #define ext4_ext_show_move(inode, path, newblock, level)
 709 #endif
 710
 711 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 712 {
 713         int depth, i;
 714
 715         if (!path)
 716                 return;
 717         depth = path->p_depth;
 718         for (i = 0; i <= depth; i++, path++)
 719                 if (path->p_bh) {
 720                         brelse(path->p_bh);
 721                         path->p_bh = NULL;
 722                 }
 723 }
 724
 725 /*
 726  * ext4_ext_binsearch_idx:
 727  * binary search for the closest index of the given block
 728  * the header must be checked before calling this
 729  */
 730 static void
 731 ext4_ext_binsearch_idx(struct inode *inode,
 732                         struct ext4_ext_path *path, ext4_lblk_t block)
 733 {
 734         struct ext4_extent_header *eh = path->p_hdr;
 735         struct ext4_extent_idx *r, *l, *m;
 736
 737
 738         ext_debug("binsearch for %u(idx):  ", block);
 739
 740         l = EXT_FIRST_INDEX(eh) + 1;
 741         r = EXT_LAST_INDEX(eh);
 742         while (l <= r) {
 743                 m = l + (r - l) / 2;
 744                 if (block < le32_to_cpu(m->ei_block))
 745                         r = m - 1;
 746                 else
 747                         l = m + 1;
 748                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 749                                 m, le32_to_cpu(m->ei_block),
 750                                 r, le32_to_cpu(r->ei_block));
 751         }
 752
 753         path->p_idx = l - 1;
 754         ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 755                   ext4_idx_pblock(path->p_idx));
 756
 757 #ifdef CHECK_BINSEARCH
 758         {
 759                 struct ext4_extent_idx *chix, *ix;
 760                 int k;
 761
 762                 chix = ix = EXT_FIRST_INDEX(eh);
 763                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 764                   if (k != 0 &&
 765                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 766                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 767                                        "first=0x%p\n", k,
 768                                        ix, EXT_FIRST_INDEX(eh));
 769                                 printk(KERN_DEBUG "%u <= %u\n",
 770                                        le32_to_cpu(ix->ei_block),
 771                                        le32_to_cpu(ix[-1].ei_block));
 772                         }
 773                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 774                                            <= le32_to_cpu(ix[-1].ei_block));
 775                         if (block < le32_to_cpu(ix->ei_block))
 776                                 break;
 777                         chix = ix;
 778                 }
 779                 BUG_ON(chix != path->p_idx);
 780         }
 781 #endif
 782
 783 }
 784
 785 /*
 786  * ext4_ext_binsearch:
 787  * binary search for closest extent of the given block
 788  * the header must be checked before calling this
 789  */
 790 static void
 791 ext4_ext_binsearch(struct inode *inode,
 792                 struct ext4_ext_path *path, ext4_lblk_t block)
 793 {
 794         struct ext4_extent_header *eh = path->p_hdr;
 795         struct ext4_extent *r, *l, *m;
 796
 797         if (eh->eh_entries == 0) {
 798                 /*
 799                  * this leaf is empty:
 800                  * we get such a leaf in split/add case
 801                  */
 802                 return;
 803         }
 804
 805         ext_debug("binsearch for %u:  ", block);
 806
 807         l = EXT_FIRST_EXTENT(eh) + 1;
 808         r = EXT_LAST_EXTENT(eh);
 809
 810         while (l <= r) {
 811                 m = l + (r - l) / 2;
 812                 if (block < le32_to_cpu(m->ee_block))
 813                         r = m - 1;
 814                 else
 815                         l = m + 1;
 816                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 817                                 m, le32_to_cpu(m->ee_block),
 818                                 r, le32_to_cpu(r->ee_block));
 819         }
 820
 821         path->p_ext = l - 1;
 822         ext_debug("  -> %d:%llu:[%d]%d ",
 823                         le32_to_cpu(path->p_ext->ee_block),
 824                         ext4_ext_pblock(path->p_ext),
 825                         ext4_ext_is_unwritten(path->p_ext),
 826                         ext4_ext_get_actual_len(path->p_ext));
 827
 828 #ifdef CHECK_BINSEARCH
 829         {
 830                 struct ext4_extent *chex, *ex;
 831                 int k;
 832
 833                 chex = ex = EXT_FIRST_EXTENT(eh);
 834                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 835                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 836                                           <= le32_to_cpu(ex[-1].ee_block));
 837                         if (block < le32_to_cpu(ex->ee_block))
 838                                 break;
 839                         chex = ex;
 840                 }
 841                 BUG_ON(chex != path->p_ext);
 842         }
 843 #endif
 844
 845 }
 846
 847 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 848 {
 849         struct ext4_extent_header *eh;
 850
 851         eh = ext_inode_hdr(inode);
 852         eh->eh_depth = 0;
 853         eh->eh_entries = 0;
 854         eh->eh_magic = EXT4_EXT_MAGIC;
 855         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 856         ext4_mark_inode_dirty(handle, inode);
 857         return 0;
 858 }
 859
 860 struct ext4_ext_path *
 861 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 862                  struct ext4_ext_path **orig_path, int flags)
 863 {
 864         struct ext4_extent_header *eh;
 865         struct buffer_head *bh;
 866         struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 867         short int depth, i, ppos = 0;
 868         int ret;
 869
 870         eh = ext_inode_hdr(inode);
 871         depth = ext_depth(inode);
 872         if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
 873                 EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
 874                                  depth);
 875                 ret = -EFSCORRUPTED;
 876                 goto err;
 877         }
 878
 879         if (path) {
 880                 ext4_ext_drop_refs(path);
 881                 if (depth > path[0].p_maxdepth) {
 882                         kfree(path);
 883                         *orig_path = path = NULL;
 884                 }
 885         }
 886         if (!path) {
 887                 /* account possible depth increase */
 888                 path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
 889                                 GFP_NOFS);
 890                 if (unlikely(!path))
 891                         return ERR_PTR(-ENOMEM);
 892                 path[0].p_maxdepth = depth + 1;
 893         }
 894         path[0].p_hdr = eh;
 895         path[0].p_bh = NULL;
 896
 897         i = depth;
 898         /* walk through the tree */
 899         while (i) {
 900                 ext_debug("depth %d: num %d, max %d\n",
 901                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 902
 903                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 904                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 905                 path[ppos].p_depth = i;
 906                 path[ppos].p_ext = NULL;
 907
 908                 bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
 909                                             flags);
 910                 if (IS_ERR(bh)) {
 911                         ret = PTR_ERR(bh);
 912                         goto err;
 913                 }
 914
 915                 eh = ext_block_hdr(bh);
 916                 ppos++;
 917                 path[ppos].p_bh = bh;
 918                 path[ppos].p_hdr = eh;
 919         }
 920
 921         path[ppos].p_depth = i;
 922         path[ppos].p_ext = NULL;
 923         path[ppos].p_idx = NULL;
 924
 925         /* find extent */
 926         ext4_ext_binsearch(inode, path + ppos, block);
 927         /* if not an empty leaf */
 928         if (path[ppos].p_ext)
 929                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 930
 931         ext4_ext_show_path(inode, path);
 932
 933         return path;
 934
 935 err:
 936         ext4_ext_drop_refs(path);
 937         kfree(path);
 938         if (orig_path)
 939                 *orig_path = NULL;
 940         return ERR_PTR(ret);
 941 }
 942
 943 /*
 944  * ext4_ext_insert_index:
 945  * insert new index [@logical;@ptr] into the block at @curp;
 946  * check where to insert: before @curp or after @curp
 947  */
 948 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 949                                  struct ext4_ext_path *curp,
 950                                  int logical, ext4_fsblk_t ptr)
 951 {
 952         struct ext4_extent_idx *ix;
 953         int len, err;
 954
 955         err = ext4_ext_get_access(handle, inode, curp);
 956         if (err)
 957                 return err;
 958
 959         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 960                 EXT4_ERROR_INODE(inode,
 961                                  "logical %d == ei_block %d!",
 962                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 963                 return -EFSCORRUPTED;
 964         }
 965
 966         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 967                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 968                 EXT4_ERROR_INODE(inode,
 969                                  "eh_entries %d >= eh_max %d!",
 970                                  le16_to_cpu(curp->p_hdr->eh_entries),
 971                                  le16_to_cpu(curp->p_hdr->eh_max));
 972                 return -EFSCORRUPTED;
 973         }
 974
 975         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 976                 /* insert after */
 977                 ext_debug("insert new index %d after: %llu\n", logical, ptr);
 978                 ix = curp->p_idx + 1;
 979         } else {
 980                 /* insert before */
 981                 ext_debug("insert new index %d before: %llu\n", logical, ptr);
 982                 ix = curp->p_idx;
 983         }
 984
 985         len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 986         BUG_ON(len < 0);
 987         if (len > 0) {
 988                 ext_debug("insert new index %d: "
 989                                 "move %d indices from 0x%p to 0x%p\n",
 990                                 logical, len, ix, ix + 1);
 991                 memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 992         }
 993
 994         if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 995                 EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 996                 return -EFSCORRUPTED;
 997         }
 998
 999         ix->ei_block = cpu_to_le32(logical);
1000         ext4_idx_store_pblock(ix, ptr);
1001         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1002
1003         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1004                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1005                 return -EFSCORRUPTED;
1006         }
1007
1008         err = ext4_ext_dirty(handle, inode, curp);
1009         ext4_std_error(inode->i_sb, err);
1010
1011         return err;
1012 }
1013
1014 /*
1015  * ext4_ext_split:
1016  * inserts new subtree into the path, using free index entry
1017  * at depth @at:
1018  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1019  * - makes decision where to split
1020  * - moves remaining extents and index entries (right to the split point)
1021  *   into the newly allocated blocks
1022  * - initializes subtree
1023  */
1024 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1025                           unsigned int flags,
1026                           struct ext4_ext_path *path,
1027                           struct ext4_extent *newext, int at)
1028 {
1029         struct buffer_head *bh = NULL;
1030         int depth = ext_depth(inode);
1031         struct ext4_extent_header *neh;
1032         struct ext4_extent_idx *fidx;
1033         int i = at, k, m, a;
1034         ext4_fsblk_t newblock, oldblock;
1035         __le32 border;
1036         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1037         int err = 0;
1038
1039         /* make decision: where to split? */
1040         /* FIXME: now decision is simplest: at current extent */
1041
1042         /* if current leaf will be split, then we should use
1043          * border from split point */
1044         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1045                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1046                 return -EFSCORRUPTED;
1047         }
1048         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1049                 border = path[depth].p_ext[1].ee_block;
1050                 ext_debug("leaf will be split."
1051                                 " next leaf starts at %d\n",
1052                                   le32_to_cpu(border));
1053         } else {
1054                 border = newext->ee_block;
1055                 ext_debug("leaf will be added."
1056                                 " next leaf starts at %d\n",
1057                                 le32_to_cpu(border));
1058         }
1059
1060         /*
1061          * If error occurs, then we break processing
1062          * and mark filesystem read-only. index won't
1063          * be inserted and tree will be in consistent
1064          * state. Next mount will repair buffers too.
1065          */
1066
1067         /*
1068          * Get array to track all allocated blocks.
1069          * We need this to handle errors and free blocks
1070          * upon them.
1071          */
1072         ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1073         if (!ablocks)
1074                 return -ENOMEM;
1075
1076         /* allocate all needed blocks */
1077         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1078         for (a = 0; a < depth - at; a++) {
1079                 newblock = ext4_ext_new_meta_block(handle, inode, path,
1080                                                    newext, &err, flags);
1081                 if (newblock == 0)
1082                         goto cleanup;
1083                 ablocks[a] = newblock;
1084         }
1085
1086         /* initialize new leaf */
1087         newblock = ablocks[--a];
1088         if (unlikely(newblock == 0)) {
1089                 EXT4_ERROR_INODE(inode, "newblock == 0!");
1090                 err = -EFSCORRUPTED;
1091                 goto cleanup;
1092         }
1093         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1094         if (unlikely(!bh)) {
1095                 err = -ENOMEM;
1096                 goto cleanup;
1097         }
1098         lock_buffer(bh);
1099
1100         err = ext4_journal_get_create_access(handle, bh);
1101         if (err)
1102                 goto cleanup;
1103
1104         neh = ext_block_hdr(bh);
1105         neh->eh_entries = 0;
1106         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1107         neh->eh_magic = EXT4_EXT_MAGIC;
1108         neh->eh_depth = 0;
1109
1110         /* move remainder of path[depth] to the new leaf */
1111         if (unlikely(path[depth].p_hdr->eh_entries !=
1112                      path[depth].p_hdr->eh_max)) {
1113                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1114                                  path[depth].p_hdr->eh_entries,
1115                                  path[depth].p_hdr->eh_max);
1116                 err = -EFSCORRUPTED;
1117                 goto cleanup;
1118         }
1119         /* start copy from next extent */
1120         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1121         ext4_ext_show_move(inode, path, newblock, depth);
1122         if (m) {
1123                 struct ext4_extent *ex;
1124                 ex = EXT_FIRST_EXTENT(neh);
1125                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1126                 le16_add_cpu(&neh->eh_entries, m);
1127         }
1128
1129         ext4_extent_block_csum_set(inode, neh);
1130         set_buffer_uptodate(bh);
1131         unlock_buffer(bh);
1132
1133         err = ext4_handle_dirty_metadata(handle, inode, bh);
1134         if (err)
1135                 goto cleanup;
1136         brelse(bh);
1137         bh = NULL;
1138
1139         /* correct old leaf */
1140         if (m) {
1141                 err = ext4_ext_get_access(handle, inode, path + depth);
1142                 if (err)
1143                         goto cleanup;
1144                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1145                 err = ext4_ext_dirty(handle, inode, path + depth);
1146                 if (err)
1147                         goto cleanup;
1148
1149         }
1150
1151         /* create intermediate indexes */
1152         k = depth - at - 1;
1153         if (unlikely(k < 0)) {
1154                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1155                 err = -EFSCORRUPTED;
1156                 goto cleanup;
1157         }
1158         if (k)
1159                 ext_debug("create %d intermediate indices\n", k);
1160         /* insert new index into current index block */
1161         /* current depth stored in i var */
1162         i = depth - 1;
1163         while (k--) {
1164                 oldblock = newblock;
1165                 newblock = ablocks[--a];
1166                 bh = sb_getblk(inode->i_sb, newblock);
1167                 if (unlikely(!bh)) {
1168                         err = -ENOMEM;
1169                         goto cleanup;
1170                 }
1171                 lock_buffer(bh);
1172
1173                 err = ext4_journal_get_create_access(handle, bh);
1174                 if (err)
1175                         goto cleanup;
1176
1177                 neh = ext_block_hdr(bh);
1178                 neh->eh_entries = cpu_to_le16(1);
1179                 neh->eh_magic = EXT4_EXT_MAGIC;
1180                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1181                 neh->eh_depth = cpu_to_le16(depth - i);
1182                 fidx = EXT_FIRST_INDEX(neh);
1183                 fidx->ei_block = border;
1184                 ext4_idx_store_pblock(fidx, oldblock);
1185
1186                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1187                                 i, newblock, le32_to_cpu(border), oldblock);
1188
1189                 /* move remainder of path[i] to the new index block */
1190                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1191                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1192                         EXT4_ERROR_INODE(inode,
1193                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1194                                          le32_to_cpu(path[i].p_ext->ee_block));
1195                         err = -EFSCORRUPTED;
1196                         goto cleanup;
1197                 }
1198                 /* start copy indexes */
1199                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1200                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1201                                 EXT_MAX_INDEX(path[i].p_hdr));
1202                 ext4_ext_show_move(inode, path, newblock, i);
1203                 if (m) {
1204                         memmove(++fidx, path[i].p_idx,
1205                                 sizeof(struct ext4_extent_idx) * m);
1206                         le16_add_cpu(&neh->eh_entries, m);
1207                 }
1208                 ext4_extent_block_csum_set(inode, neh);
1209                 set_buffer_uptodate(bh);
1210                 unlock_buffer(bh);
1211
1212                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1213                 if (err)
1214                         goto cleanup;
1215                 brelse(bh);
1216                 bh = NULL;
1217
1218                 /* correct old index */
1219                 if (m) {
1220                         err = ext4_ext_get_access(handle, inode, path + i);
1221                         if (err)
1222                                 goto cleanup;
1223                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1224                         err = ext4_ext_dirty(handle, inode, path + i);
1225                         if (err)
1226                                 goto cleanup;
1227                 }
1228
1229                 i--;
1230         }
1231
1232         /* insert new index */
1233         err = ext4_ext_insert_index(handle, inode, path + at,
1234                                     le32_to_cpu(border), newblock);
1235
1236 cleanup:
1237         if (bh) {
1238                 if (buffer_locked(bh))
1239                         unlock_buffer(bh);
1240                 brelse(bh);
1241         }
1242
1243         if (err) {
1244                 /* free all allocated blocks in error case */
1245                 for (i = 0; i < depth; i++) {
1246                         if (!ablocks[i])
1247                                 continue;
1248                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1249                                          EXT4_FREE_BLOCKS_METADATA);
1250                 }
1251         }
1252         kfree(ablocks);
1253
1254         return err;
1255 }
1256
1257 /*
1258  * ext4_ext_grow_indepth:
1259  * implements tree growing procedure:
1260  * - allocates new block
1261  * - moves top-level data (index block or leaf) into the new block
1262  * - initializes new top-level, creating index that points to the
1263  *   just created block
1264  */
1265 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1266                                  unsigned int flags)
1267 {
1268         struct ext4_extent_header *neh;
1269         struct buffer_head *bh;
1270         ext4_fsblk_t newblock, goal = 0;
1271         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1272         int err = 0;
1273
1274         /* Try to prepend new index to old one */
1275         if (ext_depth(inode))
1276                 goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1277         if (goal > le32_to_cpu(es->s_first_data_block)) {
1278                 flags |= EXT4_MB_HINT_TRY_GOAL;
1279                 goal--;
1280         } else
1281                 goal = ext4_inode_to_goal_block(inode);
1282         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1283                                         NULL, &err);
1284         if (newblock == 0)
1285                 return err;
1286
1287         bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1288         if (unlikely(!bh))
1289                 return -ENOMEM;
1290         lock_buffer(bh);
1291
1292         err = ext4_journal_get_create_access(handle, bh);
1293         if (err) {
1294                 unlock_buffer(bh);
1295                 goto out;
1296         }
1297
1298         /* move top-level index/leaf into new block */
1299         memmove(bh->b_data, EXT4_I(inode)->i_data,
1300                 sizeof(EXT4_I(inode)->i_data));
1301
1302         /* set size of new block */
1303         neh = ext_block_hdr(bh);
1304         /* old root could have indexes or leaves
1305          * so calculate e_max right way */
1306         if (ext_depth(inode))
1307                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1308         else
1309                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1310         neh->eh_magic = EXT4_EXT_MAGIC;
1311         ext4_extent_block_csum_set(inode, neh);
1312         set_buffer_uptodate(bh);
1313         unlock_buffer(bh);
1314
1315         err = ext4_handle_dirty_metadata(handle, inode, bh);
1316         if (err)
1317                 goto out;
1318
1319         /* Update top-level index: num,max,pointer */
1320         neh = ext_inode_hdr(inode);
1321         neh->eh_entries = cpu_to_le16(1);
1322         ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1323         if (neh->eh_depth == 0) {
1324                 /* Root extent block becomes index block */
1325                 neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1326                 EXT_FIRST_INDEX(neh)->ei_block =
1327                         EXT_FIRST_EXTENT(neh)->ee_block;
1328         }
1329         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1330                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1331                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1332                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1333
1334         le16_add_cpu(&neh->eh_depth, 1);
1335         ext4_mark_inode_dirty(handle, inode);
1336 out:
1337         brelse(bh);
1338
1339         return err;
1340 }
1341
1342 /*
1343  * ext4_ext_create_new_leaf:
1344  * finds empty index and adds new leaf.
1345  * if no free index is found, then it requests in-depth growing.
