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
  48 /*
  49  * ext_pblock:
  50  * combine low and high parts of physical block number into ext4_fsblk_t
  51  */
  52 ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
  53 {
  54         ext4_fsblk_t block;
  55
  56         block = le32_to_cpu(ex->ee_start_lo);
  57         block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
  58         return block;
  59 }
  60
  61 /*
  62  * idx_pblock:
  63  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
  64  */
  65 ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
  66 {
  67         ext4_fsblk_t block;
  68
  69         block = le32_to_cpu(ix->ei_leaf_lo);
  70         block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
  71         return block;
  72 }
  73
  74 /*
  75  * ext4_ext_store_pblock:
  76  * stores a large physical block number into an extent struct,
  77  * breaking it into parts
  78  */
  79 void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
  80 {
  81         ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  82         ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  83 }
  84
  85 /*
  86  * ext4_idx_store_pblock:
  87  * stores a large physical block number into an index struct,
  88  * breaking it into parts
  89  */
  90 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
  91 {
  92         ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
  93         ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
  94 }
  95
  96 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  97                                             struct inode *inode,
  98                                             int needed)
  99 {
 100         int err;
 101
 102         if (!ext4_handle_valid(handle))
 103                 return 0;
 104         if (handle->h_buffer_credits > needed)
 105                 return 0;
 106         err = ext4_journal_extend(handle, needed);
 107         if (err <= 0)
 108                 return err;
 109         err = ext4_truncate_restart_trans(handle, inode, needed);
 110         if (err == 0)
 111                 err = -EAGAIN;
 112
 113         return err;
 114 }
 115
 116 /*
 117  * could return:
 118  *  - EROFS
 119  *  - ENOMEM
 120  */
 121 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 122                                 struct ext4_ext_path *path)
 123 {
 124         if (path->p_bh) {
 125                 /* path points to block */
 126                 return ext4_journal_get_write_access(handle, path->p_bh);
 127         }
 128         /* path points to leaf/index in inode body */
 129         /* we use in-core data, no need to protect them */
 130         return 0;
 131 }
 132
 133 /*
 134  * could return:
 135  *  - EROFS
 136  *  - ENOMEM
 137  *  - EIO
 138  */
 139 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
 140                                 struct ext4_ext_path *path)
 141 {
 142         int err;
 143         if (path->p_bh) {
 144                 /* path points to block */
 145                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
 146         } else {
 147                 /* path points to leaf/index in inode body */
 148                 err = ext4_mark_inode_dirty(handle, inode);
 149         }
 150         return err;
 151 }
 152
 153 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 154                               struct ext4_ext_path *path,
 155                               ext4_lblk_t block)
 156 {
 157         struct ext4_inode_info *ei = EXT4_I(inode);
 158         ext4_fsblk_t bg_start;
 159         ext4_fsblk_t last_block;
 160         ext4_grpblk_t colour;
 161         ext4_group_t block_group;
 162         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 163         int depth;
 164
 165         if (path) {
 166                 struct ext4_extent *ex;
 167                 depth = path->p_depth;
 168
 169                 /* try to predict block placement */
 170                 ex = path[depth].p_ext;
 171                 if (ex)
 172                         return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
 173
 174                 /* it looks like index is empty;
 175                  * try to find starting block from index itself */
 176                 if (path[depth].p_bh)
 177                         return path[depth].p_bh->b_blocknr;
 178         }
 179
 180         /* OK. use inode's group */
 181         block_group = ei->i_block_group;
 182         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 183                 /*
 184                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 185                  * block groups per flexgroup, reserve the first block
 186                  * group for directories and special files.  Regular
 187                  * files will start at the second block group.  This
 188                  * tends to speed up directory access and improves
 189                  * fsck times.
 190                  */
 191                 block_group &= ~(flex_size-1);
 192                 if (S_ISREG(inode->i_mode))
 193                         block_group++;
 194         }
 195         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 196         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 197
 198         /*
 199          * If we are doing delayed allocation, we don't need take
 200          * colour into account.
 201          */
 202         if (test_opt(inode->i_sb, DELALLOC))
 203                 return bg_start;
 204
 205         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 206                 colour = (current->pid % 16) *
 207                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 208         else
 209                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 210         return bg_start + colour + block;
 211 }
 212
 213 /*
 214  * Allocation for a meta data block
 215  */
 216 static ext4_fsblk_t
 217 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 218                         struct ext4_ext_path *path,
 219                         struct ext4_extent *ex, int *err)
 220 {
 221         ext4_fsblk_t goal, newblock;
 222
 223         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 224         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
 225         return newblock;
 226 }
 227
 228 static inline int ext4_ext_space_block(struct inode *inode, int check)
 229 {
 230         int size;
 231
 232         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 233                         / sizeof(struct ext4_extent);
 234         if (!check) {
 235 #ifdef AGGRESSIVE_TEST
 236                 if (size > 6)
 237                         size = 6;
 238 #endif
 239         }
 240         return size;
 241 }
 242
 243 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 244 {
 245         int size;
 246
 247         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 248                         / sizeof(struct ext4_extent_idx);
 249         if (!check) {
 250 #ifdef AGGRESSIVE_TEST
 251                 if (size > 5)
 252                         size = 5;
 253 #endif
 254         }
 255         return size;
 256 }
 257
 258 static inline int ext4_ext_space_root(struct inode *inode, int check)
 259 {
 260         int size;
 261
 262         size = sizeof(EXT4_I(inode)->i_data);
 263         size -= sizeof(struct ext4_extent_header);
 264         size /= sizeof(struct ext4_extent);
 265         if (!check) {
 266 #ifdef AGGRESSIVE_TEST
 267                 if (size > 3)
 268                         size = 3;
 269 #endif
 270         }
 271         return size;
 272 }
 273
 274 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 275 {
 276         int size;
 277
 278         size = sizeof(EXT4_I(inode)->i_data);
 279         size -= sizeof(struct ext4_extent_header);
 280         size /= sizeof(struct ext4_extent_idx);
 281         if (!check) {
 282 #ifdef AGGRESSIVE_TEST
 283                 if (size > 4)
 284                         size = 4;
 285 #endif
 286         }
 287         return size;
 288 }
 289
 290 /*
 291  * Calculate the number of metadata blocks needed
 292  * to allocate @blocks
 293  * Worse case is one block per extent
 294  */
 295 int ext4_ext_calc_metadata_amount(struct inode *inode, sector_t lblock)
 296 {
 297         struct ext4_inode_info *ei = EXT4_I(inode);
 298         int idxs, num = 0;
 299
 300         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 301                 / sizeof(struct ext4_extent_idx));
 302
 303         /*
 304          * If the new delayed allocation block is contiguous with the
 305          * previous da block, it can share index blocks with the
 306          * previous block, so we only need to allocate a new index
 307          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 308          * an additional index block, and at ldxs**3 blocks, yet
 309          * another index blocks.
 310          */
 311         if (ei->i_da_metadata_calc_len &&
 312             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 313                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 314                         num++;
 315                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 316                         num++;
 317                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 318                         num++;
 319                         ei->i_da_metadata_calc_len = 0;
 320                 } else
 321                         ei->i_da_metadata_calc_len++;
 322                 ei->i_da_metadata_calc_last_lblock++;
 323                 return num;
 324         }
 325
 326         /*
 327          * In the worst case we need a new set of index blocks at
 328          * every level of the inode's extent tree.
 329          */
 330         ei->i_da_metadata_calc_len = 1;
 331         ei->i_da_metadata_calc_last_lblock = lblock;
 332         return ext_depth(inode) + 1;
 333 }
 334
 335 static int
 336 ext4_ext_max_entries(struct inode *inode, int depth)
 337 {
 338         int max;
 339
 340         if (depth == ext_depth(inode)) {
 341                 if (depth == 0)
 342                         max = ext4_ext_space_root(inode, 1);
 343                 else
 344                         max = ext4_ext_space_root_idx(inode, 1);
 345         } else {
 346                 if (depth == 0)
 347                         max = ext4_ext_space_block(inode, 1);
 348                 else
 349                         max = ext4_ext_space_block_idx(inode, 1);
 350         }
 351
 352         return max;
 353 }
 354
 355 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 356 {
 357         ext4_fsblk_t block = ext_pblock(ext);
 358         int len = ext4_ext_get_actual_len(ext);
 359
 360         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 361 }
 362
 363 static int ext4_valid_extent_idx(struct inode *inode,
 364                                 struct ext4_extent_idx *ext_idx)
 365 {
 366         ext4_fsblk_t block = idx_pblock(ext_idx);
 367
 368         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 369 }
 370
 371 static int ext4_valid_extent_entries(struct inode *inode,
 372                                 struct ext4_extent_header *eh,
 373                                 int depth)
 374 {
 375         struct ext4_extent *ext;
 376         struct ext4_extent_idx *ext_idx;
 377         unsigned short entries;
 378         if (eh->eh_entries == 0)
 379                 return 1;
 380
 381         entries = le16_to_cpu(eh->eh_entries);
 382
 383         if (depth == 0) {
 384                 /* leaf entries */
 385                 ext = EXT_FIRST_EXTENT(eh);
 386                 while (entries) {
 387                         if (!ext4_valid_extent(inode, ext))
 388                                 return 0;
 389                         ext++;
 390                         entries--;
 391                 }
 392         } else {
 393                 ext_idx = EXT_FIRST_INDEX(eh);
 394                 while (entries) {
 395                         if (!ext4_valid_extent_idx(inode, ext_idx))
 396                                 return 0;
 397                         ext_idx++;
 398                         entries--;
 399                 }
 400         }
 401         return 1;
 402 }
 403
 404 static int __ext4_ext_check(const char *function, struct inode *inode,
 405                                         struct ext4_extent_header *eh,
 406                                         int depth)
 407 {
 408         const char *error_msg;
 409         int max = 0;
 410
 411         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 412                 error_msg = "invalid magic";
 413                 goto corrupted;
 414         }
 415         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 416                 error_msg = "unexpected eh_depth";
 417                 goto corrupted;
 418         }
 419         if (unlikely(eh->eh_max == 0)) {
 420                 error_msg = "invalid eh_max";
 421                 goto corrupted;
 422         }
 423         max = ext4_ext_max_entries(inode, depth);
 424         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 425                 error_msg = "too large eh_max";
 426                 goto corrupted;
 427         }
 428         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 429                 error_msg = "invalid eh_entries";
 430                 goto corrupted;
 431         }
 432         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 433                 error_msg = "invalid extent entries";
 434                 goto corrupted;
 435         }
 436         return 0;
 437
 438 corrupted:
 439         ext4_error_inode(function, inode,
 440                         "bad header/extent: %s - magic %x, "
 441                         "entries %u, max %u(%u), depth %u(%u)",
 442                         error_msg, le16_to_cpu(eh->eh_magic),
 443                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 444                         max, le16_to_cpu(eh->eh_depth), depth);
 445
 446         return -EIO;
 447 }
 448
 449 #define ext4_ext_check(inode, eh, depth)        \
 450         __ext4_ext_check(__func__, inode, eh, depth)
 451
 452 int ext4_ext_check_inode(struct inode *inode)
 453 {
 454         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 455 }
 456
 457 #ifdef EXT_DEBUG
 458 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 459 {
 460         int k, l = path->p_depth;
 461
 462         ext_debug("path:");
 463         for (k = 0; k <= l; k++, path++) {
 464                 if (path->p_idx) {
 465                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 466                             idx_pblock(path->p_idx));
 467                 } else if (path->p_ext) {
 468                         ext_debug("  %d:[%d]%d:%llu ",
 469                                   le32_to_cpu(path->p_ext->ee_block),
 470                                   ext4_ext_is_uninitialized(path->p_ext),
 471                                   ext4_ext_get_actual_len(path->p_ext),
