fs/ext4/extents.c

   1 /*
   2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   3  * Written by Alex Tomas <alex@clusterfs.com>
   4  *
   5  * Architecture independence:
   6  *   Copyright (c) 2005, Bull S.A.
   7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public Licens
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  21  */
  22
  23 /*
  24  * Extents support for EXT4
  25  *
  26  * TODO:
  27  *   - ext4*_error() should be used in some situations
  28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  29  *   - smart tree reduction
  30  */
  31
  32 #include <linux/module.h>
  33 #include <linux/fs.h>
  34 #include <linux/time.h>
  35 #include <linux/jbd2.h>
  36 #include <linux/highuid.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/string.h>
  40 #include <linux/slab.h>
  41 #include <linux/falloc.h>
  42 #include <asm/uaccess.h>
  43 #include <linux/fiemap.h>
  44 #include "ext4_jbd2.h"
  45 #include "ext4_extents.h"
  46
  47 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  48                                             struct inode *inode,
  49                                             int needed)
  50 {
  51         int err;
  52
  53         if (!ext4_handle_valid(handle))
  54                 return 0;
  55         if (handle->h_buffer_credits > needed)
  56                 return 0;
  57         err = ext4_journal_extend(handle, needed);
  58         if (err <= 0)
  59                 return err;
  60         err = ext4_truncate_restart_trans(handle, inode, needed);
  61         if (err == 0)
  62                 err = -EAGAIN;
  63
  64         return err;
  65 }
  66
  67 /*
  68  * could return:
  69  *  - EROFS
  70  *  - ENOMEM
  71  */
  72 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  73                                 struct ext4_ext_path *path)
  74 {
  75         if (path->p_bh) {
  76                 /* path points to block */
  77                 return ext4_journal_get_write_access(handle, path->p_bh);
  78         }
  79         /* path points to leaf/index in inode body */
  80         /* we use in-core data, no need to protect them */
  81         return 0;
  82 }
  83
  84 /*
  85  * could return:
  86  *  - EROFS
  87  *  - ENOMEM
  88  *  - EIO
  89  */
  90 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
  91                                 struct ext4_ext_path *path)
  92 {
  93         int err;
  94         if (path->p_bh) {
  95                 /* path points to block */
  96                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
  97         } else {
  98                 /* path points to leaf/index in inode body */
  99                 err = ext4_mark_inode_dirty(handle, inode);
 100         }
 101         return err;
 102 }
 103
 104 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 105                               struct ext4_ext_path *path,
 106                               ext4_lblk_t block)
 107 {
 108         struct ext4_inode_info *ei = EXT4_I(inode);
 109         ext4_fsblk_t bg_start;
 110         ext4_fsblk_t last_block;
 111         ext4_grpblk_t colour;
 112         ext4_group_t block_group;
 113         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 114         int depth;
 115
 116         if (path) {
 117                 struct ext4_extent *ex;
 118                 depth = path->p_depth;
 119
 120                 /*
 121                  * Try to predict block placement assuming that we are
 122                  * filling in a file which will eventually be
 123                  * non-sparse --- i.e., in the case of libbfd writing
 124                  * an ELF object sections out-of-order but in a way
 125                  * the eventually results in a contiguous object or
 126                  * executable file, or some database extending a table
 127                  * space file.  However, this is actually somewhat
 128                  * non-ideal if we are writing a sparse file such as
 129                  * qemu or KVM writing a raw image file that is going
 130                  * to stay fairly sparse, since it will end up
 131                  * fragmenting the file system's free space.  Maybe we
 132                  * should have some hueristics or some way to allow
 133                  * userspace to pass a hint to file system,
 134                  * especiially if the latter case turns out to be
 135                  * common.
 136                  */
 137                 ex = path[depth].p_ext;
 138                 if (ex) {
 139                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 140                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 141
 142                         if (block > ext_block)
 143                                 return ext_pblk + (block - ext_block);
 144                         else
 145                                 return ext_pblk - (ext_block - block);
 146                 }
 147
 148                 /* it looks like index is empty;
 149                  * try to find starting block from index itself */
 150                 if (path[depth].p_bh)
 151                         return path[depth].p_bh->b_blocknr;
 152         }
 153
 154         /* OK. use inode's group */
 155         block_group = ei->i_block_group;
 156         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 157                 /*
 158                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 159                  * block groups per flexgroup, reserve the first block
 160                  * group for directories and special files.  Regular
 161                  * files will start at the second block group.  This
 162                  * tends to speed up directory access and improves
 163                  * fsck times.
 164                  */
 165                 block_group &= ~(flex_size-1);
 166                 if (S_ISREG(inode->i_mode))
 167                         block_group++;
 168         }
 169         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 170         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 171
 172         /*
 173          * If we are doing delayed allocation, we don't need take
 174          * colour into account.
 175          */
 176         if (test_opt(inode->i_sb, DELALLOC))
 177                 return bg_start;
 178
 179         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 180                 colour = (current->pid % 16) *
 181                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 182         else
 183                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 184         return bg_start + colour + block;
 185 }
 186
 187 /*
 188  * Allocation for a meta data block
 189  */
 190 static ext4_fsblk_t
 191 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 192                         struct ext4_ext_path *path,
 193                         struct ext4_extent *ex, int *err)
 194 {
 195         ext4_fsblk_t goal, newblock;
 196
 197         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 198         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
 199         return newblock;
 200 }
 201
 202 static inline int ext4_ext_space_block(struct inode *inode, int check)
 203 {
 204         int size;
 205
 206         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 207                         / sizeof(struct ext4_extent);
 208         if (!check) {
 209 #ifdef AGGRESSIVE_TEST
 210                 if (size > 6)
 211                         size = 6;
 212 #endif
 213         }
 214         return size;
 215 }
 216
 217 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 218 {
 219         int size;
 220
 221         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 222                         / sizeof(struct ext4_extent_idx);
 223         if (!check) {
 224 #ifdef AGGRESSIVE_TEST
 225                 if (size > 5)
 226                         size = 5;
 227 #endif
 228         }
 229         return size;
 230 }
 231
 232 static inline int ext4_ext_space_root(struct inode *inode, int check)
 233 {
 234         int size;
 235
 236         size = sizeof(EXT4_I(inode)->i_data);
 237         size -= sizeof(struct ext4_extent_header);
 238         size /= sizeof(struct ext4_extent);
 239         if (!check) {
 240 #ifdef AGGRESSIVE_TEST
 241                 if (size > 3)
 242                         size = 3;
 243 #endif
 244         }
 245         return size;
 246 }
 247
 248 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 249 {
 250         int size;
 251
 252         size = sizeof(EXT4_I(inode)->i_data);
 253         size -= sizeof(struct ext4_extent_header);
 254         size /= sizeof(struct ext4_extent_idx);
 255         if (!check) {
 256 #ifdef AGGRESSIVE_TEST
 257                 if (size > 4)
 258                         size = 4;
 259 #endif
 260         }
 261         return size;
 262 }
 263
 264 /*
 265  * Calculate the number of metadata blocks needed
 266  * to allocate @blocks
 267  * Worse case is one block per extent
 268  */
 269 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 270 {
 271         struct ext4_inode_info *ei = EXT4_I(inode);
 272         int idxs, num = 0;
 273
 274         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 275                 / sizeof(struct ext4_extent_idx));
 276
 277         /*
 278          * If the new delayed allocation block is contiguous with the
 279          * previous da block, it can share index blocks with the
 280          * previous block, so we only need to allocate a new index
 281          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 282          * an additional index block, and at ldxs**3 blocks, yet
 283          * another index blocks.
 284          */
 285         if (ei->i_da_metadata_calc_len &&
 286             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 287                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 288                         num++;
 289                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 290                         num++;
 291                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 292                         num++;
 293                         ei->i_da_metadata_calc_len = 0;
 294                 } else
 295                         ei->i_da_metadata_calc_len++;
 296                 ei->i_da_metadata_calc_last_lblock++;
 297                 return num;
 298         }
 299
 300         /*
 301          * In the worst case we need a new set of index blocks at
 302          * every level of the inode's extent tree.
 303          */
 304         ei->i_da_metadata_calc_len = 1;
 305         ei->i_da_metadata_calc_last_lblock = lblock;
 306         return ext_depth(inode) + 1;
 307 }
 308
 309 static int
 310 ext4_ext_max_entries(struct inode *inode, int depth)
 311 {
 312         int max;
 313
 314         if (depth == ext_depth(inode)) {
 315                 if (depth == 0)
 316                         max = ext4_ext_space_root(inode, 1);
 317                 else
 318                         max = ext4_ext_space_root_idx(inode, 1);
 319         } else {
 320                 if (depth == 0)
 321                         max = ext4_ext_space_block(inode, 1);
 322                 else
 323                         max = ext4_ext_space_block_idx(inode, 1);
 324         }
 325
 326         return max;
 327 }
 328
 329 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 330 {
 331         ext4_fsblk_t block = ext4_ext_pblock(ext);
 332         int len = ext4_ext_get_actual_len(ext);
 333
 334         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 335 }
 336
 337 static int ext4_valid_extent_idx(struct inode *inode,
 338                                 struct ext4_extent_idx *ext_idx)
 339 {
 340         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 341
 342         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 343 }
 344
 345 static int ext4_valid_extent_entries(struct inode *inode,
 346                                 struct ext4_extent_header *eh,
 347                                 int depth)
 348 {
 349         struct ext4_extent *ext;
 350         struct ext4_extent_idx *ext_idx;
 351         unsigned short entries;
 352         if (eh->eh_entries == 0)
 353                 return 1;
 354
 355         entries = le16_to_cpu(eh->eh_entries);
 356
 357         if (depth == 0) {
 358                 /* leaf entries */
 359                 ext = EXT_FIRST_EXTENT(eh);
 360                 while (entries) {
 361                         if (!ext4_valid_extent(inode, ext))
 362                                 return 0;
 363                         ext++;
 364                         entries--;
 365                 }
 366         } else {
 367                 ext_idx = EXT_FIRST_INDEX(eh);
 368                 while (entries) {
 369                         if (!ext4_valid_extent_idx(inode, ext_idx))
 370                                 return 0;
 371                         ext_idx++;
 372                         entries--;
 373                 }
 374         }
 375         return 1;
 376 }
 377
 378 static int __ext4_ext_check(const char *function, unsigned int line,
 379                             struct inode *inode, struct ext4_extent_header *eh,
 380                             int depth)
 381 {
 382         const char *error_msg;
 383         int max = 0;
 384
 385         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 386                 error_msg = "invalid magic";
 387                 goto corrupted;
 388         }
 389         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 390                 error_msg = "unexpected eh_depth";
 391                 goto corrupted;
 392         }
 393         if (unlikely(eh->eh_max == 0)) {
 394                 error_msg = "invalid eh_max";
 395                 goto corrupted;
 396         }
 397         max = ext4_ext_max_entries(inode, depth);
 398         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 399                 error_msg = "too large eh_max";
 400                 goto corrupted;
 401         }
 402         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 403                 error_msg = "invalid eh_entries";
 404                 goto corrupted;
 405         }
 406         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 407                 error_msg = "invalid extent entries";
 408                 goto corrupted;
 409         }
 410         return 0;
 411
 412 corrupted:
 413         ext4_error_inode(inode, function, line, 0,
 414                         "bad header/extent: %s - magic %x, "
 415                         "entries %u, max %u(%u), depth %u(%u)",
 416                         error_msg, le16_to_cpu(eh->eh_magic),
 417                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 418                         max, le16_to_cpu(eh->eh_depth), depth);
 419
 420         return -EIO;
 421 }
 422
 423 #define ext4_ext_check(inode, eh, depth)        \
 424         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 425
 426 int ext4_ext_check_inode(struct inode *inode)
 427 {
 428         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 429 }
 430
 431 #ifdef EXT_DEBUG
 432 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 433 {
 434         int k, l = path->p_depth;
 435
 436         ext_debug("path:");
 437         for (k = 0; k <= l; k++, path++) {
 438                 if (path->p_idx) {
 439                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 440                             ext4_idx_pblock(path->p_idx));
 441                 } else if (path->p_ext) {
 442                         ext_debug("  %d:[%d]%d:%llu ",
 443                                   le32_to_cpu(path->p_ext->ee_block),
 444                                   ext4_ext_is_uninitialized(path->p_ext),
 445                                   ext4_ext_get_actual_len(path->p_ext),
 446                                   ext4_ext_pblock(path->p_ext));
 447                 } else
 448                         ext_debug("  []");
 449         }
 450         ext_debug("\n");
 451 }
 452
 453 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 454 {
 455         int depth = ext_depth(inode);
 456         struct ext4_extent_header *eh;
 457         struct ext4_extent *ex;
 458         int i;
 459
 460         if (!path)
 461                 return;
 462
 463         eh = path[depth].p_hdr;
 464         ex = EXT_FIRST_EXTENT(eh);
 465
 466         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 467
 468         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 469                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 470                           ext4_ext_is_uninitialized(ex),
 471                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 472         }
 473         ext_debug("\n");
 474 }
 475 #else
 476 #define ext4_ext_show_path(inode, path)
 477 #define ext4_ext_show_leaf(inode, path)
 478 #endif
 479
 480 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 481 {
 482         int depth = path->p_depth;
 483         int i;
 484
 485         for (i = 0; i <= depth; i++, path++)
 486                 if (path->p_bh) {
 487                         brelse(path->p_bh);
 488                         path->p_bh = NULL;
 489                 }
 490 }
 491
 492 /*
 493  * ext4_ext_binsearch_idx:
 494  * binary search for the closest index of the given block
 495  * the header must be checked before calling this
 496  */
 497 static void
 498 ext4_ext_binsearch_idx(struct inode *inode,
 499                         struct ext4_ext_path *path, ext4_lblk_t block)
 500 {
 501         struct ext4_extent_header *eh = path->p_hdr;
 502         struct ext4_extent_idx *r, *l, *m;
 503
 504
 505         ext_debug("binsearch for %u(idx):  ", block);
 506
 507         l = EXT_FIRST_INDEX(eh) + 1;
 508         r = EXT_LAST_INDEX(eh);
 509         while (l <= r) {
 510                 m = l + (r - l) / 2;
 511                 if (block < le32_to_cpu(m->ei_block))
 512                         r = m - 1;
 513                 else
 514                         l = m + 1;
 515                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 516                                 m, le32_to_cpu(m->ei_block),
 517                                 r, le32_to_cpu(r->ei_block));
 518         }
 519
 520         path->p_idx = l - 1;
 521         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 522                   ext4_idx_pblock(path->p_idx));
 523
 524 #ifdef CHECK_BINSEARCH
 525         {
 526                 struct ext4_extent_idx *chix, *ix;
 527                 int k;
 528
 529                 chix = ix = EXT_FIRST_INDEX(eh);
 530                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 531                   if (k != 0 &&
 532                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 533                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 534                                        "first=0x%p\n", k,
 535                                        ix, EXT_FIRST_INDEX(eh));
 536                                 printk(KERN_DEBUG "%u <= %u\n",
 537                                        le32_to_cpu(ix->ei_block),
 538                                        le32_to_cpu(ix[-1].ei_block));
 539                         }
 540                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 541                                            <= le32_to_cpu(ix[-1].ei_block));
 542                         if (block < le32_to_cpu(ix->ei_block))
 543                                 break;
 544                         chix = ix;
 545                 }
 546                 BUG_ON(chix != path->p_idx);
 547         }
 548 #endif
 549
 550 }
 551
 552 /*
 553  * ext4_ext_binsearch:
 554  * binary search for closest extent of the given block
 555  * the header must be checked before calling this
 556  */
 557 static void
 558 ext4_ext_binsearch(struct inode *inode,
 559                 struct ext4_ext_path *path, ext4_lblk_t block)
 560 {
 561         struct ext4_extent_header *eh = path->p_hdr;
 562         struct ext4_extent *r, *l, *m;
 563
 564         if (eh->eh_entries == 0) {
 565                 /*
 566                  * this leaf is empty:
 567                  * we get such a leaf in split/add case
 568                  */
 569                 return;
 570         }
 571
 572         ext_debug("binsearch for %u:  ", block);
 573
 574         l = EXT_FIRST_EXTENT(eh) + 1;
 575         r = EXT_LAST_EXTENT(eh);
 576
 577         while (l <= r) {
 578                 m = l + (r - l) / 2;
 579                 if (block < le32_to_cpu(m->ee_block))
 580                         r = m - 1;
 581                 else
 582                         l = m + 1;
 583                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 584                                 m, le32_to_cpu(m->ee_block),
 585                                 r, le32_to_cpu(r->ee_block));
 586         }
 587
 588         path->p_ext = l - 1;
 589         ext_debug("  -> %d:%llu:[%d]%d ",
 590                         le32_to_cpu(path->p_ext->ee_block),
 591                         ext4_ext_pblock(path->p_ext),
 592                         ext4_ext_is_uninitialized(path->p_ext),
 593                         ext4_ext_get_actual_len(path->p_ext));
 594
 595 #ifdef CHECK_BINSEARCH
 596         {
 597                 struct ext4_extent *chex, *ex;
 598                 int k;
 599
 600                 chex = ex = EXT_FIRST_EXTENT(eh);