1346  */
1347 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1348                                     unsigned int mb_flags,
1349                                     unsigned int gb_flags,
1350                                     struct ext4_ext_path **ppath,
1351                                     struct ext4_extent *newext)
1352 {
1353         struct ext4_ext_path *path = *ppath;
1354         struct ext4_ext_path *curp;
1355         int depth, i, err = 0;
1356
1357 repeat:
1358         i = depth = ext_depth(inode);
1359
1360         /* walk up to the tree and look for free index entry */
1361         curp = path + depth;
1362         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1363                 i--;
1364                 curp--;
1365         }
1366
1367         /* we use already allocated block for index block,
1368          * so subsequent data blocks should be contiguous */
1369         if (EXT_HAS_FREE_INDEX(curp)) {
1370                 /* if we found index with free entry, then use that
1371                  * entry: create all needed subtree and add new leaf */
1372                 err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1373                 if (err)
1374                         goto out;
1375
1376                 /* refill path */
1377                 path = ext4_find_extent(inode,
1378                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1379                                     ppath, gb_flags);
1380                 if (IS_ERR(path))
1381                         err = PTR_ERR(path);
1382         } else {
1383                 /* tree is full, time to grow in depth */
1384                 err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1385                 if (err)
1386                         goto out;
1387
1388                 /* refill path */
1389                 path = ext4_find_extent(inode,
1390                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1391                                     ppath, gb_flags);
1392                 if (IS_ERR(path)) {
1393                         err = PTR_ERR(path);
1394                         goto out;
1395                 }
1396
1397                 /*
1398                  * only first (depth 0 -> 1) produces free space;
1399                  * in all other cases we have to split the grown tree
1400                  */
1401                 depth = ext_depth(inode);
1402                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1403                         /* now we need to split */
1404                         goto repeat;
1405                 }
1406         }
1407
1408 out:
1409         return err;
1410 }
1411
1412 /*
1413  * search the closest allocated block to the left for *logical
1414  * and returns it at @logical + it's physical address at @phys
1415  * if *logical is the smallest allocated block, the function
1416  * returns 0 at @phys
1417  * return value contains 0 (success) or error code
1418  */
1419 static int ext4_ext_search_left(struct inode *inode,
1420                                 struct ext4_ext_path *path,
1421                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1422 {
1423         struct ext4_extent_idx *ix;
1424         struct ext4_extent *ex;
1425         int depth, ee_len;
1426
1427         if (unlikely(path == NULL)) {
1428                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1429                 return -EFSCORRUPTED;
1430         }
1431         depth = path->p_depth;
1432         *phys = 0;
1433
1434         if (depth == 0 && path->p_ext == NULL)
1435                 return 0;
1436
1437         /* usually extent in the path covers blocks smaller
1438          * then *logical, but it can be that extent is the
1439          * first one in the file */
1440
1441         ex = path[depth].p_ext;
1442         ee_len = ext4_ext_get_actual_len(ex);
1443         if (*logical < le32_to_cpu(ex->ee_block)) {
1444                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1445                         EXT4_ERROR_INODE(inode,
1446                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1447                                          *logical, le32_to_cpu(ex->ee_block));
1448                         return -EFSCORRUPTED;
1449                 }
1450                 while (--depth >= 0) {
1451                         ix = path[depth].p_idx;
1452                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1453                                 EXT4_ERROR_INODE(inode,
1454                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1455                                   ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1456                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1457                 le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1458                                   depth);
1459                                 return -EFSCORRUPTED;
1460                         }
1461                 }
1462                 return 0;
1463         }
1464
1465         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1466                 EXT4_ERROR_INODE(inode,
1467                                  "logical %d < ee_block %d + ee_len %d!",
1468                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1469                 return -EFSCORRUPTED;
1470         }
1471
1472         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1473         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1474         return 0;
1475 }
1476
1477 /*
1478  * search the closest allocated block to the right for *logical
1479  * and returns it at @logical + it's physical address at @phys
1480  * if *logical is the largest allocated block, the function
1481  * returns 0 at @phys
1482  * return value contains 0 (success) or error code
1483  */
1484 static int ext4_ext_search_right(struct inode *inode,
1485                                  struct ext4_ext_path *path,
1486                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1487                                  struct ext4_extent **ret_ex)
1488 {
1489         struct buffer_head *bh = NULL;
1490         struct ext4_extent_header *eh;
1491         struct ext4_extent_idx *ix;
1492         struct ext4_extent *ex;
1493         ext4_fsblk_t block;
1494         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1495         int ee_len;
1496
1497         if (unlikely(path == NULL)) {
1498                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1499                 return -EFSCORRUPTED;
1500         }
1501         depth = path->p_depth;
1502         *phys = 0;
1503
1504         if (depth == 0 && path->p_ext == NULL)
1505                 return 0;
1506
1507         /* usually extent in the path covers blocks smaller
1508          * then *logical, but it can be that extent is the
1509          * first one in the file */
1510
1511         ex = path[depth].p_ext;
1512         ee_len = ext4_ext_get_actual_len(ex);
1513         if (*logical < le32_to_cpu(ex->ee_block)) {
1514                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1515                         EXT4_ERROR_INODE(inode,
1516                                          "first_extent(path[%d].p_hdr) != ex",
1517                                          depth);
1518                         return -EFSCORRUPTED;
1519                 }
1520                 while (--depth >= 0) {
1521                         ix = path[depth].p_idx;
1522                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1523                                 EXT4_ERROR_INODE(inode,
1524                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1525                                                  *logical);
1526                                 return -EFSCORRUPTED;
1527                         }
1528                 }
1529                 goto found_extent;
1530         }
1531
1532         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1533                 EXT4_ERROR_INODE(inode,
1534                                  "logical %d < ee_block %d + ee_len %d!",
1535                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1536                 return -EFSCORRUPTED;
1537         }
1538
1539         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1540                 /* next allocated block in this leaf */
1541                 ex++;
1542                 goto found_extent;
1543         }
1544
1545         /* go up and search for index to the right */
1546         while (--depth >= 0) {
1547                 ix = path[depth].p_idx;
1548                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1549                         goto got_index;
1550         }
1551
1552         /* we've gone up to the root and found no index to the right */
1553         return 0;
1554
1555 got_index:
1556         /* we've found index to the right, let's
1557          * follow it and find the closest allocated
1558          * block to the right */
1559         ix++;
1560         block = ext4_idx_pblock(ix);
1561         while (++depth < path->p_depth) {
1562                 /* subtract from p_depth to get proper eh_depth */
1563                 bh = read_extent_tree_block(inode, block,
1564                                             path->p_depth - depth, 0);
1565                 if (IS_ERR(bh))
1566                         return PTR_ERR(bh);
1567                 eh = ext_block_hdr(bh);
1568                 ix = EXT_FIRST_INDEX(eh);
1569                 block = ext4_idx_pblock(ix);
1570                 put_bh(bh);
1571         }
1572
1573         bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1574         if (IS_ERR(bh))
1575                 return PTR_ERR(bh);
1576         eh = ext_block_hdr(bh);
1577         ex = EXT_FIRST_EXTENT(eh);
1578 found_extent:
1579         *logical = le32_to_cpu(ex->ee_block);
1580         *phys = ext4_ext_pblock(ex);
1581         *ret_ex = ex;
1582         if (bh)
1583                 put_bh(bh);
1584         return 0;
1585 }
1586
1587 /*
1588  * ext4_ext_next_allocated_block:
1589  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1590  * NOTE: it considers block number from index entry as
1591  * allocated block. Thus, index entries have to be consistent
1592  * with leaves.
1593  */
1594 ext4_lblk_t
1595 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1596 {
1597         int depth;
1598
1599         BUG_ON(path == NULL);
1600         depth = path->p_depth;
1601
1602         if (depth == 0 && path->p_ext == NULL)
1603                 return EXT_MAX_BLOCKS;
1604
1605         while (depth >= 0) {
1606                 if (depth == path->p_depth) {
1607                         /* leaf */
1608                         if (path[depth].p_ext &&
1609                                 path[depth].p_ext !=
1610                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1611                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1612                 } else {
1613                         /* index */
1614                         if (path[depth].p_idx !=
1615                                         EXT_LAST_INDEX(path[depth].p_hdr))
1616                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1617                 }
1618                 depth--;
1619         }
1620
1621         return EXT_MAX_BLOCKS;
1622 }
1623
1624 /*
1625  * ext4_ext_next_leaf_block:
1626  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1627  */
1628 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1629 {
1630         int depth;
1631
1632         BUG_ON(path == NULL);
1633         depth = path->p_depth;
1634
1635         /* zero-tree has no leaf blocks at all */
1636         if (depth == 0)
1637                 return EXT_MAX_BLOCKS;
1638
1639         /* go to index block */
1640         depth--;
1641
1642         while (depth >= 0) {
1643                 if (path[depth].p_idx !=
1644                                 EXT_LAST_INDEX(path[depth].p_hdr))
1645                         return (ext4_lblk_t)
1646                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1647                 depth--;
1648         }
1649
1650         return EXT_MAX_BLOCKS;
1651 }
1652
1653 /*
1654  * ext4_ext_correct_indexes:
1655  * if leaf gets modified and modified extent is first in the leaf,
1656  * then we have to correct all indexes above.
1657  * TODO: do we need to correct tree in all cases?
1658  */
1659 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1660                                 struct ext4_ext_path *path)
1661 {
1662         struct ext4_extent_header *eh;
1663         int depth = ext_depth(inode);
1664         struct ext4_extent *ex;
1665         __le32 border;
1666         int k, err = 0;
1667
1668         eh = path[depth].p_hdr;
1669         ex = path[depth].p_ext;
1670
1671         if (unlikely(ex == NULL || eh == NULL)) {
1672                 EXT4_ERROR_INODE(inode,
1673                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1674                 return -EFSCORRUPTED;
1675         }
1676
1677         if (depth == 0) {
1678                 /* there is no tree at all */
1679                 return 0;
1680         }
1681
1682         if (ex != EXT_FIRST_EXTENT(eh)) {
1683                 /* we correct tree if first leaf got modified only */
1684                 return 0;
1685         }
1686
1687         /*
1688          * TODO: we need correction if border is smaller than current one
1689          */
1690         k = depth - 1;
1691         border = path[depth].p_ext->ee_block;
1692         err = ext4_ext_get_access(handle, inode, path + k);
1693         if (err)
1694                 return err;
1695         path[k].p_idx->ei_block = border;
1696         err = ext4_ext_dirty(handle, inode, path + k);
1697         if (err)
1698                 return err;
1699
1700         while (k--) {
1701                 /* change all left-side indexes */
1702                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1703                         break;
1704                 err = ext4_ext_get_access(handle, inode, path + k);
1705                 if (err)
1706                         break;
1707                 path[k].p_idx->ei_block = border;
1708                 err = ext4_ext_dirty(handle, inode, path + k);
1709                 if (err)
1710                         break;
1711         }
1712
1713         return err;
1714 }
1715
1716 int
1717 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1718                                 struct ext4_extent *ex2)
1719 {
1720         unsigned short ext1_ee_len, ext2_ee_len;
1721
1722         if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1723                 return 0;
1724
1725         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1726         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1727
1728         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1729                         le32_to_cpu(ex2->ee_block))
1730                 return 0;
1731
1732         /*
1733          * To allow future support for preallocated extents to be added
1734          * as an RO_COMPAT feature, refuse to merge to extents if
1735          * this can result in the top bit of ee_len being set.
1736          */
1737         if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1738                 return 0;
1739         /*
1740          * The check for IO to unwritten extent is somewhat racy as we
1741          * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1742          * dropping i_data_sem. But reserved blocks should save us in that
1743          * case.
1744          */
1745         if (ext4_ext_is_unwritten(ex1) &&
1746             (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
1747              atomic_read(&EXT4_I(inode)->i_unwritten) ||
1748              (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)))
1749                 return 0;
1750 #ifdef AGGRESSIVE_TEST
1751         if (ext1_ee_len >= 4)
1752                 return 0;
1753 #endif
1754
1755         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1756                 return 1;
1757         return 0;
1758 }
1759
1760 /*
1761  * This function tries to merge the "ex" extent to the next extent in the tree.
1762  * It always tries to merge towards right. If you want to merge towards
1763  * left, pass "ex - 1" as argument instead of "ex".
1764  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1765  * 1 if they got merged.
1766  */
1767 static int ext4_ext_try_to_merge_right(struct inode *inode,
1768                                  struct ext4_ext_path *path,
1769                                  struct ext4_extent *ex)
1770 {
1771         struct ext4_extent_header *eh;
1772         unsigned int depth, len;
1773         int merge_done = 0, unwritten;
1774
1775         depth = ext_depth(inode);
1776         BUG_ON(path[depth].p_hdr == NULL);
1777         eh = path[depth].p_hdr;
1778
1779         while (ex < EXT_LAST_EXTENT(eh)) {
1780                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1781                         break;
1782                 /* merge with next extent! */
1783                 unwritten = ext4_ext_is_unwritten(ex);
1784                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1785                                 + ext4_ext_get_actual_len(ex + 1));
1786                 if (unwritten)
1787                         ext4_ext_mark_unwritten(ex);
1788
1789                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1790                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1791                                 * sizeof(struct ext4_extent);
1792                         memmove(ex + 1, ex + 2, len);
1793                 }
1794                 le16_add_cpu(&eh->eh_entries, -1);
1795                 merge_done = 1;
1796                 WARN_ON(eh->eh_entries == 0);
1797                 if (!eh->eh_entries)
1798                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1799         }
1800
1801         return merge_done;
1802 }
1803
1804 /*
1805  * This function does a very simple check to see if we can collapse
1806  * an extent tree with a single extent tree leaf block into the inode.
1807  */
1808 static void ext4_ext_try_to_merge_up(handle_t *handle,
1809                                      struct inode *inode,
1810                                      struct ext4_ext_path *path)
1811 {
1812         size_t s;
1813         unsigned max_root = ext4_ext_space_root(inode, 0);
1814         ext4_fsblk_t blk;
1815
1816         if ((path[0].p_depth != 1) ||
1817             (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1818             (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1819                 return;
1820
1821         /*
1822          * We need to modify the block allocation bitmap and the block
1823          * group descriptor to release the extent tree block.  If we
1824          * can't get the journal credits, give up.
1825          */
1826         if (ext4_journal_extend(handle, 2))
1827                 return;
1828
1829         /*
1830          * Copy the extent data up to the inode
1831          */
1832         blk = ext4_idx_pblock(path[0].p_idx);
1833         s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1834                 sizeof(struct ext4_extent_idx);
1835         s += sizeof(struct ext4_extent_header);
1836
1837         path[1].p_maxdepth = path[0].p_maxdepth;
1838         memcpy(path[0].p_hdr, path[1].p_hdr, s);
1839         path[0].p_depth = 0;
1840         path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1841                 (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1842         path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1843
1844         brelse(path[1].p_bh);
1845         ext4_free_blocks(handle, inode, NULL, blk, 1,
1846                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1847 }
1848
1849 /*
1850  * This function tries to merge the @ex extent to neighbours in the tree.
1851  * return 1 if merge left else 0.
1852  */
1853 static void ext4_ext_try_to_merge(handle_t *handle,
1854                                   struct inode *inode,
1855                                   struct ext4_ext_path *path,
1856                                   struct ext4_extent *ex) {
1857         struct ext4_extent_header *eh;
1858         unsigned int depth;
1859         int merge_done = 0;
1860
1861         depth = ext_depth(inode);
1862         BUG_ON(path[depth].p_hdr == NULL);
1863         eh = path[depth].p_hdr;
1864
1865         if (ex > EXT_FIRST_EXTENT(eh))
1866                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1867
1868         if (!merge_done)
1869                 (void) ext4_ext_try_to_merge_right(inode, path, ex);
1870
1871         ext4_ext_try_to_merge_up(handle, inode, path);
1872 }
1873
1874 /*
1875  * check if a portion of the "newext" extent overlaps with an
1876  * existing extent.
1877  *
1878  * If there is an overlap discovered, it updates the length of the newext
1879  * such that there will be no overlap, and then returns 1.
1880  * If there is no overlap found, it returns 0.
1881  */
1882 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1883                                            struct inode *inode,
1884                                            struct ext4_extent *newext,
1885                                            struct ext4_ext_path *path)
1886 {
1887         ext4_lblk_t b1, b2;
1888         unsigned int depth, len1;
1889         unsigned int ret = 0;
1890
1891         b1 = le32_to_cpu(newext->ee_block);
1892         len1 = ext4_ext_get_actual_len(newext);
1893         depth = ext_depth(inode);
1894         if (!path[depth].p_ext)
1895                 goto out;
1896         b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1897
1898         /*
1899          * get the next allocated block if the extent in the path
1900          * is before the requested block(s)
1901          */
1902         if (b2 < b1) {
1903                 b2 = ext4_ext_next_allocated_block(path);
1904                 if (b2 == EXT_MAX_BLOCKS)
1905                         goto out;
1906                 b2 = EXT4_LBLK_CMASK(sbi, b2);
1907         }
1908
1909         /* check for wrap through zero on extent logical start block*/
1910         if (b1 + len1 < b1) {
1911                 len1 = EXT_MAX_BLOCKS - b1;
1912                 newext->ee_len = cpu_to_le16(len1);
1913                 ret = 1;
1914         }
1915
1916         /* check for overlap */
1917         if (b1 + len1 > b2) {
1918                 newext->ee_len = cpu_to_le16(b2 - b1);
1919                 ret = 1;
1920         }
1921 out:
1922         return ret;
1923 }
1924
1925 /*
1926  * ext4_ext_insert_extent:
1927  * tries to merge requsted extent into the existing extent or
1928  * inserts requested extent as new one into the tree,
1929  * creating new leaf in the no-space case.
1930  */
1931 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1932                                 struct ext4_ext_path **ppath,
1933                                 struct ext4_extent *newext, int gb_flags)
1934 {
1935         struct ext4_ext_path *path = *ppath;
1936         struct ext4_extent_header *eh;
1937         struct ext4_extent *ex, *fex;
1938         struct ext4_extent *nearex; /* nearest extent */
1939         struct ext4_ext_path *npath = NULL;
1940         int depth, len, err;
1941         ext4_lblk_t next;
1942         int mb_flags = 0, unwritten;
1943
1944         if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1945                 mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1946         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1947                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1948                 return -EFSCORRUPTED;
1949         }
1950         depth = ext_depth(inode);
1951         ex = path[depth].p_ext;
1952         eh = path[depth].p_hdr;
1953         if (unlikely(path[depth].p_hdr == NULL)) {
1954                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1955                 return -EFSCORRUPTED;
1956         }
1957
1958         /* try to insert block into found extent and return */
1959         if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1960
1961                 /*
1962                  * Try to see whether we should rather test the extent on
1963                  * right from ex, or from the left of ex. This is because
1964                  * ext4_find_extent() can return either extent on the
1965                  * left, or on the right from the searched position. This
1966                  * will make merging more effective.
1967                  */
1968                 if (ex < EXT_LAST_EXTENT(eh) &&
1969                     (le32_to_cpu(ex->ee_block) +
1970                     ext4_ext_get_actual_len(ex) <
1971                     le32_to_cpu(newext->ee_block))) {
1972                         ex += 1;
1973                         goto prepend;
1974                 } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1975                            (le32_to_cpu(newext->ee_block) +
1976                            ext4_ext_get_actual_len(newext) <
1977                            le32_to_cpu(ex->ee_block)))
1978                         ex -= 1;
1979
1980                 /* Try to append newex to the ex */
1981                 if (ext4_can_extents_be_merged(inode, ex, newext)) {
1982                         ext_debug("append [%d]%d block to %u:[%d]%d"
1983                                   "(from %llu)\n",
1984                                   ext4_ext_is_unwritten(newext),
1985                                   ext4_ext_get_actual_len(newext),
1986                                   le32_to_cpu(ex->ee_block),
1987                                   ext4_ext_is_unwritten(ex),
1988                                   ext4_ext_get_actual_len(ex),
1989                                   ext4_ext_pblock(ex));
1990                         err = ext4_ext_get_access(handle, inode,
1991                                                   path + depth);
1992                         if (err)
1993                                 return err;
1994                         unwritten = ext4_ext_is_unwritten(ex);
1995                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1996                                         + ext4_ext_get_actual_len(newext));
1997                         if (unwritten)
1998                                 ext4_ext_mark_unwritten(ex);
1999                         eh = path[depth].p_hdr;
2000                         nearex = ex;
2001                         goto merge;
2002                 }
2003
2004 prepend:
2005                 /* Try to prepend newex to the ex */
2006                 if (ext4_can_extents_be_merged(inode, newext, ex)) {
2007                         ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2008                                   "(from %llu)\n",
2009                                   le32_to_cpu(newext->ee_block),
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
2021                         unwritten = ext4_ext_is_unwritten(ex);
2022                         ex->ee_block = newext->ee_block;
2023                         ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2024                         ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2025                                         + ext4_ext_get_actual_len(newext));
2026                         if (unwritten)
2027                                 ext4_ext_mark_unwritten(ex);
2028                         eh = path[depth].p_hdr;
2029                         nearex = ex;
2030                         goto merge;
2031                 }
2032         }
2033
2034         depth = ext_depth(inode);
2035         eh = path[depth].p_hdr;
2036         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2037                 goto has_space;
2038
2039         /* probably next leaf has space for us? */
2040         fex = EXT_LAST_EXTENT(eh);
2041         next = EXT_MAX_BLOCKS;
2042         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2043                 next = ext4_ext_next_leaf_block(path);
2044         if (next != EXT_MAX_BLOCKS) {
2045                 ext_debug("next leaf block - %u\n", next);
2046                 BUG_ON(npath != NULL);
2047                 npath = ext4_find_extent(inode, next, NULL, 0);
2048                 if (IS_ERR(npath))
2049                         return PTR_ERR(npath);
2050                 BUG_ON(npath->p_depth != path->p_depth);
2051                 eh = npath[depth].p_hdr;
2052                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2053                         ext_debug("next leaf isn't full(%d)\n",
2054                                   le16_to_cpu(eh->eh_entries));
2055                         path = npath;
2056                         goto has_space;
2057                 }
2058                 ext_debug("next leaf has no free space(%d,%d)\n",
2059                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2060         }
2061
2062         /*
2063          * There is no free space in the found leaf.
2064          * We're gonna add a new leaf in the tree.