 472                                   ext_pblock(path->p_ext));
 473                 } else
 474                         ext_debug("  []");
 475         }
 476         ext_debug("\n");
 477 }
 478
 479 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 480 {
 481         int depth = ext_depth(inode);
 482         struct ext4_extent_header *eh;
 483         struct ext4_extent *ex;
 484         int i;
 485
 486         if (!path)
 487                 return;
 488
 489         eh = path[depth].p_hdr;
 490         ex = EXT_FIRST_EXTENT(eh);
 491
 492         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 493
 494         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 495                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 496                           ext4_ext_is_uninitialized(ex),
 497                           ext4_ext_get_actual_len(ex), ext_pblock(ex));
 498         }
 499         ext_debug("\n");
 500 }
 501 #else
 502 #define ext4_ext_show_path(inode, path)
 503 #define ext4_ext_show_leaf(inode, path)
 504 #endif
 505
 506 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 507 {
 508         int depth = path->p_depth;
 509         int i;
 510
 511         for (i = 0; i <= depth; i++, path++)
 512                 if (path->p_bh) {
 513                         brelse(path->p_bh);
 514                         path->p_bh = NULL;
 515                 }
 516 }
 517
 518 /*
 519  * ext4_ext_binsearch_idx:
 520  * binary search for the closest index of the given block
 521  * the header must be checked before calling this
 522  */
 523 static void
 524 ext4_ext_binsearch_idx(struct inode *inode,
 525                         struct ext4_ext_path *path, ext4_lblk_t block)
 526 {
 527         struct ext4_extent_header *eh = path->p_hdr;
 528         struct ext4_extent_idx *r, *l, *m;
 529
 530
 531         ext_debug("binsearch for %u(idx):  ", block);
 532
 533         l = EXT_FIRST_INDEX(eh) + 1;
 534         r = EXT_LAST_INDEX(eh);
 535         while (l <= r) {
 536                 m = l + (r - l) / 2;
 537                 if (block < le32_to_cpu(m->ei_block))
 538                         r = m - 1;
 539                 else
 540                         l = m + 1;
 541                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 542                                 m, le32_to_cpu(m->ei_block),
 543                                 r, le32_to_cpu(r->ei_block));
 544         }
 545
 546         path->p_idx = l - 1;
 547         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 548                   idx_pblock(path->p_idx));
 549
 550 #ifdef CHECK_BINSEARCH
 551         {
 552                 struct ext4_extent_idx *chix, *ix;
 553                 int k;
 554
 555                 chix = ix = EXT_FIRST_INDEX(eh);
 556                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 557                   if (k != 0 &&
 558                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 559                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 560                                        "first=0x%p\n", k,
 561                                        ix, EXT_FIRST_INDEX(eh));
 562                                 printk(KERN_DEBUG "%u <= %u\n",
 563                                        le32_to_cpu(ix->ei_block),
 564                                        le32_to_cpu(ix[-1].ei_block));
 565                         }
 566                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 567                                            <= le32_to_cpu(ix[-1].ei_block));
 568                         if (block < le32_to_cpu(ix->ei_block))
 569                                 break;
 570                         chix = ix;
 571                 }
 572                 BUG_ON(chix != path->p_idx);
 573         }
 574 #endif
 575
 576 }
 577
 578 /*
 579  * ext4_ext_binsearch:
 580  * binary search for closest extent of the given block
 581  * the header must be checked before calling this
 582  */
 583 static void
 584 ext4_ext_binsearch(struct inode *inode,
 585                 struct ext4_ext_path *path, ext4_lblk_t block)
 586 {
 587         struct ext4_extent_header *eh = path->p_hdr;
 588         struct ext4_extent *r, *l, *m;
 589
 590         if (eh->eh_entries == 0) {
 591                 /*
 592                  * this leaf is empty:
 593                  * we get such a leaf in split/add case
 594                  */
 595                 return;
 596         }
 597
 598         ext_debug("binsearch for %u:  ", block);
 599
 600         l = EXT_FIRST_EXTENT(eh) + 1;
 601         r = EXT_LAST_EXTENT(eh);
 602
 603         while (l <= r) {
 604                 m = l + (r - l) / 2;
 605                 if (block < le32_to_cpu(m->ee_block))
 606                         r = m - 1;
 607                 else
 608                         l = m + 1;
 609                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 610                                 m, le32_to_cpu(m->ee_block),
 611                                 r, le32_to_cpu(r->ee_block));
 612         }
 613
 614         path->p_ext = l - 1;
 615         ext_debug("  -> %d:%llu:[%d]%d ",
 616                         le32_to_cpu(path->p_ext->ee_block),
 617                         ext_pblock(path->p_ext),
 618                         ext4_ext_is_uninitialized(path->p_ext),
 619                         ext4_ext_get_actual_len(path->p_ext));
 620
 621 #ifdef CHECK_BINSEARCH
 622         {
 623                 struct ext4_extent *chex, *ex;
 624                 int k;
 625
 626                 chex = ex = EXT_FIRST_EXTENT(eh);
 627                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 628                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 629                                           <= le32_to_cpu(ex[-1].ee_block));
 630                         if (block < le32_to_cpu(ex->ee_block))
 631                                 break;
 632                         chex = ex;
 633                 }
 634                 BUG_ON(chex != path->p_ext);
 635         }
 636 #endif
 637
 638 }
 639
 640 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 641 {
 642         struct ext4_extent_header *eh;
 643
 644         eh = ext_inode_hdr(inode);
 645         eh->eh_depth = 0;
 646         eh->eh_entries = 0;
 647         eh->eh_magic = EXT4_EXT_MAGIC;
 648         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 649         ext4_mark_inode_dirty(handle, inode);
 650         ext4_ext_invalidate_cache(inode);
 651         return 0;
 652 }
 653
 654 struct ext4_ext_path *
 655 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 656                                         struct ext4_ext_path *path)
 657 {
 658         struct ext4_extent_header *eh;
 659         struct buffer_head *bh;
 660         short int depth, i, ppos = 0, alloc = 0;
 661
 662         eh = ext_inode_hdr(inode);
 663         depth = ext_depth(inode);
 664
 665         /* account possible depth increase */
 666         if (!path) {
 667                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 668                                 GFP_NOFS);
 669                 if (!path)
 670                         return ERR_PTR(-ENOMEM);
 671                 alloc = 1;
 672         }
 673         path[0].p_hdr = eh;
 674         path[0].p_bh = NULL;
 675
 676         i = depth;
 677         /* walk through the tree */
 678         while (i) {
 679                 int need_to_validate = 0;
 680
 681                 ext_debug("depth %d: num %d, max %d\n",
 682                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 683
 684                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 685                 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
 686                 path[ppos].p_depth = i;
 687                 path[ppos].p_ext = NULL;
 688
 689                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 690                 if (unlikely(!bh))
 691                         goto err;
 692                 if (!bh_uptodate_or_lock(bh)) {
 693                         if (bh_submit_read(bh) < 0) {
 694                                 put_bh(bh);
 695                                 goto err;
 696                         }
 697                         /* validate the extent entries */
 698                         need_to_validate = 1;
 699                 }
 700                 eh = ext_block_hdr(bh);
 701                 ppos++;
 702                 if (unlikely(ppos > depth)) {
 703                         put_bh(bh);
 704                         EXT4_ERROR_INODE(inode,
 705                                          "ppos %d > depth %d", ppos, depth);
 706                         goto err;
 707                 }
 708                 path[ppos].p_bh = bh;
 709                 path[ppos].p_hdr = eh;
 710                 i--;
 711
 712                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 713                         goto err;
 714         }
 715
 716         path[ppos].p_depth = i;
 717         path[ppos].p_ext = NULL;
 718         path[ppos].p_idx = NULL;
 719
 720         /* find extent */
 721         ext4_ext_binsearch(inode, path + ppos, block);
 722         /* if not an empty leaf */
 723         if (path[ppos].p_ext)
 724                 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
 725
 726         ext4_ext_show_path(inode, path);
 727
 728         return path;
 729
 730 err:
 731         ext4_ext_drop_refs(path);
 732         if (alloc)
 733                 kfree(path);
 734         return ERR_PTR(-EIO);
 735 }
 736
 737 /*
 738  * ext4_ext_insert_index:
 739  * insert new index [@logical;@ptr] into the block at @curp;
 740  * check where to insert: before @curp or after @curp
 741  */
 742 int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 743                                 struct ext4_ext_path *curp,
 744                                 int logical, ext4_fsblk_t ptr)
 745 {
 746         struct ext4_extent_idx *ix;
 747         int len, err;
 748
 749         err = ext4_ext_get_access(handle, inode, curp);
 750         if (err)
 751                 return err;
 752
 753         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 754                 EXT4_ERROR_INODE(inode,
 755                                  "logical %d == ei_block %d!",
 756                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 757                 return -EIO;
 758         }
 759         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 760         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 761                 /* insert after */
 762                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 763                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 764                         len = len < 0 ? 0 : len;
 765                         ext_debug("insert new index %d after: %llu. "
 766                                         "move %d from 0x%p to 0x%p\n",
 767                                         logical, ptr, len,
 768                                         (curp->p_idx + 1), (curp->p_idx + 2));
 769                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 770                 }
 771                 ix = curp->p_idx + 1;
 772         } else {
 773                 /* insert before */
 774                 len = len * sizeof(struct ext4_extent_idx);
 775                 len = len < 0 ? 0 : len;
 776                 ext_debug("insert new index %d before: %llu. "
 777                                 "move %d from 0x%p to 0x%p\n",
 778                                 logical, ptr, len,
 779                                 curp->p_idx, (curp->p_idx + 1));
 780                 memmove(curp->p_idx + 1, curp->p_idx, len);
 781                 ix = curp->p_idx;
 782         }
 783
 784         ix->ei_block = cpu_to_le32(logical);
 785         ext4_idx_store_pblock(ix, ptr);
 786         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 787
 788         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 789                              > le16_to_cpu(curp->p_hdr->eh_max))) {
 790                 EXT4_ERROR_INODE(inode,
 791                                  "logical %d == ei_block %d!",
 792                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 793                 return -EIO;
 794         }
 795         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 796                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 797                 return -EIO;
 798         }
 799
 800         err = ext4_ext_dirty(handle, inode, curp);
 801         ext4_std_error(inode->i_sb, err);
 802
 803         return err;
 804 }
 805
 806 /*
 807  * ext4_ext_split:
 808  * inserts new subtree into the path, using free index entry
 809  * at depth @at:
 810  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 811  * - makes decision where to split
 812  * - moves remaining extents and index entries (right to the split point)
 813  *   into the newly allocated blocks
 814  * - initializes subtree
 815  */
 816 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 817                                 struct ext4_ext_path *path,
 818                                 struct ext4_extent *newext, int at)
 819 {
 820         struct buffer_head *bh = NULL;
 821         int depth = ext_depth(inode);
 822         struct ext4_extent_header *neh;
 823         struct ext4_extent_idx *fidx;
 824         struct ext4_extent *ex;
 825         int i = at, k, m, a;
 826         ext4_fsblk_t newblock, oldblock;
 827         __le32 border;
 828         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 829         int err = 0;
 830
 831         /* make decision: where to split? */
 832         /* FIXME: now decision is simplest: at current extent */
 833
 834         /* if current leaf will be split, then we should use
 835          * border from split point */
 836         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 837                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 838                 return -EIO;
 839         }
 840         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 841                 border = path[depth].p_ext[1].ee_block;
 842                 ext_debug("leaf will be split."
 843                                 " next leaf starts at %d\n",
 844                                   le32_to_cpu(border));
 845         } else {
 846                 border = newext->ee_block;
 847                 ext_debug("leaf will be added."
 848                                 " next leaf starts at %d\n",
 849                                 le32_to_cpu(border));
 850         }
 851
 852         /*
 853          * If error occurs, then we break processing
 854          * and mark filesystem read-only. index won't
 855          * be inserted and tree will be in consistent
 856          * state. Next mount will repair buffers too.
 857          */
 858
 859         /*
 860          * Get array to track all allocated blocks.
 861          * We need this to handle errors and free blocks
 862          * upon them.