 601                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 602                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 603                                           <= le32_to_cpu(ex[-1].ee_block));
 604                         if (block < le32_to_cpu(ex->ee_block))
 605                                 break;
 606                         chex = ex;
 607                 }
 608                 BUG_ON(chex != path->p_ext);
 609         }
 610 #endif
 611
 612 }
 613
 614 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 615 {
 616         struct ext4_extent_header *eh;
 617
 618         eh = ext_inode_hdr(inode);
 619         eh->eh_depth = 0;
 620         eh->eh_entries = 0;
 621         eh->eh_magic = EXT4_EXT_MAGIC;
 622         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 623         ext4_mark_inode_dirty(handle, inode);
 624         ext4_ext_invalidate_cache(inode);
 625         return 0;
 626 }
 627
 628 struct ext4_ext_path *
 629 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 630                                         struct ext4_ext_path *path)
 631 {
 632         struct ext4_extent_header *eh;
 633         struct buffer_head *bh;
 634         short int depth, i, ppos = 0, alloc = 0;
 635
 636         eh = ext_inode_hdr(inode);
 637         depth = ext_depth(inode);
 638
 639         /* account possible depth increase */
 640         if (!path) {
 641                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 642                                 GFP_NOFS);
 643                 if (!path)
 644                         return ERR_PTR(-ENOMEM);
 645                 alloc = 1;
 646         }
 647         path[0].p_hdr = eh;
 648         path[0].p_bh = NULL;
 649
 650         i = depth;
 651         /* walk through the tree */
 652         while (i) {
 653                 int need_to_validate = 0;
 654
 655                 ext_debug("depth %d: num %d, max %d\n",
 656                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 657
 658                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 659                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 660                 path[ppos].p_depth = i;
 661                 path[ppos].p_ext = NULL;
 662
 663                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 664                 if (unlikely(!bh))
 665                         goto err;
 666                 if (!bh_uptodate_or_lock(bh)) {
 667                         if (bh_submit_read(bh) < 0) {
 668                                 put_bh(bh);
 669                                 goto err;
 670                         }
 671                         /* validate the extent entries */
 672                         need_to_validate = 1;
 673                 }
 674                 eh = ext_block_hdr(bh);
 675                 ppos++;
 676                 if (unlikely(ppos > depth)) {
 677                         put_bh(bh);
 678                         EXT4_ERROR_INODE(inode,
 679                                          "ppos %d > depth %d", ppos, depth);
 680                         goto err;
 681                 }
 682                 path[ppos].p_bh = bh;
 683                 path[ppos].p_hdr = eh;
 684                 i--;
 685
 686                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 687                         goto err;
 688         }
 689
 690         path[ppos].p_depth = i;
 691         path[ppos].p_ext = NULL;
 692         path[ppos].p_idx = NULL;
 693
 694         /* find extent */
 695         ext4_ext_binsearch(inode, path + ppos, block);
 696         /* if not an empty leaf */
 697         if (path[ppos].p_ext)
 698                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 699
 700         ext4_ext_show_path(inode, path);
 701
 702         return path;
 703
 704 err:
 705         ext4_ext_drop_refs(path);
 706         if (alloc)
 707                 kfree(path);
 708         return ERR_PTR(-EIO);
 709 }
 710
 711 /*
 712  * ext4_ext_insert_index:
 713  * insert new index [@logical;@ptr] into the block at @curp;
 714  * check where to insert: before @curp or after @curp
 715  */
 716 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 717                                  struct ext4_ext_path *curp,
 718                                  int logical, ext4_fsblk_t ptr)
 719 {
 720         struct ext4_extent_idx *ix;
 721         int len, err;
 722
 723         err = ext4_ext_get_access(handle, inode, curp);
 724         if (err)
 725                 return err;
 726
 727         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 728                 EXT4_ERROR_INODE(inode,
 729                                  "logical %d == ei_block %d!",
 730                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 731                 return -EIO;
 732         }
 733         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 734         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 735                 /* insert after */
 736                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 737                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 738                         len = len < 0 ? 0 : len;
 739                         ext_debug("insert new index %d after: %llu. "
 740                                         "move %d from 0x%p to 0x%p\n",
 741                                         logical, ptr, len,
 742                                         (curp->p_idx + 1), (curp->p_idx + 2));
 743                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 744                 }
 745                 ix = curp->p_idx + 1;
 746         } else {
 747                 /* insert before */
 748                 len = len * sizeof(struct ext4_extent_idx);
 749                 len = len < 0 ? 0 : len;
 750                 ext_debug("insert new index %d before: %llu. "
 751                                 "move %d from 0x%p to 0x%p\n",
 752                                 logical, ptr, len,
 753                                 curp->p_idx, (curp->p_idx + 1));
 754                 memmove(curp->p_idx + 1, curp->p_idx, len);
 755                 ix = curp->p_idx;
 756         }
 757
 758         ix->ei_block = cpu_to_le32(logical);
 759         ext4_idx_store_pblock(ix, ptr);
 760         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 761
 762         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 763                              > le16_to_cpu(curp->p_hdr->eh_max))) {
 764                 EXT4_ERROR_INODE(inode,
 765                                  "logical %d == ei_block %d!",
 766                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 767                 return -EIO;
 768         }
 769         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 770                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 771                 return -EIO;
 772         }
 773
 774         err = ext4_ext_dirty(handle, inode, curp);
 775         ext4_std_error(inode->i_sb, err);
 776
 777         return err;
 778 }
 779
 780 /*
 781  * ext4_ext_split:
 782  * inserts new subtree into the path, using free index entry
 783  * at depth @at:
 784  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 785  * - makes decision where to split
 786  * - moves remaining extents and index entries (right to the split point)
 787  *   into the newly allocated blocks
 788  * - initializes subtree
 789  */
 790 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 791                                 struct ext4_ext_path *path,
 792                                 struct ext4_extent *newext, int at)
 793 {
 794         struct buffer_head *bh = NULL;
 795         int depth = ext_depth(inode);
 796         struct ext4_extent_header *neh;
 797         struct ext4_extent_idx *fidx;
 798         struct ext4_extent *ex;
 799         int i = at, k, m, a;
 800         ext4_fsblk_t newblock, oldblock;
 801         __le32 border;
 802         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 803         int err = 0;
 804
 805         /* make decision: where to split? */
 806         /* FIXME: now decision is simplest: at current extent */
 807
 808         /* if current leaf will be split, then we should use
 809          * border from split point */
 810         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 811                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 812                 return -EIO;
 813         }
 814         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 815                 border = path[depth].p_ext[1].ee_block;
 816                 ext_debug("leaf will be split."
 817                                 " next leaf starts at %d\n",
 818                                   le32_to_cpu(border));
 819         } else {
 820                 border = newext->ee_block;
 821                 ext_debug("leaf will be added."
 822                                 " next leaf starts at %d\n",
 823                                 le32_to_cpu(border));
 824         }
 825
 826         /*
 827          * If error occurs, then we break processing
 828          * and mark filesystem read-only. index won't
 829          * be inserted and tree will be in consistent
 830          * state. Next mount will repair buffers too.
 831          */
 832
 833         /*
 834          * Get array to track all allocated blocks.
 835          * We need this to handle errors and free blocks
 836          * upon them.
 837          */
 838         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 839         if (!ablocks)
 840                 return -ENOMEM;
 841
 842         /* allocate all needed blocks */
 843         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 844         for (a = 0; a < depth - at; a++) {
 845                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 846                                                    newext, &err);
 847                 if (newblock == 0)
 848                         goto cleanup;
 849                 ablocks[a] = newblock;
 850         }
 851
 852         /* initialize new leaf */
 853         newblock = ablocks[--a];
 854         if (unlikely(newblock == 0)) {
 855                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 856                 err = -EIO;
 857                 goto cleanup;
 858         }
 859         bh = sb_getblk(inode->i_sb, newblock);
 860         if (!bh) {
 861                 err = -EIO;
 862                 goto cleanup;
 863         }
 864         lock_buffer(bh);
 865
 866         err = ext4_journal_get_create_access(handle, bh);
 867         if (err)
 868                 goto cleanup;
 869
 870         neh = ext_block_hdr(bh);
 871         neh->eh_entries = 0;
 872         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 873         neh->eh_magic = EXT4_EXT_MAGIC;
 874         neh->eh_depth = 0;
 875         ex = EXT_FIRST_EXTENT(neh);
 876
 877         /* move remainder of path[depth] to the new leaf */
 878         if (unlikely(path[depth].p_hdr->eh_entries !=
 879                      path[depth].p_hdr->eh_max)) {
 880                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 881                                  path[depth].p_hdr->eh_entries,
 882                                  path[depth].p_hdr->eh_max);
 883                 err = -EIO;
 884                 goto cleanup;
 885         }
 886         /* start copy from next extent */
 887         /* TODO: we could do it by single memmove */
 888         m = 0;
 889         path[depth].p_ext++;
 890         while (path[depth].p_ext <=
 891                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
 892                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 893                                 le32_to_cpu(path[depth].p_ext->ee_block),
 894                                 ext4_ext_pblock(path[depth].p_ext),
 895                                 ext4_ext_is_uninitialized(path[depth].p_ext),
 896                                 ext4_ext_get_actual_len(path[depth].p_ext),
 897                                 newblock);
 898                 /*memmove(ex++, path[depth].p_ext++,
 899                                 sizeof(struct ext4_extent));
 900                 neh->eh_entries++;*/
 901                 path[depth].p_ext++;
 902                 m++;
 903         }
 904         if (m) {
 905                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
 906                 le16_add_cpu(&neh->eh_entries, m);
 907         }
 908
 909         set_buffer_uptodate(bh);
 910         unlock_buffer(bh);
 911
 912         err = ext4_handle_dirty_metadata(handle, inode, bh);
 913         if (err)
 914                 goto cleanup;
 915         brelse(bh);
 916         bh = NULL;
 917
 918         /* correct old leaf */
 919         if (m) {
 920                 err = ext4_ext_get_access(handle, inode, path + depth);
 921                 if (err)
 922                         goto cleanup;
 923                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 924                 err = ext4_ext_dirty(handle, inode, path + depth);
 925                 if (err)
 926                         goto cleanup;
 927
 928         }
 929
 930         /* create intermediate indexes */
 931         k = depth - at - 1;
 932         if (unlikely(k < 0)) {
 933                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 934                 err = -EIO;
 935                 goto cleanup;
 936         }
 937         if (k)
 938                 ext_debug("create %d intermediate indices\n", k);
 939         /* insert new index into current index block */
 940         /* current depth stored in i var */
 941         i = depth - 1;
 942         while (k--) {
 943                 oldblock = newblock;
 944                 newblock = ablocks[--a];
 945                 bh = sb_getblk(inode->i_sb, newblock);
 946                 if (!bh) {
 947                         err = -EIO;
 948                         goto cleanup;
 949                 }
 950                 lock_buffer(bh);
 951
 952                 err = ext4_journal_get_create_access(handle, bh);
 953                 if (err)
 954                         goto cleanup;
 955
 956                 neh = ext_block_hdr(bh);
 957                 neh->eh_entries = cpu_to_le16(1);
 958                 neh->eh_magic = EXT4_EXT_MAGIC;
 959                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 960                 neh->eh_depth = cpu_to_le16(depth - i);
 961                 fidx = EXT_FIRST_INDEX(neh);
 962                 fidx->ei_block = border;
 963                 ext4_idx_store_pblock(fidx, oldblock);
 964
 965                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 966                                 i, newblock, le32_to_cpu(border), oldblock);
 967                 /* copy indexes */
 968                 m = 0;
 969                 path[i].p_idx++;
 970
 971                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 972                                 EXT_MAX_INDEX(path[i].p_hdr));
 973                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 974                                         EXT_LAST_INDEX(path[i].p_hdr))) {
 975                         EXT4_ERROR_INODE(inode,
 976                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 977                                          le32_to_cpu(path[i].p_ext->ee_block));
 978                         err = -EIO;
 979                         goto cleanup;
 980                 }
 981                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
 982                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
 983                                         le32_to_cpu(path[i].p_idx->ei_block),
 984                                         ext4_idx_pblock(path[i].p_idx),
 985                                         newblock);
 986                         /*memmove(++fidx, path[i].p_idx++,
 987                                         sizeof(struct ext4_extent_idx));
 988                         neh->eh_entries++;
 989                         BUG_ON(neh->eh_entries > neh->eh_max);*/
 990                         path[i].p_idx++;
 991                         m++;
 992                 }
 993                 if (m) {
 994                         memmove(++fidx, path[i].p_idx - m,
 995                                 sizeof(struct ext4_extent_idx) * m);
 996                         le16_add_cpu(&neh->eh_entries, m);
 997                 }
 998                 set_buffer_uptodate(bh);
 999                 unlock_buffer(bh);
1000
1001                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1002                 if (err)
1003                         goto cleanup;
1004                 brelse(bh);
1005                 bh = NULL;
1006
1007                 /* correct old index */
1008                 if (m) {
1009                         err = ext4_ext_get_access(handle, inode, path + i);
1010                         if (err)
1011                                 goto cleanup;
1012                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1013                         err = ext4_ext_dirty(handle, inode, path + i);
1014                         if (err)
1015                                 goto cleanup;
1016                 }
1017
1018                 i--;
1019         }
1020
1021         /* insert new index */
1022         err = ext4_ext_insert_index(handle, inode, path + at,
1023                                     le32_to_cpu(border), newblock);
1024
1025 cleanup:
1026         if (bh) {
1027                 if (buffer_locked(bh))
1028                         unlock_buffer(bh);
1029                 brelse(bh);
1030         }
1031
1032         if (err) {
1033                 /* free all allocated blocks in error case */
1034                 for (i = 0; i < depth; i++) {
1035                         if (!ablocks[i])
1036                                 continue;
1037                         ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1038                                          EXT4_FREE_BLOCKS_METADATA);
1039                 }
1040         }
1041         kfree(ablocks);
1042
1043         return err;
1044 }
1045
1046 /*
1047  * ext4_ext_grow_indepth:
1048  * implements tree growing procedure:
1049  * - allocates new block
1050  * - moves top-level data (index block or leaf) into the new block
1051  * - initializes new top-level, creating index that points to the
1052  *   just created block
1053  */
1054 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1055                                         struct ext4_ext_path *path,
1056                                         struct ext4_extent *newext)
1057 {
1058         struct ext4_ext_path *curp = path;
1059         struct ext4_extent_header *neh;
1060         struct buffer_head *bh;
1061         ext4_fsblk_t newblock;
1062         int err = 0;
1063
1064         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1065         if (newblock == 0)
1066                 return err;
1067
1068         bh = sb_getblk(inode->i_sb, newblock);
1069         if (!bh) {
1070                 err = -EIO;
1071                 ext4_std_error(inode->i_sb, err);
1072                 return err;
1073         }
1074         lock_buffer(bh);
1075
1076         err = ext4_journal_get_create_access(handle, bh);
1077         if (err) {
1078                 unlock_buffer(bh);
1079                 goto out;
1080         }
1081
1082         /* move top-level index/leaf into new block */
1083         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1084
1085         /* set size of new block */
1086         neh = ext_block_hdr(bh);
1087         /* old root could have indexes or leaves
1088          * so calculate e_max right way */
1089         if (ext_depth(inode))
1090                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1091         else
1092                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1093         neh->eh_magic = EXT4_EXT_MAGIC;
1094         set_buffer_uptodate(bh);
1095         unlock_buffer(bh);
1096
1097         err = ext4_handle_dirty_metadata(handle, inode, bh);
1098         if (err)
1099                 goto out;
1100
1101         /* create index in new top-level index: num,max,pointer */
1102         err = ext4_ext_get_access(handle, inode, curp);
1103         if (err)
1104                 goto out;
1105
1106         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1107         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1108         curp->p_hdr->eh_entries = cpu_to_le16(1);
1109         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1110
1111         if (path[0].p_hdr->eh_depth)
1112                 curp->p_idx->ei_block =
1113                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1114         else
1115                 curp->p_idx->ei_block =
1116                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1117         ext4_idx_store_pblock(curp->p_idx, newblock);
1118
1119         neh = ext_inode_hdr(inode);
1120         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1121                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1122                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1123                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1124
1125         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1126         err = ext4_ext_dirty(handle, inode, curp);
1127 out:
1128         brelse(bh);
1129
1130         return err;
1131 }
1132
1133 /*
1134  * ext4_ext_create_new_leaf:
1135  * finds empty index and adds new leaf.