2065          */
2066         if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2067                 mb_flags |= EXT4_MB_USE_RESERVED;
2068         err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2069                                        ppath, newext);
2070         if (err)
2071                 goto cleanup;
2072         depth = ext_depth(inode);
2073         eh = path[depth].p_hdr;
2074
2075 has_space:
2076         nearex = path[depth].p_ext;
2077
2078         err = ext4_ext_get_access(handle, inode, path + depth);
2079         if (err)
2080                 goto cleanup;
2081
2082         if (!nearex) {
2083                 /* there is no extent in this leaf, create first one */
2084                 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2085                                 le32_to_cpu(newext->ee_block),
2086                                 ext4_ext_pblock(newext),
2087                                 ext4_ext_is_unwritten(newext),
2088                                 ext4_ext_get_actual_len(newext));
2089                 nearex = EXT_FIRST_EXTENT(eh);
2090         } else {
2091                 if (le32_to_cpu(newext->ee_block)
2092                            > le32_to_cpu(nearex->ee_block)) {
2093                         /* Insert after */
2094                         ext_debug("insert %u:%llu:[%d]%d before: "
2095                                         "nearest %p\n",
2096                                         le32_to_cpu(newext->ee_block),
2097                                         ext4_ext_pblock(newext),
2098                                         ext4_ext_is_unwritten(newext),
2099                                         ext4_ext_get_actual_len(newext),
2100                                         nearex);
2101                         nearex++;
2102                 } else {
2103                         /* Insert before */
2104                         BUG_ON(newext->ee_block == nearex->ee_block);
2105                         ext_debug("insert %u:%llu:[%d]%d after: "
2106                                         "nearest %p\n",
2107                                         le32_to_cpu(newext->ee_block),
2108                                         ext4_ext_pblock(newext),
2109                                         ext4_ext_is_unwritten(newext),
2110                                         ext4_ext_get_actual_len(newext),
2111                                         nearex);
2112                 }
2113                 len = EXT_LAST_EXTENT(eh) - nearex + 1;
2114                 if (len > 0) {
2115                         ext_debug("insert %u:%llu:[%d]%d: "
2116                                         "move %d extents from 0x%p to 0x%p\n",
2117                                         le32_to_cpu(newext->ee_block),
2118                                         ext4_ext_pblock(newext),
2119                                         ext4_ext_is_unwritten(newext),
2120                                         ext4_ext_get_actual_len(newext),
2121                                         len, nearex, nearex + 1);
2122                         memmove(nearex + 1, nearex,
2123                                 len * sizeof(struct ext4_extent));
2124                 }
2125         }
2126
2127         le16_add_cpu(&eh->eh_entries, 1);
2128         path[depth].p_ext = nearex;
2129         nearex->ee_block = newext->ee_block;
2130         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2131         nearex->ee_len = newext->ee_len;
2132
2133 merge:
2134         /* try to merge extents */
2135         if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2136                 ext4_ext_try_to_merge(handle, inode, path, nearex);
2137
2138
2139         /* time to correct all indexes above */
2140         err = ext4_ext_correct_indexes(handle, inode, path);
2141         if (err)
2142                 goto cleanup;
2143
2144         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2145
2146 cleanup:
2147         ext4_ext_drop_refs(npath);
2148         kfree(npath);
2149         return err;
2150 }
2151
2152 static int ext4_fill_fiemap_extents(struct inode *inode,
2153                                     ext4_lblk_t block, ext4_lblk_t num,
2154                                     struct fiemap_extent_info *fieinfo)
2155 {
2156         struct ext4_ext_path *path = NULL;
2157         struct ext4_extent *ex;
2158         struct extent_status es;
2159         ext4_lblk_t next, next_del, start = 0, end = 0;
2160         ext4_lblk_t last = block + num;
2161         int exists, depth = 0, err = 0;
2162         unsigned int flags = 0;
2163         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2164
2165         while (block < last && block != EXT_MAX_BLOCKS) {
2166                 num = last - block;
2167                 /* find extent for this block */
2168                 down_read(&EXT4_I(inode)->i_data_sem);
2169
2170                 path = ext4_find_extent(inode, block, &path, 0);
2171                 if (IS_ERR(path)) {
2172                         up_read(&EXT4_I(inode)->i_data_sem);
2173                         err = PTR_ERR(path);
2174                         path = NULL;
2175                         break;
2176                 }
2177
2178                 depth = ext_depth(inode);
2179                 if (unlikely(path[depth].p_hdr == NULL)) {
2180                         up_read(&EXT4_I(inode)->i_data_sem);
2181                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2182                         err = -EFSCORRUPTED;
2183                         break;
2184                 }
2185                 ex = path[depth].p_ext;
2186                 next = ext4_ext_next_allocated_block(path);
2187
2188                 flags = 0;
2189                 exists = 0;
2190                 if (!ex) {
2191                         /* there is no extent yet, so try to allocate
2192                          * all requested space */
2193                         start = block;
2194                         end = block + num;
2195                 } else if (le32_to_cpu(ex->ee_block) > block) {
2196                         /* need to allocate space before found extent */
2197                         start = block;
2198                         end = le32_to_cpu(ex->ee_block);
2199                         if (block + num < end)
2200                                 end = block + num;
2201                 } else if (block >= le32_to_cpu(ex->ee_block)
2202                                         + ext4_ext_get_actual_len(ex)) {
2203                         /* need to allocate space after found extent */
2204                         start = block;
2205                         end = block + num;
2206                         if (end >= next)
2207                                 end = next;
2208                 } else if (block >= le32_to_cpu(ex->ee_block)) {
2209                         /*
2210                          * some part of requested space is covered
2211                          * by found extent
2212                          */
2213                         start = block;
2214                         end = le32_to_cpu(ex->ee_block)
2215                                 + ext4_ext_get_actual_len(ex);
2216                         if (block + num < end)
2217                                 end = block + num;
2218                         exists = 1;
2219                 } else {
2220                         BUG();
2221                 }
2222                 BUG_ON(end <= start);
2223
2224                 if (!exists) {
2225                         es.es_lblk = start;
2226                         es.es_len = end - start;
2227                         es.es_pblk = 0;
2228                 } else {
2229                         es.es_lblk = le32_to_cpu(ex->ee_block);
2230                         es.es_len = ext4_ext_get_actual_len(ex);
2231                         es.es_pblk = ext4_ext_pblock(ex);
2232                         if (ext4_ext_is_unwritten(ex))
2233                                 flags |= FIEMAP_EXTENT_UNWRITTEN;
2234                 }
2235
2236                 /*
2237                  * Find delayed extent and update es accordingly. We call
2238                  * it even in !exists case to find out whether es is the
2239                  * last existing extent or not.
2240                  */
2241                 next_del = ext4_find_delayed_extent(inode, &es);
2242                 if (!exists && next_del) {
2243                         exists = 1;
2244                         flags |= (FIEMAP_EXTENT_DELALLOC |
2245                                   FIEMAP_EXTENT_UNKNOWN);
2246                 }
2247                 up_read(&EXT4_I(inode)->i_data_sem);
2248
2249                 if (unlikely(es.es_len == 0)) {
2250                         EXT4_ERROR_INODE(inode, "es.es_len == 0");
2251                         err = -EFSCORRUPTED;
2252                         break;
2253                 }
2254
2255                 /*
2256                  * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2257                  * we need to check next == EXT_MAX_BLOCKS because it is
2258                  * possible that an extent is with unwritten and delayed
2259                  * status due to when an extent is delayed allocated and
2260                  * is allocated by fallocate status tree will track both of
2261                  * them in a extent.
2262                  *
2263                  * So we could return a unwritten and delayed extent, and
2264                  * its block is equal to 'next'.
2265                  */
2266                 if (next == next_del && next == EXT_MAX_BLOCKS) {
2267                         flags |= FIEMAP_EXTENT_LAST;
2268                         if (unlikely(next_del != EXT_MAX_BLOCKS ||
2269                                      next != EXT_MAX_BLOCKS)) {
2270                                 EXT4_ERROR_INODE(inode,
2271                                                  "next extent == %u, next "
2272                                                  "delalloc extent = %u",
2273                                                  next, next_del);
2274                                 err = -EFSCORRUPTED;
2275                                 break;
2276                         }
2277                 }
2278
2279                 if (exists) {
2280                         err = fiemap_fill_next_extent(fieinfo,
2281                                 (__u64)es.es_lblk << blksize_bits,
2282                                 (__u64)es.es_pblk << blksize_bits,
2283                                 (__u64)es.es_len << blksize_bits,
2284                                 flags);
2285                         if (err < 0)
2286                                 break;
2287                         if (err == 1) {
2288                                 err = 0;
2289                                 break;
2290                         }
2291                 }
2292
2293                 block = es.es_lblk + es.es_len;
2294         }
2295
2296         ext4_ext_drop_refs(path);
2297         kfree(path);
2298         return err;
2299 }
2300
2301 /*
2302  * ext4_ext_determine_hole - determine hole around given block
2303  * @inode:      inode we lookup in
2304  * @path:       path in extent tree to @lblk
2305  * @lblk:       pointer to logical block around which we want to determine hole
2306  *
2307  * Determine hole length (and start if easily possible) around given logical
2308  * block. We don't try too hard to find the beginning of the hole but @path
2309  * actually points to extent before @lblk, we provide it.
2310  *
2311  * The function returns the length of a hole starting at @lblk. We update @lblk
2312  * to the beginning of the hole if we managed to find it.
2313  */
2314 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2315                                            struct ext4_ext_path *path,
2316                                            ext4_lblk_t *lblk)
2317 {
2318         int depth = ext_depth(inode);
2319         struct ext4_extent *ex;
2320         ext4_lblk_t len;
2321
2322         ex = path[depth].p_ext;
2323         if (ex == NULL) {
2324                 /* there is no extent yet, so gap is [0;-] */
2325                 *lblk = 0;
2326                 len = EXT_MAX_BLOCKS;
2327         } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2328                 len = le32_to_cpu(ex->ee_block) - *lblk;
2329         } else if (*lblk >= le32_to_cpu(ex->ee_block)
2330                         + ext4_ext_get_actual_len(ex)) {
2331                 ext4_lblk_t next;
2332
2333                 *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2334                 next = ext4_ext_next_allocated_block(path);
2335                 BUG_ON(next == *lblk);
2336                 len = next - *lblk;
2337         } else {
2338                 BUG();
2339         }
2340         return len;
2341 }
2342
2343 /*
2344  * ext4_ext_put_gap_in_cache:
2345  * calculate boundaries of the gap that the requested block fits into
2346  * and cache this gap
2347  */
2348 static void
2349 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2350                           ext4_lblk_t hole_len)
2351 {
2352         struct extent_status es;
2353
2354         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2355                                   hole_start + hole_len - 1, &es);
2356         if (es.es_len) {
2357                 /* There's delayed extent containing lblock? */
2358                 if (es.es_lblk <= hole_start)
2359                         return;
2360                 hole_len = min(es.es_lblk - hole_start, hole_len);
2361         }
2362         ext_debug(" -> %u:%u\n", hole_start, hole_len);
2363         ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2364                               EXTENT_STATUS_HOLE);
2365 }
2366
2367 /*
2368  * ext4_ext_rm_idx:
2369  * removes index from the index block.
2370  */
2371 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2372                         struct ext4_ext_path *path, int depth)
2373 {
2374         int err;
2375         ext4_fsblk_t leaf;
2376
2377         /* free index block */
2378         depth--;
2379         path = path + depth;
2380         leaf = ext4_idx_pblock(path->p_idx);
2381         if (unlikely(path->p_hdr->eh_entries == 0)) {
2382                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2383                 return -EFSCORRUPTED;
2384         }
2385         err = ext4_ext_get_access(handle, inode, path);
2386         if (err)
2387                 return err;
2388
2389         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2390                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2391                 len *= sizeof(struct ext4_extent_idx);
2392                 memmove(path->p_idx, path->p_idx + 1, len);
2393         }
2394
2395         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2396         err = ext4_ext_dirty(handle, inode, path);
2397         if (err)
2398                 return err;
2399         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2400         trace_ext4_ext_rm_idx(inode, leaf);
2401
2402         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2403                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2404
2405         while (--depth >= 0) {
2406                 if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2407                         break;
2408                 path--;
2409                 err = ext4_ext_get_access(handle, inode, path);
2410                 if (err)
2411                         break;
2412                 path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2413                 err = ext4_ext_dirty(handle, inode, path);
2414                 if (err)
2415                         break;
2416         }
2417         return err;
2418 }
2419
2420 /*
2421  * ext4_ext_calc_credits_for_single_extent:
2422  * This routine returns max. credits that needed to insert an extent
2423  * to the extent tree.
2424  * When pass the actual path, the caller should calculate credits
2425  * under i_data_sem.
2426  */
2427 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2428                                                 struct ext4_ext_path *path)
2429 {
2430         if (path) {
2431                 int depth = ext_depth(inode);
2432                 int ret = 0;
2433
2434                 /* probably there is space in leaf? */
2435                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2436                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2437
2438                         /*
2439                          *  There are some space in the leaf tree, no
2440                          *  need to account for leaf block credit
2441                          *
2442                          *  bitmaps and block group descriptor blocks
2443                          *  and other metadata blocks still need to be
2444                          *  accounted.
2445                          */
2446                         /* 1 bitmap, 1 block group descriptor */
2447                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2448                         return ret;
2449                 }
2450         }
2451
2452         return ext4_chunk_trans_blocks(inode, nrblocks);
2453 }
2454
2455 /*
2456  * How many index/leaf blocks need to change/allocate to add @extents extents?
2457  *
2458  * If we add a single extent, then in the worse case, each tree level
2459  * index/leaf need to be changed in case of the tree split.
2460  *
2461  * If more extents are inserted, they could cause the whole tree split more
2462  * than once, but this is really rare.
2463  */
2464 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2465 {
2466         int index;
2467         int depth;
2468
2469         /* If we are converting the inline data, only one is needed here. */
2470         if (ext4_has_inline_data(inode))
2471                 return 1;
2472
2473         depth = ext_depth(inode);
2474
2475         if (extents <= 1)
2476                 index = depth * 2;
2477         else
2478                 index = depth * 3;
2479
2480         return index;
2481 }
2482
2483 static inline int get_default_free_blocks_flags(struct inode *inode)
2484 {
2485         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2486             ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2487                 return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2488         else if (ext4_should_journal_data(inode))
2489                 return EXT4_FREE_BLOCKS_FORGET;
2490         return 0;
2491 }
2492
2493 /*
2494  * ext4_rereserve_cluster - increment the reserved cluster count when
2495  *                          freeing a cluster with a pending reservation
2496  *
2497  * @inode - file containing the cluster
2498  * @lblk - logical block in cluster to be reserved
2499  *
2500  * Increments the reserved cluster count and adjusts quota in a bigalloc
2501  * file system when freeing a partial cluster containing at least one
2502  * delayed and unwritten block.  A partial cluster meeting that
2503  * requirement will have a pending reservation.  If so, the
2504  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2505  * defer reserved and allocated space accounting to a subsequent call
2506  * to this function.
2507  */
2508 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2509 {
2510         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2511         struct ext4_inode_info *ei = EXT4_I(inode);
2512
2513         dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2514
2515         spin_lock(&ei->i_block_reservation_lock);
2516         ei->i_reserved_data_blocks++;
2517         percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2518         spin_unlock(&ei->i_block_reservation_lock);
2519
2520         percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2521         ext4_remove_pending(inode, lblk);
2522 }
2523
2524 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2525                               struct ext4_extent *ex,
2526                               struct partial_cluster *partial,
2527                               ext4_lblk_t from, ext4_lblk_t to)
2528 {
2529         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2530         unsigned short ee_len = ext4_ext_get_actual_len(ex);
2531         ext4_fsblk_t last_pblk, pblk;
2532         ext4_lblk_t num;
2533         int flags;
2534
2535         /* only extent tail removal is allowed */
2536         if (from < le32_to_cpu(ex->ee_block) ||
2537             to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2538                 ext4_error(sbi->s_sb,
2539                            "strange request: removal(2) %u-%u from %u:%u",
2540                            from, to, le32_to_cpu(ex->ee_block), ee_len);
2541                 return 0;
2542         }
2543
2544 #ifdef EXTENTS_STATS
2545         spin_lock(&sbi->s_ext_stats_lock);
2546         sbi->s_ext_blocks += ee_len;
2547         sbi->s_ext_extents++;
2548         if (ee_len < sbi->s_ext_min)
2549                 sbi->s_ext_min = ee_len;
2550         if (ee_len > sbi->s_ext_max)
2551                 sbi->s_ext_max = ee_len;
2552         if (ext_depth(inode) > sbi->s_depth_max)
2553                 sbi->s_depth_max = ext_depth(inode);
2554         spin_unlock(&sbi->s_ext_stats_lock);
2555 #endif
2556
2557         trace_ext4_remove_blocks(inode, ex, from, to, partial);
2558
2559         /*
2560          * if we have a partial cluster, and it's different from the
2561          * cluster of the last block in the extent, we free it
2562          */
2563         last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2564
2565         if (partial->state != initial &&
2566             partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2567                 if (partial->state == tofree) {
2568                         flags = get_default_free_blocks_flags(inode);
2569                         if (ext4_is_pending(inode, partial->lblk))
2570                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2571                         ext4_free_blocks(handle, inode, NULL,
2572                                          EXT4_C2B(sbi, partial->pclu),
2573                                          sbi->s_cluster_ratio, flags);
2574                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2575                                 ext4_rereserve_cluster(inode, partial->lblk);
2576                 }
2577                 partial->state = initial;
2578         }
2579
2580         num = le32_to_cpu(ex->ee_block) + ee_len - from;
2581         pblk = ext4_ext_pblock(ex) + ee_len - num;
2582
2583         /*
2584          * We free the partial cluster at the end of the extent (if any),
2585          * unless the cluster is used by another extent (partial_cluster
2586          * state is nofree).  If a partial cluster exists here, it must be
2587          * shared with the last block in the extent.
2588          */
2589         flags = get_default_free_blocks_flags(inode);
2590
2591         /* partial, left end cluster aligned, right end unaligned */
2592         if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2593             (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2594             (partial->state != nofree)) {
2595                 if (ext4_is_pending(inode, to))
2596                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2597                 ext4_free_blocks(handle, inode, NULL,
2598                                  EXT4_PBLK_CMASK(sbi, last_pblk),
2599                                  sbi->s_cluster_ratio, flags);
2600                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2601                         ext4_rereserve_cluster(inode, to);
2602                 partial->state = initial;
2603                 flags = get_default_free_blocks_flags(inode);
2604         }
2605
2606         flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2607
2608         /*
2609          * For bigalloc file systems, we never free a partial cluster
2610          * at the beginning of the extent.  Instead, we check to see if we
2611          * need to free it on a subsequent call to ext4_remove_blocks,
2612          * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2613          */
2614         flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2615         ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2616
2617         /* reset the partial cluster if we've freed past it */
2618         if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2619                 partial->state = initial;
2620
2621         /*
2622          * If we've freed the entire extent but the beginning is not left
2623          * cluster aligned and is not marked as ineligible for freeing we
2624          * record the partial cluster at the beginning of the extent.  It
2625          * wasn't freed by the preceding ext4_free_blocks() call, and we
2626          * need to look farther to the left to determine if it's to be freed
2627          * (not shared with another extent). Else, reset the partial
2628          * cluster - we're either  done freeing or the beginning of the
2629          * extent is left cluster aligned.
2630          */
2631         if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2632                 if (partial->state == initial) {
2633                         partial->pclu = EXT4_B2C(sbi, pblk);
2634                         partial->lblk = from;
2635                         partial->state = tofree;
2636                 }
2637         } else {
2638                 partial->state = initial;
2639         }
2640
2641         return 0;
2642 }
2643
2644 /*
2645  * ext4_ext_rm_leaf() Removes the extents associated with the
2646  * blocks appearing between "start" and "end".  Both "start"
2647  * and "end" must appear in the same extent or EIO is returned.
2648  *
2649  * @handle: The journal handle
2650  * @inode:  The files inode
2651  * @path:   The path to the leaf
2652  * @partial_cluster: The cluster which we'll have to free if all extents
2653  *                   has been released from it.  However, if this value is
2654  *                   negative, it's a cluster just to the right of the
2655  *                   punched region and it must not be freed.
2656  * @start:  The first block to remove
2657  * @end:   The last block to remove
2658  */
2659 static int
2660 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2661                  struct ext4_ext_path *path,
2662                  struct partial_cluster *partial,
2663                  ext4_lblk_t start, ext4_lblk_t end)
2664 {
2665         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2666         int err = 0, correct_index = 0;
2667         int depth = ext_depth(inode), credits;
2668         struct ext4_extent_header *eh;
2669         ext4_lblk_t a, b;
2670         unsigned num;
2671         ext4_lblk_t ex_ee_block;
2672         unsigned short ex_ee_len;
2673         unsigned unwritten = 0;
2674         struct ext4_extent *ex;
2675         ext4_fsblk_t pblk;
2676
2677         /* the header must be checked already in ext4_ext_remove_space() */
2678         ext_debug("truncate since %u in leaf to %u\n", start, end);
2679         if (!path[depth].p_hdr)
2680                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2681         eh = path[depth].p_hdr;
2682         if (unlikely(path[depth].p_hdr == NULL)) {
2683                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2684                 return -EFSCORRUPTED;
2685         }
2686         /* find where to start removing */
2687         ex = path[depth].p_ext;
2688         if (!ex)
2689                 ex = EXT_LAST_EXTENT(eh);
2690
2691         ex_ee_block = le32_to_cpu(ex->ee_block);
2692         ex_ee_len = ext4_ext_get_actual_len(ex);
2693
2694         trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2695
2696         while (ex >= EXT_FIRST_EXTENT(eh) &&
2697                         ex_ee_block + ex_ee_len > start) {
2698
2699                 if (ext4_ext_is_unwritten(ex))
2700                         unwritten = 1;
2701                 else
2702                         unwritten = 0;
2703
2704                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2705                           unwritten, ex_ee_len);
2706                 path[depth].p_ext = ex;
2707
2708                 a = ex_ee_block > start ? ex_ee_block : start;
2709                 b = ex_ee_block+ex_ee_len - 1 < end ?
2710                         ex_ee_block+ex_ee_len - 1 : end;
2711
2712                 ext_debug("  border %u:%u\n", a, b);
2713
2714                 /* If this extent is beyond the end of the hole, skip it */
2715                 if (end < ex_ee_block) {
2716                         /*
2717                          * We're going to skip this extent and move to another,
2718                          * so note that its first cluster is in use to avoid
2719                          * freeing it when removing blocks.  Eventually, the
2720                          * right edge of the truncated/punched region will
2721                          * be just to the left.