 863          */
 864         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 865         if (!ablocks)
 866                 return -ENOMEM;
 867
 868         /* allocate all needed blocks */
 869         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 870         for (a = 0; a < depth - at; a++) {
 871                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 872                                                    newext, &err);
 873                 if (newblock == 0)
 874                         goto cleanup;
 875                 ablocks[a] = newblock;
 876         }
 877
 878         /* initialize new leaf */
 879         newblock = ablocks[--a];
 880         if (unlikely(newblock == 0)) {
 881                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 882                 err = -EIO;
 883                 goto cleanup;
 884         }
 885         bh = sb_getblk(inode->i_sb, newblock);
 886         if (!bh) {
 887                 err = -EIO;
 888                 goto cleanup;
 889         }
 890         lock_buffer(bh);
 891
 892         err = ext4_journal_get_create_access(handle, bh);
 893         if (err)
 894                 goto cleanup;
 895
 896         neh = ext_block_hdr(bh);
 897         neh->eh_entries = 0;
 898         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 899         neh->eh_magic = EXT4_EXT_MAGIC;
 900         neh->eh_depth = 0;
 901         ex = EXT_FIRST_EXTENT(neh);
 902
 903         /* move remainder of path[depth] to the new leaf */
 904         if (unlikely(path[depth].p_hdr->eh_entries !=
 905                      path[depth].p_hdr->eh_max)) {
 906                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 907                                  path[depth].p_hdr->eh_entries,
 908                                  path[depth].p_hdr->eh_max);
 909                 err = -EIO;
 910                 goto cleanup;
 911         }
 912         /* start copy from next extent */
 913         /* TODO: we could do it by single memmove */
 914         m = 0;
 915         path[depth].p_ext++;
 916         while (path[depth].p_ext <=
 917                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
 918                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 919                                 le32_to_cpu(path[depth].p_ext->ee_block),
 920                                 ext_pblock(path[depth].p_ext),
 921                                 ext4_ext_is_uninitialized(path[depth].p_ext),
 922                                 ext4_ext_get_actual_len(path[depth].p_ext),
 923                                 newblock);
 924                 /*memmove(ex++, path[depth].p_ext++,
 925                                 sizeof(struct ext4_extent));
 926                 neh->eh_entries++;*/
 927                 path[depth].p_ext++;
 928                 m++;
 929         }
 930         if (m) {
 931                 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
 932                 le16_add_cpu(&neh->eh_entries, m);
 933         }
 934
 935         set_buffer_uptodate(bh);
 936         unlock_buffer(bh);
 937
 938         err = ext4_handle_dirty_metadata(handle, inode, bh);
 939         if (err)
 940                 goto cleanup;
 941         brelse(bh);
 942         bh = NULL;
 943
 944         /* correct old leaf */
 945         if (m) {
 946                 err = ext4_ext_get_access(handle, inode, path + depth);
 947                 if (err)
 948                         goto cleanup;
 949                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 950                 err = ext4_ext_dirty(handle, inode, path + depth);
 951                 if (err)
 952                         goto cleanup;
 953
 954         }
 955
 956         /* create intermediate indexes */
 957         k = depth - at - 1;
 958         if (unlikely(k < 0)) {
 959                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 960                 err = -EIO;
 961                 goto cleanup;
 962         }
 963         if (k)
 964                 ext_debug("create %d intermediate indices\n", k);
 965         /* insert new index into current index block */
 966         /* current depth stored in i var */
 967         i = depth - 1;
 968         while (k--) {
 969                 oldblock = newblock;
 970                 newblock = ablocks[--a];
 971                 bh = sb_getblk(inode->i_sb, newblock);
 972                 if (!bh) {
 973                         err = -EIO;
 974                         goto cleanup;
 975                 }
 976                 lock_buffer(bh);
 977
 978                 err = ext4_journal_get_create_access(handle, bh);
 979                 if (err)
 980                         goto cleanup;
 981
 982                 neh = ext_block_hdr(bh);
 983                 neh->eh_entries = cpu_to_le16(1);
 984                 neh->eh_magic = EXT4_EXT_MAGIC;
 985                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 986                 neh->eh_depth = cpu_to_le16(depth - i);
 987                 fidx = EXT_FIRST_INDEX(neh);
 988                 fidx->ei_block = border;
 989                 ext4_idx_store_pblock(fidx, oldblock);
 990
 991                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 992                                 i, newblock, le32_to_cpu(border), oldblock);
 993                 /* copy indexes */
 994                 m = 0;
 995                 path[i].p_idx++;
 996
 997                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 998                                 EXT_MAX_INDEX(path[i].p_hdr));
 999                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1000                                         EXT_LAST_INDEX(path[i].p_hdr))) {
1001                         EXT4_ERROR_INODE(inode,
1002                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1003                                          le32_to_cpu(path[i].p_ext->ee_block));
1004                         err = -EIO;
1005                         goto cleanup;
1006                 }
1007                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
1008                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
1009                                         le32_to_cpu(path[i].p_idx->ei_block),
1010                                         idx_pblock(path[i].p_idx),
1011                                         newblock);
1012                         /*memmove(++fidx, path[i].p_idx++,
1013                                         sizeof(struct ext4_extent_idx));
1014                         neh->eh_entries++;
1015                         BUG_ON(neh->eh_entries > neh->eh_max);*/
1016                         path[i].p_idx++;
1017                         m++;
1018                 }
1019                 if (m) {
1020                         memmove(++fidx, path[i].p_idx - m,
1021                                 sizeof(struct ext4_extent_idx) * m);
1022                         le16_add_cpu(&neh->eh_entries, m);
1023                 }
1024                 set_buffer_uptodate(bh);
1025                 unlock_buffer(bh);
1026
1027                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1028                 if (err)
1029                         goto cleanup;
1030                 brelse(bh);
1031                 bh = NULL;
1032
1033                 /* correct old index */
1034                 if (m) {
1035                         err = ext4_ext_get_access(handle, inode, path + i);
1036                         if (err)
1037                                 goto cleanup;
1038                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1039                         err = ext4_ext_dirty(handle, inode, path + i);
1040                         if (err)
1041                                 goto cleanup;
1042                 }
1043
1044                 i--;
1045         }
1046
1047         /* insert new index */
1048         err = ext4_ext_insert_index(handle, inode, path + at,
1049                                     le32_to_cpu(border), newblock);
1050
1051 cleanup:
1052         if (bh) {
1053                 if (buffer_locked(bh))
1054                         unlock_buffer(bh);
1055                 brelse(bh);
1056         }
1057
1058         if (err) {
1059                 /* free all allocated blocks in error case */
1060                 for (i = 0; i < depth; i++) {
1061                         if (!ablocks[i])
1062                                 continue;
1063                         ext4_free_blocks(handle, inode, 0, ablocks[i], 1,
1064                                          EXT4_FREE_BLOCKS_METADATA);
1065                 }
1066         }
1067         kfree(ablocks);
1068
1069         return err;
1070 }
1071
1072 /*
1073  * ext4_ext_grow_indepth:
1074  * implements tree growing procedure:
1075  * - allocates new block
1076  * - moves top-level data (index block or leaf) into the new block
1077  * - initializes new top-level, creating index that points to the
1078  *   just created block
1079  */
1080 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1081                                         struct ext4_ext_path *path,
1082                                         struct ext4_extent *newext)
1083 {
1084         struct ext4_ext_path *curp = path;
1085         struct ext4_extent_header *neh;
1086         struct ext4_extent_idx *fidx;
1087         struct buffer_head *bh;
1088         ext4_fsblk_t newblock;
1089         int err = 0;
1090
1091         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1092         if (newblock == 0)
1093                 return err;
1094
1095         bh = sb_getblk(inode->i_sb, newblock);
1096         if (!bh) {
1097                 err = -EIO;
1098                 ext4_std_error(inode->i_sb, err);
1099                 return err;
1100         }
1101         lock_buffer(bh);
1102
1103         err = ext4_journal_get_create_access(handle, bh);
1104         if (err) {
1105                 unlock_buffer(bh);
1106                 goto out;
1107         }
1108
1109         /* move top-level index/leaf into new block */
1110         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1111
1112         /* set size of new block */
1113         neh = ext_block_hdr(bh);
1114         /* old root could have indexes or leaves
1115          * so calculate e_max right way */
1116         if (ext_depth(inode))
1117                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1118         else
1119                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1120         neh->eh_magic = EXT4_EXT_MAGIC;
1121         set_buffer_uptodate(bh);
1122         unlock_buffer(bh);
1123
1124         err = ext4_handle_dirty_metadata(handle, inode, bh);
1125         if (err)
1126                 goto out;
1127
1128         /* create index in new top-level index: num,max,pointer */
1129         err = ext4_ext_get_access(handle, inode, curp);
1130         if (err)
1131                 goto out;
1132
1133         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1134         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1135         curp->p_hdr->eh_entries = cpu_to_le16(1);
1136         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1137
1138         if (path[0].p_hdr->eh_depth)
1139                 curp->p_idx->ei_block =
1140                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1141         else
1142                 curp->p_idx->ei_block =
1143                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1144         ext4_idx_store_pblock(curp->p_idx, newblock);
1145
1146         neh = ext_inode_hdr(inode);
1147         fidx = EXT_FIRST_INDEX(neh);
1148         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1149                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1150                   le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
1151
1152         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1153         err = ext4_ext_dirty(handle, inode, curp);
1154 out:
1155         brelse(bh);
1156
1157         return err;
1158 }
1159
1160 /*
1161  * ext4_ext_create_new_leaf:
1162  * finds empty index and adds new leaf.
1163  * if no free index is found, then it requests in-depth growing.
1164  */
1165 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1166                                         struct ext4_ext_path *path,
1167                                         struct ext4_extent *newext)
1168 {
1169         struct ext4_ext_path *curp;
1170         int depth, i, err = 0;
1171
1172 repeat:
1173         i = depth = ext_depth(inode);
1174
1175         /* walk up to the tree and look for free index entry */
1176         curp = path + depth;
1177         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1178                 i--;
1179                 curp--;
1180         }
1181
1182         /* we use already allocated block for index block,
1183          * so subsequent data blocks should be contiguous */
1184         if (EXT_HAS_FREE_INDEX(curp)) {
1185                 /* if we found index with free entry, then use that
1186                  * entry: create all needed subtree and add new leaf */
1187                 err = ext4_ext_split(handle, inode, path, newext, i);
1188                 if (err)
1189                         goto out;
1190
1191                 /* refill path */
1192                 ext4_ext_drop_refs(path);
1193                 path = ext4_ext_find_extent(inode,
1194                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1195                                     path);
1196                 if (IS_ERR(path))
1197                         err = PTR_ERR(path);
1198         } else {
1199                 /* tree is full, time to grow in depth */
1200                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1201                 if (err)
1202                         goto out;
1203
1204                 /* refill path */
1205                 ext4_ext_drop_refs(path);
1206                 path = ext4_ext_find_extent(inode,
1207                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1208                                     path);
1209                 if (IS_ERR(path)) {
1210                         err = PTR_ERR(path);
1211                         goto out;
1212                 }
1213
1214                 /*
1215                  * only first (depth 0 -> 1) produces free space;
1216                  * in all other cases we have to split the grown tree
1217                  */
1218                 depth = ext_depth(inode);
1219                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1220                         /* now we need to split */
1221                         goto repeat;
1222                 }
1223         }
1224
1225 out:
1226         return err;
1227 }
1228
1229 /*
1230  * search the closest allocated block to the left for *logical
1231  * and returns it at @logical + it's physical address at @phys
1232  * if *logical is the smallest allocated block, the function
1233  * returns 0 at @phys
1234  * return value contains 0 (success) or error code
1235  */
1236 int
1237 ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1238                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1239 {
1240         struct ext4_extent_idx *ix;
1241         struct ext4_extent *ex;
1242         int depth, ee_len;
1243
1244         if (unlikely(path == NULL)) {
1245                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1246                 return -EIO;
1247         }
1248         depth = path->p_depth;
1249         *phys = 0;
1250
1251         if (depth == 0 && path->p_ext == NULL)
1252                 return 0;
1253
1254         /* usually extent in the path covers blocks smaller
1255          * then *logical, but it can be that extent is the
1256          * first one in the file */
1257
1258         ex = path[depth].p_ext;
1259         ee_len = ext4_ext_get_actual_len(ex);
1260         if (*logical < le32_to_cpu(ex->ee_block)) {
1261                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1262                         EXT4_ERROR_INODE(inode,
1263                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1264                                          *logical, le32_to_cpu(ex->ee_block));
1265                         return -EIO;
1266                 }
1267                 while (--depth >= 0) {
1268                         ix = path[depth].p_idx;
1269                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1270                                 EXT4_ERROR_INODE(inode,
1271                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1272                                   ix != NULL ? ix->ei_block : 0,
1273                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1274                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1275                                   depth);
1276                                 return -EIO;
1277                         }
1278                 }
1279                 return 0;
1280         }
1281
1282         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1283                 EXT4_ERROR_INODE(inode,
1284                                  "logical %d < ee_block %d + ee_len %d!",
1285                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1286                 return -EIO;
1287         }
1288
1289         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1290         *phys = ext_pblock(ex) + ee_len - 1;
1291         return 0;
1292 }
1293
1294 /*
1295  * search the closest allocated block to the right for *logical
1296  * and returns it at @logical + it's physical address at @phys
1297  * if *logical is the smallest allocated block, the function
1298  * returns 0 at @phys
1299  * return value contains 0 (success) or error code
1300  */
1301 int
1302 ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1303                         ext4_lblk_t *logical, ext4_fsblk_t *phys)
1304 {
1305         struct buffer_head *bh = NULL;
1306         struct ext4_extent_header *eh;
1307         struct ext4_extent_idx *ix;
1308         struct ext4_extent *ex;
1309         ext4_fsblk_t block;
1310         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1311         int ee_len;
1312
1313         if (unlikely(path == NULL)) {
1314                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1315                 return -EIO;
1316         }
1317         depth = path->p_depth;
1318         *phys = 0;
1319
1320         if (depth == 0 && path->p_ext == NULL)
1321                 return 0;
1322
1323         /* usually extent in the path covers blocks smaller
1324          * then *logical, but it can be that extent is the
1325          * first one in the file */
1326
1327         ex = path[depth].p_ext;
1328         ee_len = ext4_ext_get_actual_len(ex);
1329         if (*logical < le32_to_cpu(ex->ee_block)) {
1330                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1331                         EXT4_ERROR_INODE(inode,
1332                                          "first_extent(path[%d].p_hdr) != ex",
1333                                          depth);
1334                         return -EIO;
1335                 }
1336                 while (--depth >= 0) {
1337                         ix = path[depth].p_idx;
1338                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1339                                 EXT4_ERROR_INODE(inode,
1340                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1341                                                  *logical);
1342                                 return -EIO;
1343                         }
1344                 }
1345                 *logical = le32_to_cpu(ex->ee_block);
1346                 *phys = ext_pblock(ex);
1347                 return 0;
1348         }
1349
1350         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1351                 EXT4_ERROR_INODE(inode,
1352                                  "logical %d < ee_block %d + ee_len %d!",
1353                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1354                 return -EIO;
1355         }
1356
1357         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1358                 /* next allocated block in this leaf */
1359                 ex++;
1360                 *logical = le32_to_cpu(ex->ee_block);
1361                 *phys = ext_pblock(ex);
1362                 return 0;
1363         }
1364
1365         /* go up and search for index to the right */
1366         while (--depth >= 0) {
1367                 ix = path[depth].p_idx;
1368                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1369                         goto got_index;
1370         }
1371
1372         /* we've gone up to the root and found no index to the right */
1373         return 0;
1374
1375 got_index:
1376         /* we've found index to the right, let's
1377          * follow it and find the closest allocated
1378          * block to the right */
1379         ix++;
1380         block = idx_pblock(ix);
1381         while (++depth < path->p_depth) {
1382                 bh = sb_bread(inode->i_sb, block);
1383                 if (bh == NULL)
1384                         return -EIO;
1385                 eh = ext_block_hdr(bh);
1386                 /* subtract from p_depth to get proper eh_depth */
1387                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1388                         put_bh(bh);
1389                         return -EIO;
1390                 }
1391                 ix = EXT_FIRST_INDEX(eh);
1392                 block = idx_pblock(ix);
1393                 put_bh(bh);
1394         }
1395
1396         bh = sb_bread(inode->i_sb, block);
1397         if (bh == NULL)
1398                 return -EIO;
1399         eh = ext_block_hdr(bh);
1400         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1401                 put_bh(bh);
1402                 return -EIO;
1403         }
1404         ex = EXT_FIRST_EXTENT(eh);
1405         *logical = le32_to_cpu(ex->ee_block);
1406         *phys = ext_pblock(ex);
1407         put_bh(bh);
1408         return 0;
1409 }
1410
1411 /*
1412  * ext4_ext_next_allocated_block:
1413  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1414  * NOTE: it considers block number from index entry as
1415  * allocated block. Thus, index entries have to be consistent
1416  * with leaves.
1417  */
1418 static ext4_lblk_t
1419 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1420 {
1421         int depth;
1422
1423         BUG_ON(path == NULL);
1424         depth = path->p_depth;
1425
1426         if (depth == 0 && path->p_ext == NULL)
1427                 return EXT_MAX_BLOCK;
1428
1429         while (depth >= 0) {
1430                 if (depth == path->p_depth) {
1431                         /* leaf */
1432                         if (path[depth].p_ext !=
1433                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1434                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1435                 } else {
1436                         /* index */
1437                         if (path[depth].p_idx !=
1438                                         EXT_LAST_INDEX(path[depth].p_hdr))
1439                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1440                 }
1441                 depth--;
1442         }
1443
1444         return EXT_MAX_BLOCK;
1445 }
1446
1447 /*
1448  * ext4_ext_next_leaf_block:
1449  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1450  */
1451 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1452                                         struct ext4_ext_path *path)
1453 {
1454         int depth;
1455
1456         BUG_ON(path == NULL);
1457         depth = path->p_depth;
1458
1459         /* zero-tree has no leaf blocks at all */
1460         if (depth == 0)
1461                 return EXT_MAX_BLOCK;
1462
1463         /* go to index block */
1464         depth--;
1465
1466         while (depth >= 0) {
1467                 if (path[depth].p_idx !=
1468                                 EXT_LAST_INDEX(path[depth].p_hdr))
1469                         return (ext4_lblk_t)
1470                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1471                 depth--;
1472         }
1473
1474         return EXT_MAX_BLOCK;
1475 }
1476
1477 /*
1478  * ext4_ext_correct_indexes:
1479  * if leaf gets modified and modified extent is first in the leaf,
1480  * then we have to correct all indexes above.