1136  * if no free index is found, then it requests in-depth growing.
1137  */
1138 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1139                                         struct ext4_ext_path *path,
1140                                         struct ext4_extent *newext)
1141 {
1142         struct ext4_ext_path *curp;
1143         int depth, i, err = 0;
1144
1145 repeat:
1146         i = depth = ext_depth(inode);
1147
1148         /* walk up to the tree and look for free index entry */
1149         curp = path + depth;
1150         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1151                 i--;
1152                 curp--;
1153         }
1154
1155         /* we use already allocated block for index block,
1156          * so subsequent data blocks should be contiguous */
1157         if (EXT_HAS_FREE_INDEX(curp)) {
1158                 /* if we found index with free entry, then use that
1159                  * entry: create all needed subtree and add new leaf */
1160                 err = ext4_ext_split(handle, inode, path, newext, i);
1161                 if (err)
1162                         goto out;
1163
1164                 /* refill path */
1165                 ext4_ext_drop_refs(path);
1166                 path = ext4_ext_find_extent(inode,
1167                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1168                                     path);
1169                 if (IS_ERR(path))
1170                         err = PTR_ERR(path);
1171         } else {
1172                 /* tree is full, time to grow in depth */
1173                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1174                 if (err)
1175                         goto out;
1176
1177                 /* refill path */
1178                 ext4_ext_drop_refs(path);
1179                 path = ext4_ext_find_extent(inode,
1180                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1181                                     path);
1182                 if (IS_ERR(path)) {
1183                         err = PTR_ERR(path);
1184                         goto out;
1185                 }
1186
1187                 /*
1188                  * only first (depth 0 -> 1) produces free space;
1189                  * in all other cases we have to split the grown tree
1190                  */
1191                 depth = ext_depth(inode);
1192                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1193                         /* now we need to split */
1194                         goto repeat;
1195                 }
1196         }
1197
1198 out:
1199         return err;
1200 }
1201
1202 /*
1203  * search the closest allocated block to the left for *logical
1204  * and returns it at @logical + it's physical address at @phys
1205  * if *logical is the smallest allocated block, the function
1206  * returns 0 at @phys
1207  * return value contains 0 (success) or error code
1208  */
1209 static int ext4_ext_search_left(struct inode *inode,
1210                                 struct ext4_ext_path *path,
1211                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1212 {
1213         struct ext4_extent_idx *ix;
1214         struct ext4_extent *ex;
1215         int depth, ee_len;
1216
1217         if (unlikely(path == NULL)) {
1218                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1219                 return -EIO;
1220         }
1221         depth = path->p_depth;
1222         *phys = 0;
1223
1224         if (depth == 0 && path->p_ext == NULL)
1225                 return 0;
1226
1227         /* usually extent in the path covers blocks smaller
1228          * then *logical, but it can be that extent is the
1229          * first one in the file */
1230
1231         ex = path[depth].p_ext;
1232         ee_len = ext4_ext_get_actual_len(ex);
1233         if (*logical < le32_to_cpu(ex->ee_block)) {
1234                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1235                         EXT4_ERROR_INODE(inode,
1236                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1237                                          *logical, le32_to_cpu(ex->ee_block));
1238                         return -EIO;
1239                 }
1240                 while (--depth >= 0) {
1241                         ix = path[depth].p_idx;
1242                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1243                                 EXT4_ERROR_INODE(inode,
1244                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1245                                   ix != NULL ? ix->ei_block : 0,
1246                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1247                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1248                                   depth);
1249                                 return -EIO;
1250                         }
1251                 }
1252                 return 0;
1253         }
1254
1255         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1256                 EXT4_ERROR_INODE(inode,
1257                                  "logical %d < ee_block %d + ee_len %d!",
1258                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1259                 return -EIO;
1260         }
1261
1262         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1263         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1264         return 0;
1265 }
1266
1267 /*
1268  * search the closest allocated block to the right for *logical
1269  * and returns it at @logical + it's physical address at @phys
1270  * if *logical is the smallest allocated block, the function
1271  * returns 0 at @phys
1272  * return value contains 0 (success) or error code
1273  */
1274 static int ext4_ext_search_right(struct inode *inode,
1275                                  struct ext4_ext_path *path,
1276                                  ext4_lblk_t *logical, ext4_fsblk_t *phys)
1277 {
1278         struct buffer_head *bh = NULL;
1279         struct ext4_extent_header *eh;
1280         struct ext4_extent_idx *ix;
1281         struct ext4_extent *ex;
1282         ext4_fsblk_t block;
1283         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1284         int ee_len;
1285
1286         if (unlikely(path == NULL)) {
1287                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1288                 return -EIO;
1289         }
1290         depth = path->p_depth;
1291         *phys = 0;
1292
1293         if (depth == 0 && path->p_ext == NULL)
1294                 return 0;
1295
1296         /* usually extent in the path covers blocks smaller
1297          * then *logical, but it can be that extent is the
1298          * first one in the file */
1299
1300         ex = path[depth].p_ext;
1301         ee_len = ext4_ext_get_actual_len(ex);
1302         if (*logical < le32_to_cpu(ex->ee_block)) {
1303                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1304                         EXT4_ERROR_INODE(inode,
1305                                          "first_extent(path[%d].p_hdr) != ex",
1306                                          depth);
1307                         return -EIO;
1308                 }
1309                 while (--depth >= 0) {
1310                         ix = path[depth].p_idx;
1311                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1312                                 EXT4_ERROR_INODE(inode,
1313                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1314                                                  *logical);
1315                                 return -EIO;
1316                         }
1317                 }
1318                 *logical = le32_to_cpu(ex->ee_block);
1319                 *phys = ext4_ext_pblock(ex);
1320                 return 0;
1321         }
1322
1323         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1324                 EXT4_ERROR_INODE(inode,
1325                                  "logical %d < ee_block %d + ee_len %d!",
1326                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1327                 return -EIO;
1328         }
1329
1330         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1331                 /* next allocated block in this leaf */
1332                 ex++;
1333                 *logical = le32_to_cpu(ex->ee_block);
1334                 *phys = ext4_ext_pblock(ex);
1335                 return 0;
1336         }
1337
1338         /* go up and search for index to the right */
1339         while (--depth >= 0) {
1340                 ix = path[depth].p_idx;
1341                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1342                         goto got_index;
1343         }
1344
1345         /* we've gone up to the root and found no index to the right */
1346         return 0;
1347
1348 got_index:
1349         /* we've found index to the right, let's
1350          * follow it and find the closest allocated
1351          * block to the right */
1352         ix++;
1353         block = ext4_idx_pblock(ix);
1354         while (++depth < path->p_depth) {
1355                 bh = sb_bread(inode->i_sb, block);
1356                 if (bh == NULL)
1357                         return -EIO;
1358                 eh = ext_block_hdr(bh);
1359                 /* subtract from p_depth to get proper eh_depth */
1360                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1361                         put_bh(bh);
1362                         return -EIO;
1363                 }
1364                 ix = EXT_FIRST_INDEX(eh);
1365                 block = ext4_idx_pblock(ix);
1366                 put_bh(bh);
1367         }
1368
1369         bh = sb_bread(inode->i_sb, block);
1370         if (bh == NULL)
1371                 return -EIO;
1372         eh = ext_block_hdr(bh);
1373         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1374                 put_bh(bh);
1375                 return -EIO;
1376         }
1377         ex = EXT_FIRST_EXTENT(eh);
1378         *logical = le32_to_cpu(ex->ee_block);
1379         *phys = ext4_ext_pblock(ex);
1380         put_bh(bh);
1381         return 0;
1382 }
1383
1384 /*
1385  * ext4_ext_next_allocated_block:
1386  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1387  * NOTE: it considers block number from index entry as
1388  * allocated block. Thus, index entries have to be consistent
1389  * with leaves.
1390  */
1391 static ext4_lblk_t
1392 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1393 {
1394         int depth;
1395
1396         BUG_ON(path == NULL);
1397         depth = path->p_depth;
1398
1399         if (depth == 0 && path->p_ext == NULL)
1400                 return EXT_MAX_BLOCK;
1401
1402         while (depth >= 0) {
1403                 if (depth == path->p_depth) {
1404                         /* leaf */
1405                         if (path[depth].p_ext !=
1406                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1407                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1408                 } else {
1409                         /* index */
1410                         if (path[depth].p_idx !=
1411                                         EXT_LAST_INDEX(path[depth].p_hdr))
1412                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1413                 }
1414                 depth--;
1415         }
1416
1417         return EXT_MAX_BLOCK;
1418 }
1419
1420 /*
1421  * ext4_ext_next_leaf_block:
1422  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1423  */
1424 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1425                                         struct ext4_ext_path *path)
1426 {
1427         int depth;
1428
1429         BUG_ON(path == NULL);
1430         depth = path->p_depth;
1431
1432         /* zero-tree has no leaf blocks at all */
1433         if (depth == 0)
1434                 return EXT_MAX_BLOCK;
1435
1436         /* go to index block */
1437         depth--;
1438
1439         while (depth >= 0) {
1440                 if (path[depth].p_idx !=
1441                                 EXT_LAST_INDEX(path[depth].p_hdr))
1442                         return (ext4_lblk_t)
1443                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1444                 depth--;
1445         }
1446
1447         return EXT_MAX_BLOCK;
1448 }
1449
1450 /*
1451  * ext4_ext_correct_indexes:
1452  * if leaf gets modified and modified extent is first in the leaf,
1453  * then we have to correct all indexes above.
1454  * TODO: do we need to correct tree in all cases?
1455  */
1456 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1457                                 struct ext4_ext_path *path)
1458 {
1459         struct ext4_extent_header *eh;
1460         int depth = ext_depth(inode);
1461         struct ext4_extent *ex;
1462         __le32 border;
1463         int k, err = 0;
1464
1465         eh = path[depth].p_hdr;
1466         ex = path[depth].p_ext;
1467
1468         if (unlikely(ex == NULL || eh == NULL)) {
1469                 EXT4_ERROR_INODE(inode,
1470                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1471                 return -EIO;
1472         }
1473
1474         if (depth == 0) {
1475                 /* there is no tree at all */
1476                 return 0;
1477         }
1478
1479         if (ex != EXT_FIRST_EXTENT(eh)) {
1480                 /* we correct tree if first leaf got modified only */
1481                 return 0;
1482         }
1483
1484         /*
1485          * TODO: we need correction if border is smaller than current one
1486          */
1487         k = depth - 1;
1488         border = path[depth].p_ext->ee_block;
1489         err = ext4_ext_get_access(handle, inode, path + k);
1490         if (err)
1491                 return err;
1492         path[k].p_idx->ei_block = border;
1493         err = ext4_ext_dirty(handle, inode, path + k);
1494         if (err)
1495                 return err;
1496
1497         while (k--) {
1498                 /* change all left-side indexes */
1499                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1500                         break;
1501                 err = ext4_ext_get_access(handle, inode, path + k);
1502                 if (err)
1503                         break;
1504                 path[k].p_idx->ei_block = border;
1505                 err = ext4_ext_dirty(handle, inode, path + k);
1506                 if (err)
1507                         break;
1508         }
1509
1510         return err;
1511 }
1512
1513 int
1514 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1515                                 struct ext4_extent *ex2)
1516 {
1517         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1518
1519         /*
1520          * Make sure that either both extents are uninitialized, or
1521          * both are _not_.
1522          */
1523         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1524                 return 0;
1525
1526         if (ext4_ext_is_uninitialized(ex1))
1527                 max_len = EXT_UNINIT_MAX_LEN;
1528         else
1529                 max_len = EXT_INIT_MAX_LEN;
1530
1531         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1532         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1533
1534         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1535                         le32_to_cpu(ex2->ee_block))
1536                 return 0;
1537
1538         /*
1539          * To allow future support for preallocated extents to be added
1540          * as an RO_COMPAT feature, refuse to merge to extents if
1541          * this can result in the top bit of ee_len being set.
1542          */
1543         if (ext1_ee_len + ext2_ee_len > max_len)
1544                 return 0;
1545 #ifdef AGGRESSIVE_TEST
1546         if (ext1_ee_len >= 4)
1547                 return 0;
1548 #endif
1549
1550         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1551                 return 1;
1552         return 0;
1553 }
1554
1555 /*
1556  * This function tries to merge the "ex" extent to the next extent in the tree.
1557  * It always tries to merge towards right. If you want to merge towards
1558  * left, pass "ex - 1" as argument instead of "ex".
1559  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1560  * 1 if they got merged.