2722                          */
2723                         if (sbi->s_cluster_ratio > 1) {
2724                                 pblk = ext4_ext_pblock(ex);
2725                                 partial->pclu = EXT4_B2C(sbi, pblk);
2726                                 partial->state = nofree;
2727                         }
2728                         ex--;
2729                         ex_ee_block = le32_to_cpu(ex->ee_block);
2730                         ex_ee_len = ext4_ext_get_actual_len(ex);
2731                         continue;
2732                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2733                         EXT4_ERROR_INODE(inode,
2734                                          "can not handle truncate %u:%u "
2735                                          "on extent %u:%u",
2736                                          start, end, ex_ee_block,
2737                                          ex_ee_block + ex_ee_len - 1);
2738                         err = -EFSCORRUPTED;
2739                         goto out;
2740                 } else if (a != ex_ee_block) {
2741                         /* remove tail of the extent */
2742                         num = a - ex_ee_block;
2743                 } else {
2744                         /* remove whole extent: excellent! */
2745                         num = 0;
2746                 }
2747                 /*
2748                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2749                  * descriptor) for each block group; assume two block
2750                  * groups plus ex_ee_len/blocks_per_block_group for
2751                  * the worst case
2752                  */
2753                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2754                 if (ex == EXT_FIRST_EXTENT(eh)) {
2755                         correct_index = 1;
2756                         credits += (ext_depth(inode)) + 1;
2757                 }
2758                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2759
2760                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2761                 if (err)
2762                         goto out;
2763
2764                 err = ext4_ext_get_access(handle, inode, path + depth);
2765                 if (err)
2766                         goto out;
2767
2768                 err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2769                 if (err)
2770                         goto out;
2771
2772                 if (num == 0)
2773                         /* this extent is removed; mark slot entirely unused */
2774                         ext4_ext_store_pblock(ex, 0);
2775
2776                 ex->ee_len = cpu_to_le16(num);
2777                 /*
2778                  * Do not mark unwritten if all the blocks in the
2779                  * extent have been removed.
2780                  */
2781                 if (unwritten && num)
2782                         ext4_ext_mark_unwritten(ex);
2783                 /*
2784                  * If the extent was completely released,
2785                  * we need to remove it from the leaf
2786                  */
2787                 if (num == 0) {
2788                         if (end != EXT_MAX_BLOCKS - 1) {
2789                                 /*
2790                                  * For hole punching, we need to scoot all the
2791                                  * extents up when an extent is removed so that
2792                                  * we dont have blank extents in the middle
2793                                  */
2794                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2795                                         sizeof(struct ext4_extent));
2796
2797                                 /* Now get rid of the one at the end */
2798                                 memset(EXT_LAST_EXTENT(eh), 0,
2799                                         sizeof(struct ext4_extent));
2800                         }
2801                         le16_add_cpu(&eh->eh_entries, -1);
2802                 }
2803
2804                 err = ext4_ext_dirty(handle, inode, path + depth);
2805                 if (err)
2806                         goto out;
2807
2808                 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2809                                 ext4_ext_pblock(ex));
2810                 ex--;
2811                 ex_ee_block = le32_to_cpu(ex->ee_block);
2812                 ex_ee_len = ext4_ext_get_actual_len(ex);
2813         }
2814
2815         if (correct_index && eh->eh_entries)
2816                 err = ext4_ext_correct_indexes(handle, inode, path);
2817
2818         /*
2819          * If there's a partial cluster and at least one extent remains in
2820          * the leaf, free the partial cluster if it isn't shared with the
2821          * current extent.  If it is shared with the current extent
2822          * we reset the partial cluster because we've reached the start of the
2823          * truncated/punched region and we're done removing blocks.
2824          */
2825         if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2826                 pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2827                 if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2828                         int flags = get_default_free_blocks_flags(inode);
2829
2830                         if (ext4_is_pending(inode, partial->lblk))
2831                                 flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2832                         ext4_free_blocks(handle, inode, NULL,
2833                                          EXT4_C2B(sbi, partial->pclu),
2834                                          sbi->s_cluster_ratio, flags);
2835                         if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2836                                 ext4_rereserve_cluster(inode, partial->lblk);
2837                 }
2838                 partial->state = initial;
2839         }
2840
2841         /* if this leaf is free, then we should
2842          * remove it from index block above */
2843         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2844                 err = ext4_ext_rm_idx(handle, inode, path, depth);
2845
2846 out:
2847         return err;
2848 }
2849
2850 /*
2851  * ext4_ext_more_to_rm:
2852  * returns 1 if current index has to be freed (even partial)
2853  */
2854 static int
2855 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2856 {
2857         BUG_ON(path->p_idx == NULL);
2858
2859         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2860                 return 0;
2861
2862         /*
2863          * if truncate on deeper level happened, it wasn't partial,
2864          * so we have to consider current index for truncation
2865          */
2866         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2867                 return 0;
2868         return 1;
2869 }
2870
2871 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2872                           ext4_lblk_t end)
2873 {
2874         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2875         int depth = ext_depth(inode);
2876         struct ext4_ext_path *path = NULL;
2877         struct partial_cluster partial;
2878         handle_t *handle;
2879         int i = 0, err = 0;
2880
2881         partial.pclu = 0;
2882         partial.lblk = 0;
2883         partial.state = initial;
2884
2885         ext_debug("truncate since %u to %u\n", start, end);
2886
2887         /* probably first extent we're gonna free will be last in block */
2888         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1);
2889         if (IS_ERR(handle))
2890                 return PTR_ERR(handle);
2891
2892 again:
2893         trace_ext4_ext_remove_space(inode, start, end, depth);
2894
2895         /*
2896          * Check if we are removing extents inside the extent tree. If that
2897          * is the case, we are going to punch a hole inside the extent tree
2898          * so we have to check whether we need to split the extent covering
2899          * the last block to remove so we can easily remove the part of it
2900          * in ext4_ext_rm_leaf().
2901          */
2902         if (end < EXT_MAX_BLOCKS - 1) {
2903                 struct ext4_extent *ex;
2904                 ext4_lblk_t ee_block, ex_end, lblk;
2905                 ext4_fsblk_t pblk;
2906
2907                 /* find extent for or closest extent to this block */
2908                 path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2909                 if (IS_ERR(path)) {
2910                         ext4_journal_stop(handle);
2911                         return PTR_ERR(path);
2912                 }
2913                 depth = ext_depth(inode);
2914                 /* Leaf not may not exist only if inode has no blocks at all */
2915                 ex = path[depth].p_ext;
2916                 if (!ex) {
2917                         if (depth) {
2918                                 EXT4_ERROR_INODE(inode,
2919                                                  "path[%d].p_hdr == NULL",
2920                                                  depth);
2921                                 err = -EFSCORRUPTED;
2922                         }
2923                         goto out;
2924                 }
2925
2926                 ee_block = le32_to_cpu(ex->ee_block);
2927                 ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2928
2929                 /*
2930                  * See if the last block is inside the extent, if so split
2931                  * the extent at 'end' block so we can easily remove the
2932                  * tail of the first part of the split extent in
2933                  * ext4_ext_rm_leaf().
2934                  */
2935                 if (end >= ee_block && end < ex_end) {
2936
2937                         /*
2938                          * If we're going to split the extent, note that
2939                          * the cluster containing the block after 'end' is
2940                          * in use to avoid freeing it when removing blocks.
2941                          */
2942                         if (sbi->s_cluster_ratio > 1) {
2943                                 pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2944                                 partial.pclu = EXT4_B2C(sbi, pblk);
2945                                 partial.state = nofree;
2946                         }
2947
2948                         /*
2949                          * Split the extent in two so that 'end' is the last
2950                          * block in the first new extent. Also we should not
2951                          * fail removing space due to ENOSPC so try to use
2952                          * reserved block if that happens.
2953                          */
2954                         err = ext4_force_split_extent_at(handle, inode, &path,
2955                                                          end + 1, 1);
2956                         if (err < 0)
2957                                 goto out;
2958
2959                 } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2960                            partial.state == initial) {
2961                         /*
2962                          * If we're punching, there's an extent to the right.
2963                          * If the partial cluster hasn't been set, set it to
2964                          * that extent's first cluster and its state to nofree
2965                          * so it won't be freed should it contain blocks to be
2966                          * removed. If it's already set (tofree/nofree), we're
2967                          * retrying and keep the original partial cluster info
2968                          * so a cluster marked tofree as a result of earlier
2969                          * extent removal is not lost.
2970                          */
2971                         lblk = ex_end + 1;
2972                         err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2973                                                     &ex);
2974                         if (err)
2975                                 goto out;
2976                         if (pblk) {
2977                                 partial.pclu = EXT4_B2C(sbi, pblk);
2978                                 partial.state = nofree;
2979                         }
2980                 }
2981         }
2982         /*
2983          * We start scanning from right side, freeing all the blocks
2984          * after i_size and walking into the tree depth-wise.
2985          */
2986         depth = ext_depth(inode);
2987         if (path) {
2988                 int k = i = depth;
2989                 while (--k > 0)
2990                         path[k].p_block =
2991                                 le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2992         } else {
2993                 path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2994                                GFP_NOFS);
2995                 if (path == NULL) {
2996                         ext4_journal_stop(handle);
2997                         return -ENOMEM;
2998                 }
2999                 path[0].p_maxdepth = path[0].p_depth = depth;
3000                 path[0].p_hdr = ext_inode_hdr(inode);
3001                 i = 0;
3002
3003                 if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3004                         err = -EFSCORRUPTED;
3005                         goto out;
3006                 }
3007         }
3008         err = 0;
3009
3010         while (i >= 0 && err == 0) {
3011                 if (i == depth) {
3012                         /* this is leaf block */
3013                         err = ext4_ext_rm_leaf(handle, inode, path,
3014                                                &partial, start, end);
3015                         /* root level has p_bh == NULL, brelse() eats this */
3016                         brelse(path[i].p_bh);
3017                         path[i].p_bh = NULL;
3018                         i--;
3019                         continue;
3020                 }
3021
3022                 /* this is index block */
3023                 if (!path[i].p_hdr) {
3024                         ext_debug("initialize header\n");
3025                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3026                 }
3027
3028                 if (!path[i].p_idx) {
3029                         /* this level hasn't been touched yet */
3030                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3031                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3032                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
3033                                   path[i].p_hdr,
3034                                   le16_to_cpu(path[i].p_hdr->eh_entries));
3035                 } else {
3036                         /* we were already here, see at next index */
3037                         path[i].p_idx--;
3038                 }
3039
3040                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3041                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
3042                                 path[i].p_idx);
3043                 if (ext4_ext_more_to_rm(path + i)) {
3044                         struct buffer_head *bh;
3045                         /* go to the next level */
3046                         ext_debug("move to level %d (block %llu)\n",
3047                                   i + 1, ext4_idx_pblock(path[i].p_idx));
3048                         memset(path + i + 1, 0, sizeof(*path));
3049                         bh = read_extent_tree_block(inode,
3050                                 ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3051                                 EXT4_EX_NOCACHE);
3052                         if (IS_ERR(bh)) {
3053                                 /* should we reset i_size? */
3054                                 err = PTR_ERR(bh);
3055                                 break;
3056                         }
3057                         /* Yield here to deal with large extent trees.
3058                          * Should be a no-op if we did IO above. */
3059                         cond_resched();
3060                         if (WARN_ON(i + 1 > depth)) {
3061                                 err = -EFSCORRUPTED;
3062                                 break;
3063                         }
3064                         path[i + 1].p_bh = bh;
3065
3066                         /* save actual number of indexes since this
3067                          * number is changed at the next iteration */
3068                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3069                         i++;
3070                 } else {
3071                         /* we finished processing this index, go up */
3072                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3073                                 /* index is empty, remove it;
3074                                  * handle must be already prepared by the
3075                                  * truncatei_leaf() */
3076                                 err = ext4_ext_rm_idx(handle, inode, path, i);
3077                         }
3078                         /* root level has p_bh == NULL, brelse() eats this */
3079                         brelse(path[i].p_bh);
3080                         path[i].p_bh = NULL;
3081                         i--;
3082                         ext_debug("return to level %d\n", i);
3083                 }
3084         }
3085
3086         trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3087                                          path->p_hdr->eh_entries);
3088
3089         /*
3090          * if there's a partial cluster and we have removed the first extent
3091          * in the file, then we also free the partial cluster, if any
3092          */
3093         if (partial.state == tofree && err == 0) {
3094                 int flags = get_default_free_blocks_flags(inode);
3095
3096                 if (ext4_is_pending(inode, partial.lblk))
3097                         flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3098                 ext4_free_blocks(handle, inode, NULL,
3099                                  EXT4_C2B(sbi, partial.pclu),
3100                                  sbi->s_cluster_ratio, flags);
3101                 if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3102                         ext4_rereserve_cluster(inode, partial.lblk);
3103                 partial.state = initial;
3104         }
3105
3106         /* TODO: flexible tree reduction should be here */
3107         if (path->p_hdr->eh_entries == 0) {
3108                 /*
3109                  * truncate to zero freed all the tree,
3110                  * so we need to correct eh_depth
3111                  */
3112                 err = ext4_ext_get_access(handle, inode, path);
3113                 if (err == 0) {
3114                         ext_inode_hdr(inode)->eh_depth = 0;
3115                         ext_inode_hdr(inode)->eh_max =
3116                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
3117                         err = ext4_ext_dirty(handle, inode, path);
3118                 }
3119         }
3120 out:
3121         ext4_ext_drop_refs(path);
3122         kfree(path);
3123         path = NULL;
3124         if (err == -EAGAIN)
3125                 goto again;
3126         ext4_journal_stop(handle);
3127
3128         return err;
3129 }
3130
3131 /*
3132  * called at mount time
3133  */
3134 void ext4_ext_init(struct super_block *sb)
3135 {
3136         /*
3137          * possible initialization would be here
3138          */
3139
3140         if (ext4_has_feature_extents(sb)) {
3141 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3142                 printk(KERN_INFO "EXT4-fs: file extents enabled"
3143 #ifdef AGGRESSIVE_TEST
3144                        ", aggressive tests"
3145 #endif
3146 #ifdef CHECK_BINSEARCH
3147                        ", check binsearch"
3148 #endif
3149 #ifdef EXTENTS_STATS
3150                        ", stats"
3151 #endif
3152                        "\n");
3153 #endif
3154 #ifdef EXTENTS_STATS
3155                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3156                 EXT4_SB(sb)->s_ext_min = 1 << 30;
3157                 EXT4_SB(sb)->s_ext_max = 0;
3158 #endif
3159         }
3160 }
3161
3162 /*
3163  * called at umount time
3164  */
3165 void ext4_ext_release(struct super_block *sb)
3166 {
3167         if (!ext4_has_feature_extents(sb))
3168                 return;
3169
3170 #ifdef EXTENTS_STATS
3171         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3172                 struct ext4_sb_info *sbi = EXT4_SB(sb);
3173                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3174                         sbi->s_ext_blocks, sbi->s_ext_extents,
3175                         sbi->s_ext_blocks / sbi->s_ext_extents);
3176                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3177                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3178         }
3179 #endif
3180 }
3181
3182 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3183 {
3184         ext4_lblk_t  ee_block;
3185         ext4_fsblk_t ee_pblock;
3186         unsigned int ee_len;
3187
3188         ee_block  = le32_to_cpu(ex->ee_block);
3189         ee_len    = ext4_ext_get_actual_len(ex);
3190         ee_pblock = ext4_ext_pblock(ex);
3191
3192         if (ee_len == 0)
3193                 return 0;
3194
3195         return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3196                                      EXTENT_STATUS_WRITTEN);
3197 }
3198
3199 /* FIXME!! we need to try to merge to left or right after zero-out  */
3200 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3201 {
3202         ext4_fsblk_t ee_pblock;
3203         unsigned int ee_len;
3204
3205         ee_len    = ext4_ext_get_actual_len(ex);
3206         ee_pblock = ext4_ext_pblock(ex);
3207         return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3208                                   ee_len);
3209 }
3210
3211 /*
3212  * ext4_split_extent_at() splits an extent at given block.
3213  *
3214  * @handle: the journal handle
3215  * @inode: the file inode
3216  * @path: the path to the extent
3217  * @split: the logical block where the extent is splitted.
3218  * @split_flags: indicates if the extent could be zeroout if split fails, and
3219  *               the states(init or unwritten) of new extents.
3220  * @flags: flags used to insert new extent to extent tree.
3221  *
3222  *
3223  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3224  * of which are deterimined by split_flag.
3225  *
3226  * There are two cases:
3227  *  a> the extent are splitted into two extent.
3228  *  b> split is not needed, and just mark the extent.
3229  *
3230  * return 0 on success.
3231  */
3232 static int ext4_split_extent_at(handle_t *handle,
3233                              struct inode *inode,
3234                              struct ext4_ext_path **ppath,
3235                              ext4_lblk_t split,
3236                              int split_flag,
3237                              int flags)
3238 {
3239         struct ext4_ext_path *path = *ppath;
3240         ext4_fsblk_t newblock;
3241         ext4_lblk_t ee_block;
3242         struct ext4_extent *ex, newex, orig_ex, zero_ex;
3243         struct ext4_extent *ex2 = NULL;
3244         unsigned int ee_len, depth;
3245         int err = 0;
3246
3247         BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3248                (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3249
3250         ext_debug("ext4_split_extents_at: inode %lu, logical"
3251                 "block %llu\n", inode->i_ino, (unsigned long long)split);
3252
3253         ext4_ext_show_leaf(inode, path);
3254
3255         depth = ext_depth(inode);
3256         ex = path[depth].p_ext;
3257         ee_block = le32_to_cpu(ex->ee_block);
3258         ee_len = ext4_ext_get_actual_len(ex);
3259         newblock = split - ee_block + ext4_ext_pblock(ex);
3260
3261         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3262         BUG_ON(!ext4_ext_is_unwritten(ex) &&
3263                split_flag & (EXT4_EXT_MAY_ZEROOUT |
3264                              EXT4_EXT_MARK_UNWRIT1 |
3265                              EXT4_EXT_MARK_UNWRIT2));
3266
3267         err = ext4_ext_get_access(handle, inode, path + depth);
3268         if (err)
3269                 goto out;
3270
3271         if (split == ee_block) {
3272                 /*
3273                  * case b: block @split is the block that the extent begins with
3274                  * then we just change the state of the extent, and splitting
3275                  * is not needed.
3276                  */
3277                 if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3278                         ext4_ext_mark_unwritten(ex);
3279                 else
3280                         ext4_ext_mark_initialized(ex);
3281
3282                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3283                         ext4_ext_try_to_merge(handle, inode, path, ex);
3284
3285                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3286                 goto out;
3287         }
3288
3289         /* case a */
3290         memcpy(&orig_ex, ex, sizeof(orig_ex));
3291         ex->ee_len = cpu_to_le16(split - ee_block);
3292         if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3293                 ext4_ext_mark_unwritten(ex);
3294
3295         /*
3296          * path may lead to new leaf, not to original leaf any more
3297          * after ext4_ext_insert_extent() returns,
3298          */
3299         err = ext4_ext_dirty(handle, inode, path + depth);
3300         if (err)
3301                 goto fix_extent_len;
3302
3303         ex2 = &newex;
3304         ex2->ee_block = cpu_to_le32(split);
3305         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3306         ext4_ext_store_pblock(ex2, newblock);
3307         if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3308                 ext4_ext_mark_unwritten(ex2);
3309
3310         err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3311         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3312                 if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3313                         if (split_flag & EXT4_EXT_DATA_VALID1) {
3314                                 err = ext4_ext_zeroout(inode, ex2);
3315                                 zero_ex.ee_block = ex2->ee_block;
3316                                 zero_ex.ee_len = cpu_to_le16(
3317                                                 ext4_ext_get_actual_len(ex2));
3318                                 ext4_ext_store_pblock(&zero_ex,
3319                                                       ext4_ext_pblock(ex2));
3320                         } else {
3321                                 err = ext4_ext_zeroout(inode, ex);
3322                                 zero_ex.ee_block = ex->ee_block;
3323                                 zero_ex.ee_len = cpu_to_le16(
3324                                                 ext4_ext_get_actual_len(ex));
3325                                 ext4_ext_store_pblock(&zero_ex,
3326                                                       ext4_ext_pblock(ex));
3327                         }
3328                 } else {
3329                         err = ext4_ext_zeroout(inode, &orig_ex);
3330                         zero_ex.ee_block = orig_ex.ee_block;
3331                         zero_ex.ee_len = cpu_to_le16(
3332                                                 ext4_ext_get_actual_len(&orig_ex));
3333                         ext4_ext_store_pblock(&zero_ex,
3334                                               ext4_ext_pblock(&orig_ex));
3335                 }
3336
3337                 if (err)
3338                         goto fix_extent_len;
3339                 /* update the extent length and mark as initialized */
3340                 ex->ee_len = cpu_to_le16(ee_len);
3341                 ext4_ext_try_to_merge(handle, inode, path, ex);
3342                 err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3343                 if (err)
3344                         goto fix_extent_len;
3345
3346                 /* update extent status tree */
3347                 err = ext4_zeroout_es(inode, &zero_ex);
3348
3349                 goto out;
3350         } else if (err)
3351                 goto fix_extent_len;
3352
3353 out:
3354         ext4_ext_show_leaf(inode, path);
3355         return err;
3356
3357 fix_extent_len:
3358         ex->ee_len = orig_ex.ee_len;
3359         ext4_ext_dirty(handle, inode, path + path->p_depth);
3360         return err;
3361 }
3362
3363 /*
3364  * ext4_split_extents() splits an extent and mark extent which is covered
3365  * by @map as split_flags indicates
3366  *
3367  * It may result in splitting the extent into multiple extents (up to three)
3368  * There are three possibilities:
3369  *   a> There is no split required
3370  *   b> Splits in two extents: Split is happening at either end of the extent
3371  *   c> Splits in three extents: Somone is splitting in middle of the extent
3372  *
3373  */
3374 static int ext4_split_extent(handle_t *handle,
3375                               struct inode *inode,
3376                               struct ext4_ext_path **ppath,
3377                               struct ext4_map_blocks *map,
3378                               int split_flag,
3379                               int flags)
3380 {
3381         struct ext4_ext_path *path = *ppath;
3382         ext4_lblk_t ee_block;
3383         struct ext4_extent *ex;
3384         unsigned int ee_len, depth;
3385         int err = 0;
3386         int unwritten;
3387         int split_flag1, flags1;
3388         int allocated = map->m_len;
3389
3390         depth = ext_depth(inode);
3391         ex = path[depth].p_ext;
3392         ee_block = le32_to_cpu(ex->ee_block);
3393         ee_len = ext4_ext_get_actual_len(ex);
3394         unwritten = ext4_ext_is_unwritten(ex);
3395
3396         if (map->m_lblk + map->m_len < ee_block + ee_len) {
3397                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3398                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3399                 if (unwritten)
3400                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3401                                        EXT4_EXT_MARK_UNWRIT2;
3402                 if (split_flag & EXT4_EXT_DATA_VALID2)
3403                         split_flag1 |= EXT4_EXT_DATA_VALID1;
3404                 err = ext4_split_extent_at(handle, inode, ppath,
3405                                 map->m_lblk + map->m_len, split_flag1, flags1);
3406                 if (err)
3407                         goto out;
3408         } else {
3409                 allocated = ee_len - (map->m_lblk - ee_block);
3410         }
3411         /*
3412          * Update path is required because previous ext4_split_extent_at() may
3413          * result in split of original leaf or extent zeroout.