1481  * TODO: do we need to correct tree in all cases?
1482  */
1483 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1484                                 struct ext4_ext_path *path)
1485 {
1486         struct ext4_extent_header *eh;
1487         int depth = ext_depth(inode);
1488         struct ext4_extent *ex;
1489         __le32 border;
1490         int k, err = 0;
1491
1492         eh = path[depth].p_hdr;
1493         ex = path[depth].p_ext;
1494
1495         if (unlikely(ex == NULL || eh == NULL)) {
1496                 EXT4_ERROR_INODE(inode,
1497                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1498                 return -EIO;
1499         }
1500
1501         if (depth == 0) {
1502                 /* there is no tree at all */
1503                 return 0;
1504         }
1505
1506         if (ex != EXT_FIRST_EXTENT(eh)) {
1507                 /* we correct tree if first leaf got modified only */
1508                 return 0;
1509         }
1510
1511         /*
1512          * TODO: we need correction if border is smaller than current one
1513          */
1514         k = depth - 1;
1515         border = path[depth].p_ext->ee_block;
1516         err = ext4_ext_get_access(handle, inode, path + k);
1517         if (err)
1518                 return err;
1519         path[k].p_idx->ei_block = border;
1520         err = ext4_ext_dirty(handle, inode, path + k);
1521         if (err)
1522                 return err;
1523
1524         while (k--) {
1525                 /* change all left-side indexes */
1526                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1527                         break;
1528                 err = ext4_ext_get_access(handle, inode, path + k);
1529                 if (err)
1530                         break;
1531                 path[k].p_idx->ei_block = border;
1532                 err = ext4_ext_dirty(handle, inode, path + k);
1533                 if (err)
1534                         break;
1535         }
1536
1537         return err;
1538 }
1539
1540 int
1541 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1542                                 struct ext4_extent *ex2)
1543 {
1544         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1545
1546         /*
1547          * Make sure that either both extents are uninitialized, or
1548          * both are _not_.
1549          */
1550         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1551                 return 0;
1552
1553         if (ext4_ext_is_uninitialized(ex1))
1554                 max_len = EXT_UNINIT_MAX_LEN;
1555         else
1556                 max_len = EXT_INIT_MAX_LEN;
1557
1558         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1559         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1560
1561         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1562                         le32_to_cpu(ex2->ee_block))
1563                 return 0;
1564
1565         /*
1566          * To allow future support for preallocated extents to be added
1567          * as an RO_COMPAT feature, refuse to merge to extents if
1568          * this can result in the top bit of ee_len being set.
1569          */
1570         if (ext1_ee_len + ext2_ee_len > max_len)
1571                 return 0;
1572 #ifdef AGGRESSIVE_TEST
1573         if (ext1_ee_len >= 4)
1574                 return 0;
1575 #endif
1576
1577         if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
1578                 return 1;
1579         return 0;
1580 }
1581
1582 /*
1583  * This function tries to merge the "ex" extent to the next extent in the tree.
1584  * It always tries to merge towards right. If you want to merge towards
1585  * left, pass "ex - 1" as argument instead of "ex".
1586  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1587  * 1 if they got merged.
1588  */
1589 int ext4_ext_try_to_merge(struct inode *inode,
1590                           struct ext4_ext_path *path,
1591                           struct ext4_extent *ex)
1592 {
1593         struct ext4_extent_header *eh;
1594         unsigned int depth, len;
1595         int merge_done = 0;
1596         int uninitialized = 0;
1597
1598         depth = ext_depth(inode);
1599         BUG_ON(path[depth].p_hdr == NULL);
1600         eh = path[depth].p_hdr;
1601
1602         while (ex < EXT_LAST_EXTENT(eh)) {
1603                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1604                         break;
1605                 /* merge with next extent! */
1606                 if (ext4_ext_is_uninitialized(ex))
1607                         uninitialized = 1;
1608                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1609                                 + ext4_ext_get_actual_len(ex + 1));
1610                 if (uninitialized)
1611                         ext4_ext_mark_uninitialized(ex);
1612
1613                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1614                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1615                                 * sizeof(struct ext4_extent);
1616                         memmove(ex + 1, ex + 2, len);
1617                 }
1618                 le16_add_cpu(&eh->eh_entries, -1);
1619                 merge_done = 1;
1620                 WARN_ON(eh->eh_entries == 0);
1621                 if (!eh->eh_entries)
1622                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1623         }
1624
1625         return merge_done;
1626 }
1627
1628 /*
1629  * check if a portion of the "newext" extent overlaps with an
1630  * existing extent.
1631  *
1632  * If there is an overlap discovered, it updates the length of the newext
1633  * such that there will be no overlap, and then returns 1.
1634  * If there is no overlap found, it returns 0.
1635  */
1636 unsigned int ext4_ext_check_overlap(struct inode *inode,
1637                                     struct ext4_extent *newext,
1638                                     struct ext4_ext_path *path)
1639 {
1640         ext4_lblk_t b1, b2;
1641         unsigned int depth, len1;
1642         unsigned int ret = 0;
1643
1644         b1 = le32_to_cpu(newext->ee_block);
1645         len1 = ext4_ext_get_actual_len(newext);
1646         depth = ext_depth(inode);
1647         if (!path[depth].p_ext)
1648                 goto out;
1649         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1650
1651         /*
1652          * get the next allocated block if the extent in the path
1653          * is before the requested block(s)
1654          */
1655         if (b2 < b1) {
1656                 b2 = ext4_ext_next_allocated_block(path);
1657                 if (b2 == EXT_MAX_BLOCK)
1658                         goto out;
1659         }
1660
1661         /* check for wrap through zero on extent logical start block*/
1662         if (b1 + len1 < b1) {
1663                 len1 = EXT_MAX_BLOCK - b1;
1664                 newext->ee_len = cpu_to_le16(len1);
1665                 ret = 1;
1666         }
1667
1668         /* check for overlap */
1669         if (b1 + len1 > b2) {
1670                 newext->ee_len = cpu_to_le16(b2 - b1);
1671                 ret = 1;
1672         }
1673 out:
1674         return ret;
1675 }
1676
1677 /*
1678  * ext4_ext_insert_extent:
1679  * tries to merge requsted extent into the existing extent or
1680  * inserts requested extent as new one into the tree,
1681  * creating new leaf in the no-space case.
1682  */
1683 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1684                                 struct ext4_ext_path *path,
1685                                 struct ext4_extent *newext, int flag)
1686 {
1687         struct ext4_extent_header *eh;
1688         struct ext4_extent *ex, *fex;
1689         struct ext4_extent *nearex; /* nearest extent */
1690         struct ext4_ext_path *npath = NULL;
1691         int depth, len, err;
1692         ext4_lblk_t next;
1693         unsigned uninitialized = 0;
1694
1695         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1696                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1697                 return -EIO;
1698         }
1699         depth = ext_depth(inode);
1700         ex = path[depth].p_ext;
1701         if (unlikely(path[depth].p_hdr == NULL)) {
1702                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1703                 return -EIO;
1704         }
1705
1706         /* try to insert block into found extent and return */
1707         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1708                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1709                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1710                                 ext4_ext_is_uninitialized(newext),
1711                                 ext4_ext_get_actual_len(newext),
1712                                 le32_to_cpu(ex->ee_block),
1713                                 ext4_ext_is_uninitialized(ex),
1714                                 ext4_ext_get_actual_len(ex), ext_pblock(ex));
1715                 err = ext4_ext_get_access(handle, inode, path + depth);
1716                 if (err)
1717                         return err;
1718
1719                 /*
1720                  * ext4_can_extents_be_merged should have checked that either
1721                  * both extents are uninitialized, or both aren't. Thus we
1722                  * need to check only one of them here.
1723                  */
1724                 if (ext4_ext_is_uninitialized(ex))
1725                         uninitialized = 1;
1726                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1727                                         + ext4_ext_get_actual_len(newext));
1728                 if (uninitialized)
1729                         ext4_ext_mark_uninitialized(ex);
1730                 eh = path[depth].p_hdr;
1731                 nearex = ex;
1732                 goto merge;
1733         }
1734
1735 repeat:
1736         depth = ext_depth(inode);
1737         eh = path[depth].p_hdr;
1738         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1739                 goto has_space;
1740
1741         /* probably next leaf has space for us? */
1742         fex = EXT_LAST_EXTENT(eh);
1743         next = ext4_ext_next_leaf_block(inode, path);
1744         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1745             && next != EXT_MAX_BLOCK) {
1746                 ext_debug("next leaf block - %d\n", next);
1747                 BUG_ON(npath != NULL);
1748                 npath = ext4_ext_find_extent(inode, next, NULL);
1749                 if (IS_ERR(npath))
1750                         return PTR_ERR(npath);
1751                 BUG_ON(npath->p_depth != path->p_depth);
1752                 eh = npath[depth].p_hdr;
1753                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1754                         ext_debug("next leaf isnt full(%d)\n",
1755                                   le16_to_cpu(eh->eh_entries));
1756                         path = npath;
1757                         goto repeat;
1758                 }
1759                 ext_debug("next leaf has no free space(%d,%d)\n",
1760                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1761         }
1762
1763         /*
1764          * There is no free space in the found leaf.
1765          * We're gonna add a new leaf in the tree.