1561  */
1562 static int ext4_ext_try_to_merge(struct inode *inode,
1563                                  struct ext4_ext_path *path,
1564                                  struct ext4_extent *ex)
1565 {
1566         struct ext4_extent_header *eh;
1567         unsigned int depth, len;
1568         int merge_done = 0;
1569         int uninitialized = 0;
1570
1571         depth = ext_depth(inode);
1572         BUG_ON(path[depth].p_hdr == NULL);
1573         eh = path[depth].p_hdr;
1574
1575         while (ex < EXT_LAST_EXTENT(eh)) {
1576                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1577                         break;
1578                 /* merge with next extent! */
1579                 if (ext4_ext_is_uninitialized(ex))
1580                         uninitialized = 1;
1581                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1582                                 + ext4_ext_get_actual_len(ex + 1));
1583                 if (uninitialized)
1584                         ext4_ext_mark_uninitialized(ex);
1585
1586                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1587                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1588                                 * sizeof(struct ext4_extent);
1589                         memmove(ex + 1, ex + 2, len);
1590                 }
1591                 le16_add_cpu(&eh->eh_entries, -1);
1592                 merge_done = 1;
1593                 WARN_ON(eh->eh_entries == 0);
1594                 if (!eh->eh_entries)
1595                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1596         }
1597
1598         return merge_done;
1599 }
1600
1601 /*
1602  * check if a portion of the "newext" extent overlaps with an
1603  * existing extent.
1604  *
1605  * If there is an overlap discovered, it updates the length of the newext
1606  * such that there will be no overlap, and then returns 1.
1607  * If there is no overlap found, it returns 0.
1608  */
1609 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1610                                            struct ext4_extent *newext,
1611                                            struct ext4_ext_path *path)
1612 {
1613         ext4_lblk_t b1, b2;
1614         unsigned int depth, len1;
1615         unsigned int ret = 0;
1616
1617         b1 = le32_to_cpu(newext->ee_block);
1618         len1 = ext4_ext_get_actual_len(newext);
1619         depth = ext_depth(inode);
1620         if (!path[depth].p_ext)
1621                 goto out;
1622         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1623
1624         /*
1625          * get the next allocated block if the extent in the path
1626          * is before the requested block(s)
1627          */
1628         if (b2 < b1) {
1629                 b2 = ext4_ext_next_allocated_block(path);
1630                 if (b2 == EXT_MAX_BLOCK)
1631                         goto out;
1632         }
1633
1634         /* check for wrap through zero on extent logical start block*/
1635         if (b1 + len1 < b1) {
1636                 len1 = EXT_MAX_BLOCK - b1;
1637                 newext->ee_len = cpu_to_le16(len1);
1638                 ret = 1;
1639         }
1640
1641         /* check for overlap */
1642         if (b1 + len1 > b2) {
1643                 newext->ee_len = cpu_to_le16(b2 - b1);
1644                 ret = 1;
1645         }
1646 out:
1647         return ret;
1648 }
1649
1650 /*
1651  * ext4_ext_insert_extent:
1652  * tries to merge requsted extent into the existing extent or
1653  * inserts requested extent as new one into the tree,
1654  * creating new leaf in the no-space case.
1655  */
1656 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1657                                 struct ext4_ext_path *path,
1658                                 struct ext4_extent *newext, int flag)
1659 {
1660         struct ext4_extent_header *eh;
1661         struct ext4_extent *ex, *fex;
1662         struct ext4_extent *nearex; /* nearest extent */
1663         struct ext4_ext_path *npath = NULL;
1664         int depth, len, err;
1665         ext4_lblk_t next;
1666         unsigned uninitialized = 0;
1667
1668         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1669                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1670                 return -EIO;
1671         }
1672         depth = ext_depth(inode);
1673         ex = path[depth].p_ext;
1674         if (unlikely(path[depth].p_hdr == NULL)) {
1675                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1676                 return -EIO;
1677         }
1678
1679         /* try to insert block into found extent and return */
1680         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1681                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1682                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1683                           ext4_ext_is_uninitialized(newext),
1684                           ext4_ext_get_actual_len(newext),
1685                           le32_to_cpu(ex->ee_block),
1686                           ext4_ext_is_uninitialized(ex),
1687                           ext4_ext_get_actual_len(ex),
1688                           ext4_ext_pblock(ex));
1689                 err = ext4_ext_get_access(handle, inode, path + depth);
1690                 if (err)
1691                         return err;
1692
1693                 /*
1694                  * ext4_can_extents_be_merged should have checked that either
1695                  * both extents are uninitialized, or both aren't. Thus we
1696                  * need to check only one of them here.
1697                  */
1698                 if (ext4_ext_is_uninitialized(ex))
1699                         uninitialized = 1;
1700                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1701                                         + ext4_ext_get_actual_len(newext));
1702                 if (uninitialized)
1703                         ext4_ext_mark_uninitialized(ex);
1704                 eh = path[depth].p_hdr;
1705                 nearex = ex;
1706                 goto merge;
1707         }
1708
1709 repeat:
1710         depth = ext_depth(inode);
1711         eh = path[depth].p_hdr;
1712         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1713                 goto has_space;
1714
1715         /* probably next leaf has space for us? */
1716         fex = EXT_LAST_EXTENT(eh);
1717         next = ext4_ext_next_leaf_block(inode, path);
1718         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1719             && next != EXT_MAX_BLOCK) {
1720                 ext_debug("next leaf block - %d\n", next);
1721                 BUG_ON(npath != NULL);
1722                 npath = ext4_ext_find_extent(inode, next, NULL);
1723                 if (IS_ERR(npath))
1724                         return PTR_ERR(npath);
1725                 BUG_ON(npath->p_depth != path->p_depth);
1726                 eh = npath[depth].p_hdr;
1727                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1728                         ext_debug("next leaf isnt full(%d)\n",
1729                                   le16_to_cpu(eh->eh_entries));
1730                         path = npath;
1731                         goto repeat;
1732                 }
1733                 ext_debug("next leaf has no free space(%d,%d)\n",
1734                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1735         }
1736
1737         /*
1738          * There is no free space in the found leaf.
1739          * We're gonna add a new leaf in the tree.
1740          */
1741         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1742         if (err)
1743                 goto cleanup;
1744         depth = ext_depth(inode);
1745         eh = path[depth].p_hdr;
1746
1747 has_space:
1748         nearex = path[depth].p_ext;
1749
1750         err = ext4_ext_get_access(handle, inode, path + depth);
1751         if (err)
1752                 goto cleanup;
1753
1754         if (!nearex) {
1755                 /* there is no extent in this leaf, create first one */
1756                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1757                                 le32_to_cpu(newext->ee_block),
1758                                 ext4_ext_pblock(newext),
1759                                 ext4_ext_is_uninitialized(newext),
1760                                 ext4_ext_get_actual_len(newext));
1761                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1762         } else if (le32_to_cpu(newext->ee_block)
1763                            > le32_to_cpu(nearex->ee_block)) {
1764 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1765                 if (nearex != EXT_LAST_EXTENT(eh)) {
1766                         len = EXT_MAX_EXTENT(eh) - nearex;
1767                         len = (len - 1) * sizeof(struct ext4_extent);
1768                         len = len < 0 ? 0 : len;
1769                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1770                                         "move %d from 0x%p to 0x%p\n",
1771                                         le32_to_cpu(newext->ee_block),
1772                                         ext4_ext_pblock(newext),
1773                                         ext4_ext_is_uninitialized(newext),
1774                                         ext4_ext_get_actual_len(newext),
1775                                         nearex, len, nearex + 1, nearex + 2);
1776                         memmove(nearex + 2, nearex + 1, len);
1777                 }
1778                 path[depth].p_ext = nearex + 1;
1779         } else {
1780                 BUG_ON(newext->ee_block == nearex->ee_block);
1781                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1782                 len = len < 0 ? 0 : len;
1783                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1784                                 "move %d from 0x%p to 0x%p\n",
1785                                 le32_to_cpu(newext->ee_block),
1786                                 ext4_ext_pblock(newext),
1787                                 ext4_ext_is_uninitialized(newext),
1788                                 ext4_ext_get_actual_len(newext),
1789                                 nearex, len, nearex + 1, nearex + 2);
1790                 memmove(nearex + 1, nearex, len);
1791                 path[depth].p_ext = nearex;
1792         }
1793
1794         le16_add_cpu(&eh->eh_entries, 1);
1795         nearex = path[depth].p_ext;
1796         nearex->ee_block = newext->ee_block;
1797         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1798         nearex->ee_len = newext->ee_len;
1799
1800 merge:
1801         /* try to merge extents to the right */
1802         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1803                 ext4_ext_try_to_merge(inode, path, nearex);
1804
1805         /* try to merge extents to the left */
1806
1807         /* time to correct all indexes above */
1808         err = ext4_ext_correct_indexes(handle, inode, path);
1809         if (err)
1810                 goto cleanup;
1811
1812         err = ext4_ext_dirty(handle, inode, path + depth);
1813
1814 cleanup:
1815         if (npath) {
1816                 ext4_ext_drop_refs(npath);
1817                 kfree(npath);
1818         }
1819         ext4_ext_invalidate_cache(inode);
1820         return err;
1821 }
1822
1823 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1824                                ext4_lblk_t num, ext_prepare_callback func,
1825                                void *cbdata)
1826 {
1827         struct ext4_ext_path *path = NULL;
1828         struct ext4_ext_cache cbex;
1829         struct ext4_extent *ex;
1830         ext4_lblk_t next, start = 0, end = 0;
1831         ext4_lblk_t last = block + num;
1832         int depth, exists, err = 0;
1833
1834         BUG_ON(func == NULL);
1835         BUG_ON(inode == NULL);
1836
1837         while (block < last && block != EXT_MAX_BLOCK) {
1838                 num = last - block;
1839                 /* find extent for this block */
1840                 down_read(&EXT4_I(inode)->i_data_sem);
1841                 path = ext4_ext_find_extent(inode, block, path);
1842                 up_read(&EXT4_I(inode)->i_data_sem);
1843                 if (IS_ERR(path)) {
1844                         err = PTR_ERR(path);
1845                         path = NULL;
1846                         break;
1847                 }
1848
1849                 depth = ext_depth(inode);
1850                 if (unlikely(path[depth].p_hdr == NULL)) {
1851                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1852                         err = -EIO;
1853                         break;
1854                 }
1855                 ex = path[depth].p_ext;
1856                 next = ext4_ext_next_allocated_block(path);
1857
1858                 exists = 0;
1859                 if (!ex) {
1860                         /* there is no extent yet, so try to allocate
1861                          * all requested space */
1862                         start = block;
1863                         end = block + num;
1864                 } else if (le32_to_cpu(ex->ee_block) > block) {
1865                         /* need to allocate space before found extent */
1866                         start = block;
1867                         end = le32_to_cpu(ex->ee_block);
1868                         if (block + num < end)
1869                                 end = block + num;
1870                 } else if (block >= le32_to_cpu(ex->ee_block)
1871                                         + ext4_ext_get_actual_len(ex)) {
1872                         /* need to allocate space after found extent */
1873                         start = block;
1874                         end = block + num;
1875                         if (end >= next)
1876                                 end = next;
1877                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1878                         /*
1879                          * some part of requested space is covered
1880                          * by found extent
1881                          */
1882                         start = block;
1883                         end = le32_to_cpu(ex->ee_block)
1884                                 + ext4_ext_get_actual_len(ex);
1885                         if (block + num < end)
1886                                 end = block + num;
1887                         exists = 1;
1888                 } else {
1889                         BUG();
1890                 }
1891                 BUG_ON(end <= start);
1892
1893                 if (!exists) {
1894                         cbex.ec_block = start;
1895                         cbex.ec_len = end - start;
1896                         cbex.ec_start = 0;
1897                 } else {
1898                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1899                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1900                         cbex.ec_start = ext4_ext_pblock(ex);
1901                 }
1902
1903                 if (unlikely(cbex.ec_len == 0)) {
1904                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1905                         err = -EIO;
1906                         break;
1907                 }
1908                 err = func(inode, path, &cbex, ex, cbdata);
1909                 ext4_ext_drop_refs(path);
1910
1911                 if (err < 0)
1912                         break;
1913
1914                 if (err == EXT_REPEAT)
1915                         continue;
1916                 else if (err == EXT_BREAK) {
1917                         err = 0;
1918                         break;
1919                 }
1920
1921                 if (ext_depth(inode) != depth) {
1922                         /* depth was changed. we have to realloc path */
1923                         kfree(path);
1924                         path = NULL;
1925                 }
1926
1927                 block = cbex.ec_block + cbex.ec_len;
1928         }
1929
1930         if (path) {
1931                 ext4_ext_drop_refs(path);
1932                 kfree(path);
1933         }
1934
1935         return err;
1936 }
1937
1938 static void
1939 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1940                         __u32 len, ext4_fsblk_t start)
1941 {
1942         struct ext4_ext_cache *cex;
1943         BUG_ON(len == 0);
1944         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1945         cex = &EXT4_I(inode)->i_cached_extent;
1946         cex->ec_block = block;
1947         cex->ec_len = len;
1948         cex->ec_start = start;
1949         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1950 }
1951
1952 /*
1953  * ext4_ext_put_gap_in_cache:
1954  * calculate boundaries of the gap that the requested block fits into
1955  * and cache this gap
1956  */
1957 static void
1958 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1959                                 ext4_lblk_t block)
1960 {
1961         int depth = ext_depth(inode);
1962         unsigned long len;
1963         ext4_lblk_t lblock;
1964         struct ext4_extent *ex;
1965
1966         ex = path[depth].p_ext;
1967         if (ex == NULL) {
1968                 /* there is no extent yet, so gap is [0;-] */
1969                 lblock = 0;
1970                 len = EXT_MAX_BLOCK;
1971                 ext_debug("cache gap(whole file):");
1972         } else if (block < le32_to_cpu(ex->ee_block)) {
1973                 lblock = block;
1974                 len = le32_to_cpu(ex->ee_block) - block;
1975                 ext_debug("cache gap(before): %u [%u:%u]",
1976                                 block,
1977                                 le32_to_cpu(ex->ee_block),
1978                                  ext4_ext_get_actual_len(ex));
1979         } else if (block >= le32_to_cpu(ex->ee_block)
1980                         + ext4_ext_get_actual_len(ex)) {
1981                 ext4_lblk_t next;
1982                 lblock = le32_to_cpu(ex->ee_block)
1983                         + ext4_ext_get_actual_len(ex);
1984
1985                 next = ext4_ext_next_allocated_block(path);
1986                 ext_debug("cache gap(after): [%u:%u] %u",
1987                                 le32_to_cpu(ex->ee_block),
1988                                 ext4_ext_get_actual_len(ex),
1989                                 block);
1990                 BUG_ON(next == lblock);
1991                 len = next - lblock;
1992         } else {
1993                 lblock = len = 0;
1994                 BUG();
1995         }
1996
1997         ext_debug(" -> %u:%lu\n", lblock, len);
1998         ext4_ext_put_in_cache(inode, lblock, len, 0);
1999 }
2000
2001 /*
2002  * Return 0 if cache is invalid; 1 if the cache is valid
2003  */
2004 static int
2005 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2006                         struct ext4_extent *ex)
2007 {
2008         struct ext4_ext_cache *cex;
2009         int ret = 0;
2010
2011         /*
2012          * We borrow i_block_reservation_lock to protect i_cached_extent
2013          */
2014         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2015         cex = &EXT4_I(inode)->i_cached_extent;
2016
2017         /* has cache valid data? */
2018         if (cex->ec_len == 0)
2019                 goto errout;
2020
2021         if (in_range(block, cex->ec_block, cex->ec_len)) {
2022                 ex->ee_block = cpu_to_le32(cex->ec_block);
2023                 ext4_ext_store_pblock(ex, cex->ec_start);
2024                 ex->ee_len = cpu_to_le16(cex->ec_len);
2025                 ext_debug("%u cached by %u:%u:%llu\n",
2026                                 block,
2027                                 cex->ec_block, cex->ec_len, cex->ec_start);
2028                 ret = 1;
2029         }
2030 errout:
2031         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2032         return ret;
2033 }
2034
2035 /*
2036  * ext4_ext_rm_idx:
2037  * removes index from the index block.