3414          */
3415         path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3416         if (IS_ERR(path))
3417                 return PTR_ERR(path);
3418         depth = ext_depth(inode);
3419         ex = path[depth].p_ext;
3420         if (!ex) {
3421                 EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3422                                  (unsigned long) map->m_lblk);
3423                 return -EFSCORRUPTED;
3424         }
3425         unwritten = ext4_ext_is_unwritten(ex);
3426         split_flag1 = 0;
3427
3428         if (map->m_lblk >= ee_block) {
3429                 split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3430                 if (unwritten) {
3431                         split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3432                         split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3433                                                      EXT4_EXT_MARK_UNWRIT2);
3434                 }
3435                 err = ext4_split_extent_at(handle, inode, ppath,
3436                                 map->m_lblk, split_flag1, flags);
3437                 if (err)
3438                         goto out;
3439         }
3440
3441         ext4_ext_show_leaf(inode, path);
3442 out:
3443         return err ? err : allocated;
3444 }
3445
3446 /*
3447  * This function is called by ext4_ext_map_blocks() if someone tries to write
3448  * to an unwritten extent. It may result in splitting the unwritten
3449  * extent into multiple extents (up to three - one initialized and two
3450  * unwritten).
3451  * There are three possibilities:
3452  *   a> There is no split required: Entire extent should be initialized
3453  *   b> Splits in two extents: Write is happening at either end of the extent
3454  *   c> Splits in three extents: Somone is writing in middle of the extent
3455  *
3456  * Pre-conditions:
3457  *  - The extent pointed to by 'path' is unwritten.
3458  *  - The extent pointed to by 'path' contains a superset
3459  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3460  *
3461  * Post-conditions on success:
3462  *  - the returned value is the number of blocks beyond map->l_lblk
3463  *    that are allocated and initialized.
3464  *    It is guaranteed to be >= map->m_len.
3465  */
3466 static int ext4_ext_convert_to_initialized(handle_t *handle,
3467                                            struct inode *inode,
3468                                            struct ext4_map_blocks *map,
3469                                            struct ext4_ext_path **ppath,
3470                                            int flags)
3471 {
3472         struct ext4_ext_path *path = *ppath;
3473         struct ext4_sb_info *sbi;
3474         struct ext4_extent_header *eh;
3475         struct ext4_map_blocks split_map;
3476         struct ext4_extent zero_ex1, zero_ex2;
3477         struct ext4_extent *ex, *abut_ex;
3478         ext4_lblk_t ee_block, eof_block;
3479         unsigned int ee_len, depth, map_len = map->m_len;
3480         int allocated = 0, max_zeroout = 0;
3481         int err = 0;
3482         int split_flag = EXT4_EXT_DATA_VALID2;
3483
3484         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3485                 "block %llu, max_blocks %u\n", inode->i_ino,
3486                 (unsigned long long)map->m_lblk, map_len);
3487
3488         sbi = EXT4_SB(inode->i_sb);
3489         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3490                 inode->i_sb->s_blocksize_bits;
3491         if (eof_block < map->m_lblk + map_len)
3492                 eof_block = map->m_lblk + map_len;
3493
3494         depth = ext_depth(inode);
3495         eh = path[depth].p_hdr;
3496         ex = path[depth].p_ext;
3497         ee_block = le32_to_cpu(ex->ee_block);
3498         ee_len = ext4_ext_get_actual_len(ex);
3499         zero_ex1.ee_len = 0;
3500         zero_ex2.ee_len = 0;
3501
3502         trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3503
3504         /* Pre-conditions */
3505         BUG_ON(!ext4_ext_is_unwritten(ex));
3506         BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3507
3508         /*
3509          * Attempt to transfer newly initialized blocks from the currently
3510          * unwritten extent to its neighbor. This is much cheaper
3511          * than an insertion followed by a merge as those involve costly
3512          * memmove() calls. Transferring to the left is the common case in
3513          * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3514          * followed by append writes.
3515          *
3516          * Limitations of the current logic:
3517          *  - L1: we do not deal with writes covering the whole extent.
3518          *    This would require removing the extent if the transfer
3519          *    is possible.
3520          *  - L2: we only attempt to merge with an extent stored in the
3521          *    same extent tree node.
3522          */
3523         if ((map->m_lblk == ee_block) &&
3524                 /* See if we can merge left */
3525                 (map_len < ee_len) &&           /*L1*/
3526                 (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3527                 ext4_lblk_t prev_lblk;
3528                 ext4_fsblk_t prev_pblk, ee_pblk;
3529                 unsigned int prev_len;
3530
3531                 abut_ex = ex - 1;
3532                 prev_lblk = le32_to_cpu(abut_ex->ee_block);
3533                 prev_len = ext4_ext_get_actual_len(abut_ex);
3534                 prev_pblk = ext4_ext_pblock(abut_ex);
3535                 ee_pblk = ext4_ext_pblock(ex);
3536
3537                 /*
3538                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3539                  * upon those conditions:
3540                  * - C1: abut_ex is initialized,
3541                  * - C2: abut_ex is logically abutting ex,
3542                  * - C3: abut_ex is physically abutting ex,
3543                  * - C4: abut_ex can receive the additional blocks without
3544                  *   overflowing the (initialized) length limit.
3545                  */
3546                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3547                         ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3548                         ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3549                         (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3550                         err = ext4_ext_get_access(handle, inode, path + depth);
3551                         if (err)
3552                                 goto out;
3553
3554                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3555                                 map, ex, abut_ex);
3556
3557                         /* Shift the start of ex by 'map_len' blocks */
3558                         ex->ee_block = cpu_to_le32(ee_block + map_len);
3559                         ext4_ext_store_pblock(ex, ee_pblk + map_len);
3560                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3561                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3562
3563                         /* Extend abut_ex by 'map_len' blocks */
3564                         abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3565
3566                         /* Result: number of initialized blocks past m_lblk */
3567                         allocated = map_len;
3568                 }
3569         } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3570                    (map_len < ee_len) &&        /*L1*/
3571                    ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3572                 /* See if we can merge right */
3573                 ext4_lblk_t next_lblk;
3574                 ext4_fsblk_t next_pblk, ee_pblk;
3575                 unsigned int next_len;
3576
3577                 abut_ex = ex + 1;
3578                 next_lblk = le32_to_cpu(abut_ex->ee_block);
3579                 next_len = ext4_ext_get_actual_len(abut_ex);
3580                 next_pblk = ext4_ext_pblock(abut_ex);
3581                 ee_pblk = ext4_ext_pblock(ex);
3582
3583                 /*
3584                  * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3585                  * upon those conditions:
3586                  * - C1: abut_ex is initialized,
3587                  * - C2: abut_ex is logically abutting ex,
3588                  * - C3: abut_ex is physically abutting ex,
3589                  * - C4: abut_ex can receive the additional blocks without
3590                  *   overflowing the (initialized) length limit.
3591                  */
3592                 if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3593                     ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3594                     ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3595                     (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3596                         err = ext4_ext_get_access(handle, inode, path + depth);
3597                         if (err)
3598                                 goto out;
3599
3600                         trace_ext4_ext_convert_to_initialized_fastpath(inode,
3601                                 map, ex, abut_ex);
3602
3603                         /* Shift the start of abut_ex by 'map_len' blocks */
3604                         abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3605                         ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3606                         ex->ee_len = cpu_to_le16(ee_len - map_len);
3607                         ext4_ext_mark_unwritten(ex); /* Restore the flag */
3608
3609                         /* Extend abut_ex by 'map_len' blocks */
3610                         abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3611
3612                         /* Result: number of initialized blocks past m_lblk */
3613                         allocated = map_len;
3614                 }
3615         }
3616         if (allocated) {
3617                 /* Mark the block containing both extents as dirty */
3618                 ext4_ext_dirty(handle, inode, path + depth);
3619
3620                 /* Update path to point to the right extent */
3621                 path[depth].p_ext = abut_ex;
3622                 goto out;
3623         } else
3624                 allocated = ee_len - (map->m_lblk - ee_block);
3625
3626         WARN_ON(map->m_lblk < ee_block);
3627         /*
3628          * It is safe to convert extent to initialized via explicit
3629          * zeroout only if extent is fully inside i_size or new_size.
3630          */
3631         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3632
3633         if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3634                 max_zeroout = sbi->s_extent_max_zeroout_kb >>
3635                         (inode->i_sb->s_blocksize_bits - 10);
3636
3637         if (IS_ENCRYPTED(inode))
3638                 max_zeroout = 0;
3639
3640         /*
3641          * five cases:
3642          * 1. split the extent into three extents.
3643          * 2. split the extent into two extents, zeroout the head of the first
3644          *    extent.
3645          * 3. split the extent into two extents, zeroout the tail of the second
3646          *    extent.
3647          * 4. split the extent into two extents with out zeroout.
3648          * 5. no splitting needed, just possibly zeroout the head and / or the
3649          *    tail of the extent.
3650          */
3651         split_map.m_lblk = map->m_lblk;
3652         split_map.m_len = map->m_len;
3653
3654         if (max_zeroout && (allocated > split_map.m_len)) {
3655                 if (allocated <= max_zeroout) {
3656                         /* case 3 or 5 */
3657                         zero_ex1.ee_block =
3658                                  cpu_to_le32(split_map.m_lblk +
3659                                              split_map.m_len);
3660                         zero_ex1.ee_len =
3661                                 cpu_to_le16(allocated - split_map.m_len);
3662                         ext4_ext_store_pblock(&zero_ex1,
3663                                 ext4_ext_pblock(ex) + split_map.m_lblk +
3664                                 split_map.m_len - ee_block);
3665                         err = ext4_ext_zeroout(inode, &zero_ex1);
3666                         if (err)
3667                                 goto out;
3668                         split_map.m_len = allocated;
3669                 }
3670                 if (split_map.m_lblk - ee_block + split_map.m_len <
3671                                                                 max_zeroout) {
3672                         /* case 2 or 5 */
3673                         if (split_map.m_lblk != ee_block) {
3674                                 zero_ex2.ee_block = ex->ee_block;
3675                                 zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3676                                                         ee_block);
3677                                 ext4_ext_store_pblock(&zero_ex2,
3678                                                       ext4_ext_pblock(ex));
3679                                 err = ext4_ext_zeroout(inode, &zero_ex2);
3680                                 if (err)
3681                                         goto out;
3682                         }
3683
3684                         split_map.m_len += split_map.m_lblk - ee_block;
3685                         split_map.m_lblk = ee_block;
3686                         allocated = map->m_len;
3687                 }
3688         }
3689
3690         err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3691                                 flags);
3692         if (err > 0)
3693                 err = 0;
3694 out:
3695         /* If we have gotten a failure, don't zero out status tree */
3696         if (!err) {
3697                 err = ext4_zeroout_es(inode, &zero_ex1);
3698                 if (!err)
3699                         err = ext4_zeroout_es(inode, &zero_ex2);
3700         }
3701         return err ? err : allocated;
3702 }
3703
3704 /*
3705  * This function is called by ext4_ext_map_blocks() from
3706  * ext4_get_blocks_dio_write() when DIO to write
3707  * to an unwritten extent.
3708  *
3709  * Writing to an unwritten extent may result in splitting the unwritten
3710  * extent into multiple initialized/unwritten extents (up to three)
3711  * There are three possibilities:
3712  *   a> There is no split required: Entire extent should be unwritten
3713  *   b> Splits in two extents: Write is happening at either end of the extent
3714  *   c> Splits in three extents: Somone is writing in middle of the extent
3715  *
3716  * This works the same way in the case of initialized -> unwritten conversion.
3717  *
3718  * One of more index blocks maybe needed if the extent tree grow after
3719  * the unwritten extent split. To prevent ENOSPC occur at the IO
3720  * complete, we need to split the unwritten extent before DIO submit
3721  * the IO. The unwritten extent called at this time will be split
3722  * into three unwritten extent(at most). After IO complete, the part
3723  * being filled will be convert to initialized by the end_io callback function
3724  * via ext4_convert_unwritten_extents().
3725  *
3726  * Returns the size of unwritten extent to be written on success.
3727  */
3728 static int ext4_split_convert_extents(handle_t *handle,
3729                                         struct inode *inode,
3730                                         struct ext4_map_blocks *map,
3731                                         struct ext4_ext_path **ppath,
3732                                         int flags)
3733 {
3734         struct ext4_ext_path *path = *ppath;
3735         ext4_lblk_t eof_block;
3736         ext4_lblk_t ee_block;
3737         struct ext4_extent *ex;
3738         unsigned int ee_len;
3739         int split_flag = 0, depth;
3740
3741         ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3742                   __func__, inode->i_ino,
3743                   (unsigned long long)map->m_lblk, map->m_len);
3744
3745         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3746                 inode->i_sb->s_blocksize_bits;
3747         if (eof_block < map->m_lblk + map->m_len)
3748                 eof_block = map->m_lblk + map->m_len;
3749         /*
3750          * It is safe to convert extent to initialized via explicit
3751          * zeroout only if extent is fully insde i_size or new_size.
3752          */
3753         depth = ext_depth(inode);
3754         ex = path[depth].p_ext;
3755         ee_block = le32_to_cpu(ex->ee_block);
3756         ee_len = ext4_ext_get_actual_len(ex);
3757
3758         /* Convert to unwritten */
3759         if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3760                 split_flag |= EXT4_EXT_DATA_VALID1;
3761         /* Convert to initialized */
3762         } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3763                 split_flag |= ee_block + ee_len <= eof_block ?
3764                               EXT4_EXT_MAY_ZEROOUT : 0;
3765                 split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3766         }
3767         flags |= EXT4_GET_BLOCKS_PRE_IO;
3768         return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3769 }
3770
3771 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3772                                                 struct inode *inode,
3773                                                 struct ext4_map_blocks *map,
3774                                                 struct ext4_ext_path **ppath)
3775 {
3776         struct ext4_ext_path *path = *ppath;
3777         struct ext4_extent *ex;
3778         ext4_lblk_t ee_block;
3779         unsigned int ee_len;
3780         int depth;
3781         int err = 0;
3782
3783         depth = ext_depth(inode);
3784         ex = path[depth].p_ext;
3785         ee_block = le32_to_cpu(ex->ee_block);
3786         ee_len = ext4_ext_get_actual_len(ex);
3787
3788         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3789                 "block %llu, max_blocks %u\n", inode->i_ino,
3790                   (unsigned long long)ee_block, ee_len);
3791
3792         /* If extent is larger than requested it is a clear sign that we still
3793          * have some extent state machine issues left. So extent_split is still
3794          * required.
3795          * TODO: Once all related issues will be fixed this situation should be
3796          * illegal.
3797          */
3798         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3799 #ifdef EXT4_DEBUG
3800                 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3801                              " len %u; IO logical block %llu, len %u",
3802                              inode->i_ino, (unsigned long long)ee_block, ee_len,
3803                              (unsigned long long)map->m_lblk, map->m_len);
3804 #endif
3805                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3806                                                  EXT4_GET_BLOCKS_CONVERT);
3807                 if (err < 0)
3808                         return err;
3809                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3810                 if (IS_ERR(path))
3811                         return PTR_ERR(path);
3812                 depth = ext_depth(inode);
3813                 ex = path[depth].p_ext;
3814         }
3815
3816         err = ext4_ext_get_access(handle, inode, path + depth);
3817         if (err)
3818                 goto out;
3819         /* first mark the extent as initialized */
3820         ext4_ext_mark_initialized(ex);
3821
3822         /* note: ext4_ext_correct_indexes() isn't needed here because
3823          * borders are not changed
3824          */
3825         ext4_ext_try_to_merge(handle, inode, path, ex);
3826
3827         /* Mark modified extent as dirty */
3828         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3829 out:
3830         ext4_ext_show_leaf(inode, path);
3831         return err;
3832 }
3833
3834 /*
3835  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3836  */
3837 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3838                               ext4_lblk_t lblk,
3839                               struct ext4_ext_path *path,
3840                               unsigned int len)
3841 {
3842         int i, depth;
3843         struct ext4_extent_header *eh;
3844         struct ext4_extent *last_ex;
3845
3846         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3847                 return 0;
3848
3849         depth = ext_depth(inode);
3850         eh = path[depth].p_hdr;
3851
3852         /*
3853          * We're going to remove EOFBLOCKS_FL entirely in future so we
3854          * do not care for this case anymore. Simply remove the flag
3855          * if there are no extents.
3856          */
3857         if (unlikely(!eh->eh_entries))
3858                 goto out;
3859         last_ex = EXT_LAST_EXTENT(eh);
3860         /*
3861          * We should clear the EOFBLOCKS_FL flag if we are writing the
3862          * last block in the last extent in the file.  We test this by
3863          * first checking to see if the caller to
3864          * ext4_ext_get_blocks() was interested in the last block (or
3865          * a block beyond the last block) in the current extent.  If
3866          * this turns out to be false, we can bail out from this
3867          * function immediately.
3868          */
3869         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3870             ext4_ext_get_actual_len(last_ex))
3871                 return 0;
3872         /*
3873          * If the caller does appear to be planning to write at or
3874          * beyond the end of the current extent, we then test to see
3875          * if the current extent is the last extent in the file, by
3876          * checking to make sure it was reached via the rightmost node
3877          * at each level of the tree.
3878          */
3879         for (i = depth-1; i >= 0; i--)
3880                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3881                         return 0;
3882 out:
3883         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3884         return ext4_mark_inode_dirty(handle, inode);
3885 }
3886
3887 static int
3888 convert_initialized_extent(handle_t *handle, struct inode *inode,
3889                            struct ext4_map_blocks *map,
3890                            struct ext4_ext_path **ppath,
3891                            unsigned int allocated)
3892 {
3893         struct ext4_ext_path *path = *ppath;
3894         struct ext4_extent *ex;
3895         ext4_lblk_t ee_block;
3896         unsigned int ee_len;
3897         int depth;
3898         int err = 0;
3899
3900         /*
3901          * Make sure that the extent is no bigger than we support with
3902          * unwritten extent
3903          */
3904         if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3905                 map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3906
3907         depth = ext_depth(inode);
3908         ex = path[depth].p_ext;
3909         ee_block = le32_to_cpu(ex->ee_block);
3910         ee_len = ext4_ext_get_actual_len(ex);
3911
3912         ext_debug("%s: inode %lu, logical"
3913                 "block %llu, max_blocks %u\n", __func__, inode->i_ino,
3914                   (unsigned long long)ee_block, ee_len);
3915
3916         if (ee_block != map->m_lblk || ee_len > map->m_len) {
3917                 err = ext4_split_convert_extents(handle, inode, map, ppath,
3918                                 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3919                 if (err < 0)
3920                         return err;
3921                 path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3922                 if (IS_ERR(path))
3923                         return PTR_ERR(path);
3924                 depth = ext_depth(inode);
3925                 ex = path[depth].p_ext;
3926                 if (!ex) {
3927                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3928                                          (unsigned long) map->m_lblk);
3929                         return -EFSCORRUPTED;
3930                 }
3931         }
3932
3933         err = ext4_ext_get_access(handle, inode, path + depth);
3934         if (err)
3935                 return err;
3936         /* first mark the extent as unwritten */
3937         ext4_ext_mark_unwritten(ex);
3938
3939         /* note: ext4_ext_correct_indexes() isn't needed here because
3940          * borders are not changed
3941          */
3942         ext4_ext_try_to_merge(handle, inode, path, ex);
3943
3944         /* Mark modified extent as dirty */
3945         err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3946         if (err)
3947                 return err;
3948         ext4_ext_show_leaf(inode, path);
3949
3950         ext4_update_inode_fsync_trans(handle, inode, 1);
3951         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3952         if (err)
3953                 return err;
3954         map->m_flags |= EXT4_MAP_UNWRITTEN;
3955         if (allocated > map->m_len)
3956                 allocated = map->m_len;
3957         map->m_len = allocated;
3958         return allocated;
3959 }
3960
3961 static int
3962 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3963                         struct ext4_map_blocks *map,
3964                         struct ext4_ext_path **ppath, int flags,
3965                         unsigned int allocated, ext4_fsblk_t newblock)
3966 {
3967         struct ext4_ext_path *path = *ppath;
3968         int ret = 0;
3969         int err = 0;
3970
3971         ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
3972                   "block %llu, max_blocks %u, flags %x, allocated %u\n",
3973                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3974                   flags, allocated);
3975         ext4_ext_show_leaf(inode, path);
3976
3977         /*
3978          * When writing into unwritten space, we should not fail to
3979          * allocate metadata blocks for the new extent block if needed.