1766          */
1767         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1768         if (err)
1769                 goto cleanup;
1770         depth = ext_depth(inode);
1771         eh = path[depth].p_hdr;
1772
1773 has_space:
1774         nearex = path[depth].p_ext;
1775
1776         err = ext4_ext_get_access(handle, inode, path + depth);
1777         if (err)
1778                 goto cleanup;
1779
1780         if (!nearex) {
1781                 /* there is no extent in this leaf, create first one */
1782                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1783                                 le32_to_cpu(newext->ee_block),
1784                                 ext_pblock(newext),
1785                                 ext4_ext_is_uninitialized(newext),
1786                                 ext4_ext_get_actual_len(newext));
1787                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1788         } else if (le32_to_cpu(newext->ee_block)
1789                            > le32_to_cpu(nearex->ee_block)) {
1790 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1791                 if (nearex != EXT_LAST_EXTENT(eh)) {
1792                         len = EXT_MAX_EXTENT(eh) - nearex;
1793                         len = (len - 1) * sizeof(struct ext4_extent);
1794                         len = len < 0 ? 0 : len;
1795                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1796                                         "move %d from 0x%p to 0x%p\n",
1797                                         le32_to_cpu(newext->ee_block),
1798                                         ext_pblock(newext),
1799                                         ext4_ext_is_uninitialized(newext),
1800                                         ext4_ext_get_actual_len(newext),
1801                                         nearex, len, nearex + 1, nearex + 2);
1802                         memmove(nearex + 2, nearex + 1, len);
1803                 }
1804                 path[depth].p_ext = nearex + 1;
1805         } else {
1806                 BUG_ON(newext->ee_block == nearex->ee_block);
1807                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1808                 len = len < 0 ? 0 : len;
1809                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1810                                 "move %d from 0x%p to 0x%p\n",
1811                                 le32_to_cpu(newext->ee_block),
1812                                 ext_pblock(newext),
1813                                 ext4_ext_is_uninitialized(newext),
1814                                 ext4_ext_get_actual_len(newext),
1815                                 nearex, len, nearex + 1, nearex + 2);
1816                 memmove(nearex + 1, nearex, len);
1817                 path[depth].p_ext = nearex;
1818         }
1819
1820         le16_add_cpu(&eh->eh_entries, 1);
1821         nearex = path[depth].p_ext;
1822         nearex->ee_block = newext->ee_block;
1823         ext4_ext_store_pblock(nearex, ext_pblock(newext));
1824         nearex->ee_len = newext->ee_len;
1825
1826 merge:
1827         /* try to merge extents to the right */
1828         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1829                 ext4_ext_try_to_merge(inode, path, nearex);
1830
1831         /* try to merge extents to the left */
1832
1833         /* time to correct all indexes above */
1834         err = ext4_ext_correct_indexes(handle, inode, path);
1835         if (err)
1836                 goto cleanup;
1837
1838         err = ext4_ext_dirty(handle, inode, path + depth);
1839
1840 cleanup:
1841         if (npath) {
1842                 ext4_ext_drop_refs(npath);
1843                 kfree(npath);
1844         }
1845         ext4_ext_invalidate_cache(inode);
1846         return err;
1847 }
1848
1849 int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1850                         ext4_lblk_t num, ext_prepare_callback func,
1851                         void *cbdata)
1852 {
1853         struct ext4_ext_path *path = NULL;
1854         struct ext4_ext_cache cbex;
1855         struct ext4_extent *ex;
1856         ext4_lblk_t next, start = 0, end = 0;
1857         ext4_lblk_t last = block + num;
1858         int depth, exists, err = 0;
1859
1860         BUG_ON(func == NULL);
1861         BUG_ON(inode == NULL);
1862
1863         while (block < last && block != EXT_MAX_BLOCK) {
1864                 num = last - block;
1865                 /* find extent for this block */
1866                 down_read(&EXT4_I(inode)->i_data_sem);
1867                 path = ext4_ext_find_extent(inode, block, path);
1868                 up_read(&EXT4_I(inode)->i_data_sem);
1869                 if (IS_ERR(path)) {
1870                         err = PTR_ERR(path);
1871                         path = NULL;
1872                         break;
1873                 }
1874
1875                 depth = ext_depth(inode);
1876                 if (unlikely(path[depth].p_hdr == NULL)) {
1877                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1878                         err = -EIO;
1879                         break;
1880                 }
1881                 ex = path[depth].p_ext;
1882                 next = ext4_ext_next_allocated_block(path);
1883
1884                 exists = 0;
1885                 if (!ex) {
1886                         /* there is no extent yet, so try to allocate
1887                          * all requested space */
1888                         start = block;
1889                         end = block + num;
1890                 } else if (le32_to_cpu(ex->ee_block) > block) {
1891                         /* need to allocate space before found extent */
1892                         start = block;
1893                         end = le32_to_cpu(ex->ee_block);
1894                         if (block + num < end)
1895                                 end = block + num;
1896                 } else if (block >= le32_to_cpu(ex->ee_block)
1897                                         + ext4_ext_get_actual_len(ex)) {
1898                         /* need to allocate space after found extent */
1899                         start = block;
1900                         end = block + num;
1901                         if (end >= next)
1902                                 end = next;
1903                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1904                         /*
1905                          * some part of requested space is covered
1906                          * by found extent
1907                          */
1908                         start = block;
1909                         end = le32_to_cpu(ex->ee_block)
1910                                 + ext4_ext_get_actual_len(ex);
1911                         if (block + num < end)
1912                                 end = block + num;
1913                         exists = 1;
1914                 } else {
1915                         BUG();
1916                 }
1917                 BUG_ON(end <= start);
1918
1919                 if (!exists) {
1920                         cbex.ec_block = start;
1921                         cbex.ec_len = end - start;
1922                         cbex.ec_start = 0;
1923                         cbex.ec_type = EXT4_EXT_CACHE_GAP;
1924                 } else {
1925                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1926                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1927                         cbex.ec_start = ext_pblock(ex);
1928                         cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1929                 }
1930
1931                 if (unlikely(cbex.ec_len == 0)) {
1932                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1933                         err = -EIO;
1934                         break;
1935                 }
1936                 err = func(inode, path, &cbex, ex, cbdata);
1937                 ext4_ext_drop_refs(path);
1938
1939                 if (err < 0)
1940                         break;
1941
1942                 if (err == EXT_REPEAT)
1943                         continue;
1944                 else if (err == EXT_BREAK) {
1945                         err = 0;
1946                         break;
1947                 }
1948
1949                 if (ext_depth(inode) != depth) {
1950                         /* depth was changed. we have to realloc path */
1951                         kfree(path);
1952                         path = NULL;
1953                 }
1954
1955                 block = cbex.ec_block + cbex.ec_len;
1956         }
1957
1958         if (path) {
1959                 ext4_ext_drop_refs(path);
1960                 kfree(path);
1961         }
1962
1963         return err;
1964 }
1965
1966 static void
1967 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1968                         __u32 len, ext4_fsblk_t start, int type)
1969 {
1970         struct ext4_ext_cache *cex;
1971         BUG_ON(len == 0);
1972         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1973         cex = &EXT4_I(inode)->i_cached_extent;
1974         cex->ec_type = type;
1975         cex->ec_block = block;
1976         cex->ec_len = len;
1977         cex->ec_start = start;
1978         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1979 }
1980
1981 /*
1982  * ext4_ext_put_gap_in_cache:
1983  * calculate boundaries of the gap that the requested block fits into
1984  * and cache this gap
1985  */
1986 static void
1987 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1988                                 ext4_lblk_t block)
1989 {
1990         int depth = ext_depth(inode);
1991         unsigned long len;
1992         ext4_lblk_t lblock;
1993         struct ext4_extent *ex;
1994
1995         ex = path[depth].p_ext;
1996         if (ex == NULL) {
1997                 /* there is no extent yet, so gap is [0;-] */
1998                 lblock = 0;
1999                 len = EXT_MAX_BLOCK;
2000                 ext_debug("cache gap(whole file):");
2001         } else if (block < le32_to_cpu(ex->ee_block)) {
2002                 lblock = block;
2003                 len = le32_to_cpu(ex->ee_block) - block;
2004                 ext_debug("cache gap(before): %u [%u:%u]",
2005                                 block,
2006                                 le32_to_cpu(ex->ee_block),
2007                                  ext4_ext_get_actual_len(ex));
2008         } else if (block >= le32_to_cpu(ex->ee_block)
2009                         + ext4_ext_get_actual_len(ex)) {
2010                 ext4_lblk_t next;
2011                 lblock = le32_to_cpu(ex->ee_block)
2012                         + ext4_ext_get_actual_len(ex);
2013
2014                 next = ext4_ext_next_allocated_block(path);
2015                 ext_debug("cache gap(after): [%u:%u] %u",
2016                                 le32_to_cpu(ex->ee_block),
2017                                 ext4_ext_get_actual_len(ex),
2018                                 block);
2019                 BUG_ON(next == lblock);
2020                 len = next - lblock;
2021         } else {
2022                 lblock = len = 0;
2023                 BUG();
2024         }
2025
2026         ext_debug(" -> %u:%lu\n", lblock, len);
2027         ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
2028 }
2029
2030 static int
2031 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2032                         struct ext4_extent *ex)
2033 {
2034         struct ext4_ext_cache *cex;
2035         int ret = EXT4_EXT_CACHE_NO;
2036
2037         /*
2038          * We borrow i_block_reservation_lock to protect i_cached_extent
2039          */
2040         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2041         cex = &EXT4_I(inode)->i_cached_extent;
2042
2043         /* has cache valid data? */
2044         if (cex->ec_type == EXT4_EXT_CACHE_NO)
2045                 goto errout;
2046
2047         BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
2048                         cex->ec_type != EXT4_EXT_CACHE_EXTENT);
2049         if (in_range(block, cex->ec_block, cex->ec_len)) {
2050                 ex->ee_block = cpu_to_le32(cex->ec_block);
2051                 ext4_ext_store_pblock(ex, cex->ec_start);
2052                 ex->ee_len = cpu_to_le16(cex->ec_len);
2053                 ext_debug("%u cached by %u:%u:%llu\n",
2054                                 block,
2055                                 cex->ec_block, cex->ec_len, cex->ec_start);
2056                 ret = cex->ec_type;
2057         }
2058 errout:
2059         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2060         return ret;
2061 }
2062
2063 /*
2064  * ext4_ext_rm_idx:
2065  * removes index from the index block.
2066  * It's used in truncate case only, thus all requests are for
2067  * last index in the block only.
2068  */
2069 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2070                         struct ext4_ext_path *path)
2071 {
2072         int err;
2073         ext4_fsblk_t leaf;
2074
2075         /* free index block */
2076         path--;
2077         leaf = idx_pblock(path->p_idx);
2078         if (unlikely(path->p_hdr->eh_entries == 0)) {
2079                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2080                 return -EIO;
2081         }
2082         err = ext4_ext_get_access(handle, inode, path);
2083         if (err)
2084                 return err;
2085         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2086         err = ext4_ext_dirty(handle, inode, path);
2087         if (err)
2088                 return err;
2089         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2090         ext4_free_blocks(handle, inode, 0, leaf, 1,
2091                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2092         return err;
2093 }
2094
2095 /*
2096  * ext4_ext_calc_credits_for_single_extent:
2097  * This routine returns max. credits that needed to insert an extent
2098  * to the extent tree.
2099  * When pass the actual path, the caller should calculate credits
2100  * under i_data_sem.
2101  */
2102 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2103                                                 struct ext4_ext_path *path)
2104 {
2105         if (path) {
2106                 int depth = ext_depth(inode);
2107                 int ret = 0;
2108
2109                 /* probably there is space in leaf? */
2110                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2111                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2112
2113                         /*
2114                          *  There are some space in the leaf tree, no
2115                          *  need to account for leaf block credit
2116                          *
2117                          *  bitmaps and block group descriptor blocks
2118                          *  and other metadat blocks still need to be
2119                          *  accounted.
2120                          */
2121                         /* 1 bitmap, 1 block group descriptor */
2122                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2123                         return ret;
2124                 }
2125         }
2126
2127         return ext4_chunk_trans_blocks(inode, nrblocks);
2128 }
2129
2130 /*
2131  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2132  *
2133  * if nrblocks are fit in a single extent (chunk flag is 1), then
2134  * in the worse case, each tree level index/leaf need to be changed
2135  * if the tree split due to insert a new extent, then the old tree
2136  * index/leaf need to be updated too
2137  *
2138  * If the nrblocks are discontiguous, they could cause
2139  * the whole tree split more than once, but this is really rare.
2140  */
2141 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2142 {
2143         int index;
2144         int depth = ext_depth(inode);
2145
2146         if (chunk)
2147                 index = depth * 2;
2148         else
2149                 index = depth * 3;
2150
2151         return index;
2152 }
2153
2154 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2155                                 struct ext4_extent *ex,
2156                                 ext4_lblk_t from, ext4_lblk_t to)
2157 {
2158         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2159         int flags = EXT4_FREE_BLOCKS_FORGET;
2160
2161         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2162                 flags |= EXT4_FREE_BLOCKS_METADATA;
2163 #ifdef EXTENTS_STATS
2164         {
2165                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2166                 spin_lock(&sbi->s_ext_stats_lock);
2167                 sbi->s_ext_blocks += ee_len;
2168                 sbi->s_ext_extents++;
2169                 if (ee_len < sbi->s_ext_min)
2170                         sbi->s_ext_min = ee_len;
2171                 if (ee_len > sbi->s_ext_max)
2172                         sbi->s_ext_max = ee_len;
2173                 if (ext_depth(inode) > sbi->s_depth_max)
2174                         sbi->s_depth_max = ext_depth(inode);
2175                 spin_unlock(&sbi->s_ext_stats_lock);
2176         }
2177 #endif
2178         if (from >= le32_to_cpu(ex->ee_block)
2179             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2180                 /* tail removal */
2181                 ext4_lblk_t num;
2182                 ext4_fsblk_t start;
2183
2184                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2185                 start = ext_pblock(ex) + ee_len - num;
2186                 ext_debug("free last %u blocks starting %llu\n", num, start);
2187                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2188         } else if (from == le32_to_cpu(ex->ee_block)
2189                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2190                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2191                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2192         } else {
2193                 printk(KERN_INFO "strange request: removal(2) "
2194                                 "%u-%u from %u:%u\n",
2195                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2196         }
2197         return 0;
2198 }
2199
2200 static int
2201 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2202                 struct ext4_ext_path *path, ext4_lblk_t start)
2203 {
2204         int err = 0, correct_index = 0;
2205         int depth = ext_depth(inode), credits;
2206         struct ext4_extent_header *eh;
2207         ext4_lblk_t a, b, block;
2208         unsigned num;
2209         ext4_lblk_t ex_ee_block;
2210         unsigned short ex_ee_len;
2211         unsigned uninitialized = 0;
2212         struct ext4_extent *ex;
2213
2214         /* the header must be checked already in ext4_ext_remove_space() */
2215         ext_debug("truncate since %u in leaf\n", start);
2216         if (!path[depth].p_hdr)
2217                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2218         eh = path[depth].p_hdr;
2219         if (unlikely(path[depth].p_hdr == NULL)) {
2220                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2221                 return -EIO;
2222         }
2223         /* find where to start removing */
2224         ex = EXT_LAST_EXTENT(eh);
2225
2226         ex_ee_block = le32_to_cpu(ex->ee_block);
2227         ex_ee_len = ext4_ext_get_actual_len(ex);
2228
2229         while (ex >= EXT_FIRST_EXTENT(eh) &&
2230                         ex_ee_block + ex_ee_len > start) {
2231
2232                 if (ext4_ext_is_uninitialized(ex))
2233                         uninitialized = 1;
2234                 else
2235                         uninitialized = 0;
2236
2237                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2238                          uninitialized, ex_ee_len);
2239                 path[depth].p_ext = ex;
2240
2241                 a = ex_ee_block > start ? ex_ee_block : start;
2242                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2243                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2244
2245                 ext_debug("  border %u:%u\n", a, b);
2246
2247                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2248                         block = 0;
2249                         num = 0;
2250                         BUG();
2251                 } else if (a != ex_ee_block) {
2252                         /* remove tail of the extent */
2253                         block = ex_ee_block;
2254                         num = a - block;
2255                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2256                         /* remove head of the extent */
2257                         block = a;
2258                         num = b - a;
2259                         /* there is no "make a hole" API yet */
2260                         BUG();
2261                 } else {
2262                         /* remove whole extent: excellent! */
2263                         block = ex_ee_block;
2264                         num = 0;
2265                         BUG_ON(a != ex_ee_block);
2266                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2267                 }
2268
2269                 /*
2270                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2271                  * descriptor) for each block group; assume two block
2272                  * groups plus ex_ee_len/blocks_per_block_group for
2273                  * the worst case
2274                  */
2275                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2276                 if (ex == EXT_FIRST_EXTENT(eh)) {
2277                         correct_index = 1;
2278                         credits += (ext_depth(inode)) + 1;
2279                 }
2280                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2281
2282                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2283                 if (err)
2284                         goto out;
2285
2286                 err = ext4_ext_get_access(handle, inode, path + depth);
2287                 if (err)
2288                         goto out;
2289
2290                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2291                 if (err)
2292                         goto out;
2293
2294                 if (num == 0) {
2295                         /* this extent is removed; mark slot entirely unused */
2296                         ext4_ext_store_pblock(ex, 0);
2297                         le16_add_cpu(&eh->eh_entries, -1);
2298                 }
2299
2300                 ex->ee_block = cpu_to_le32(block);
2301                 ex->ee_len = cpu_to_le16(num);
2302                 /*
2303                  * Do not mark uninitialized if all the blocks in the
2304                  * extent have been removed.