2038  * It's used in truncate case only, thus all requests are for
2039  * last index in the block only.
2040  */
2041 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2042                         struct ext4_ext_path *path)
2043 {
2044         int err;
2045         ext4_fsblk_t leaf;
2046
2047         /* free index block */
2048         path--;
2049         leaf = ext4_idx_pblock(path->p_idx);
2050         if (unlikely(path->p_hdr->eh_entries == 0)) {
2051                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2052                 return -EIO;
2053         }
2054         err = ext4_ext_get_access(handle, inode, path);
2055         if (err)
2056                 return err;
2057         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2058         err = ext4_ext_dirty(handle, inode, path);
2059         if (err)
2060                 return err;
2061         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2062         ext4_free_blocks(handle, inode, 0, leaf, 1,
2063                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2064         return err;
2065 }
2066
2067 /*
2068  * ext4_ext_calc_credits_for_single_extent:
2069  * This routine returns max. credits that needed to insert an extent
2070  * to the extent tree.
2071  * When pass the actual path, the caller should calculate credits
2072  * under i_data_sem.
2073  */
2074 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2075                                                 struct ext4_ext_path *path)
2076 {
2077         if (path) {
2078                 int depth = ext_depth(inode);
2079                 int ret = 0;
2080
2081                 /* probably there is space in leaf? */
2082                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2083                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2084
2085                         /*
2086                          *  There are some space in the leaf tree, no
2087                          *  need to account for leaf block credit
2088                          *
2089                          *  bitmaps and block group descriptor blocks
2090                          *  and other metadat blocks still need to be
2091                          *  accounted.
2092                          */
2093                         /* 1 bitmap, 1 block group descriptor */
2094                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2095                         return ret;
2096                 }
2097         }
2098
2099         return ext4_chunk_trans_blocks(inode, nrblocks);
2100 }
2101
2102 /*
2103  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2104  *
2105  * if nrblocks are fit in a single extent (chunk flag is 1), then
2106  * in the worse case, each tree level index/leaf need to be changed
2107  * if the tree split due to insert a new extent, then the old tree
2108  * index/leaf need to be updated too
2109  *
2110  * If the nrblocks are discontiguous, they could cause
2111  * the whole tree split more than once, but this is really rare.
2112  */
2113 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2114 {
2115         int index;
2116         int depth = ext_depth(inode);
2117
2118         if (chunk)
2119                 index = depth * 2;
2120         else
2121                 index = depth * 3;
2122
2123         return index;
2124 }
2125
2126 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2127                                 struct ext4_extent *ex,
2128                                 ext4_lblk_t from, ext4_lblk_t to)
2129 {
2130         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2131         int flags = EXT4_FREE_BLOCKS_FORGET;
2132
2133         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2134                 flags |= EXT4_FREE_BLOCKS_METADATA;
2135 #ifdef EXTENTS_STATS
2136         {
2137                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2138                 spin_lock(&sbi->s_ext_stats_lock);
2139                 sbi->s_ext_blocks += ee_len;
2140                 sbi->s_ext_extents++;
2141                 if (ee_len < sbi->s_ext_min)
2142                         sbi->s_ext_min = ee_len;
2143                 if (ee_len > sbi->s_ext_max)
2144                         sbi->s_ext_max = ee_len;
2145                 if (ext_depth(inode) > sbi->s_depth_max)
2146                         sbi->s_depth_max = ext_depth(inode);
2147                 spin_unlock(&sbi->s_ext_stats_lock);
2148         }
2149 #endif
2150         if (from >= le32_to_cpu(ex->ee_block)
2151             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2152                 /* tail removal */
2153                 ext4_lblk_t num;
2154                 ext4_fsblk_t start;
2155
2156                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2157                 start = ext4_ext_pblock(ex) + ee_len - num;
2158                 ext_debug("free last %u blocks starting %llu\n", num, start);
2159                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2160         } else if (from == le32_to_cpu(ex->ee_block)
2161                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2162                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2163                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2164         } else {
2165                 printk(KERN_INFO "strange request: removal(2) "
2166                                 "%u-%u from %u:%u\n",
2167                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2168         }
2169         return 0;
2170 }
2171
2172 static int
2173 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2174                 struct ext4_ext_path *path, ext4_lblk_t start)
2175 {
2176         int err = 0, correct_index = 0;
2177         int depth = ext_depth(inode), credits;
2178         struct ext4_extent_header *eh;
2179         ext4_lblk_t a, b, block;
2180         unsigned num;
2181         ext4_lblk_t ex_ee_block;
2182         unsigned short ex_ee_len;
2183         unsigned uninitialized = 0;
2184         struct ext4_extent *ex;
2185
2186         /* the header must be checked already in ext4_ext_remove_space() */
2187         ext_debug("truncate since %u in leaf\n", start);
2188         if (!path[depth].p_hdr)
2189                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2190         eh = path[depth].p_hdr;
2191         if (unlikely(path[depth].p_hdr == NULL)) {
2192                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2193                 return -EIO;
2194         }
2195         /* find where to start removing */
2196         ex = EXT_LAST_EXTENT(eh);
2197
2198         ex_ee_block = le32_to_cpu(ex->ee_block);
2199         ex_ee_len = ext4_ext_get_actual_len(ex);
2200
2201         while (ex >= EXT_FIRST_EXTENT(eh) &&
2202                         ex_ee_block + ex_ee_len > start) {
2203
2204                 if (ext4_ext_is_uninitialized(ex))
2205                         uninitialized = 1;
2206                 else
2207                         uninitialized = 0;
2208
2209                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2210                          uninitialized, ex_ee_len);
2211                 path[depth].p_ext = ex;
2212
2213                 a = ex_ee_block > start ? ex_ee_block : start;
2214                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2215                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2216
2217                 ext_debug("  border %u:%u\n", a, b);
2218
2219                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2220                         block = 0;
2221                         num = 0;
2222                         BUG();
2223                 } else if (a != ex_ee_block) {
2224                         /* remove tail of the extent */
2225                         block = ex_ee_block;
2226                         num = a - block;
2227                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2228                         /* remove head of the extent */
2229                         block = a;
2230                         num = b - a;
2231                         /* there is no "make a hole" API yet */
2232                         BUG();
2233                 } else {
2234                         /* remove whole extent: excellent! */
2235                         block = ex_ee_block;
2236                         num = 0;
2237                         BUG_ON(a != ex_ee_block);
2238                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2239                 }
2240
2241                 /*
2242                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2243                  * descriptor) for each block group; assume two block
2244                  * groups plus ex_ee_len/blocks_per_block_group for
2245                  * the worst case
2246                  */
2247                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2248                 if (ex == EXT_FIRST_EXTENT(eh)) {
2249                         correct_index = 1;
2250                         credits += (ext_depth(inode)) + 1;
2251                 }
2252                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2253
2254                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2255                 if (err)
2256                         goto out;
2257
2258                 err = ext4_ext_get_access(handle, inode, path + depth);
2259                 if (err)
2260                         goto out;
2261
2262                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2263                 if (err)
2264                         goto out;
2265
2266                 if (num == 0) {
2267                         /* this extent is removed; mark slot entirely unused */
2268                         ext4_ext_store_pblock(ex, 0);
2269                         le16_add_cpu(&eh->eh_entries, -1);
2270                 }
2271
2272                 ex->ee_block = cpu_to_le32(block);
2273                 ex->ee_len = cpu_to_le16(num);
2274                 /*
2275                  * Do not mark uninitialized if all the blocks in the
2276                  * extent have been removed.
2277                  */
2278                 if (uninitialized && num)
2279                         ext4_ext_mark_uninitialized(ex);
2280
2281                 err = ext4_ext_dirty(handle, inode, path + depth);
2282                 if (err)
2283                         goto out;
2284
2285                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2286                                 ext4_ext_pblock(ex));
2287                 ex--;
2288                 ex_ee_block = le32_to_cpu(ex->ee_block);
2289                 ex_ee_len = ext4_ext_get_actual_len(ex);
2290         }
2291
2292         if (correct_index && eh->eh_entries)
2293                 err = ext4_ext_correct_indexes(handle, inode, path);
2294
2295         /* if this leaf is free, then we should
2296          * remove it from index block above */
2297         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2298                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2299
2300 out:
2301         return err;
2302 }
2303
2304 /*
2305  * ext4_ext_more_to_rm:
2306  * returns 1 if current index has to be freed (even partial)
2307  */
2308 static int
2309 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2310 {
2311         BUG_ON(path->p_idx == NULL);
2312
2313         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2314                 return 0;
2315
2316         /*
2317          * if truncate on deeper level happened, it wasn't partial,
2318          * so we have to consider current index for truncation
2319          */
2320         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2321                 return 0;
2322         return 1;
2323 }
2324
2325 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2326 {
2327         struct super_block *sb = inode->i_sb;
2328         int depth = ext_depth(inode);
2329         struct ext4_ext_path *path;
2330         handle_t *handle;
2331         int i, err;
2332
2333         ext_debug("truncate since %u\n", start);
2334
2335         /* probably first extent we're gonna free will be last in block */
2336         handle = ext4_journal_start(inode, depth + 1);
2337         if (IS_ERR(handle))
2338                 return PTR_ERR(handle);
2339
2340 again:
2341         ext4_ext_invalidate_cache(inode);
2342
2343         /*
2344          * We start scanning from right side, freeing all the blocks
2345          * after i_size and walking into the tree depth-wise.
2346          */
2347         depth = ext_depth(inode);
2348         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2349         if (path == NULL) {
2350                 ext4_journal_stop(handle);
2351                 return -ENOMEM;
2352         }
2353         path[0].p_depth = depth;
2354         path[0].p_hdr = ext_inode_hdr(inode);
2355         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2356                 err = -EIO;
2357                 goto out;
2358         }
2359         i = err = 0;
2360
2361         while (i >= 0 && err == 0) {
2362                 if (i == depth) {
2363                         /* this is leaf block */
2364                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2365                         /* root level has p_bh == NULL, brelse() eats this */
2366                         brelse(path[i].p_bh);
2367                         path[i].p_bh = NULL;
2368                         i--;
2369                         continue;
2370                 }
2371
2372                 /* this is index block */
2373                 if (!path[i].p_hdr) {
2374                         ext_debug("initialize header\n");
2375                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2376                 }
2377
2378                 if (!path[i].p_idx) {
2379                         /* this level hasn't been touched yet */
2380                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2381                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2382                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2383                                   path[i].p_hdr,
2384                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2385                 } else {
2386                         /* we were already here, see at next index */
2387                         path[i].p_idx--;
2388                 }
2389
2390                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2391                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2392                                 path[i].p_idx);
2393                 if (ext4_ext_more_to_rm(path + i)) {
2394                         struct buffer_head *bh;
2395                         /* go to the next level */
2396                         ext_debug("move to level %d (block %llu)\n",
2397                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2398                         memset(path + i + 1, 0, sizeof(*path));
2399                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2400                         if (!bh) {
2401                                 /* should we reset i_size? */
2402                                 err = -EIO;
2403                                 break;
2404                         }
2405                         if (WARN_ON(i + 1 > depth)) {
2406                                 err = -EIO;
2407                                 break;
2408                         }
2409                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2410                                                         depth - i - 1)) {
2411                                 err = -EIO;
2412                                 break;
2413                         }
2414                         path[i + 1].p_bh = bh;
2415
2416                         /* save actual number of indexes since this
2417                          * number is changed at the next iteration */
2418                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2419                         i++;
2420                 } else {
2421                         /* we finished processing this index, go up */
2422                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2423                                 /* index is empty, remove it;
2424                                  * handle must be already prepared by the
2425                                  * truncatei_leaf() */
2426                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2427                         }
2428                         /* root level has p_bh == NULL, brelse() eats this */
2429                         brelse(path[i].p_bh);
2430                         path[i].p_bh = NULL;
2431                         i--;
2432                         ext_debug("return to level %d\n", i);
2433                 }
2434         }
2435
2436         /* TODO: flexible tree reduction should be here */
2437         if (path->p_hdr->eh_entries == 0) {
2438                 /*
2439                  * truncate to zero freed all the tree,
2440                  * so we need to correct eh_depth
2441                  */
2442                 err = ext4_ext_get_access(handle, inode, path);
2443                 if (err == 0) {
2444                         ext_inode_hdr(inode)->eh_depth = 0;
2445                         ext_inode_hdr(inode)->eh_max =
2446                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2447                         err = ext4_ext_dirty(handle, inode, path);
2448                 }
2449         }
2450 out:
2451         ext4_ext_drop_refs(path);
2452         kfree(path);
2453         if (err == -EAGAIN)
2454                 goto again;
2455         ext4_journal_stop(handle);
2456
2457         return err;
2458 }
2459
2460 /*
2461  * called at mount time
2462  */
2463 void ext4_ext_init(struct super_block *sb)
2464 {
2465         /*
2466          * possible initialization would be here
2467          */
2468
2469         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2470 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2471                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2472 #ifdef AGGRESSIVE_TEST
2473                 printk(", aggressive tests");
2474 #endif
2475 #ifdef CHECK_BINSEARCH
2476                 printk(", check binsearch");
2477 #endif
2478 #ifdef EXTENTS_STATS
2479                 printk(", stats");
2480 #endif
2481                 printk("\n");
2482 #endif
2483 #ifdef EXTENTS_STATS
2484                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2485                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2486                 EXT4_SB(sb)->s_ext_max = 0;
2487 #endif
2488         }
2489 }
2490
2491 /*
2492  * called at umount time
2493  */
2494 void ext4_ext_release(struct super_block *sb)
2495 {
2496         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2497                 return;
2498
2499 #ifdef EXTENTS_STATS
2500         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2501                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2502                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2503                         sbi->s_ext_blocks, sbi->s_ext_extents,
2504                         sbi->s_ext_blocks / sbi->s_ext_extents);
2505                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2506                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2507         }
2508 #endif
2509 }
2510
2511 /* FIXME!! we need to try to merge to left or right after zero-out  */
2512 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2513 {
2514         ext4_fsblk_t ee_pblock;
2515         unsigned int ee_len;
2516         int ret;
2517
2518         ee_len    = ext4_ext_get_actual_len(ex);
2519         ee_pblock = ext4_ext_pblock(ex);
2520
2521         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2522         if (ret > 0)
2523                 ret = 0;
2524
2525         return ret;
2526 }
2527
2528 #define EXT4_EXT_ZERO_LEN 7
2529 /*
2530  * This function is called by ext4_ext_map_blocks() if someone tries to write
2531  * to an uninitialized extent. It may result in splitting the uninitialized
2532  * extent into multiple extents (upto three - one initialized and two
2533  * uninitialized).