3980          */
3981         flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3982
3983         trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3984                                                     allocated, newblock);
3985
3986         /* get_block() before submit the IO, split the extent */
3987         if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3988                 ret = ext4_split_convert_extents(handle, inode, map, ppath,
3989                                          flags | EXT4_GET_BLOCKS_CONVERT);
3990                 if (ret <= 0)
3991                         goto out;
3992                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3993                 goto out;
3994         }
3995         /* IO end_io complete, convert the filled extent to written */
3996         if (flags & EXT4_GET_BLOCKS_CONVERT) {
3997                 if (flags & EXT4_GET_BLOCKS_ZERO) {
3998                         if (allocated > map->m_len)
3999                                 allocated = map->m_len;
4000                         err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4001                                                  allocated);
4002                         if (err < 0)
4003                                 goto out2;
4004                 }
4005                 ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4006                                                            ppath);
4007                 if (ret >= 0) {
4008                         ext4_update_inode_fsync_trans(handle, inode, 1);
4009                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
4010                                                  path, map->m_len);
4011                 } else
4012                         err = ret;
4013                 map->m_flags |= EXT4_MAP_MAPPED;
4014                 map->m_pblk = newblock;
4015                 if (allocated > map->m_len)
4016                         allocated = map->m_len;
4017                 map->m_len = allocated;
4018                 goto out2;
4019         }
4020         /* buffered IO case */
4021         /*
4022          * repeat fallocate creation request
4023          * we already have an unwritten extent
4024          */
4025         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4026                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4027                 goto map_out;
4028         }
4029
4030         /* buffered READ or buffered write_begin() lookup */
4031         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4032                 /*
4033                  * We have blocks reserved already.  We
4034                  * return allocated blocks so that delalloc
4035                  * won't do block reservation for us.  But
4036                  * the buffer head will be unmapped so that
4037                  * a read from the block returns 0s.
4038                  */
4039                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4040                 goto out1;
4041         }
4042
4043         /* buffered write, writepage time, convert*/
4044         ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4045         if (ret >= 0)
4046                 ext4_update_inode_fsync_trans(handle, inode, 1);
4047 out:
4048         if (ret <= 0) {
4049                 err = ret;
4050                 goto out2;
4051         } else
4052                 allocated = ret;
4053         map->m_flags |= EXT4_MAP_NEW;
4054         if (allocated > map->m_len)
4055                 allocated = map->m_len;
4056         map->m_len = allocated;
4057
4058 map_out:
4059         map->m_flags |= EXT4_MAP_MAPPED;
4060         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4061                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4062                                          map->m_len);
4063                 if (err < 0)
4064                         goto out2;
4065         }
4066 out1:
4067         if (allocated > map->m_len)
4068                 allocated = map->m_len;
4069         ext4_ext_show_leaf(inode, path);
4070         map->m_pblk = newblock;
4071         map->m_len = allocated;
4072 out2:
4073         return err ? err : allocated;
4074 }
4075
4076 /*
4077  * get_implied_cluster_alloc - check to see if the requested
4078  * allocation (in the map structure) overlaps with a cluster already
4079  * allocated in an extent.
4080  *      @sb     The filesystem superblock structure
4081  *      @map    The requested lblk->pblk mapping
4082  *      @ex     The extent structure which might contain an implied
4083  *                      cluster allocation
4084  *
4085  * This function is called by ext4_ext_map_blocks() after we failed to
4086  * find blocks that were already in the inode's extent tree.  Hence,
4087  * we know that the beginning of the requested region cannot overlap
4088  * the extent from the inode's extent tree.  There are three cases we
4089  * want to catch.  The first is this case:
4090  *
4091  *               |--- cluster # N--|
4092  *    |--- extent ---|  |---- requested region ---|
4093  *                      |==========|
4094  *
4095  * The second case that we need to test for is this one:
4096  *
4097  *   |--------- cluster # N ----------------|
4098  *         |--- requested region --|   |------- extent ----|
4099  *         |=======================|
4100  *
4101  * The third case is when the requested region lies between two extents
4102  * within the same cluster:
4103  *          |------------- cluster # N-------------|
4104  * |----- ex -----|                  |---- ex_right ----|
4105  *                  |------ requested region ------|
4106  *                  |================|
4107  *
4108  * In each of the above cases, we need to set the map->m_pblk and
4109  * map->m_len so it corresponds to the return the extent labelled as
4110  * "|====|" from cluster #N, since it is already in use for data in
4111  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
4112  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4113  * as a new "allocated" block region.  Otherwise, we will return 0 and
4114  * ext4_ext_map_blocks() will then allocate one or more new clusters
4115  * by calling ext4_mb_new_blocks().
4116  */
4117 static int get_implied_cluster_alloc(struct super_block *sb,
4118                                      struct ext4_map_blocks *map,
4119                                      struct ext4_extent *ex,
4120                                      struct ext4_ext_path *path)
4121 {
4122         struct ext4_sb_info *sbi = EXT4_SB(sb);
4123         ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4124         ext4_lblk_t ex_cluster_start, ex_cluster_end;
4125         ext4_lblk_t rr_cluster_start;
4126         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4127         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4128         unsigned short ee_len = ext4_ext_get_actual_len(ex);
4129
4130         /* The extent passed in that we are trying to match */
4131         ex_cluster_start = EXT4_B2C(sbi, ee_block);
4132         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4133
4134         /* The requested region passed into ext4_map_blocks() */
4135         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4136
4137         if ((rr_cluster_start == ex_cluster_end) ||
4138             (rr_cluster_start == ex_cluster_start)) {
4139                 if (rr_cluster_start == ex_cluster_end)
4140                         ee_start += ee_len - 1;
4141                 map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4142                 map->m_len = min(map->m_len,
4143                                  (unsigned) sbi->s_cluster_ratio - c_offset);
4144                 /*
4145                  * Check for and handle this case:
4146                  *
4147                  *   |--------- cluster # N-------------|
4148                  *                     |------- extent ----|
4149                  *         |--- requested region ---|
4150                  *         |===========|
4151                  */
4152
4153                 if (map->m_lblk < ee_block)
4154                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
4155
4156                 /*
4157                  * Check for the case where there is already another allocated
4158                  * block to the right of 'ex' but before the end of the cluster.
4159                  *
4160                  *          |------------- cluster # N-------------|
4161                  * |----- ex -----|                  |---- ex_right ----|
4162                  *                  |------ requested region ------|
4163                  *                  |================|
4164                  */
4165                 if (map->m_lblk > ee_block) {
4166                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4167                         map->m_len = min(map->m_len, next - map->m_lblk);
4168                 }
4169
4170                 trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4171                 return 1;
4172         }
4173
4174         trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4175         return 0;
4176 }
4177
4178
4179 /*
4180  * Block allocation/map/preallocation routine for extents based files
4181  *
4182  *
4183  * Need to be called with
4184  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4185  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4186  *
4187  * return > 0, number of of blocks already mapped/allocated
4188  *          if create == 0 and these are pre-allocated blocks
4189  *              buffer head is unmapped
4190  *          otherwise blocks are mapped
4191  *
4192  * return = 0, if plain look up failed (blocks have not been allocated)
4193  *          buffer head is unmapped
4194  *
4195  * return < 0, error case.
4196  */
4197 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4198                         struct ext4_map_blocks *map, int flags)
4199 {
4200         struct ext4_ext_path *path = NULL;
4201         struct ext4_extent newex, *ex, *ex2;
4202         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4203         ext4_fsblk_t newblock = 0;
4204         int free_on_err = 0, err = 0, depth, ret;
4205         unsigned int allocated = 0, offset = 0;
4206         unsigned int allocated_clusters = 0;
4207         struct ext4_allocation_request ar;
4208         ext4_lblk_t cluster_offset;
4209         bool map_from_cluster = false;
4210
4211         ext_debug("blocks %u/%u requested for inode %lu\n",
4212                   map->m_lblk, map->m_len, inode->i_ino);
4213         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4214
4215         /* find extent for this block */
4216         path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4217         if (IS_ERR(path)) {
4218                 err = PTR_ERR(path);
4219                 path = NULL;
4220                 goto out2;
4221         }
4222
4223         depth = ext_depth(inode);
4224
4225         /*
4226          * consistent leaf must not be empty;
4227          * this situation is possible, though, _during_ tree modification;
4228          * this is why assert can't be put in ext4_find_extent()
4229          */
4230         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4231                 EXT4_ERROR_INODE(inode, "bad extent address "
4232                                  "lblock: %lu, depth: %d pblock %lld",
4233                                  (unsigned long) map->m_lblk, depth,
4234                                  path[depth].p_block);
4235                 err = -EFSCORRUPTED;
4236                 goto out2;
4237         }
4238
4239         ex = path[depth].p_ext;
4240         if (ex) {
4241                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4242                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4243                 unsigned short ee_len;
4244
4245
4246                 /*
4247                  * unwritten extents are treated as holes, except that
4248                  * we split out initialized portions during a write.
4249                  */
4250                 ee_len = ext4_ext_get_actual_len(ex);
4251
4252                 trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4253
4254                 /* if found extent covers block, simply return it */
4255                 if (in_range(map->m_lblk, ee_block, ee_len)) {
4256                         newblock = map->m_lblk - ee_block + ee_start;
4257                         /* number of remaining blocks in the extent */
4258                         allocated = ee_len - (map->m_lblk - ee_block);
4259                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4260                                   ee_block, ee_len, newblock);
4261
4262                         /*
4263                          * If the extent is initialized check whether the
4264                          * caller wants to convert it to unwritten.
4265                          */
4266                         if ((!ext4_ext_is_unwritten(ex)) &&
4267                             (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4268                                 allocated = convert_initialized_extent(
4269                                                 handle, inode, map, &path,
4270                                                 allocated);
4271                                 goto out2;
4272                         } else if (!ext4_ext_is_unwritten(ex))
4273                                 goto out;
4274
4275                         ret = ext4_ext_handle_unwritten_extents(
4276                                 handle, inode, map, &path, flags,
4277                                 allocated, newblock);
4278                         if (ret < 0)
4279                                 err = ret;
4280                         else
4281                                 allocated = ret;
4282                         goto out2;
4283                 }
4284         }
4285
4286         /*
4287          * requested block isn't allocated yet;
4288          * we couldn't try to create block if create flag is zero
4289          */
4290         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4291                 ext4_lblk_t hole_start, hole_len;
4292
4293                 hole_start = map->m_lblk;
4294                 hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4295                 /*
4296                  * put just found gap into cache to speed up
4297                  * subsequent requests
4298                  */
4299                 ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4300
4301                 /* Update hole_len to reflect hole size after map->m_lblk */
4302                 if (hole_start != map->m_lblk)
4303                         hole_len -= map->m_lblk - hole_start;
4304                 map->m_pblk = 0;
4305                 map->m_len = min_t(unsigned int, map->m_len, hole_len);
4306
4307                 goto out2;
4308         }
4309
4310         /*
4311          * Okay, we need to do block allocation.
4312          */
4313         newex.ee_block = cpu_to_le32(map->m_lblk);
4314         cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4315
4316         /*
4317          * If we are doing bigalloc, check to see if the extent returned
4318          * by ext4_find_extent() implies a cluster we can use.
4319          */
4320         if (cluster_offset && ex &&
4321             get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4322                 ar.len = allocated = map->m_len;
4323                 newblock = map->m_pblk;
4324                 map_from_cluster = true;
4325                 goto got_allocated_blocks;
4326         }
4327
4328         /* find neighbour allocated blocks */
4329         ar.lleft = map->m_lblk;
4330         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4331         if (err)
4332                 goto out2;
4333         ar.lright = map->m_lblk;
4334         ex2 = NULL;
4335         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4336         if (err)
4337                 goto out2;
4338
4339         /* Check if the extent after searching to the right implies a
4340          * cluster we can use. */
4341         if ((sbi->s_cluster_ratio > 1) && ex2 &&
4342             get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4343                 ar.len = allocated = map->m_len;
4344                 newblock = map->m_pblk;
4345                 map_from_cluster = true;
4346                 goto got_allocated_blocks;
4347         }
4348
4349         /*
4350          * See if request is beyond maximum number of blocks we can have in
4351          * a single extent. For an initialized extent this limit is
4352          * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4353          * EXT_UNWRITTEN_MAX_LEN.
4354          */
4355         if (map->m_len > EXT_INIT_MAX_LEN &&
4356             !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4357                 map->m_len = EXT_INIT_MAX_LEN;
4358         else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4359                  (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4360                 map->m_len = EXT_UNWRITTEN_MAX_LEN;
4361
4362         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4363         newex.ee_len = cpu_to_le16(map->m_len);
4364         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4365         if (err)
4366                 allocated = ext4_ext_get_actual_len(&newex);
4367         else
4368                 allocated = map->m_len;
4369
4370         /* allocate new block */
4371         ar.inode = inode;
4372         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4373         ar.logical = map->m_lblk;
4374         /*
4375          * We calculate the offset from the beginning of the cluster
4376          * for the logical block number, since when we allocate a
4377          * physical cluster, the physical block should start at the
4378          * same offset from the beginning of the cluster.  This is
4379          * needed so that future calls to get_implied_cluster_alloc()
4380          * work correctly.
4381          */
4382         offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4383         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4384         ar.goal -= offset;
4385         ar.logical -= offset;
4386         if (S_ISREG(inode->i_mode))
4387                 ar.flags = EXT4_MB_HINT_DATA;
4388         else
4389                 /* disable in-core preallocation for non-regular files */
4390                 ar.flags = 0;
4391         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4392                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4393         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4394                 ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4395         if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4396                 ar.flags |= EXT4_MB_USE_RESERVED;
4397         newblock = ext4_mb_new_blocks(handle, &ar, &err);
4398         if (!newblock)
4399                 goto out2;
4400         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4401                   ar.goal, newblock, allocated);
4402         free_on_err = 1;
4403         allocated_clusters = ar.len;
4404         ar.len = EXT4_C2B(sbi, ar.len) - offset;
4405         if (ar.len > allocated)
4406                 ar.len = allocated;
4407
4408 got_allocated_blocks:
4409         /* try to insert new extent into found leaf and return */
4410         ext4_ext_store_pblock(&newex, newblock + offset);
4411         newex.ee_len = cpu_to_le16(ar.len);
4412         /* Mark unwritten */
4413         if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4414                 ext4_ext_mark_unwritten(&newex);
4415                 map->m_flags |= EXT4_MAP_UNWRITTEN;
4416         }
4417
4418         err = 0;
4419         if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4420                 err = check_eofblocks_fl(handle, inode, map->m_lblk,
4421                                          path, ar.len);
4422         if (!err)
4423                 err = ext4_ext_insert_extent(handle, inode, &path,
4424                                              &newex, flags);
4425
4426         if (err && free_on_err) {
4427                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4428                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4429                 /* free data blocks we just allocated */
4430                 /* not a good idea to call discard here directly,
4431                  * but otherwise we'd need to call it every free() */
4432                 ext4_discard_preallocations(inode);
4433                 ext4_free_blocks(handle, inode, NULL, newblock,
4434                                  EXT4_C2B(sbi, allocated_clusters), fb_flags);
4435                 goto out2;
4436         }
4437
4438         /* previous routine could use block we allocated */
4439         newblock = ext4_ext_pblock(&newex);
4440         allocated = ext4_ext_get_actual_len(&newex);
4441         if (allocated > map->m_len)
4442                 allocated = map->m_len;
4443         map->m_flags |= EXT4_MAP_NEW;
4444
4445         /*
4446          * Reduce the reserved cluster count to reflect successful deferred
4447          * allocation of delayed allocated clusters or direct allocation of
4448          * clusters discovered to be delayed allocated.  Once allocated, a
4449          * cluster is not included in the reserved count.
4450          */
4451         if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4452                 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4453                         /*
4454                          * When allocating delayed allocated clusters, simply
4455                          * reduce the reserved cluster count and claim quota
4456                          */
4457                         ext4_da_update_reserve_space(inode, allocated_clusters,
4458                                                         1);
4459                 } else {
4460                         ext4_lblk_t lblk, len;
4461                         unsigned int n;
4462
4463                         /*
4464                          * When allocating non-delayed allocated clusters
4465                          * (from fallocate, filemap, DIO, or clusters
4466                          * allocated when delalloc has been disabled by
4467                          * ext4_nonda_switch), reduce the reserved cluster
4468                          * count by the number of allocated clusters that
4469                          * have previously been delayed allocated.  Quota
4470                          * has been claimed by ext4_mb_new_blocks() above,
4471                          * so release the quota reservations made for any
4472                          * previously delayed allocated clusters.
4473                          */
4474                         lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4475                         len = allocated_clusters << sbi->s_cluster_bits;
4476                         n = ext4_es_delayed_clu(inode, lblk, len);
4477                         if (n > 0)
4478                                 ext4_da_update_reserve_space(inode, (int) n, 0);
4479                 }
4480         }
4481
4482         /*
4483          * Cache the extent and update transaction to commit on fdatasync only
4484          * when it is _not_ an unwritten extent.
4485          */
4486         if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4487                 ext4_update_inode_fsync_trans(handle, inode, 1);
4488         else
4489                 ext4_update_inode_fsync_trans(handle, inode, 0);
4490 out:
4491         if (allocated > map->m_len)
4492                 allocated = map->m_len;
4493         ext4_ext_show_leaf(inode, path);
4494         map->m_flags |= EXT4_MAP_MAPPED;
4495         map->m_pblk = newblock;
4496         map->m_len = allocated;
4497 out2:
4498         ext4_ext_drop_refs(path);
4499         kfree(path);
4500
4501         trace_ext4_ext_map_blocks_exit(inode, flags, map,
4502                                        err ? err : allocated);
4503         return err ? err : allocated;
4504 }
4505
4506 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4507 {
4508         struct super_block *sb = inode->i_sb;
4509         ext4_lblk_t last_block;
4510         int err = 0;
4511
4512         /*
4513          * TODO: optimization is possible here.
4514          * Probably we need not scan at all,
4515          * because page truncation is enough.
4516          */
4517
4518         /* we have to know where to truncate from in crash case */
4519         EXT4_I(inode)->i_disksize = inode->i_size;
4520         err = ext4_mark_inode_dirty(handle, inode);
4521         if (err)
4522                 return err;
4523
4524         last_block = (inode->i_size + sb->s_blocksize - 1)
4525                         >> EXT4_BLOCK_SIZE_BITS(sb);
4526 retry:
4527         err = ext4_es_remove_extent(inode, last_block,
4528                                     EXT_MAX_BLOCKS - last_block);
4529         if (err == -ENOMEM) {
4530                 cond_resched();
4531                 congestion_wait(BLK_RW_ASYNC, HZ/50);
4532                 goto retry;
4533         }
4534         if (err)
4535                 return err;
4536         return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4537 }
4538
4539 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4540                                   ext4_lblk_t len, loff_t new_size,
4541                                   int flags)
4542 {
4543         struct inode *inode = file_inode(file);
4544         handle_t *handle;
4545         int ret = 0;
4546         int ret2 = 0;
4547         int retries = 0;
4548         int depth = 0;
4549         struct ext4_map_blocks map;
4550         unsigned int credits;
4551         loff_t epos;
4552
4553         BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4554         map.m_lblk = offset;
4555         map.m_len = len;
4556         /*
4557          * Don't normalize the request if it can fit in one extent so
4558          * that it doesn't get unnecessarily split into multiple
4559          * extents.
4560          */
4561         if (len <= EXT_UNWRITTEN_MAX_LEN)
4562                 flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4563
4564         /*
4565          * credits to insert 1 extent into extent tree
4566          */
4567         credits = ext4_chunk_trans_blocks(inode, len);
4568         depth = ext_depth(inode);
4569
4570 retry:
4571         while (ret >= 0 && len) {
4572                 /*
4573                  * Recalculate credits when extent tree depth changes.
4574                  */
4575                 if (depth != ext_depth(inode)) {
4576                         credits = ext4_chunk_trans_blocks(inode, len);
4577                         depth = ext_depth(inode);
4578                 }
4579
4580                 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4581                                             credits);
4582                 if (IS_ERR(handle)) {
4583                         ret = PTR_ERR(handle);
4584                         break;
4585                 }
4586                 ret = ext4_map_blocks(handle, inode, &map, flags);
4587                 if (ret <= 0) {
4588                         ext4_debug("inode #%lu: block %u: len %u: "
4589                                    "ext4_ext_map_blocks returned %d",
4590                                    inode->i_ino, map.m_lblk,
4591                                    map.m_len, ret);
4592                         ext4_mark_inode_dirty(handle, inode);
4593                         ret2 = ext4_journal_stop(handle);
4594                         break;
4595                 }
4596                 map.m_lblk += ret;
4597                 map.m_len = len = len - ret;
4598                 epos = (loff_t)map.m_lblk << inode->i_blkbits;
4599                 inode->i_ctime = current_time(inode);
4600                 if (new_size) {
4601                         if (epos > new_size)
4602                                 epos = new_size;
4603                         if (ext4_update_inode_size(inode, epos) & 0x1)
4604                                 inode->i_mtime = inode->i_ctime;
4605                 } else {
4606                         if (epos > inode->i_size)
4607                                 ext4_set_inode_flag(inode,
4608                                                     EXT4_INODE_EOFBLOCKS);
4609                 }
4610                 ext4_mark_inode_dirty(handle, inode);
4611                 ext4_update_inode_fsync_trans(handle, inode, 1);
4612                 ret2 = ext4_journal_stop(handle);
4613                 if (ret2)
4614                         break;
4615         }
4616         if (ret == -ENOSPC &&
4617                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
4618                 ret = 0;
4619                 goto retry;
4620         }
4621
4622         return ret > 0 ? ret2 : ret;
4623 }
4624
4625 static long ext4_zero_range(struct file *file, loff_t offset,
4626                             loff_t len, int mode)
4627 {
4628         struct inode *inode = file_inode(file);
4629         handle_t *handle = NULL;
4630         unsigned int max_blocks;
4631         loff_t new_size = 0;
4632         int ret = 0;
4633         int flags;
4634         int credits;
4635         int partial_begin, partial_end;
4636         loff_t start, end;
4637         ext4_lblk_t lblk;
4638         unsigned int blkbits = inode->i_blkbits;
4639
4640         trace_ext4_zero_range(inode, offset, len, mode);
4641
4642         if (!S_ISREG(inode->i_mode))
4643                 return -EINVAL;
4644
4645         /* Call ext4_force_commit to flush all data in case of data=journal. */
4646         if (ext4_should_journal_data(inode)) {
4647                 ret = ext4_force_commit(inode->i_sb);
4648                 if (ret)
4649                         return ret;
4650         }
4651
4652         /*
4653          * Round up offset. This is not fallocate, we neet to zero out
4654          * blocks, so convert interior block aligned part of the range to
4655          * unwritten and possibly manually zero out unaligned parts of the
4656          * range.