2305                  */
2306                 if (uninitialized && num)
2307                         ext4_ext_mark_uninitialized(ex);
2308
2309                 err = ext4_ext_dirty(handle, inode, path + depth);
2310                 if (err)
2311                         goto out;
2312
2313                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2314                                 ext_pblock(ex));
2315                 ex--;
2316                 ex_ee_block = le32_to_cpu(ex->ee_block);
2317                 ex_ee_len = ext4_ext_get_actual_len(ex);
2318         }
2319
2320         if (correct_index && eh->eh_entries)
2321                 err = ext4_ext_correct_indexes(handle, inode, path);
2322
2323         /* if this leaf is free, then we should
2324          * remove it from index block above */
2325         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2326                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2327
2328 out:
2329         return err;
2330 }
2331
2332 /*
2333  * ext4_ext_more_to_rm:
2334  * returns 1 if current index has to be freed (even partial)
2335  */
2336 static int
2337 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2338 {
2339         BUG_ON(path->p_idx == NULL);
2340
2341         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2342                 return 0;
2343
2344         /*
2345          * if truncate on deeper level happened, it wasn't partial,
2346          * so we have to consider current index for truncation
2347          */
2348         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2349                 return 0;
2350         return 1;
2351 }
2352
2353 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2354 {
2355         struct super_block *sb = inode->i_sb;
2356         int depth = ext_depth(inode);
2357         struct ext4_ext_path *path;
2358         handle_t *handle;
2359         int i, err;
2360
2361         ext_debug("truncate since %u\n", start);
2362
2363         /* probably first extent we're gonna free will be last in block */
2364         handle = ext4_journal_start(inode, depth + 1);
2365         if (IS_ERR(handle))
2366                 return PTR_ERR(handle);
2367
2368 again:
2369         ext4_ext_invalidate_cache(inode);
2370
2371         /*
2372          * We start scanning from right side, freeing all the blocks
2373          * after i_size and walking into the tree depth-wise.
2374          */
2375         depth = ext_depth(inode);
2376         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2377         if (path == NULL) {
2378                 ext4_journal_stop(handle);
2379                 return -ENOMEM;
2380         }
2381         path[0].p_depth = depth;
2382         path[0].p_hdr = ext_inode_hdr(inode);
2383         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2384                 err = -EIO;
2385                 goto out;
2386         }
2387         i = err = 0;
2388
2389         while (i >= 0 && err == 0) {
2390                 if (i == depth) {
2391                         /* this is leaf block */
2392                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2393                         /* root level has p_bh == NULL, brelse() eats this */
2394                         brelse(path[i].p_bh);
2395                         path[i].p_bh = NULL;
2396                         i--;
2397                         continue;
2398                 }
2399
2400                 /* this is index block */
2401                 if (!path[i].p_hdr) {
2402                         ext_debug("initialize header\n");
2403                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2404                 }
2405
2406                 if (!path[i].p_idx) {
2407                         /* this level hasn't been touched yet */
2408                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2409                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2410                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2411                                   path[i].p_hdr,
2412                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2413                 } else {
2414                         /* we were already here, see at next index */
2415                         path[i].p_idx--;
2416                 }
2417
2418                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2419                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2420                                 path[i].p_idx);
2421                 if (ext4_ext_more_to_rm(path + i)) {
2422                         struct buffer_head *bh;
2423                         /* go to the next level */
2424                         ext_debug("move to level %d (block %llu)\n",
2425                                   i + 1, idx_pblock(path[i].p_idx));
2426                         memset(path + i + 1, 0, sizeof(*path));
2427                         bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2428                         if (!bh) {
2429                                 /* should we reset i_size? */
2430                                 err = -EIO;
2431                                 break;
2432                         }
2433                         if (WARN_ON(i + 1 > depth)) {
2434                                 err = -EIO;
2435                                 break;
2436                         }
2437                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2438                                                         depth - i - 1)) {
2439                                 err = -EIO;
2440                                 break;
2441                         }
2442                         path[i + 1].p_bh = bh;
2443
2444                         /* save actual number of indexes since this
2445                          * number is changed at the next iteration */
2446                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2447                         i++;
2448                 } else {
2449                         /* we finished processing this index, go up */
2450                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2451                                 /* index is empty, remove it;
2452                                  * handle must be already prepared by the
2453                                  * truncatei_leaf() */
2454                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2455                         }
2456                         /* root level has p_bh == NULL, brelse() eats this */
2457                         brelse(path[i].p_bh);
2458                         path[i].p_bh = NULL;
2459                         i--;
2460                         ext_debug("return to level %d\n", i);
2461                 }
2462         }
2463
2464         /* TODO: flexible tree reduction should be here */
2465         if (path->p_hdr->eh_entries == 0) {
2466                 /*
2467                  * truncate to zero freed all the tree,
2468                  * so we need to correct eh_depth
2469                  */
2470                 err = ext4_ext_get_access(handle, inode, path);
2471                 if (err == 0) {
2472                         ext_inode_hdr(inode)->eh_depth = 0;
2473                         ext_inode_hdr(inode)->eh_max =
2474                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2475                         err = ext4_ext_dirty(handle, inode, path);
2476                 }
2477         }
2478 out:
2479         ext4_ext_drop_refs(path);
2480         kfree(path);
2481         if (err == -EAGAIN)
2482                 goto again;
2483         ext4_journal_stop(handle);
2484
2485         return err;
2486 }
2487
2488 /*
2489  * called at mount time
2490  */
2491 void ext4_ext_init(struct super_block *sb)
2492 {
2493         /*
2494          * possible initialization would be here
2495          */
2496
2497         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2498 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2499                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2500 #ifdef AGGRESSIVE_TEST
2501                 printk(", aggressive tests");
2502 #endif
2503 #ifdef CHECK_BINSEARCH
2504                 printk(", check binsearch");
2505 #endif
2506 #ifdef EXTENTS_STATS
2507                 printk(", stats");
2508 #endif
2509                 printk("\n");
2510 #endif
2511 #ifdef EXTENTS_STATS
2512                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2513                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2514                 EXT4_SB(sb)->s_ext_max = 0;
2515 #endif
2516         }
2517 }
2518
2519 /*
2520  * called at umount time
2521  */
2522 void ext4_ext_release(struct super_block *sb)
2523 {
2524         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2525                 return;
2526
2527 #ifdef EXTENTS_STATS
2528         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2529                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2530                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2531                         sbi->s_ext_blocks, sbi->s_ext_extents,
2532                         sbi->s_ext_blocks / sbi->s_ext_extents);
2533                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2534                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2535         }
2536 #endif
2537 }
2538
2539 static void bi_complete(struct bio *bio, int error)
2540 {
2541         complete((struct completion *)bio->bi_private);
2542 }
2543
2544 /* FIXME!! we need to try to merge to left or right after zero-out  */
2545 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2546 {
2547         int ret;
2548         struct bio *bio;
2549         int blkbits, blocksize;
2550         sector_t ee_pblock;
2551         struct completion event;
2552         unsigned int ee_len, len, done, offset;
2553
2554
2555         blkbits   = inode->i_blkbits;
2556         blocksize = inode->i_sb->s_blocksize;
2557         ee_len    = ext4_ext_get_actual_len(ex);
2558         ee_pblock = ext_pblock(ex);
2559
2560         /* convert ee_pblock to 512 byte sectors */
2561         ee_pblock = ee_pblock << (blkbits - 9);
2562
2563         while (ee_len > 0) {
2564
2565                 if (ee_len > BIO_MAX_PAGES)
2566                         len = BIO_MAX_PAGES;
2567                 else
2568                         len = ee_len;
2569
2570                 bio = bio_alloc(GFP_NOIO, len);
2571                 if (!bio)
2572                         return -ENOMEM;
2573
2574                 bio->bi_sector = ee_pblock;
2575                 bio->bi_bdev   = inode->i_sb->s_bdev;
2576
2577                 done = 0;
2578                 offset = 0;
2579                 while (done < len) {
2580                         ret = bio_add_page(bio, ZERO_PAGE(0),
2581                                                         blocksize, offset);
2582                         if (ret != blocksize) {
2583                                 /*
2584                                  * We can't add any more pages because of
2585                                  * hardware limitations.  Start a new bio.
2586                                  */
2587                                 break;
2588                         }
2589                         done++;
2590                         offset += blocksize;
2591                         if (offset >= PAGE_CACHE_SIZE)
2592                                 offset = 0;
2593                 }
2594
2595                 init_completion(&event);
2596                 bio->bi_private = &event;
2597                 bio->bi_end_io = bi_complete;
2598                 submit_bio(WRITE, bio);
2599                 wait_for_completion(&event);
2600
2601                 if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2602                         bio_put(bio);
2603                         return -EIO;
2604                 }
2605                 bio_put(bio);
2606                 ee_len    -= done;
2607                 ee_pblock += done  << (blkbits - 9);
2608         }
2609         return 0;
2610 }
2611
2612 #define EXT4_EXT_ZERO_LEN 7
2613 /*
2614  * This function is called by ext4_ext_map_blocks() if someone tries to write
2615  * to an uninitialized extent. It may result in splitting the uninitialized
2616  * extent into multiple extents (upto three - one initialized and two
2617  * uninitialized).
2618  * There are three possibilities:
2619  *   a> There is no split required: Entire extent should be initialized
2620  *   b> Splits in two extents: Write is happening at either end of the extent
2621  *   c> Splits in three extents: Somone is writing in middle of the extent
2622  */
2623 static int ext4_ext_convert_to_initialized(handle_t *handle,
2624                                            struct inode *inode,
2625                                            struct ext4_map_blocks *map,
2626                                            struct ext4_ext_path *path)
2627 {
2628         struct ext4_extent *ex, newex, orig_ex;
2629         struct ext4_extent *ex1 = NULL;
2630         struct ext4_extent *ex2 = NULL;
2631         struct ext4_extent *ex3 = NULL;
2632         struct ext4_extent_header *eh;
2633         ext4_lblk_t ee_block, eof_block;
2634         unsigned int allocated, ee_len, depth;
2635         ext4_fsblk_t newblock;
2636         int err = 0;
2637         int ret = 0;
2638         int may_zeroout;
2639
2640         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2641                 "block %llu, max_blocks %u\n", inode->i_ino,
2642                 (unsigned long long)map->m_lblk, map->m_len);
2643
2644         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2645                 inode->i_sb->s_blocksize_bits;
2646         if (eof_block < map->m_lblk + map->m_len)
2647                 eof_block = map->m_lblk + map->m_len;
2648
2649         depth = ext_depth(inode);
2650         eh = path[depth].p_hdr;
2651         ex = path[depth].p_ext;
2652         ee_block = le32_to_cpu(ex->ee_block);
2653         ee_len = ext4_ext_get_actual_len(ex);
2654         allocated = ee_len - (map->m_lblk - ee_block);
2655         newblock = map->m_lblk - ee_block + ext_pblock(ex);
2656
2657         ex2 = ex;
2658         orig_ex.ee_block = ex->ee_block;
2659         orig_ex.ee_len   = cpu_to_le16(ee_len);
2660         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2661
2662         /*
2663          * It is safe to convert extent to initialized via explicit
2664          * zeroout only if extent is fully insde i_size or new_size.
2665          */
2666         may_zeroout = ee_block + ee_len <= eof_block;
2667
2668         err = ext4_ext_get_access(handle, inode, path + depth);
2669         if (err)
2670                 goto out;
2671         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2672         if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2673                 err =  ext4_ext_zeroout(inode, &orig_ex);
2674                 if (err)
2675                         goto fix_extent_len;
2676                 /* update the extent length and mark as initialized */
2677                 ex->ee_block = orig_ex.ee_block;
2678                 ex->ee_len   = orig_ex.ee_len;
2679                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2680                 ext4_ext_dirty(handle, inode, path + depth);
2681                 /* zeroed the full extent */
2682                 return allocated;
2683         }
2684
2685         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2686         if (map->m_lblk > ee_block) {
2687                 ex1 = ex;
2688                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2689                 ext4_ext_mark_uninitialized(ex1);
2690                 ex2 = &newex;
2691         }
2692         /*
2693          * for sanity, update the length of the ex2 extent before
2694          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2695          * overlap of blocks.
2696          */
2697         if (!ex1 && allocated > map->m_len)
2698                 ex2->ee_len = cpu_to_le16(map->m_len);
2699         /* ex3: to ee_block + ee_len : uninitialised */
2700         if (allocated > map->m_len) {
2701                 unsigned int newdepth;
2702                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2703                 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2704                         /*
2705                          * map->m_lblk == ee_block is handled by the zerouout
2706                          * at the beginning.
2707                          * Mark first half uninitialized.
2708                          * Mark second half initialized and zero out the
2709                          * initialized extent
2710                          */
2711                         ex->ee_block = orig_ex.ee_block;
2712                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2713                         ext4_ext_mark_uninitialized(ex);
2714                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2715                         ext4_ext_dirty(handle, inode, path + depth);
2716
2717                         ex3 = &newex;
2718                         ex3->ee_block = cpu_to_le32(map->m_lblk);
2719                         ext4_ext_store_pblock(ex3, newblock);
2720                         ex3->ee_len = cpu_to_le16(allocated);
2721                         err = ext4_ext_insert_extent(handle, inode, path,
2722                                                         ex3, 0);
2723                         if (err == -ENOSPC) {
2724                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2725                                 if (err)
2726                                         goto fix_extent_len;
2727                                 ex->ee_block = orig_ex.ee_block;
2728                                 ex->ee_len   = orig_ex.ee_len;
2729                                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2730                                 ext4_ext_dirty(handle, inode, path + depth);
2731                                 /* blocks available from map->m_lblk */
2732                                 return allocated;
2733
2734                         } else if (err)
2735                                 goto fix_extent_len;
2736
2737                         /*
2738                          * We need to zero out the second half because
2739                          * an fallocate request can update file size and
2740                          * converting the second half to initialized extent
2741                          * implies that we can leak some junk data to user
2742                          * space.