2534  * There are three possibilities:
2535  *   a> There is no split required: Entire extent should be initialized
2536  *   b> Splits in two extents: Write is happening at either end of the extent
2537  *   c> Splits in three extents: Somone is writing in middle of the extent
2538  */
2539 static int ext4_ext_convert_to_initialized(handle_t *handle,
2540                                            struct inode *inode,
2541                                            struct ext4_map_blocks *map,
2542                                            struct ext4_ext_path *path)
2543 {
2544         struct ext4_extent *ex, newex, orig_ex;
2545         struct ext4_extent *ex1 = NULL;
2546         struct ext4_extent *ex2 = NULL;
2547         struct ext4_extent *ex3 = NULL;
2548         struct ext4_extent_header *eh;
2549         ext4_lblk_t ee_block, eof_block;
2550         unsigned int allocated, ee_len, depth;
2551         ext4_fsblk_t newblock;
2552         int err = 0;
2553         int ret = 0;
2554         int may_zeroout;
2555
2556         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2557                 "block %llu, max_blocks %u\n", inode->i_ino,
2558                 (unsigned long long)map->m_lblk, map->m_len);
2559
2560         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2561                 inode->i_sb->s_blocksize_bits;
2562         if (eof_block < map->m_lblk + map->m_len)
2563                 eof_block = map->m_lblk + map->m_len;
2564
2565         depth = ext_depth(inode);
2566         eh = path[depth].p_hdr;
2567         ex = path[depth].p_ext;
2568         ee_block = le32_to_cpu(ex->ee_block);
2569         ee_len = ext4_ext_get_actual_len(ex);
2570         allocated = ee_len - (map->m_lblk - ee_block);
2571         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2572
2573         ex2 = ex;
2574         orig_ex.ee_block = ex->ee_block;
2575         orig_ex.ee_len   = cpu_to_le16(ee_len);
2576         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2577
2578         /*
2579          * It is safe to convert extent to initialized via explicit
2580          * zeroout only if extent is fully insde i_size or new_size.
2581          */
2582         may_zeroout = ee_block + ee_len <= eof_block;
2583
2584         err = ext4_ext_get_access(handle, inode, path + depth);
2585         if (err)
2586                 goto out;
2587         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2588         if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2589                 err =  ext4_ext_zeroout(inode, &orig_ex);
2590                 if (err)
2591                         goto fix_extent_len;
2592                 /* update the extent length and mark as initialized */
2593                 ex->ee_block = orig_ex.ee_block;
2594                 ex->ee_len   = orig_ex.ee_len;
2595                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2596                 ext4_ext_dirty(handle, inode, path + depth);
2597                 /* zeroed the full extent */
2598                 return allocated;
2599         }
2600
2601         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2602         if (map->m_lblk > ee_block) {
2603                 ex1 = ex;
2604                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2605                 ext4_ext_mark_uninitialized(ex1);
2606                 ex2 = &newex;
2607         }
2608         /*
2609          * for sanity, update the length of the ex2 extent before
2610          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2611          * overlap of blocks.
2612          */
2613         if (!ex1 && allocated > map->m_len)
2614                 ex2->ee_len = cpu_to_le16(map->m_len);
2615         /* ex3: to ee_block + ee_len : uninitialised */
2616         if (allocated > map->m_len) {
2617                 unsigned int newdepth;
2618                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2619                 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2620                         /*
2621                          * map->m_lblk == ee_block is handled by the zerouout
2622                          * at the beginning.
2623                          * Mark first half uninitialized.
2624                          * Mark second half initialized and zero out the
2625                          * initialized extent
2626                          */
2627                         ex->ee_block = orig_ex.ee_block;
2628                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2629                         ext4_ext_mark_uninitialized(ex);
2630                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2631                         ext4_ext_dirty(handle, inode, path + depth);
2632
2633                         ex3 = &newex;
2634                         ex3->ee_block = cpu_to_le32(map->m_lblk);
2635                         ext4_ext_store_pblock(ex3, newblock);
2636                         ex3->ee_len = cpu_to_le16(allocated);
2637                         err = ext4_ext_insert_extent(handle, inode, path,
2638                                                         ex3, 0);
2639                         if (err == -ENOSPC) {
2640                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2641                                 if (err)
2642                                         goto fix_extent_len;
2643                                 ex->ee_block = orig_ex.ee_block;
2644                                 ex->ee_len   = orig_ex.ee_len;
2645                                 ext4_ext_store_pblock(ex,
2646                                         ext4_ext_pblock(&orig_ex));
2647                                 ext4_ext_dirty(handle, inode, path + depth);
2648                                 /* blocks available from map->m_lblk */
2649                                 return allocated;
2650
2651                         } else if (err)
2652                                 goto fix_extent_len;
2653
2654                         /*
2655                          * We need to zero out the second half because
2656                          * an fallocate request can update file size and
2657                          * converting the second half to initialized extent
2658                          * implies that we can leak some junk data to user
2659                          * space.
2660                          */
2661                         err =  ext4_ext_zeroout(inode, ex3);
2662                         if (err) {
2663                                 /*
2664                                  * We should actually mark the
2665                                  * second half as uninit and return error
2666                                  * Insert would have changed the extent
2667                                  */
2668                                 depth = ext_depth(inode);
2669                                 ext4_ext_drop_refs(path);
2670                                 path = ext4_ext_find_extent(inode, map->m_lblk,
2671                                                             path);
2672                                 if (IS_ERR(path)) {
2673                                         err = PTR_ERR(path);
2674                                         return err;
2675                                 }
2676                                 /* get the second half extent details */
2677                                 ex = path[depth].p_ext;
2678                                 err = ext4_ext_get_access(handle, inode,
2679                                                                 path + depth);
2680                                 if (err)
2681                                         return err;
2682                                 ext4_ext_mark_uninitialized(ex);
2683                                 ext4_ext_dirty(handle, inode, path + depth);
2684                                 return err;
2685                         }
2686
2687                         /* zeroed the second half */
2688                         return allocated;
2689                 }
2690                 ex3 = &newex;
2691                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2692                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2693                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2694                 ext4_ext_mark_uninitialized(ex3);
2695                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2696                 if (err == -ENOSPC && may_zeroout) {
2697                         err =  ext4_ext_zeroout(inode, &orig_ex);
2698                         if (err)
2699                                 goto fix_extent_len;
2700                         /* update the extent length and mark as initialized */
2701                         ex->ee_block = orig_ex.ee_block;
2702                         ex->ee_len   = orig_ex.ee_len;
2703                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2704                         ext4_ext_dirty(handle, inode, path + depth);
2705                         /* zeroed the full extent */
2706                         /* blocks available from map->m_lblk */
2707                         return allocated;
2708
2709                 } else if (err)
2710                         goto fix_extent_len;
2711                 /*
2712                  * The depth, and hence eh & ex might change
2713                  * as part of the insert above.
2714                  */
2715                 newdepth = ext_depth(inode);
2716                 /*
2717                  * update the extent length after successful insert of the
2718                  * split extent
2719                  */
2720                 ee_len -= ext4_ext_get_actual_len(ex3);
2721                 orig_ex.ee_len = cpu_to_le16(ee_len);
2722                 may_zeroout = ee_block + ee_len <= eof_block;
2723
2724                 depth = newdepth;
2725                 ext4_ext_drop_refs(path);
2726                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2727                 if (IS_ERR(path)) {
2728                         err = PTR_ERR(path);
2729                         goto out;
2730                 }
2731                 eh = path[depth].p_hdr;
2732                 ex = path[depth].p_ext;
2733                 if (ex2 != &newex)
2734                         ex2 = ex;
2735
2736                 err = ext4_ext_get_access(handle, inode, path + depth);
2737                 if (err)
2738                         goto out;
2739
2740                 allocated = map->m_len;
2741
2742                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2743                  * to insert a extent in the middle zerout directly
2744                  * otherwise give the extent a chance to merge to left
2745                  */
2746                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2747                         map->m_lblk != ee_block && may_zeroout) {
2748                         err =  ext4_ext_zeroout(inode, &orig_ex);
2749                         if (err)
2750                                 goto fix_extent_len;
2751                         /* update the extent length and mark as initialized */
2752                         ex->ee_block = orig_ex.ee_block;
2753                         ex->ee_len   = orig_ex.ee_len;
2754                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2755                         ext4_ext_dirty(handle, inode, path + depth);
2756                         /* zero out the first half */
2757                         /* blocks available from map->m_lblk */
2758                         return allocated;
2759                 }
2760         }
2761         /*
2762          * If there was a change of depth as part of the
2763          * insertion of ex3 above, we need to update the length
2764          * of the ex1 extent again here
2765          */
2766         if (ex1 && ex1 != ex) {
2767                 ex1 = ex;
2768                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2769                 ext4_ext_mark_uninitialized(ex1);
2770                 ex2 = &newex;
2771         }
2772         /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2773         ex2->ee_block = cpu_to_le32(map->m_lblk);
2774         ext4_ext_store_pblock(ex2, newblock);
2775         ex2->ee_len = cpu_to_le16(allocated);
2776         if (ex2 != ex)
2777                 goto insert;
2778         /*
2779          * New (initialized) extent starts from the first block
2780          * in the current extent. i.e., ex2 == ex
2781          * We have to see if it can be merged with the extent
2782          * on the left.
2783          */
2784         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2785                 /*
2786                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2787                  * since it merges towards right _only_.
2788                  */
2789                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2790                 if (ret) {
2791                         err = ext4_ext_correct_indexes(handle, inode, path);
2792                         if (err)
2793                                 goto out;
2794                         depth = ext_depth(inode);
2795                         ex2--;
2796                 }
2797         }
2798         /*
2799          * Try to Merge towards right. This might be required
2800          * only when the whole extent is being written to.
2801          * i.e. ex2 == ex and ex3 == NULL.
2802          */
2803         if (!ex3) {
2804                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2805                 if (ret) {
2806                         err = ext4_ext_correct_indexes(handle, inode, path);
2807                         if (err)
2808                                 goto out;
2809                 }
2810         }
2811         /* Mark modified extent as dirty */
2812         err = ext4_ext_dirty(handle, inode, path + depth);
2813         goto out;
2814 insert:
2815         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2816         if (err == -ENOSPC && may_zeroout) {
2817                 err =  ext4_ext_zeroout(inode, &orig_ex);
2818                 if (err)
2819                         goto fix_extent_len;
2820                 /* update the extent length and mark as initialized */
2821                 ex->ee_block = orig_ex.ee_block;
2822                 ex->ee_len   = orig_ex.ee_len;
2823                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2824                 ext4_ext_dirty(handle, inode, path + depth);
2825                 /* zero out the first half */
2826                 return allocated;
2827         } else if (err)
2828                 goto fix_extent_len;
2829 out:
2830         ext4_ext_show_leaf(inode, path);
2831         return err ? err : allocated;
2832
2833 fix_extent_len:
2834         ex->ee_block = orig_ex.ee_block;
2835         ex->ee_len   = orig_ex.ee_len;
2836         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2837         ext4_ext_mark_uninitialized(ex);
2838         ext4_ext_dirty(handle, inode, path + depth);
2839         return err;
2840 }
2841
2842 /*
2843  * This function is called by ext4_ext_map_blocks() from
2844  * ext4_get_blocks_dio_write() when DIO to write
2845  * to an uninitialized extent.
2846  *
2847  * Writing to an uninitized extent may result in splitting the uninitialized
2848  * extent into multiple /intialized unintialized extents (up to three)
2849  * There are three possibilities:
2850  *   a> There is no split required: Entire extent should be uninitialized
2851  *   b> Splits in two extents: Write is happening at either end of the extent
2852  *   c> Splits in three extents: Somone is writing in middle of the extent
2853  *
2854  * One of more index blocks maybe needed if the extent tree grow after
2855  * the unintialized extent split. To prevent ENOSPC occur at the IO
2856  * complete, we need to split the uninitialized extent before DIO submit
2857  * the IO. The uninitialized extent called at this time will be split
2858  * into three uninitialized extent(at most). After IO complete, the part
2859  * being filled will be convert to initialized by the end_io callback function
2860  * via ext4_convert_unwritten_extents().
2861  *
2862  * Returns the size of uninitialized extent to be written on success.
2863  */
2864 static int ext4_split_unwritten_extents(handle_t *handle,
2865                                         struct inode *inode,
2866                                         struct ext4_map_blocks *map,
2867                                         struct ext4_ext_path *path,
2868                                         int flags)
2869 {
2870         struct ext4_extent *ex, newex, orig_ex;
2871         struct ext4_extent *ex1 = NULL;
2872         struct ext4_extent *ex2 = NULL;
2873         struct ext4_extent *ex3 = NULL;
2874         ext4_lblk_t ee_block, eof_block;
2875         unsigned int allocated, ee_len, depth;
2876         ext4_fsblk_t newblock;
2877         int err = 0;
2878         int may_zeroout;
2879
2880         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2881                 "block %llu, max_blocks %u\n", inode->i_ino,
2882                 (unsigned long long)map->m_lblk, map->m_len);
2883
2884         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2885                 inode->i_sb->s_blocksize_bits;
2886         if (eof_block < map->m_lblk + map->m_len)
2887                 eof_block = map->m_lblk + map->m_len;
2888
2889         depth = ext_depth(inode);
2890         ex = path[depth].p_ext;
2891         ee_block = le32_to_cpu(ex->ee_block);
2892         ee_len = ext4_ext_get_actual_len(ex);
2893         allocated = ee_len - (map->m_lblk - ee_block);
2894         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2895
2896         ex2 = ex;
2897         orig_ex.ee_block = ex->ee_block;
2898         orig_ex.ee_len   = cpu_to_le16(ee_len);
2899         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2900
2901         /*
2902          * It is safe to convert extent to initialized via explicit
2903          * zeroout only if extent is fully insde i_size or new_size.
2904          */
2905         may_zeroout = ee_block + ee_len <= eof_block;
2906
2907         /*
2908          * If the uninitialized extent begins at the same logical
2909          * block where the write begins, and the write completely
2910          * covers the extent, then we don't need to split it.
2911          */
2912         if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2913                 return allocated;
2914
2915         err = ext4_ext_get_access(handle, inode, path + depth);
2916         if (err)
2917                 goto out;
2918         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2919         if (map->m_lblk > ee_block) {
2920                 ex1 = ex;
2921                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2922                 ext4_ext_mark_uninitialized(ex1);
2923                 ex2 = &newex;
2924         }
2925         /*
2926          * for sanity, update the length of the ex2 extent before
2927          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2928          * overlap of blocks.
2929          */
2930         if (!ex1 && allocated > map->m_len)
2931                 ex2->ee_len = cpu_to_le16(map->m_len);
2932         /* ex3: to ee_block + ee_len : uninitialised */
2933         if (allocated > map->m_len) {
2934                 unsigned int newdepth;
2935                 ex3 = &newex;
2936                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2937                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2938                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2939                 ext4_ext_mark_uninitialized(ex3);
2940                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2941                 if (err == -ENOSPC && may_zeroout) {
2942                         err =  ext4_ext_zeroout(inode, &orig_ex);
2943                         if (err)
2944                                 goto fix_extent_len;
2945                         /* update the extent length and mark as initialized */
2946                         ex->ee_block = orig_ex.ee_block;
2947                         ex->ee_len   = orig_ex.ee_len;
2948                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2949                         ext4_ext_dirty(handle, inode, path + depth);
2950                         /* zeroed the full extent */
2951                         /* blocks available from map->m_lblk */
2952                         return allocated;
2953
2954                 } else if (err)
2955                         goto fix_extent_len;
2956                 /*
2957                  * The depth, and hence eh & ex might change
2958                  * as part of the insert above.