4657          */
4658         start = round_up(offset, 1 << blkbits);
4659         end = round_down((offset + len), 1 << blkbits);
4660
4661         if (start < offset || end > offset + len)
4662                 return -EINVAL;
4663         partial_begin = offset & ((1 << blkbits) - 1);
4664         partial_end = (offset + len) & ((1 << blkbits) - 1);
4665
4666         lblk = start >> blkbits;
4667         max_blocks = (end >> blkbits);
4668         if (max_blocks < lblk)
4669                 max_blocks = 0;
4670         else
4671                 max_blocks -= lblk;
4672
4673         inode_lock(inode);
4674
4675         /*
4676          * Indirect files do not support unwritten extnets
4677          */
4678         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4679                 ret = -EOPNOTSUPP;
4680                 goto out_mutex;
4681         }
4682
4683         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4684             (offset + len > i_size_read(inode) ||
4685              offset + len > EXT4_I(inode)->i_disksize)) {
4686                 new_size = offset + len;
4687                 ret = inode_newsize_ok(inode, new_size);
4688                 if (ret)
4689                         goto out_mutex;
4690         }
4691
4692         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4693         if (mode & FALLOC_FL_KEEP_SIZE)
4694                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4695
4696         /* Wait all existing dio workers, newcomers will block on i_mutex */
4697         inode_dio_wait(inode);
4698
4699         /* Preallocate the range including the unaligned edges */
4700         if (partial_begin || partial_end) {
4701                 ret = ext4_alloc_file_blocks(file,
4702                                 round_down(offset, 1 << blkbits) >> blkbits,
4703                                 (round_up((offset + len), 1 << blkbits) -
4704                                  round_down(offset, 1 << blkbits)) >> blkbits,
4705                                 new_size, flags);
4706                 if (ret)
4707                         goto out_mutex;
4708
4709         }
4710
4711         /* Zero range excluding the unaligned edges */
4712         if (max_blocks > 0) {
4713                 flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4714                           EXT4_EX_NOCACHE);
4715
4716                 /*
4717                  * Prevent page faults from reinstantiating pages we have
4718                  * released from page cache.
4719                  */
4720                 down_write(&EXT4_I(inode)->i_mmap_sem);
4721
4722                 ret = ext4_break_layouts(inode);
4723                 if (ret) {
4724                         up_write(&EXT4_I(inode)->i_mmap_sem);
4725                         goto out_mutex;
4726                 }
4727
4728                 ret = ext4_update_disksize_before_punch(inode, offset, len);
4729                 if (ret) {
4730                         up_write(&EXT4_I(inode)->i_mmap_sem);
4731                         goto out_mutex;
4732                 }
4733                 /* Now release the pages and zero block aligned part of pages */
4734                 truncate_pagecache_range(inode, start, end - 1);
4735                 inode->i_mtime = inode->i_ctime = current_time(inode);
4736
4737                 ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4738                                              flags);
4739                 up_write(&EXT4_I(inode)->i_mmap_sem);
4740                 if (ret)
4741                         goto out_mutex;
4742         }
4743         if (!partial_begin && !partial_end)
4744                 goto out_mutex;
4745
4746         /*
4747          * In worst case we have to writeout two nonadjacent unwritten
4748          * blocks and update the inode
4749          */
4750         credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4751         if (ext4_should_journal_data(inode))
4752                 credits += 2;
4753         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4754         if (IS_ERR(handle)) {
4755                 ret = PTR_ERR(handle);
4756                 ext4_std_error(inode->i_sb, ret);
4757                 goto out_mutex;
4758         }
4759
4760         inode->i_mtime = inode->i_ctime = current_time(inode);
4761         if (new_size) {
4762                 ext4_update_inode_size(inode, new_size);
4763         } else {
4764                 /*
4765                 * Mark that we allocate beyond EOF so the subsequent truncate
4766                 * can proceed even if the new size is the same as i_size.
4767                 */
4768                 if ((offset + len) > i_size_read(inode))
4769                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4770         }
4771         ext4_mark_inode_dirty(handle, inode);
4772
4773         /* Zero out partial block at the edges of the range */
4774         ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4775         if (ret >= 0)
4776                 ext4_update_inode_fsync_trans(handle, inode, 1);
4777
4778         if (file->f_flags & O_SYNC)
4779                 ext4_handle_sync(handle);
4780
4781         ext4_journal_stop(handle);
4782 out_mutex:
4783         inode_unlock(inode);
4784         return ret;
4785 }
4786
4787 /*
4788  * preallocate space for a file. This implements ext4's fallocate file
4789  * operation, which gets called from sys_fallocate system call.
4790  * For block-mapped files, posix_fallocate should fall back to the method
4791  * of writing zeroes to the required new blocks (the same behavior which is
4792  * expected for file systems which do not support fallocate() system call).
4793  */
4794 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4795 {
4796         struct inode *inode = file_inode(file);
4797         loff_t new_size = 0;
4798         unsigned int max_blocks;
4799         int ret = 0;
4800         int flags;
4801         ext4_lblk_t lblk;
4802         unsigned int blkbits = inode->i_blkbits;
4803
4804         /*
4805          * Encrypted inodes can't handle collapse range or insert
4806          * range since we would need to re-encrypt blocks with a
4807          * different IV or XTS tweak (which are based on the logical
4808          * block number).
4809          *
4810          * XXX It's not clear why zero range isn't working, but we'll
4811          * leave it disabled for encrypted inodes for now.  This is a
4812          * bug we should fix....
4813          */
4814         if (IS_ENCRYPTED(inode) &&
4815             (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
4816                      FALLOC_FL_ZERO_RANGE)))
4817                 return -EOPNOTSUPP;
4818
4819         /* Return error if mode is not supported */
4820         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4821                      FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4822                      FALLOC_FL_INSERT_RANGE))
4823                 return -EOPNOTSUPP;
4824
4825         if (mode & FALLOC_FL_PUNCH_HOLE)
4826                 return ext4_punch_hole(inode, offset, len);
4827
4828         ret = ext4_convert_inline_data(inode);
4829         if (ret)
4830                 return ret;
4831
4832         if (mode & FALLOC_FL_COLLAPSE_RANGE)
4833                 return ext4_collapse_range(inode, offset, len);
4834
4835         if (mode & FALLOC_FL_INSERT_RANGE)
4836                 return ext4_insert_range(inode, offset, len);
4837
4838         if (mode & FALLOC_FL_ZERO_RANGE)
4839                 return ext4_zero_range(file, offset, len, mode);
4840
4841         trace_ext4_fallocate_enter(inode, offset, len, mode);
4842         lblk = offset >> blkbits;
4843
4844         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4845         flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4846         if (mode & FALLOC_FL_KEEP_SIZE)
4847                 flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4848
4849         inode_lock(inode);
4850
4851         /*
4852          * We only support preallocation for extent-based files only
4853          */
4854         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4855                 ret = -EOPNOTSUPP;
4856                 goto out;
4857         }
4858
4859         if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4860             (offset + len > i_size_read(inode) ||
4861              offset + len > EXT4_I(inode)->i_disksize)) {
4862                 new_size = offset + len;
4863                 ret = inode_newsize_ok(inode, new_size);
4864                 if (ret)
4865                         goto out;
4866         }
4867
4868         /* Wait all existing dio workers, newcomers will block on i_mutex */
4869         inode_dio_wait(inode);
4870
4871         ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4872         if (ret)
4873                 goto out;
4874
4875         if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4876                 ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4877                                                 EXT4_I(inode)->i_sync_tid);
4878         }
4879 out:
4880         inode_unlock(inode);
4881         trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4882         return ret;
4883 }
4884
4885 /*
4886  * This function convert a range of blocks to written extents
4887  * The caller of this function will pass the start offset and the size.
4888  * all unwritten extents within this range will be converted to
4889  * written extents.
4890  *
4891  * This function is called from the direct IO end io call back
4892  * function, to convert the fallocated extents after IO is completed.
4893  * Returns 0 on success.
4894  */
4895 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4896                                    loff_t offset, ssize_t len)
4897 {
4898         unsigned int max_blocks;
4899         int ret = 0;
4900         int ret2 = 0;
4901         struct ext4_map_blocks map;
4902         unsigned int credits, blkbits = inode->i_blkbits;
4903
4904         map.m_lblk = offset >> blkbits;
4905         max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4906
4907         /*
4908          * This is somewhat ugly but the idea is clear: When transaction is
4909          * reserved, everything goes into it. Otherwise we rather start several
4910          * smaller transactions for conversion of each extent separately.
4911          */
4912         if (handle) {
4913                 handle = ext4_journal_start_reserved(handle,
4914                                                      EXT4_HT_EXT_CONVERT);
4915                 if (IS_ERR(handle))
4916                         return PTR_ERR(handle);
4917                 credits = 0;
4918         } else {
4919                 /*
4920                  * credits to insert 1 extent into extent tree
4921                  */
4922                 credits = ext4_chunk_trans_blocks(inode, max_blocks);
4923         }
4924         while (ret >= 0 && ret < max_blocks) {
4925                 map.m_lblk += ret;
4926                 map.m_len = (max_blocks -= ret);
4927                 if (credits) {
4928                         handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4929                                                     credits);
4930                         if (IS_ERR(handle)) {
4931                                 ret = PTR_ERR(handle);
4932                                 break;
4933                         }
4934                 }
4935                 ret = ext4_map_blocks(handle, inode, &map,
4936                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4937                 if (ret <= 0)
4938                         ext4_warning(inode->i_sb,
4939                                      "inode #%lu: block %u: len %u: "
4940                                      "ext4_ext_map_blocks returned %d",
4941                                      inode->i_ino, map.m_lblk,
4942                                      map.m_len, ret);
4943                 ext4_mark_inode_dirty(handle, inode);
4944                 if (credits)
4945                         ret2 = ext4_journal_stop(handle);
4946                 if (ret <= 0 || ret2)
4947                         break;
4948         }
4949         if (!credits)
4950                 ret2 = ext4_journal_stop(handle);
4951         return ret > 0 ? ret2 : ret;
4952 }
4953
4954 /*
4955  * If newes is not existing extent (newes->ec_pblk equals zero) find
4956  * delayed extent at start of newes and update newes accordingly and
4957  * return start of the next delayed extent.
4958  *
4959  * If newes is existing extent (newes->ec_pblk is not equal zero)
4960  * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4961  * extent found. Leave newes unmodified.
4962  */
4963 static int ext4_find_delayed_extent(struct inode *inode,
4964                                     struct extent_status *newes)
4965 {
4966         struct extent_status es;
4967         ext4_lblk_t block, next_del;
4968
4969         if (newes->es_pblk == 0) {
4970                 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
4971                                           newes->es_lblk,
4972                                           newes->es_lblk + newes->es_len - 1,
4973                                           &es);
4974
4975                 /*
4976                  * No extent in extent-tree contains block @newes->es_pblk,
4977                  * then the block may stay in 1)a hole or 2)delayed-extent.
4978                  */
4979                 if (es.es_len == 0)
4980                         /* A hole found. */
4981                         return 0;
4982
4983                 if (es.es_lblk > newes->es_lblk) {
4984                         /* A hole found. */
4985                         newes->es_len = min(es.es_lblk - newes->es_lblk,
4986                                             newes->es_len);
4987                         return 0;
4988                 }
4989
4990                 newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
4991         }
4992
4993         block = newes->es_lblk + newes->es_len;
4994         ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
4995                                   EXT_MAX_BLOCKS, &es);
4996         if (es.es_len == 0)
4997                 next_del = EXT_MAX_BLOCKS;
4998         else
4999                 next_del = es.es_lblk;
5000
5001         return next_del;
5002 }
5003 /* fiemap flags we can handle specified here */
5004 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5005
5006 static int ext4_xattr_fiemap(struct inode *inode,
5007                                 struct fiemap_extent_info *fieinfo)
5008 {
5009         __u64 physical = 0;
5010         __u64 length;
5011         __u32 flags = FIEMAP_EXTENT_LAST;
5012         int blockbits = inode->i_sb->s_blocksize_bits;
5013         int error = 0;
5014
5015         /* in-inode? */
5016         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5017                 struct ext4_iloc iloc;
5018                 int offset;     /* offset of xattr in inode */
5019
5020                 error = ext4_get_inode_loc(inode, &iloc);
5021                 if (error)
5022                         return error;
5023                 physical = (__u64)iloc.bh->b_blocknr << blockbits;
5024                 offset = EXT4_GOOD_OLD_INODE_SIZE +
5025                                 EXT4_I(inode)->i_extra_isize;
5026                 physical += offset;
5027                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5028                 flags |= FIEMAP_EXTENT_DATA_INLINE;
5029                 brelse(iloc.bh);
5030         } else { /* external block */
5031                 physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5032                 length = inode->i_sb->s_blocksize;
5033         }
5034
5035         if (physical)
5036                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
5037                                                 length, flags);
5038         return (error < 0 ? error : 0);
5039 }
5040
5041 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5042                 __u64 start, __u64 len)
5043 {
5044         ext4_lblk_t start_blk;
5045         int error = 0;
5046
5047         if (ext4_has_inline_data(inode)) {
5048                 int has_inline = 1;
5049
5050                 error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5051                                                 start, len);
5052
5053                 if (has_inline)
5054                         return error;
5055         }
5056
5057         if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5058                 error = ext4_ext_precache(inode);
5059                 if (error)
5060                         return error;
5061         }
5062
5063         /* fallback to generic here if not in extents fmt */
5064         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5065                 return generic_block_fiemap(inode, fieinfo, start, len,
5066                         ext4_get_block);
5067
5068         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5069                 return -EBADR;
5070
5071         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5072                 error = ext4_xattr_fiemap(inode, fieinfo);
5073         } else {
5074                 ext4_lblk_t len_blks;
5075                 __u64 last_blk;
5076
5077                 start_blk = start >> inode->i_sb->s_blocksize_bits;
5078                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5079                 if (last_blk >= EXT_MAX_BLOCKS)
5080                         last_blk = EXT_MAX_BLOCKS-1;
5081                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5082
5083                 /*
5084                  * Walk the extent tree gathering extent information
5085                  * and pushing extents back to the user.
5086                  */
5087                 error = ext4_fill_fiemap_extents(inode, start_blk,
5088                                                  len_blks, fieinfo);
5089         }
5090         return error;
5091 }
5092
5093 /*
5094  * ext4_access_path:
5095  * Function to access the path buffer for marking it dirty.
5096  * It also checks if there are sufficient credits left in the journal handle
5097  * to update path.
5098  */
5099 static int
5100 ext4_access_path(handle_t *handle, struct inode *inode,
5101                 struct ext4_ext_path *path)
5102 {
5103         int credits, err;
5104
5105         if (!ext4_handle_valid(handle))
5106                 return 0;
5107
5108         /*
5109          * Check if need to extend journal credits
5110          * 3 for leaf, sb, and inode plus 2 (bmap and group
5111          * descriptor) for each block group; assume two block
5112          * groups
5113          */
5114         if (handle->h_buffer_credits < 7) {
5115                 credits = ext4_writepage_trans_blocks(inode);
5116                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
5117                 /* EAGAIN is success */
5118                 if (err && err != -EAGAIN)
5119                         return err;
5120         }
5121
5122         err = ext4_ext_get_access(handle, inode, path);
5123         return err;
5124 }
5125
5126 /*
5127  * ext4_ext_shift_path_extents:
5128  * Shift the extents of a path structure lying between path[depth].p_ext
5129  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5130  * if it is right shift or left shift operation.
5131  */
5132 static int
5133 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5134                             struct inode *inode, handle_t *handle,
5135                             enum SHIFT_DIRECTION SHIFT)
5136 {
5137         int depth, err = 0;
5138         struct ext4_extent *ex_start, *ex_last;
5139         bool update = 0;
5140         depth = path->p_depth;
5141
5142         while (depth >= 0) {
5143                 if (depth == path->p_depth) {
5144                         ex_start = path[depth].p_ext;
5145                         if (!ex_start)
5146                                 return -EFSCORRUPTED;
5147
5148                         ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5149
5150                         err = ext4_access_path(handle, inode, path + depth);
5151                         if (err)
5152                                 goto out;
5153
5154                         if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5155                                 update = 1;
5156
5157                         while (ex_start <= ex_last) {
5158                                 if (SHIFT == SHIFT_LEFT) {
5159                                         le32_add_cpu(&ex_start->ee_block,
5160                                                 -shift);
5161                                         /* Try to merge to the left. */
5162                                         if ((ex_start >
5163                                             EXT_FIRST_EXTENT(path[depth].p_hdr))
5164                                             &&
5165                                             ext4_ext_try_to_merge_right(inode,
5166                                             path, ex_start - 1))
5167                                                 ex_last--;
5168                                         else
5169                                                 ex_start++;
5170                                 } else {
5171                                         le32_add_cpu(&ex_last->ee_block, shift);
5172                                         ext4_ext_try_to_merge_right(inode, path,
5173                                                 ex_last);
5174                                         ex_last--;
5175                                 }
5176                         }
5177                         err = ext4_ext_dirty(handle, inode, path + depth);
5178                         if (err)
5179                                 goto out;
5180
5181                         if (--depth < 0 || !update)
5182                                 break;
5183                 }
5184
5185                 /* Update index too */
5186                 err = ext4_access_path(handle, inode, path + depth);
5187                 if (err)
5188                         goto out;
5189
5190                 if (SHIFT == SHIFT_LEFT)
5191                         le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5192                 else
5193                         le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5194                 err = ext4_ext_dirty(handle, inode, path + depth);
5195                 if (err)
5196                         goto out;
5197
5198                 /* we are done if current index is not a starting index */
5199                 if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5200                         break;
5201
5202                 depth--;
5203         }
5204
5205 out:
5206         return err;
5207 }
5208
5209 /*
5210  * ext4_ext_shift_extents:
5211  * All the extents which lies in the range from @start to the last allocated
5212  * block for the @inode are shifted either towards left or right (depending
5213  * upon @SHIFT) by @shift blocks.
5214  * On success, 0 is returned, error otherwise.
5215  */
5216 static int
5217 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5218                        ext4_lblk_t start, ext4_lblk_t shift,
5219                        enum SHIFT_DIRECTION SHIFT)
5220 {
5221         struct ext4_ext_path *path;
5222         int ret = 0, depth;
5223         struct ext4_extent *extent;
5224         ext4_lblk_t stop, *iterator, ex_start, ex_end;
5225
5226         /* Let path point to the last extent */
5227         path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5228                                 EXT4_EX_NOCACHE);
5229         if (IS_ERR(path))
5230                 return PTR_ERR(path);
5231
5232         depth = path->p_depth;
5233         extent = path[depth].p_ext;
5234         if (!extent)
5235                 goto out;
5236
5237         stop = le32_to_cpu(extent->ee_block);
5238
5239        /*
5240         * For left shifts, make sure the hole on the left is big enough to
5241         * accommodate the shift.  For right shifts, make sure the last extent
5242         * won't be shifted beyond EXT_MAX_BLOCKS.
5243         */
5244         if (SHIFT == SHIFT_LEFT) {
5245                 path = ext4_find_extent(inode, start - 1, &path,
5246                                         EXT4_EX_NOCACHE);
5247                 if (IS_ERR(path))
5248                         return PTR_ERR(path);
5249                 depth = path->p_depth;
5250                 extent =  path[depth].p_ext;
5251                 if (extent) {
5252                         ex_start = le32_to_cpu(extent->ee_block);
5253                         ex_end = le32_to_cpu(extent->ee_block) +
5254                                 ext4_ext_get_actual_len(extent);
5255                 } else {
5256                         ex_start = 0;
5257                         ex_end = 0;
5258                 }
5259
5260                 if ((start == ex_start && shift > ex_start) ||
5261                     (shift > start - ex_end)) {
5262                         ret = -EINVAL;
5263                         goto out;
5264                 }
5265         } else {
5266                 if (shift > EXT_MAX_BLOCKS -
5267                     (stop + ext4_ext_get_actual_len(extent))) {
5268                         ret = -EINVAL;
5269                         goto out;
5270                 }
5271         }
5272
5273         /*
5274          * In case of left shift, iterator points to start and it is increased
5275          * till we reach stop. In case of right shift, iterator points to stop
5276          * and it is decreased till we reach start.
5277          */
5278         if (SHIFT == SHIFT_LEFT)
5279                 iterator = &start;
5280         else
5281                 iterator = &stop;
5282
5283         /*
5284          * Its safe to start updating extents.  Start and stop are unsigned, so
5285          * in case of right shift if extent with 0 block is reached, iterator
5286          * becomes NULL to indicate the end of the loop.