2743                          */
2744                         err =  ext4_ext_zeroout(inode, ex3);
2745                         if (err) {
2746                                 /*
2747                                  * We should actually mark the
2748                                  * second half as uninit and return error
2749                                  * Insert would have changed the extent
2750                                  */
2751                                 depth = ext_depth(inode);
2752                                 ext4_ext_drop_refs(path);
2753                                 path = ext4_ext_find_extent(inode, map->m_lblk,
2754                                                             path);
2755                                 if (IS_ERR(path)) {
2756                                         err = PTR_ERR(path);
2757                                         return err;
2758                                 }
2759                                 /* get the second half extent details */
2760                                 ex = path[depth].p_ext;
2761                                 err = ext4_ext_get_access(handle, inode,
2762                                                                 path + depth);
2763                                 if (err)
2764                                         return err;
2765                                 ext4_ext_mark_uninitialized(ex);
2766                                 ext4_ext_dirty(handle, inode, path + depth);
2767                                 return err;
2768                         }
2769
2770                         /* zeroed the second half */
2771                         return allocated;
2772                 }
2773                 ex3 = &newex;
2774                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2775                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2776                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2777                 ext4_ext_mark_uninitialized(ex3);
2778                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2779                 if (err == -ENOSPC && may_zeroout) {
2780                         err =  ext4_ext_zeroout(inode, &orig_ex);
2781                         if (err)
2782                                 goto fix_extent_len;
2783                         /* update the extent length and mark as initialized */
2784                         ex->ee_block = orig_ex.ee_block;
2785                         ex->ee_len   = orig_ex.ee_len;
2786                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2787                         ext4_ext_dirty(handle, inode, path + depth);
2788                         /* zeroed the full extent */
2789                         /* blocks available from map->m_lblk */
2790                         return allocated;
2791
2792                 } else if (err)
2793                         goto fix_extent_len;
2794                 /*
2795                  * The depth, and hence eh & ex might change
2796                  * as part of the insert above.
2797                  */
2798                 newdepth = ext_depth(inode);
2799                 /*
2800                  * update the extent length after successful insert of the
2801                  * split extent
2802                  */
2803                 ee_len -= ext4_ext_get_actual_len(ex3);
2804                 orig_ex.ee_len = cpu_to_le16(ee_len);
2805                 may_zeroout = ee_block + ee_len <= eof_block;
2806
2807                 depth = newdepth;
2808                 ext4_ext_drop_refs(path);
2809                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2810                 if (IS_ERR(path)) {
2811                         err = PTR_ERR(path);
2812                         goto out;
2813                 }
2814                 eh = path[depth].p_hdr;
2815                 ex = path[depth].p_ext;
2816                 if (ex2 != &newex)
2817                         ex2 = ex;
2818
2819                 err = ext4_ext_get_access(handle, inode, path + depth);
2820                 if (err)
2821                         goto out;
2822
2823                 allocated = map->m_len;
2824
2825                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2826                  * to insert a extent in the middle zerout directly
2827                  * otherwise give the extent a chance to merge to left
2828                  */
2829                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2830                         map->m_lblk != ee_block && may_zeroout) {
2831                         err =  ext4_ext_zeroout(inode, &orig_ex);
2832                         if (err)
2833                                 goto fix_extent_len;
2834                         /* update the extent length and mark as initialized */
2835                         ex->ee_block = orig_ex.ee_block;
2836                         ex->ee_len   = orig_ex.ee_len;
2837                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2838                         ext4_ext_dirty(handle, inode, path + depth);
2839                         /* zero out the first half */
2840                         /* blocks available from map->m_lblk */
2841                         return allocated;
2842                 }
2843         }
2844         /*
2845          * If there was a change of depth as part of the
2846          * insertion of ex3 above, we need to update the length
2847          * of the ex1 extent again here
2848          */
2849         if (ex1 && ex1 != ex) {
2850                 ex1 = ex;
2851                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2852                 ext4_ext_mark_uninitialized(ex1);
2853                 ex2 = &newex;
2854         }
2855         /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2856         ex2->ee_block = cpu_to_le32(map->m_lblk);
2857         ext4_ext_store_pblock(ex2, newblock);
2858         ex2->ee_len = cpu_to_le16(allocated);
2859         if (ex2 != ex)
2860                 goto insert;
2861         /*
2862          * New (initialized) extent starts from the first block
2863          * in the current extent. i.e., ex2 == ex
2864          * We have to see if it can be merged with the extent
2865          * on the left.
2866          */
2867         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2868                 /*
2869                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2870                  * since it merges towards right _only_.
2871                  */
2872                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2873                 if (ret) {
2874                         err = ext4_ext_correct_indexes(handle, inode, path);
2875                         if (err)
2876                                 goto out;
2877                         depth = ext_depth(inode);
2878                         ex2--;
2879                 }
2880         }
2881         /*
2882          * Try to Merge towards right. This might be required
2883          * only when the whole extent is being written to.
2884          * i.e. ex2 == ex and ex3 == NULL.
2885          */
2886         if (!ex3) {
2887                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2888                 if (ret) {
2889                         err = ext4_ext_correct_indexes(handle, inode, path);
2890                         if (err)
2891                                 goto out;
2892                 }
2893         }
2894         /* Mark modified extent as dirty */
2895         err = ext4_ext_dirty(handle, inode, path + depth);
2896         goto out;
2897 insert:
2898         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2899         if (err == -ENOSPC && may_zeroout) {
2900                 err =  ext4_ext_zeroout(inode, &orig_ex);
2901                 if (err)
2902                         goto fix_extent_len;
2903                 /* update the extent length and mark as initialized */
2904                 ex->ee_block = orig_ex.ee_block;
2905                 ex->ee_len   = orig_ex.ee_len;
2906                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2907                 ext4_ext_dirty(handle, inode, path + depth);
2908                 /* zero out the first half */
2909                 return allocated;
2910         } else if (err)
2911                 goto fix_extent_len;
2912 out:
2913         ext4_ext_show_leaf(inode, path);
2914         return err ? err : allocated;
2915
2916 fix_extent_len:
2917         ex->ee_block = orig_ex.ee_block;
2918         ex->ee_len   = orig_ex.ee_len;
2919         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2920         ext4_ext_mark_uninitialized(ex);
2921         ext4_ext_dirty(handle, inode, path + depth);
2922         return err;
2923 }
2924
2925 /*
2926  * This function is called by ext4_ext_map_blocks() from
2927  * ext4_get_blocks_dio_write() when DIO to write
2928  * to an uninitialized extent.
2929  *
2930  * Writing to an uninitized extent may result in splitting the uninitialized
2931  * extent into multiple /intialized unintialized extents (up to three)
2932  * There are three possibilities:
2933  *   a> There is no split required: Entire extent should be uninitialized
2934  *   b> Splits in two extents: Write is happening at either end of the extent
2935  *   c> Splits in three extents: Somone is writing in middle of the extent
2936  *
2937  * One of more index blocks maybe needed if the extent tree grow after
2938  * the unintialized extent split. To prevent ENOSPC occur at the IO
2939  * complete, we need to split the uninitialized extent before DIO submit
2940  * the IO. The uninitilized extent called at this time will be split
2941  * into three uninitialized extent(at most). After IO complete, the part
2942  * being filled will be convert to initialized by the end_io callback function
2943  * via ext4_convert_unwritten_extents().
2944  *
2945  * Returns the size of uninitialized extent to be written on success.
2946  */
2947 static int ext4_split_unwritten_extents(handle_t *handle,
2948                                         struct inode *inode,
2949                                         struct ext4_map_blocks *map,
2950                                         struct ext4_ext_path *path,
2951                                         int flags)
2952 {
2953         struct ext4_extent *ex, newex, orig_ex;
2954         struct ext4_extent *ex1 = NULL;
2955         struct ext4_extent *ex2 = NULL;
2956         struct ext4_extent *ex3 = NULL;
2957         struct ext4_extent_header *eh;
2958         ext4_lblk_t ee_block, eof_block;
2959         unsigned int allocated, ee_len, depth;
2960         ext4_fsblk_t newblock;
2961         int err = 0;
2962         int may_zeroout;
2963
2964         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2965                 "block %llu, max_blocks %u\n", inode->i_ino,
2966                 (unsigned long long)map->m_lblk, map->m_len);
2967
2968         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2969                 inode->i_sb->s_blocksize_bits;
2970         if (eof_block < map->m_lblk + map->m_len)
2971                 eof_block = map->m_lblk + map->m_len;
2972
2973         depth = ext_depth(inode);
2974         eh = path[depth].p_hdr;
2975         ex = path[depth].p_ext;
2976         ee_block = le32_to_cpu(ex->ee_block);
2977         ee_len = ext4_ext_get_actual_len(ex);
2978         allocated = ee_len - (map->m_lblk - ee_block);
2979         newblock = map->m_lblk - ee_block + ext_pblock(ex);
2980
2981         ex2 = ex;
2982         orig_ex.ee_block = ex->ee_block;
2983         orig_ex.ee_len   = cpu_to_le16(ee_len);
2984         ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
2985
2986         /*
2987          * It is safe to convert extent to initialized via explicit
2988          * zeroout only if extent is fully insde i_size or new_size.
2989          */
2990         may_zeroout = ee_block + ee_len <= eof_block;
2991
2992         /*
2993          * If the uninitialized extent begins at the same logical
2994          * block where the write begins, and the write completely
2995          * covers the extent, then we don't need to split it.
2996          */
2997         if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2998                 return allocated;
2999
3000         err = ext4_ext_get_access(handle, inode, path + depth);
3001         if (err)
3002                 goto out;
3003         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
3004         if (map->m_lblk > ee_block) {
3005                 ex1 = ex;
3006                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
3007                 ext4_ext_mark_uninitialized(ex1);
3008                 ex2 = &newex;
3009         }
3010         /*
3011          * for sanity, update the length of the ex2 extent before
3012          * we insert ex3, if ex1 is NULL. This is to avoid temporary
3013          * overlap of blocks.
3014          */
3015         if (!ex1 && allocated > map->m_len)
3016                 ex2->ee_len = cpu_to_le16(map->m_len);
3017         /* ex3: to ee_block + ee_len : uninitialised */
3018         if (allocated > map->m_len) {
3019                 unsigned int newdepth;
3020                 ex3 = &newex;
3021                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
3022                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
3023                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
3024                 ext4_ext_mark_uninitialized(ex3);
3025                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
3026                 if (err == -ENOSPC && may_zeroout) {
3027                         err =  ext4_ext_zeroout(inode, &orig_ex);
3028                         if (err)
3029                                 goto fix_extent_len;
3030                         /* update the extent length and mark as initialized */
3031                         ex->ee_block = orig_ex.ee_block;
3032                         ex->ee_len   = orig_ex.ee_len;
3033                         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3034                         ext4_ext_dirty(handle, inode, path + depth);
3035                         /* zeroed the full extent */
3036                         /* blocks available from map->m_lblk */
3037                         return allocated;
3038
3039                 } else if (err)
3040                         goto fix_extent_len;
3041                 /*
3042                  * The depth, and hence eh & ex might change
3043                  * as part of the insert above.
3044                  */
3045                 newdepth = ext_depth(inode);
3046                 /*
3047                  * update the extent length after successful insert of the
3048                  * split extent
3049                  */
3050                 ee_len -= ext4_ext_get_actual_len(ex3);
3051                 orig_ex.ee_len = cpu_to_le16(ee_len);
3052                 may_zeroout = ee_block + ee_len <= eof_block;
3053
3054                 depth = newdepth;
3055                 ext4_ext_drop_refs(path);
3056                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
3057                 if (IS_ERR(path)) {
3058                         err = PTR_ERR(path);
3059                         goto out;
3060                 }
3061                 eh = path[depth].p_hdr;
3062                 ex = path[depth].p_ext;
3063                 if (ex2 != &newex)
3064                         ex2 = ex;
3065
3066                 err = ext4_ext_get_access(handle, inode, path + depth);
3067                 if (err)
3068                         goto out;
3069
3070                 allocated = map->m_len;
3071         }
3072         /*
3073          * If there was a change of depth as part of the
3074          * insertion of ex3 above, we need to update the length
3075          * of the ex1 extent again here
3076          */
3077         if (ex1 && ex1 != ex) {
3078                 ex1 = ex;
3079                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
3080                 ext4_ext_mark_uninitialized(ex1);
3081                 ex2 = &newex;
3082         }
3083         /*
3084          * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
3085          * using direct I/O, uninitialised still.
3086          */
3087         ex2->ee_block = cpu_to_le32(map->m_lblk);
3088         ext4_ext_store_pblock(ex2, newblock);
3089         ex2->ee_len = cpu_to_le16(allocated);
3090         ext4_ext_mark_uninitialized(ex2);
3091         if (ex2 != ex)
3092                 goto insert;
3093         /* Mark modified extent as dirty */
3094         err = ext4_ext_dirty(handle, inode, path + depth);
3095         ext_debug("out here\n");
3096         goto out;
3097 insert:
3098         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3099         if (err == -ENOSPC && may_zeroout) {
3100                 err =  ext4_ext_zeroout(inode, &orig_ex);
3101                 if (err)
3102                         goto fix_extent_len;
3103                 /* update the extent length and mark as initialized */
3104                 ex->ee_block = orig_ex.ee_block;
3105                 ex->ee_len   = orig_ex.ee_len;
3106                 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3107                 ext4_ext_dirty(handle, inode, path + depth);
3108                 /* zero out the first half */
3109                 return allocated;
3110         } else if (err)
3111                 goto fix_extent_len;
3112 out:
3113         ext4_ext_show_leaf(inode, path);
3114         return err ? err : allocated;
3115
3116 fix_extent_len:
3117         ex->ee_block = orig_ex.ee_block;
3118         ex->ee_len   = orig_ex.ee_len;
3119         ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
3120         ext4_ext_mark_uninitialized(ex);
3121         ext4_ext_dirty(handle, inode, path + depth);
3122         return err;
3123 }
3124 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3125                                               struct inode *inode,
3126                                               struct ext4_ext_path *path)
3127 {
3128         struct ext4_extent *ex;
3129         struct ext4_extent_header *eh;
3130         int depth;
3131         int err = 0;
3132         int ret = 0;
3133
3134         depth = ext_depth(inode);
3135         eh = path[depth].p_hdr;
3136         ex = path[depth].p_ext;
3137
3138         err = ext4_ext_get_access(handle, inode, path + depth);
3139         if (err)
3140                 goto out;
3141         /* first mark the extent as initialized */
3142         ext4_ext_mark_initialized(ex);
3143
3144         /*
3145          * We have to see if it can be merged with the extent
3146          * on the left.