2959                  */
2960                 newdepth = ext_depth(inode);
2961                 /*
2962                  * update the extent length after successful insert of the
2963                  * split extent
2964                  */
2965                 ee_len -= ext4_ext_get_actual_len(ex3);
2966                 orig_ex.ee_len = cpu_to_le16(ee_len);
2967                 may_zeroout = ee_block + ee_len <= eof_block;
2968
2969                 depth = newdepth;
2970                 ext4_ext_drop_refs(path);
2971                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2972                 if (IS_ERR(path)) {
2973                         err = PTR_ERR(path);
2974                         goto out;
2975                 }
2976                 ex = path[depth].p_ext;
2977                 if (ex2 != &newex)
2978                         ex2 = ex;
2979
2980                 err = ext4_ext_get_access(handle, inode, path + depth);
2981                 if (err)
2982                         goto out;
2983
2984                 allocated = map->m_len;
2985         }
2986         /*
2987          * If there was a change of depth as part of the
2988          * insertion of ex3 above, we need to update the length
2989          * of the ex1 extent again here
2990          */
2991         if (ex1 && ex1 != ex) {
2992                 ex1 = ex;
2993                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2994                 ext4_ext_mark_uninitialized(ex1);
2995                 ex2 = &newex;
2996         }
2997         /*
2998          * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
2999          * using direct I/O, uninitialised still.
3000          */
3001         ex2->ee_block = cpu_to_le32(map->m_lblk);
3002         ext4_ext_store_pblock(ex2, newblock);
3003         ex2->ee_len = cpu_to_le16(allocated);
3004         ext4_ext_mark_uninitialized(ex2);
3005         if (ex2 != ex)
3006                 goto insert;
3007         /* Mark modified extent as dirty */
3008         err = ext4_ext_dirty(handle, inode, path + depth);
3009         ext_debug("out here\n");
3010         goto out;
3011 insert:
3012         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3013         if (err == -ENOSPC && may_zeroout) {
3014                 err =  ext4_ext_zeroout(inode, &orig_ex);
3015                 if (err)
3016                         goto fix_extent_len;
3017                 /* update the extent length and mark as initialized */
3018                 ex->ee_block = orig_ex.ee_block;
3019                 ex->ee_len   = orig_ex.ee_len;
3020                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3021                 ext4_ext_dirty(handle, inode, path + depth);
3022                 /* zero out the first half */
3023                 return allocated;
3024         } else if (err)
3025                 goto fix_extent_len;
3026 out:
3027         ext4_ext_show_leaf(inode, path);
3028         return err ? err : allocated;
3029
3030 fix_extent_len:
3031         ex->ee_block = orig_ex.ee_block;
3032         ex->ee_len   = orig_ex.ee_len;
3033         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3034         ext4_ext_mark_uninitialized(ex);
3035         ext4_ext_dirty(handle, inode, path + depth);
3036         return err;
3037 }
3038 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3039                                               struct inode *inode,
3040                                               struct ext4_ext_path *path)
3041 {
3042         struct ext4_extent *ex;
3043         struct ext4_extent_header *eh;
3044         int depth;
3045         int err = 0;
3046         int ret = 0;
3047
3048         depth = ext_depth(inode);
3049         eh = path[depth].p_hdr;
3050         ex = path[depth].p_ext;
3051
3052         err = ext4_ext_get_access(handle, inode, path + depth);
3053         if (err)
3054                 goto out;
3055         /* first mark the extent as initialized */
3056         ext4_ext_mark_initialized(ex);
3057
3058         /*
3059          * We have to see if it can be merged with the extent
3060          * on the left.
3061          */
3062         if (ex > EXT_FIRST_EXTENT(eh)) {
3063                 /*
3064                  * To merge left, pass "ex - 1" to try_to_merge(),
3065                  * since it merges towards right _only_.
3066                  */
3067                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3068                 if (ret) {
3069                         err = ext4_ext_correct_indexes(handle, inode, path);
3070                         if (err)
3071                                 goto out;
3072                         depth = ext_depth(inode);
3073                         ex--;
3074                 }
3075         }
3076         /*
3077          * Try to Merge towards right.
3078          */
3079         ret = ext4_ext_try_to_merge(inode, path, ex);
3080         if (ret) {
3081                 err = ext4_ext_correct_indexes(handle, inode, path);
3082                 if (err)
3083                         goto out;
3084                 depth = ext_depth(inode);
3085         }
3086         /* Mark modified extent as dirty */
3087         err = ext4_ext_dirty(handle, inode, path + depth);
3088 out:
3089         ext4_ext_show_leaf(inode, path);
3090         return err;
3091 }
3092
3093 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3094                         sector_t block, int count)
3095 {
3096         int i;
3097         for (i = 0; i < count; i++)
3098                 unmap_underlying_metadata(bdev, block + i);
3099 }
3100
3101 /*
3102  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3103  */
3104 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3105                               ext4_lblk_t lblk,
3106                               struct ext4_ext_path *path,
3107                               unsigned int len)
3108 {
3109         int i, depth;
3110         struct ext4_extent_header *eh;
3111         struct ext4_extent *ex, *last_ex;
3112
3113         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3114                 return 0;
3115
3116         depth = ext_depth(inode);
3117         eh = path[depth].p_hdr;
3118         ex = path[depth].p_ext;
3119
3120         if (unlikely(!eh->eh_entries)) {
3121                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3122                                  "EOFBLOCKS_FL set");
3123                 return -EIO;
3124         }
3125         last_ex = EXT_LAST_EXTENT(eh);
3126         /*
3127          * We should clear the EOFBLOCKS_FL flag if we are writing the
3128          * last block in the last extent in the file.  We test this by
3129          * first checking to see if the caller to
3130          * ext4_ext_get_blocks() was interested in the last block (or
3131          * a block beyond the last block) in the current extent.  If
3132          * this turns out to be false, we can bail out from this
3133          * function immediately.
3134          */
3135         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3136             ext4_ext_get_actual_len(last_ex))
3137                 return 0;
3138         /*
3139          * If the caller does appear to be planning to write at or
3140          * beyond the end of the current extent, we then test to see
3141          * if the current extent is the last extent in the file, by
3142          * checking to make sure it was reached via the rightmost node
3143          * at each level of the tree.
3144          */
3145         for (i = depth-1; i >= 0; i--)
3146                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3147                         return 0;
3148         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3149         return ext4_mark_inode_dirty(handle, inode);
3150 }
3151
3152 static int
3153 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3154                         struct ext4_map_blocks *map,
3155                         struct ext4_ext_path *path, int flags,
3156                         unsigned int allocated, ext4_fsblk_t newblock)
3157 {
3158         int ret = 0;
3159         int err = 0;
3160         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3161
3162         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3163                   "block %llu, max_blocks %u, flags %d, allocated %u",
3164                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3165                   flags, allocated);
3166         ext4_ext_show_leaf(inode, path);
3167
3168         /* get_block() before submit the IO, split the extent */
3169         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3170                 ret = ext4_split_unwritten_extents(handle, inode, map,
3171                                                    path, flags);
3172                 /*
3173                  * Flag the inode(non aio case) or end_io struct (aio case)
3174                  * that this IO needs to convertion to written when IO is
3175                  * completed
3176                  */
3177                 if (io)
3178                         io->flag = EXT4_IO_END_UNWRITTEN;
3179                 else
3180                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3181                 if (ext4_should_dioread_nolock(inode))
3182                         map->m_flags |= EXT4_MAP_UNINIT;
3183                 goto out;
3184         }
3185         /* IO end_io complete, convert the filled extent to written */
3186         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3187                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3188                                                         path);
3189                 if (ret >= 0) {
3190                         ext4_update_inode_fsync_trans(handle, inode, 1);
3191                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3192                                                  path, map->m_len);
3193                 } else
3194                         err = ret;
3195                 goto out2;
3196         }
3197         /* buffered IO case */
3198         /*
3199          * repeat fallocate creation request
3200          * we already have an unwritten extent
3201          */
3202         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3203                 goto map_out;
3204
3205         /* buffered READ or buffered write_begin() lookup */
3206         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3207                 /*
3208                  * We have blocks reserved already.  We
3209                  * return allocated blocks so that delalloc
3210                  * won't do block reservation for us.  But
3211                  * the buffer head will be unmapped so that
3212                  * a read from the block returns 0s.
3213                  */
3214                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3215                 goto out1;
3216         }
3217
3218         /* buffered write, writepage time, convert*/
3219         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3220         if (ret >= 0) {
3221                 ext4_update_inode_fsync_trans(handle, inode, 1);
3222                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3223                                          map->m_len);
3224                 if (err < 0)
3225                         goto out2;
3226         }
3227
3228 out:
3229         if (ret <= 0) {
3230                 err = ret;
3231                 goto out2;
3232         } else
3233                 allocated = ret;
3234         map->m_flags |= EXT4_MAP_NEW;
3235         /*
3236          * if we allocated more blocks than requested
3237          * we need to make sure we unmap the extra block
3238          * allocated. The actual needed block will get
3239          * unmapped later when we find the buffer_head marked
3240          * new.
3241          */
3242         if (allocated > map->m_len) {
3243                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3244                                         newblock + map->m_len,
3245                                         allocated - map->m_len);
3246                 allocated = map->m_len;
3247         }
3248
3249         /*
3250          * If we have done fallocate with the offset that is already
3251          * delayed allocated, we would have block reservation
3252          * and quota reservation done in the delayed write path.
3253          * But fallocate would have already updated quota and block
3254          * count for this offset. So cancel these reservation
3255          */
3256         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3257                 ext4_da_update_reserve_space(inode, allocated, 0);
3258
3259 map_out:
3260         map->m_flags |= EXT4_MAP_MAPPED;
3261 out1:
3262         if (allocated > map->m_len)
3263                 allocated = map->m_len;
3264         ext4_ext_show_leaf(inode, path);
3265         map->m_pblk = newblock;
3266         map->m_len = allocated;
3267 out2:
3268         if (path) {
3269                 ext4_ext_drop_refs(path);
3270                 kfree(path);
3271         }
3272         return err ? err : allocated;
3273 }
3274
3275 /*
3276  * Block allocation/map/preallocation routine for extents based files
3277  *
3278  *
3279  * Need to be called with
3280  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3281  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3282  *
3283  * return > 0, number of of blocks already mapped/allocated
3284  *          if create == 0 and these are pre-allocated blocks
3285  *              buffer head is unmapped
3286  *          otherwise blocks are mapped
3287  *
3288  * return = 0, if plain look up failed (blocks have not been allocated)
3289  *          buffer head is unmapped
3290  *
3291  * return < 0, error case.
3292  */
3293 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3294                         struct ext4_map_blocks *map, int flags)
3295 {
3296         struct ext4_ext_path *path = NULL;
3297         struct ext4_extent_header *eh;
3298         struct ext4_extent newex, *ex;
3299         ext4_fsblk_t newblock;
3300         int err = 0, depth, ret;
3301         unsigned int allocated = 0;
3302         struct ext4_allocation_request ar;
3303         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3304
3305         ext_debug("blocks %u/%u requested for inode %lu\n",
3306                   map->m_lblk, map->m_len, inode->i_ino);
3307
3308         /* check in cache */
3309         if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3310                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3311                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3312                                 /*
3313                                  * block isn't allocated yet and
3314                                  * user doesn't want to allocate it
3315                                  */
3316                                 goto out2;
3317                         }
3318                         /* we should allocate requested block */
3319                 } else {
3320                         /* block is already allocated */
3321                         newblock = map->m_lblk
3322                                    - le32_to_cpu(newex.ee_block)
3323                                    + ext4_ext_pblock(&newex);
3324                         /* number of remaining blocks in the extent */
3325                         allocated = ext4_ext_get_actual_len(&newex) -
3326                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3327                         goto out;
3328                 }
3329         }
3330
3331         /* find extent for this block */
3332         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3333         if (IS_ERR(path)) {
3334                 err = PTR_ERR(path);
3335                 path = NULL;
3336                 goto out2;
3337         }
3338
3339         depth = ext_depth(inode);
3340
3341         /*
3342          * consistent leaf must not be empty;
3343          * this situation is possible, though, _during_ tree modification;
3344          * this is why assert can't be put in ext4_ext_find_extent()
3345          */
3346         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3347                 EXT4_ERROR_INODE(inode, "bad extent address "
3348                                  "lblock: %lu, depth: %d pblock %lld",
3349                                  (unsigned long) map->m_lblk, depth,
3350                                  path[depth].p_block);
3351                 err = -EIO;
3352                 goto out2;
3353         }
3354         eh = path[depth].p_hdr;
3355
3356         ex = path[depth].p_ext;
3357         if (ex) {
3358                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3359                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3360                 unsigned short ee_len;
3361
3362                 /*
3363                  * Uninitialized extents are treated as holes, except that
3364                  * we split out initialized portions during a write.
3365                  */
3366                 ee_len = ext4_ext_get_actual_len(ex);
3367                 /* if found extent covers block, simply return it */
3368                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3369                         newblock = map->m_lblk - ee_block + ee_start;
3370                         /* number of remaining blocks in the extent */
3371                         allocated = ee_len - (map->m_lblk - ee_block);
3372                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3373                                   ee_block, ee_len, newblock);
3374
3375                         /* Do not put uninitialized extent in the cache */
3376                         if (!ext4_ext_is_uninitialized(ex)) {
3377                                 ext4_ext_put_in_cache(inode, ee_block,
3378                                                         ee_len, ee_start);
3379                                 goto out;
3380                         }
3381                         ret = ext4_ext_handle_uninitialized_extents(handle,
3382                                         inode, map, path, flags, allocated,
3383                                         newblock);
3384                         return ret;
3385                 }
3386         }
3387
3388         /*
3389          * requested block isn't allocated yet;
3390          * we couldn't try to create block if create flag is zero
3391          */
3392         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3393                 /*
3394                  * put just found gap into cache to speed up
3395                  * subsequent requests
3396                  */
3397                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3398                 goto out2;
3399         }
3400         /*
3401          * Okay, we need to do block allocation.
3402          */
3403
3404         /* find neighbour allocated blocks */
3405         ar.lleft = map->m_lblk;
3406         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3407         if (err)
3408                 goto out2;
3409         ar.lright = map->m_lblk;
3410         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3411         if (err)
3412                 goto out2;
3413
3414         /*
3415          * See if request is beyond maximum number of blocks we can have in
3416          * a single extent. For an initialized extent this limit is
3417          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3418          * EXT_UNINIT_MAX_LEN.
3419          */
3420         if (map->m_len > EXT_INIT_MAX_LEN &&
3421             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3422                 map->m_len = EXT_INIT_MAX_LEN;
3423         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3424                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3425                 map->m_len = EXT_UNINIT_MAX_LEN;
3426
3427         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3428         newex.ee_block = cpu_to_le32(map->m_lblk);
3429         newex.ee_len = cpu_to_le16(map->m_len);
3430         err = ext4_ext_check_overlap(inode, &newex, path);
3431         if (err)
3432                 allocated = ext4_ext_get_actual_len(&newex);
3433         else
3434                 allocated = map->m_len;
3435
3436         /* allocate new block */
3437         ar.inode = inode;
3438         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3439         ar.logical = map->m_lblk;
3440         ar.len = allocated;
3441         if (S_ISREG(inode->i_mode))
3442                 ar.flags = EXT4_MB_HINT_DATA;
3443         else
3444                 /* disable in-core preallocation for non-regular files */
3445                 ar.flags = 0;
3446         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3447         if (!newblock)
3448                 goto out2;
3449         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3450                   ar.goal, newblock, allocated);
3451
3452         /* try to insert new extent into found leaf and return */
3453         ext4_ext_store_pblock(&newex, newblock);
3454         newex.ee_len = cpu_to_le16(ar.len);
3455         /* Mark uninitialized */
3456         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3457                 ext4_ext_mark_uninitialized(&newex);
3458                 /*
3459                  * io_end structure was created for every IO write to an
3460                  * uninitialized extent. To avoid unecessary conversion,
3461                  * here we flag the IO that really needs the conversion.