5287          */
5288         while (iterator && start <= stop) {
5289                 path = ext4_find_extent(inode, *iterator, &path,
5290                                         EXT4_EX_NOCACHE);
5291                 if (IS_ERR(path))
5292                         return PTR_ERR(path);
5293                 depth = path->p_depth;
5294                 extent = path[depth].p_ext;
5295                 if (!extent) {
5296                         EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5297                                          (unsigned long) *iterator);
5298                         return -EFSCORRUPTED;
5299                 }
5300                 if (SHIFT == SHIFT_LEFT && *iterator >
5301                     le32_to_cpu(extent->ee_block)) {
5302                         /* Hole, move to the next extent */
5303                         if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5304                                 path[depth].p_ext++;
5305                         } else {
5306                                 *iterator = ext4_ext_next_allocated_block(path);
5307                                 continue;
5308                         }
5309                 }
5310
5311                 if (SHIFT == SHIFT_LEFT) {
5312                         extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5313                         *iterator = le32_to_cpu(extent->ee_block) +
5314                                         ext4_ext_get_actual_len(extent);
5315                 } else {
5316                         extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5317                         if (le32_to_cpu(extent->ee_block) > 0)
5318                                 *iterator = le32_to_cpu(extent->ee_block) - 1;
5319                         else
5320                                 /* Beginning is reached, end of the loop */
5321                                 iterator = NULL;
5322                         /* Update path extent in case we need to stop */
5323                         while (le32_to_cpu(extent->ee_block) < start)
5324                                 extent++;
5325                         path[depth].p_ext = extent;
5326                 }
5327                 ret = ext4_ext_shift_path_extents(path, shift, inode,
5328                                 handle, SHIFT);
5329                 if (ret)
5330                         break;
5331         }
5332 out:
5333         ext4_ext_drop_refs(path);
5334         kfree(path);
5335         return ret;
5336 }
5337
5338 /*
5339  * ext4_collapse_range:
5340  * This implements the fallocate's collapse range functionality for ext4
5341  * Returns: 0 and non-zero on error.
5342  */
5343 int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5344 {
5345         struct super_block *sb = inode->i_sb;
5346         ext4_lblk_t punch_start, punch_stop;
5347         handle_t *handle;
5348         unsigned int credits;
5349         loff_t new_size, ioffset;
5350         int ret;
5351
5352         /*
5353          * We need to test this early because xfstests assumes that a
5354          * collapse range of (0, 1) will return EOPNOTSUPP if the file
5355          * system does not support collapse range.
5356          */
5357         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5358                 return -EOPNOTSUPP;
5359
5360         /* Collapse range works only on fs block size aligned offsets. */
5361         if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5362             len & (EXT4_CLUSTER_SIZE(sb) - 1))
5363                 return -EINVAL;
5364
5365         if (!S_ISREG(inode->i_mode))
5366                 return -EINVAL;
5367
5368         trace_ext4_collapse_range(inode, offset, len);
5369
5370         punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5371         punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5372
5373         /* Call ext4_force_commit to flush all data in case of data=journal. */
5374         if (ext4_should_journal_data(inode)) {
5375                 ret = ext4_force_commit(inode->i_sb);
5376                 if (ret)
5377                         return ret;
5378         }
5379
5380         inode_lock(inode);
5381         /*
5382          * There is no need to overlap collapse range with EOF, in which case
5383          * it is effectively a truncate operation
5384          */
5385         if (offset + len >= i_size_read(inode)) {
5386                 ret = -EINVAL;
5387                 goto out_mutex;
5388         }
5389
5390         /* Currently just for extent based files */
5391         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5392                 ret = -EOPNOTSUPP;
5393                 goto out_mutex;
5394         }
5395
5396         /* Wait for existing dio to complete */
5397         inode_dio_wait(inode);
5398
5399         /*
5400          * Prevent page faults from reinstantiating pages we have released from
5401          * page cache.
5402          */
5403         down_write(&EXT4_I(inode)->i_mmap_sem);
5404
5405         ret = ext4_break_layouts(inode);
5406         if (ret)
5407                 goto out_mmap;
5408
5409         /*
5410          * Need to round down offset to be aligned with page size boundary
5411          * for page size > block size.
5412          */
5413         ioffset = round_down(offset, PAGE_SIZE);
5414         /*
5415          * Write tail of the last page before removed range since it will get
5416          * removed from the page cache below.
5417          */
5418         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5419         if (ret)
5420                 goto out_mmap;
5421         /*
5422          * Write data that will be shifted to preserve them when discarding
5423          * page cache below. We are also protected from pages becoming dirty
5424          * by i_mmap_sem.
5425          */
5426         ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5427                                            LLONG_MAX);
5428         if (ret)
5429                 goto out_mmap;
5430         truncate_pagecache(inode, ioffset);
5431
5432         credits = ext4_writepage_trans_blocks(inode);
5433         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5434         if (IS_ERR(handle)) {
5435                 ret = PTR_ERR(handle);
5436                 goto out_mmap;
5437         }
5438
5439         down_write(&EXT4_I(inode)->i_data_sem);
5440         ext4_discard_preallocations(inode);
5441
5442         ret = ext4_es_remove_extent(inode, punch_start,
5443                                     EXT_MAX_BLOCKS - punch_start);
5444         if (ret) {
5445                 up_write(&EXT4_I(inode)->i_data_sem);
5446                 goto out_stop;
5447         }
5448
5449         ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5450         if (ret) {
5451                 up_write(&EXT4_I(inode)->i_data_sem);
5452                 goto out_stop;
5453         }
5454         ext4_discard_preallocations(inode);
5455
5456         ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5457                                      punch_stop - punch_start, SHIFT_LEFT);
5458         if (ret) {
5459                 up_write(&EXT4_I(inode)->i_data_sem);
5460                 goto out_stop;
5461         }
5462
5463         new_size = i_size_read(inode) - len;
5464         i_size_write(inode, new_size);
5465         EXT4_I(inode)->i_disksize = new_size;
5466
5467         up_write(&EXT4_I(inode)->i_data_sem);
5468         if (IS_SYNC(inode))
5469                 ext4_handle_sync(handle);
5470         inode->i_mtime = inode->i_ctime = current_time(inode);
5471         ext4_mark_inode_dirty(handle, inode);
5472         ext4_update_inode_fsync_trans(handle, inode, 1);
5473
5474 out_stop:
5475         ext4_journal_stop(handle);
5476 out_mmap:
5477         up_write(&EXT4_I(inode)->i_mmap_sem);
5478 out_mutex:
5479         inode_unlock(inode);
5480         return ret;
5481 }
5482
5483 /*
5484  * ext4_insert_range:
5485  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5486  * The data blocks starting from @offset to the EOF are shifted by @len
5487  * towards right to create a hole in the @inode. Inode size is increased
5488  * by len bytes.
5489  * Returns 0 on success, error otherwise.
5490  */
5491 int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5492 {
5493         struct super_block *sb = inode->i_sb;
5494         handle_t *handle;
5495         struct ext4_ext_path *path;
5496         struct ext4_extent *extent;
5497         ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5498         unsigned int credits, ee_len;
5499         int ret = 0, depth, split_flag = 0;
5500         loff_t ioffset;
5501
5502         /*
5503          * We need to test this early because xfstests assumes that an
5504          * insert range of (0, 1) will return EOPNOTSUPP if the file
5505          * system does not support insert range.
5506          */
5507         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5508                 return -EOPNOTSUPP;
5509
5510         /* Insert range works only on fs block size aligned offsets. */
5511         if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) ||
5512                         len & (EXT4_CLUSTER_SIZE(sb) - 1))
5513                 return -EINVAL;
5514
5515         if (!S_ISREG(inode->i_mode))
5516                 return -EOPNOTSUPP;
5517
5518         trace_ext4_insert_range(inode, offset, len);
5519
5520         offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5521         len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5522
5523         /* Call ext4_force_commit to flush all data in case of data=journal */
5524         if (ext4_should_journal_data(inode)) {
5525                 ret = ext4_force_commit(inode->i_sb);
5526                 if (ret)
5527                         return ret;
5528         }
5529
5530         inode_lock(inode);
5531         /* Currently just for extent based files */
5532         if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5533                 ret = -EOPNOTSUPP;
5534                 goto out_mutex;
5535         }
5536
5537         /* Check for wrap through zero */
5538         if (inode->i_size + len > inode->i_sb->s_maxbytes) {
5539                 ret = -EFBIG;
5540                 goto out_mutex;
5541         }
5542
5543         /* Offset should be less than i_size */
5544         if (offset >= i_size_read(inode)) {
5545                 ret = -EINVAL;
5546                 goto out_mutex;
5547         }
5548
5549         /* Wait for existing dio to complete */
5550         inode_dio_wait(inode);
5551
5552         /*
5553          * Prevent page faults from reinstantiating pages we have released from
5554          * page cache.
5555          */
5556         down_write(&EXT4_I(inode)->i_mmap_sem);
5557
5558         ret = ext4_break_layouts(inode);
5559         if (ret)
5560                 goto out_mmap;
5561
5562         /*
5563          * Need to round down to align start offset to page size boundary
5564          * for page size > block size.
5565          */
5566         ioffset = round_down(offset, PAGE_SIZE);
5567         /* Write out all dirty pages */
5568         ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5569                         LLONG_MAX);
5570         if (ret)
5571                 goto out_mmap;
5572         truncate_pagecache(inode, ioffset);
5573
5574         credits = ext4_writepage_trans_blocks(inode);
5575         handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5576         if (IS_ERR(handle)) {
5577                 ret = PTR_ERR(handle);
5578                 goto out_mmap;
5579         }
5580
5581         /* Expand file to avoid data loss if there is error while shifting */
5582         inode->i_size += len;
5583         EXT4_I(inode)->i_disksize += len;
5584         inode->i_mtime = inode->i_ctime = current_time(inode);
5585         ret = ext4_mark_inode_dirty(handle, inode);
5586         if (ret)
5587                 goto out_stop;
5588
5589         down_write(&EXT4_I(inode)->i_data_sem);
5590         ext4_discard_preallocations(inode);
5591
5592         path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5593         if (IS_ERR(path)) {
5594                 up_write(&EXT4_I(inode)->i_data_sem);
5595                 goto out_stop;
5596         }
5597
5598         depth = ext_depth(inode);
5599         extent = path[depth].p_ext;
5600         if (extent) {
5601                 ee_start_lblk = le32_to_cpu(extent->ee_block);
5602                 ee_len = ext4_ext_get_actual_len(extent);
5603
5604                 /*
5605                  * If offset_lblk is not the starting block of extent, split
5606                  * the extent @offset_lblk
5607                  */
5608                 if ((offset_lblk > ee_start_lblk) &&
5609                                 (offset_lblk < (ee_start_lblk + ee_len))) {
5610                         if (ext4_ext_is_unwritten(extent))
5611                                 split_flag = EXT4_EXT_MARK_UNWRIT1 |
5612                                         EXT4_EXT_MARK_UNWRIT2;
5613                         ret = ext4_split_extent_at(handle, inode, &path,
5614                                         offset_lblk, split_flag,
5615                                         EXT4_EX_NOCACHE |
5616                                         EXT4_GET_BLOCKS_PRE_IO |
5617                                         EXT4_GET_BLOCKS_METADATA_NOFAIL);
5618                 }
5619
5620                 ext4_ext_drop_refs(path);
5621                 kfree(path);
5622                 if (ret < 0) {
5623                         up_write(&EXT4_I(inode)->i_data_sem);
5624                         goto out_stop;
5625                 }
5626         } else {
5627                 ext4_ext_drop_refs(path);
5628                 kfree(path);
5629         }
5630
5631         ret = ext4_es_remove_extent(inode, offset_lblk,
5632                         EXT_MAX_BLOCKS - offset_lblk);
5633         if (ret) {
5634                 up_write(&EXT4_I(inode)->i_data_sem);
5635                 goto out_stop;
5636         }
5637
5638         /*
5639          * if offset_lblk lies in a hole which is at start of file, use
5640          * ee_start_lblk to shift extents
5641          */
5642         ret = ext4_ext_shift_extents(inode, handle,
5643                 ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5644                 len_lblk, SHIFT_RIGHT);
5645
5646         up_write(&EXT4_I(inode)->i_data_sem);
5647         if (IS_SYNC(inode))
5648                 ext4_handle_sync(handle);
5649         if (ret >= 0)
5650                 ext4_update_inode_fsync_trans(handle, inode, 1);
5651
5652 out_stop:
5653         ext4_journal_stop(handle);
5654 out_mmap:
5655         up_write(&EXT4_I(inode)->i_mmap_sem);
5656 out_mutex:
5657         inode_unlock(inode);
5658         return ret;
5659 }
5660
5661 /**
5662  * ext4_swap_extents - Swap extents between two inodes
5663  *
5664  * @inode1:     First inode
5665  * @inode2:     Second inode
5666  * @lblk1:      Start block for first inode
5667  * @lblk2:      Start block for second inode
5668  * @count:      Number of blocks to swap
5669  * @unwritten: Mark second inode's extents as unwritten after swap
5670  * @erp:        Pointer to save error value
5671  *
5672  * This helper routine does exactly what is promise "swap extents". All other
5673  * stuff such as page-cache locking consistency, bh mapping consistency or
5674  * extent's data copying must be performed by caller.
5675  * Locking:
5676  *              i_mutex is held for both inodes
5677  *              i_data_sem is locked for write for both inodes
5678  * Assumptions:
5679  *              All pages from requested range are locked for both inodes
5680  */
5681 int
5682 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5683                   struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5684                   ext4_lblk_t count, int unwritten, int *erp)
5685 {
5686         struct ext4_ext_path *path1 = NULL;
5687         struct ext4_ext_path *path2 = NULL;
5688         int replaced_count = 0;
5689
5690         BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5691         BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5692         BUG_ON(!inode_is_locked(inode1));
5693         BUG_ON(!inode_is_locked(inode2));
5694
5695         *erp = ext4_es_remove_extent(inode1, lblk1, count);
5696         if (unlikely(*erp))
5697                 return 0;
5698         *erp = ext4_es_remove_extent(inode2, lblk2, count);
5699         if (unlikely(*erp))
5700                 return 0;
5701
5702         while (count) {
5703                 struct ext4_extent *ex1, *ex2, tmp_ex;
5704                 ext4_lblk_t e1_blk, e2_blk;
5705                 int e1_len, e2_len, len;
5706                 int split = 0;
5707
5708                 path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5709                 if (IS_ERR(path1)) {
5710                         *erp = PTR_ERR(path1);
5711                         path1 = NULL;
5712                 finish:
5713                         count = 0;
5714                         goto repeat;
5715                 }
5716                 path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5717                 if (IS_ERR(path2)) {
5718                         *erp = PTR_ERR(path2);
5719                         path2 = NULL;
5720                         goto finish;
5721                 }
5722                 ex1 = path1[path1->p_depth].p_ext;
5723                 ex2 = path2[path2->p_depth].p_ext;
5724                 /* Do we have somthing to swap ? */
5725                 if (unlikely(!ex2 || !ex1))
5726                         goto finish;
5727
5728                 e1_blk = le32_to_cpu(ex1->ee_block);
5729                 e2_blk = le32_to_cpu(ex2->ee_block);
5730                 e1_len = ext4_ext_get_actual_len(ex1);
5731                 e2_len = ext4_ext_get_actual_len(ex2);
5732
5733                 /* Hole handling */
5734                 if (!in_range(lblk1, e1_blk, e1_len) ||
5735                     !in_range(lblk2, e2_blk, e2_len)) {
5736                         ext4_lblk_t next1, next2;
5737
5738                         /* if hole after extent, then go to next extent */
5739                         next1 = ext4_ext_next_allocated_block(path1);
5740                         next2 = ext4_ext_next_allocated_block(path2);
5741                         /* If hole before extent, then shift to that extent */
5742                         if (e1_blk > lblk1)
5743                                 next1 = e1_blk;
5744                         if (e2_blk > lblk2)
5745                                 next2 = e2_blk;
5746                         /* Do we have something to swap */
5747                         if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5748                                 goto finish;
5749                         /* Move to the rightest boundary */
5750                         len = next1 - lblk1;
5751                         if (len < next2 - lblk2)
5752                                 len = next2 - lblk2;
5753                         if (len > count)
5754                                 len = count;
5755                         lblk1 += len;
5756                         lblk2 += len;
5757                         count -= len;
5758                         goto repeat;
5759                 }
5760
5761                 /* Prepare left boundary */
5762                 if (e1_blk < lblk1) {
5763                         split = 1;
5764                         *erp = ext4_force_split_extent_at(handle, inode1,
5765                                                 &path1, lblk1, 0);
5766                         if (unlikely(*erp))
5767                                 goto finish;
5768                 }
5769                 if (e2_blk < lblk2) {
5770                         split = 1;
5771                         *erp = ext4_force_split_extent_at(handle, inode2,
5772                                                 &path2,  lblk2, 0);
5773                         if (unlikely(*erp))
5774                                 goto finish;
5775                 }
5776                 /* ext4_split_extent_at() may result in leaf extent split,
5777                  * path must to be revalidated. */
5778                 if (split)
5779                         goto repeat;
5780
5781                 /* Prepare right boundary */
5782                 len = count;
5783                 if (len > e1_blk + e1_len - lblk1)
5784                         len = e1_blk + e1_len - lblk1;
5785                 if (len > e2_blk + e2_len - lblk2)
5786                         len = e2_blk + e2_len - lblk2;
5787
5788                 if (len != e1_len) {
5789                         split = 1;
5790                         *erp = ext4_force_split_extent_at(handle, inode1,
5791                                                 &path1, lblk1 + len, 0);
5792                         if (unlikely(*erp))
5793                                 goto finish;
5794                 }
5795                 if (len != e2_len) {
5796                         split = 1;
5797                         *erp = ext4_force_split_extent_at(handle, inode2,
5798                                                 &path2, lblk2 + len, 0);
5799                         if (*erp)
5800                                 goto finish;
5801                 }
5802                 /* ext4_split_extent_at() may result in leaf extent split,
5803                  * path must to be revalidated. */
5804                 if (split)
5805                         goto repeat;
5806
5807                 BUG_ON(e2_len != e1_len);
5808                 *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5809                 if (unlikely(*erp))
5810                         goto finish;
5811                 *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5812                 if (unlikely(*erp))
5813                         goto finish;
5814
5815                 /* Both extents are fully inside boundaries. Swap it now */
5816                 tmp_ex = *ex1;
5817                 ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5818                 ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5819                 ex1->ee_len = cpu_to_le16(e2_len);
5820                 ex2->ee_len = cpu_to_le16(e1_len);
5821                 if (unwritten)
5822                         ext4_ext_mark_unwritten(ex2);
5823                 if (ext4_ext_is_unwritten(&tmp_ex))
5824                         ext4_ext_mark_unwritten(ex1);
5825
5826                 ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5827                 ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5828                 *erp = ext4_ext_dirty(handle, inode2, path2 +
5829                                       path2->p_depth);
5830                 if (unlikely(*erp))
5831                         goto finish;
5832                 *erp = ext4_ext_dirty(handle, inode1, path1 +
5833                                       path1->p_depth);
5834                 /*
5835                  * Looks scarry ah..? second inode already points to new blocks,
5836                  * and it was successfully dirtied. But luckily error may happen
5837                  * only due to journal error, so full transaction will be
5838                  * aborted anyway.
5839                  */
5840                 if (unlikely(*erp))
5841                         goto finish;
5842                 lblk1 += len;
5843                 lblk2 += len;
5844                 replaced_count += len;
5845                 count -= len;
5846
5847         repeat:
5848                 ext4_ext_drop_refs(path1);
5849                 kfree(path1);
5850                 ext4_ext_drop_refs(path2);
5851                 kfree(path2);
5852                 path1 = path2 = NULL;
5853         }
5854         return replaced_count;
5855 }
5856
5857 /*
5858  * ext4_clu_mapped - determine whether any block in a logical cluster has
5859  *                   been mapped to a physical cluster
5860  *
5861  * @inode - file containing the logical cluster
5862  * @lclu - logical cluster of interest
5863  *
5864  * Returns 1 if any block in the logical cluster is mapped, signifying
5865  * that a physical cluster has been allocated for it.  Otherwise,
5866  * returns 0.  Can also return negative error codes.  Derived from
5867  * ext4_ext_map_blocks().
5868  */
5869 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5870 {
5871         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5872         struct ext4_ext_path *path;
5873         int depth, mapped = 0, err = 0;
5874         struct ext4_extent *extent;
5875         ext4_lblk_t first_lblk, first_lclu, last_lclu;
5876
5877         /* search for the extent closest to the first block in the cluster */
5878         path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5879         if (IS_ERR(path)) {
5880                 err = PTR_ERR(path);
5881                 path = NULL;
5882                 goto out;
5883         }
5884
5885         depth = ext_depth(inode);
5886
5887         /*
5888          * A consistent leaf must not be empty.  This situation is possible,
5889          * though, _during_ tree modification, and it's why an assert can't
5890          * be put in ext4_find_extent().
5891          */
5892         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5893                 EXT4_ERROR_INODE(inode,
5894                     "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5895                                  (unsigned long) EXT4_C2B(sbi, lclu),
5896                                  depth, path[depth].p_block);
5897                 err = -EFSCORRUPTED;
5898                 goto out;
5899         }
5900
5901         extent = path[depth].p_ext;
5902
5903         /* can't be mapped if the extent tree is empty */
5904         if (extent == NULL)
5905                 goto out;
5906
5907         first_lblk = le32_to_cpu(extent->ee_block);
5908         first_lclu = EXT4_B2C(sbi, first_lblk);
5909
5910         /*
5911          * Three possible outcomes at this point - found extent spanning
5912          * the target cluster, to the left of the target cluster, or to the
5913          * right of the target cluster.  The first two cases are handled here.
5914          * The last case indicates the target cluster is not mapped.
5915          */
5916         if (lclu >= first_lclu) {
5917                 last_lclu = EXT4_B2C(sbi, first_lblk +
5918                                      ext4_ext_get_actual_len(extent) - 1);
5919                 if (lclu <= last_lclu) {
5920                         mapped = 1;
5921                 } else {
5922                         first_lblk = ext4_ext_next_allocated_block(path);
5923                         first_lclu = EXT4_B2C(sbi, first_lblk);
5924                         if (lclu == first_lclu)
5925                                 mapped = 1;
5926                 }
5927         }
5928
5929 out:
5930         ext4_ext_drop_refs(path);
5931         kfree(path);
5932
5933         return err ? err : mapped;
5934 }