3462                  * For non asycn direct IO case, flag the inode state
3463                  * that we need to perform convertion when IO is done.
3464                  */
3465                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3466                         if (io)
3467                                 io->flag = EXT4_IO_END_UNWRITTEN;
3468                         else
3469                                 ext4_set_inode_state(inode,
3470                                                      EXT4_STATE_DIO_UNWRITTEN);
3471                 }
3472                 if (ext4_should_dioread_nolock(inode))
3473                         map->m_flags |= EXT4_MAP_UNINIT;
3474         }
3475
3476         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3477         if (err)
3478                 goto out2;
3479
3480         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3481         if (err) {
3482                 /* free data blocks we just allocated */
3483                 /* not a good idea to call discard here directly,
3484                  * but otherwise we'd need to call it every free() */
3485                 ext4_discard_preallocations(inode);
3486                 ext4_free_blocks(handle, inode, 0, ext4_ext_pblock(&newex),
3487                                  ext4_ext_get_actual_len(&newex), 0);
3488                 goto out2;
3489         }
3490
3491         /* previous routine could use block we allocated */
3492         newblock = ext4_ext_pblock(&newex);
3493         allocated = ext4_ext_get_actual_len(&newex);
3494         if (allocated > map->m_len)
3495                 allocated = map->m_len;
3496         map->m_flags |= EXT4_MAP_NEW;
3497
3498         /*
3499          * Update reserved blocks/metadata blocks after successful
3500          * block allocation which had been deferred till now.
3501          */
3502         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3503                 ext4_da_update_reserve_space(inode, allocated, 1);
3504
3505         /*
3506          * Cache the extent and update transaction to commit on fdatasync only
3507          * when it is _not_ an uninitialized extent.
3508          */
3509         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3510                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3511                 ext4_update_inode_fsync_trans(handle, inode, 1);
3512         } else
3513                 ext4_update_inode_fsync_trans(handle, inode, 0);
3514 out:
3515         if (allocated > map->m_len)
3516                 allocated = map->m_len;
3517         ext4_ext_show_leaf(inode, path);
3518         map->m_flags |= EXT4_MAP_MAPPED;
3519         map->m_pblk = newblock;
3520         map->m_len = allocated;
3521 out2:
3522         if (path) {
3523                 ext4_ext_drop_refs(path);
3524                 kfree(path);
3525         }
3526         return err ? err : allocated;
3527 }
3528
3529 void ext4_ext_truncate(struct inode *inode)
3530 {
3531         struct address_space *mapping = inode->i_mapping;
3532         struct super_block *sb = inode->i_sb;
3533         ext4_lblk_t last_block;
3534         handle_t *handle;
3535         int err = 0;
3536
3537         /*
3538          * finish any pending end_io work so we won't run the risk of
3539          * converting any truncated blocks to initialized later
3540          */
3541         ext4_flush_completed_IO(inode);
3542
3543         /*
3544          * probably first extent we're gonna free will be last in block
3545          */
3546         err = ext4_writepage_trans_blocks(inode);
3547         handle = ext4_journal_start(inode, err);
3548         if (IS_ERR(handle))
3549                 return;
3550
3551         if (inode->i_size & (sb->s_blocksize - 1))
3552                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3553
3554         if (ext4_orphan_add(handle, inode))
3555                 goto out_stop;
3556
3557         down_write(&EXT4_I(inode)->i_data_sem);
3558         ext4_ext_invalidate_cache(inode);
3559
3560         ext4_discard_preallocations(inode);
3561
3562         /*
3563          * TODO: optimization is possible here.
3564          * Probably we need not scan at all,
3565          * because page truncation is enough.
3566          */
3567
3568         /* we have to know where to truncate from in crash case */
3569         EXT4_I(inode)->i_disksize = inode->i_size;
3570         ext4_mark_inode_dirty(handle, inode);
3571
3572         last_block = (inode->i_size + sb->s_blocksize - 1)
3573                         >> EXT4_BLOCK_SIZE_BITS(sb);
3574         err = ext4_ext_remove_space(inode, last_block);
3575
3576         /* In a multi-transaction truncate, we only make the final
3577          * transaction synchronous.
3578          */
3579         if (IS_SYNC(inode))
3580                 ext4_handle_sync(handle);
3581
3582 out_stop:
3583         up_write(&EXT4_I(inode)->i_data_sem);
3584         /*
3585          * If this was a simple ftruncate() and the file will remain alive,
3586          * then we need to clear up the orphan record which we created above.
3587          * However, if this was a real unlink then we were called by
3588          * ext4_delete_inode(), and we allow that function to clean up the
3589          * orphan info for us.
3590          */
3591         if (inode->i_nlink)
3592                 ext4_orphan_del(handle, inode);
3593
3594         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3595         ext4_mark_inode_dirty(handle, inode);
3596         ext4_journal_stop(handle);
3597 }
3598
3599 static void ext4_falloc_update_inode(struct inode *inode,
3600                                 int mode, loff_t new_size, int update_ctime)
3601 {
3602         struct timespec now;
3603
3604         if (update_ctime) {
3605                 now = current_fs_time(inode->i_sb);
3606                 if (!timespec_equal(&inode->i_ctime, &now))
3607                         inode->i_ctime = now;
3608         }
3609         /*
3610          * Update only when preallocation was requested beyond
3611          * the file size.
3612          */
3613         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3614                 if (new_size > i_size_read(inode))
3615                         i_size_write(inode, new_size);
3616                 if (new_size > EXT4_I(inode)->i_disksize)
3617                         ext4_update_i_disksize(inode, new_size);
3618         } else {
3619                 /*
3620                  * Mark that we allocate beyond EOF so the subsequent truncate
3621                  * can proceed even if the new size is the same as i_size.
3622                  */
3623                 if (new_size > i_size_read(inode))
3624                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3625         }
3626
3627 }
3628
3629 /*
3630  * preallocate space for a file. This implements ext4's fallocate inode
3631  * operation, which gets called from sys_fallocate system call.
3632  * For block-mapped files, posix_fallocate should fall back to the method
3633  * of writing zeroes to the required new blocks (the same behavior which is
3634  * expected for file systems which do not support fallocate() system call).
3635  */
3636 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3637 {
3638         handle_t *handle;
3639         loff_t new_size;
3640         unsigned int max_blocks;
3641         int ret = 0;
3642         int ret2 = 0;
3643         int retries = 0;
3644         struct ext4_map_blocks map;
3645         unsigned int credits, blkbits = inode->i_blkbits;
3646
3647         /*
3648          * currently supporting (pre)allocate mode for extent-based
3649          * files _only_
3650          */
3651         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3652                 return -EOPNOTSUPP;
3653
3654         /* preallocation to directories is currently not supported */
3655         if (S_ISDIR(inode->i_mode))
3656                 return -ENODEV;
3657
3658         map.m_lblk = offset >> blkbits;
3659         /*
3660          * We can't just convert len to max_blocks because
3661          * If blocksize = 4096 offset = 3072 and len = 2048
3662          */
3663         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3664                 - map.m_lblk;
3665         /*
3666          * credits to insert 1 extent into extent tree
3667          */
3668         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3669         mutex_lock(&inode->i_mutex);
3670         ret = inode_newsize_ok(inode, (len + offset));
3671         if (ret) {
3672                 mutex_unlock(&inode->i_mutex);
3673                 return ret;
3674         }
3675 retry:
3676         while (ret >= 0 && ret < max_blocks) {
3677                 map.m_lblk = map.m_lblk + ret;
3678                 map.m_len = max_blocks = max_blocks - ret;
3679                 handle = ext4_journal_start(inode, credits);
3680                 if (IS_ERR(handle)) {
3681                         ret = PTR_ERR(handle);
3682                         break;
3683                 }
3684                 ret = ext4_map_blocks(handle, inode, &map,
3685                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3686                 if (ret <= 0) {
3687 #ifdef EXT4FS_DEBUG
3688                         WARN_ON(ret <= 0);
3689                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3690                                     "returned error inode#%lu, block=%u, "
3691                                     "max_blocks=%u", __func__,
3692                                     inode->i_ino, map.m_lblk, max_blocks);
3693 #endif
3694                         ext4_mark_inode_dirty(handle, inode);
3695                         ret2 = ext4_journal_stop(handle);
3696                         break;
3697                 }
3698                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3699                                                 blkbits) >> blkbits))
3700                         new_size = offset + len;
3701                 else
3702                         new_size = (map.m_lblk + ret) << blkbits;
3703
3704                 ext4_falloc_update_inode(inode, mode, new_size,
3705                                          (map.m_flags & EXT4_MAP_NEW));
3706                 ext4_mark_inode_dirty(handle, inode);
3707                 ret2 = ext4_journal_stop(handle);
3708                 if (ret2)
3709                         break;
3710         }
3711         if (ret == -ENOSPC &&
3712                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3713                 ret = 0;
3714                 goto retry;
3715         }
3716         mutex_unlock(&inode->i_mutex);
3717         return ret > 0 ? ret2 : ret;
3718 }
3719
3720 /*
3721  * This function convert a range of blocks to written extents
3722  * The caller of this function will pass the start offset and the size.
3723  * all unwritten extents within this range will be converted to
3724  * written extents.
3725  *
3726  * This function is called from the direct IO end io call back
3727  * function, to convert the fallocated extents after IO is completed.
3728  * Returns 0 on success.
3729  */
3730 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3731                                     ssize_t len)
3732 {
3733         handle_t *handle;
3734         unsigned int max_blocks;
3735         int ret = 0;
3736         int ret2 = 0;
3737         struct ext4_map_blocks map;
3738         unsigned int credits, blkbits = inode->i_blkbits;
3739
3740         map.m_lblk = offset >> blkbits;
3741         /*
3742          * We can't just convert len to max_blocks because
3743          * If blocksize = 4096 offset = 3072 and len = 2048
3744          */
3745         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3746                       map.m_lblk);
3747         /*
3748          * credits to insert 1 extent into extent tree
3749          */
3750         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3751         while (ret >= 0 && ret < max_blocks) {
3752                 map.m_lblk += ret;
3753                 map.m_len = (max_blocks -= ret);
3754                 handle = ext4_journal_start(inode, credits);
3755                 if (IS_ERR(handle)) {
3756                         ret = PTR_ERR(handle);
3757                         break;
3758                 }
3759                 ret = ext4_map_blocks(handle, inode, &map,
3760                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3761                 if (ret <= 0) {
3762                         WARN_ON(ret <= 0);
3763                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3764                                     "returned error inode#%lu, block=%u, "
3765                                     "max_blocks=%u", __func__,
3766                                     inode->i_ino, map.m_lblk, map.m_len);
3767                 }
3768                 ext4_mark_inode_dirty(handle, inode);
3769                 ret2 = ext4_journal_stop(handle);
3770                 if (ret <= 0 || ret2 )
3771                         break;
3772         }
3773         return ret > 0 ? ret2 : ret;
3774 }
3775 /*
3776  * Callback function called for each extent to gather FIEMAP information.
3777  */
3778 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3779                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3780                        void *data)
3781 {
3782         struct fiemap_extent_info *fieinfo = data;
3783         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3784         __u64   logical;
3785         __u64   physical;
3786         __u64   length;
3787         __u32   flags = 0;
3788         int     error;
3789
3790         logical =  (__u64)newex->ec_block << blksize_bits;
3791
3792         if (newex->ec_start == 0) {
3793                 pgoff_t offset;
3794                 struct page *page;
3795                 struct buffer_head *bh = NULL;
3796
3797                 offset = logical >> PAGE_SHIFT;
3798                 page = find_get_page(inode->i_mapping, offset);
3799                 if (!page || !page_has_buffers(page))
3800                         return EXT_CONTINUE;
3801
3802                 bh = page_buffers(page);
3803
3804                 if (!bh)
3805                         return EXT_CONTINUE;
3806
3807                 if (buffer_delay(bh)) {
3808                         flags |= FIEMAP_EXTENT_DELALLOC;
3809                         page_cache_release(page);
3810                 } else {
3811                         page_cache_release(page);
3812                         return EXT_CONTINUE;
3813                 }
3814         }
3815
3816         physical = (__u64)newex->ec_start << blksize_bits;
3817         length =   (__u64)newex->ec_len << blksize_bits;
3818
3819         if (ex && ext4_ext_is_uninitialized(ex))
3820                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3821
3822         /*
3823          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3824          *
3825          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3826          * this also indicates no more allocated blocks.
3827          *
3828          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3829          */
3830         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3831             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3832                 loff_t size = i_size_read(inode);
3833                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3834
3835                 flags |= FIEMAP_EXTENT_LAST;
3836                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3837                     logical+length > size)
3838                         length = (size - logical + bs - 1) & ~(bs-1);
3839         }
3840
3841         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3842                                         length, flags);
3843         if (error < 0)
3844                 return error;
3845         if (error == 1)
3846                 return EXT_BREAK;
3847
3848         return EXT_CONTINUE;
3849 }
3850
3851 /* fiemap flags we can handle specified here */
3852 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3853
3854 static int ext4_xattr_fiemap(struct inode *inode,
3855                                 struct fiemap_extent_info *fieinfo)
3856 {
3857         __u64 physical = 0;
3858         __u64 length;
3859         __u32 flags = FIEMAP_EXTENT_LAST;
3860         int blockbits = inode->i_sb->s_blocksize_bits;
3861         int error = 0;
3862
3863         /* in-inode? */
3864         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3865                 struct ext4_iloc iloc;
3866                 int offset;     /* offset of xattr in inode */
3867
3868                 error = ext4_get_inode_loc(inode, &iloc);
3869                 if (error)
3870                         return error;
3871                 physical = iloc.bh->b_blocknr << blockbits;
3872                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3873                                 EXT4_I(inode)->i_extra_isize;
3874                 physical += offset;
3875                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3876                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3877                 brelse(iloc.bh);
3878         } else { /* external block */
3879                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3880                 length = inode->i_sb->s_blocksize;
3881         }
3882
3883         if (physical)
3884                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3885                                                 length, flags);
3886         return (error < 0 ? error : 0);
3887 }
3888
3889 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3890                 __u64 start, __u64 len)
3891 {
3892         ext4_lblk_t start_blk;
3893         int error = 0;
3894
3895         /* fallback to generic here if not in extents fmt */
3896         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3897                 return generic_block_fiemap(inode, fieinfo, start, len,
3898                         ext4_get_block);
3899
3900         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3901                 return -EBADR;
3902
3903         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3904                 error = ext4_xattr_fiemap(inode, fieinfo);
3905         } else {
3906                 ext4_lblk_t len_blks;
3907                 __u64 last_blk;
3908
3909                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3910                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3911                 if (last_blk >= EXT_MAX_BLOCK)
3912                         last_blk = EXT_MAX_BLOCK-1;
3913                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3914
3915                 /*
3916                  * Walk the extent tree gathering extent information.
3917                  * ext4_ext_fiemap_cb will push extents back to user.
3918                  */
3919                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3920                                           ext4_ext_fiemap_cb, fieinfo);
3921         }
3922
3923         return error;
3924 }
3925