3147          */
3148         if (ex > EXT_FIRST_EXTENT(eh)) {
3149                 /*
3150                  * To merge left, pass "ex - 1" to try_to_merge(),
3151                  * since it merges towards right _only_.
3152                  */
3153                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3154                 if (ret) {
3155                         err = ext4_ext_correct_indexes(handle, inode, path);
3156                         if (err)
3157                                 goto out;
3158                         depth = ext_depth(inode);
3159                         ex--;
3160                 }
3161         }
3162         /*
3163          * Try to Merge towards right.
3164          */
3165         ret = ext4_ext_try_to_merge(inode, path, ex);
3166         if (ret) {
3167                 err = ext4_ext_correct_indexes(handle, inode, path);
3168                 if (err)
3169                         goto out;
3170                 depth = ext_depth(inode);
3171         }
3172         /* Mark modified extent as dirty */
3173         err = ext4_ext_dirty(handle, inode, path + depth);
3174 out:
3175         ext4_ext_show_leaf(inode, path);
3176         return err;
3177 }
3178
3179 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3180                         sector_t block, int count)
3181 {
3182         int i;
3183         for (i = 0; i < count; i++)
3184                 unmap_underlying_metadata(bdev, block + i);
3185 }
3186
3187 static int
3188 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3189                         struct ext4_map_blocks *map,
3190                         struct ext4_ext_path *path, int flags,
3191                         unsigned int allocated, ext4_fsblk_t newblock)
3192 {
3193         int ret = 0;
3194         int err = 0;
3195         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3196
3197         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3198                   "block %llu, max_blocks %u, flags %d, allocated %u",
3199                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3200                   flags, allocated);
3201         ext4_ext_show_leaf(inode, path);
3202
3203         /* get_block() before submit the IO, split the extent */
3204         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3205                 ret = ext4_split_unwritten_extents(handle, inode, map,
3206                                                    path, flags);
3207                 /*
3208                  * Flag the inode(non aio case) or end_io struct (aio case)
3209                  * that this IO needs to convertion to written when IO is
3210                  * completed
3211                  */
3212                 if (io)
3213                         io->flag = EXT4_IO_UNWRITTEN;
3214                 else
3215                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3216                 if (ext4_should_dioread_nolock(inode))
3217                         map->m_flags |= EXT4_MAP_UNINIT;
3218                 goto out;
3219         }
3220         /* IO end_io complete, convert the filled extent to written */
3221         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3222                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3223                                                         path);
3224                 if (ret >= 0)
3225                         ext4_update_inode_fsync_trans(handle, inode, 1);
3226                 goto out2;
3227         }
3228         /* buffered IO case */
3229         /*
3230          * repeat fallocate creation request
3231          * we already have an unwritten extent
3232          */
3233         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3234                 goto map_out;
3235
3236         /* buffered READ or buffered write_begin() lookup */
3237         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3238                 /*
3239                  * We have blocks reserved already.  We
3240                  * return allocated blocks so that delalloc
3241                  * won't do block reservation for us.  But
3242                  * the buffer head will be unmapped so that
3243                  * a read from the block returns 0s.
3244                  */
3245                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3246                 goto out1;
3247         }
3248
3249         /* buffered write, writepage time, convert*/
3250         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3251         if (ret >= 0)
3252                 ext4_update_inode_fsync_trans(handle, inode, 1);
3253 out:
3254         if (ret <= 0) {
3255                 err = ret;
3256                 goto out2;
3257         } else
3258                 allocated = ret;
3259         map->m_flags |= EXT4_MAP_NEW;
3260         /*
3261          * if we allocated more blocks than requested
3262          * we need to make sure we unmap the extra block
3263          * allocated. The actual needed block will get
3264          * unmapped later when we find the buffer_head marked
3265          * new.
3266          */
3267         if (allocated > map->m_len) {
3268                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3269                                         newblock + map->m_len,
3270                                         allocated - map->m_len);
3271                 allocated = map->m_len;
3272         }
3273
3274         /*
3275          * If we have done fallocate with the offset that is already
3276          * delayed allocated, we would have block reservation
3277          * and quota reservation done in the delayed write path.
3278          * But fallocate would have already updated quota and block
3279          * count for this offset. So cancel these reservation
3280          */
3281         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3282                 ext4_da_update_reserve_space(inode, allocated, 0);
3283
3284 map_out:
3285         map->m_flags |= EXT4_MAP_MAPPED;
3286 out1:
3287         if (allocated > map->m_len)
3288                 allocated = map->m_len;
3289         ext4_ext_show_leaf(inode, path);
3290         map->m_pblk = newblock;
3291         map->m_len = allocated;
3292 out2:
3293         if (path) {
3294                 ext4_ext_drop_refs(path);
3295                 kfree(path);
3296         }
3297         return err ? err : allocated;
3298 }
3299 /*
3300  * Block allocation/map/preallocation routine for extents based files
3301  *
3302  *
3303  * Need to be called with
3304  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3305  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3306  *
3307  * return > 0, number of of blocks already mapped/allocated
3308  *          if create == 0 and these are pre-allocated blocks
3309  *              buffer head is unmapped
3310  *          otherwise blocks are mapped
3311  *
3312  * return = 0, if plain look up failed (blocks have not been allocated)
3313  *          buffer head is unmapped
3314  *
3315  * return < 0, error case.
3316  */
3317 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3318                         struct ext4_map_blocks *map, int flags)
3319 {
3320         struct ext4_ext_path *path = NULL;
3321         struct ext4_extent_header *eh;
3322         struct ext4_extent newex, *ex, *last_ex;
3323         ext4_fsblk_t newblock;
3324         int i, err = 0, depth, ret, cache_type;
3325         unsigned int allocated = 0;
3326         struct ext4_allocation_request ar;
3327         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3328
3329         ext_debug("blocks %u/%u requested for inode %lu\n",
3330                   map->m_lblk, map->m_len, inode->i_ino);
3331
3332         /* check in cache */
3333         cache_type = ext4_ext_in_cache(inode, map->m_lblk, &newex);
3334         if (cache_type) {
3335                 if (cache_type == EXT4_EXT_CACHE_GAP) {
3336                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3337                                 /*
3338                                  * block isn't allocated yet and
3339                                  * user doesn't want to allocate it
3340                                  */
3341                                 goto out2;
3342                         }
3343                         /* we should allocate requested block */
3344                 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
3345                         /* block is already allocated */
3346                         newblock = map->m_lblk
3347                                    - le32_to_cpu(newex.ee_block)
3348                                    + ext_pblock(&newex);
3349                         /* number of remaining blocks in the extent */
3350                         allocated = ext4_ext_get_actual_len(&newex) -
3351                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3352                         goto out;
3353                 } else {
3354                         BUG();
3355                 }
3356         }
3357
3358         /* find extent for this block */
3359         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3360         if (IS_ERR(path)) {
3361                 err = PTR_ERR(path);
3362                 path = NULL;
3363                 goto out2;
3364         }
3365
3366         depth = ext_depth(inode);
3367
3368         /*
3369          * consistent leaf must not be empty;
3370          * this situation is possible, though, _during_ tree modification;
3371          * this is why assert can't be put in ext4_ext_find_extent()
3372          */
3373         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3374                 EXT4_ERROR_INODE(inode, "bad extent address "
3375                                  "lblock: %lu, depth: %d pblock %lld",
3376                                  (unsigned long) map->m_lblk, depth,
3377                                  path[depth].p_block);
3378                 err = -EIO;
3379                 goto out2;
3380         }
3381         eh = path[depth].p_hdr;
3382
3383         ex = path[depth].p_ext;
3384         if (ex) {
3385                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3386                 ext4_fsblk_t ee_start = ext_pblock(ex);
3387                 unsigned short ee_len;
3388
3389                 /*
3390                  * Uninitialized extents are treated as holes, except that
3391                  * we split out initialized portions during a write.
3392                  */
3393                 ee_len = ext4_ext_get_actual_len(ex);
3394                 /* if found extent covers block, simply return it */
3395                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3396                         newblock = map->m_lblk - ee_block + ee_start;
3397                         /* number of remaining blocks in the extent */
3398                         allocated = ee_len - (map->m_lblk - ee_block);
3399                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3400                                   ee_block, ee_len, newblock);
3401
3402                         /* Do not put uninitialized extent in the cache */
3403                         if (!ext4_ext_is_uninitialized(ex)) {
3404                                 ext4_ext_put_in_cache(inode, ee_block,
3405                                                         ee_len, ee_start,
3406                                                         EXT4_EXT_CACHE_EXTENT);
3407                                 goto out;
3408                         }
3409                         ret = ext4_ext_handle_uninitialized_extents(handle,
3410                                         inode, map, path, flags, allocated,
3411                                         newblock);
3412                         return ret;
3413                 }
3414         }
3415
3416         /*
3417          * requested block isn't allocated yet;
3418          * we couldn't try to create block if create flag is zero
3419          */
3420         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3421                 /*
3422                  * put just found gap into cache to speed up
3423                  * subsequent requests
3424                  */
3425                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3426                 goto out2;
3427         }
3428         /*
3429          * Okay, we need to do block allocation.
3430          */
3431
3432         /* find neighbour allocated blocks */
3433         ar.lleft = map->m_lblk;
3434         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3435         if (err)
3436                 goto out2;
3437         ar.lright = map->m_lblk;
3438         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3439         if (err)
3440                 goto out2;
3441
3442         /*
3443          * See if request is beyond maximum number of blocks we can have in
3444          * a single extent. For an initialized extent this limit is
3445          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3446          * EXT_UNINIT_MAX_LEN.
3447          */
3448         if (map->m_len > EXT_INIT_MAX_LEN &&
3449             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3450                 map->m_len = EXT_INIT_MAX_LEN;
3451         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3452                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3453                 map->m_len = EXT_UNINIT_MAX_LEN;
3454
3455         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3456         newex.ee_block = cpu_to_le32(map->m_lblk);
3457         newex.ee_len = cpu_to_le16(map->m_len);
3458         err = ext4_ext_check_overlap(inode, &newex, path);
3459         if (err)
3460                 allocated = ext4_ext_get_actual_len(&newex);
3461         else
3462                 allocated = map->m_len;
3463
3464         /* allocate new block */
3465         ar.inode = inode;
3466         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3467         ar.logical = map->m_lblk;
3468         ar.len = allocated;
3469         if (S_ISREG(inode->i_mode))
3470                 ar.flags = EXT4_MB_HINT_DATA;
3471         else
3472                 /* disable in-core preallocation for non-regular files */
3473                 ar.flags = 0;
3474         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3475         if (!newblock)
3476                 goto out2;
3477         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3478                   ar.goal, newblock, allocated);
3479
3480         /* try to insert new extent into found leaf and return */
3481         ext4_ext_store_pblock(&newex, newblock);
3482         newex.ee_len = cpu_to_le16(ar.len);
3483         /* Mark uninitialized */
3484         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3485                 ext4_ext_mark_uninitialized(&newex);
3486                 /*
3487                  * io_end structure was created for every IO write to an
3488                  * uninitialized extent. To avoid unecessary conversion,
3489                  * here we flag the IO that really needs the conversion.
3490                  * For non asycn direct IO case, flag the inode state
3491                  * that we need to perform convertion when IO is done.
3492                  */
3493                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3494                         if (io)
3495                                 io->flag = EXT4_IO_UNWRITTEN;
3496                         else
3497                                 ext4_set_inode_state(inode,
3498                                                      EXT4_STATE_DIO_UNWRITTEN);
3499                 }
3500                 if (ext4_should_dioread_nolock(inode))
3501                         map->m_flags |= EXT4_MAP_UNINIT;
3502         }
3503
3504         if (unlikely(ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) {
3505                 if (unlikely(!eh->eh_entries)) {
3506                         EXT4_ERROR_INODE(inode,
3507                                          "eh->eh_entries == 0 and "
3508                                          "EOFBLOCKS_FL set");
3509                         err = -EIO;
3510                         goto out2;
3511                 }
3512                 last_ex = EXT_LAST_EXTENT(eh);
3513                 /*
3514                  * If the current leaf block was reached by looking at
3515                  * the last index block all the way down the tree, and
3516                  * we are extending the inode beyond the last extent
3517                  * in the current leaf block, then clear the
3518                  * EOFBLOCKS_FL flag.
3519                  */
3520                 for (i = depth-1; i >= 0; i--) {
3521                         if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3522                                 break;
3523                 }
3524                 if ((i < 0) &&
3525                     (map->m_lblk + ar.len > le32_to_cpu(last_ex->ee_block) +
3526                      ext4_ext_get_actual_len(last_ex)))
3527                         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3528         }
3529         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3530         if (err) {
3531                 /* free data blocks we just allocated */
3532                 /* not a good idea to call discard here directly,
3533                  * but otherwise we'd need to call it every free() */
3534                 ext4_discard_preallocations(inode);
3535                 ext4_free_blocks(handle, inode, 0, ext_pblock(&newex),
3536                                  ext4_ext_get_actual_len(&newex), 0);
3537                 goto out2;
3538         }
3539
3540         /* previous routine could use block we allocated */
3541         newblock = ext_pblock(&newex);
3542         allocated = ext4_ext_get_actual_len(&newex);
3543         if (allocated > map->m_len)
3544                 allocated = map->m_len;
3545         map->m_flags |= EXT4_MAP_NEW;
3546
3547         /*
3548          * Update reserved blocks/metadata blocks after successful
3549          * block allocation which had been deferred till now.
3550          */
3551         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3552                 ext4_da_update_reserve_space(inode, allocated, 1);
3553
3554         /*
3555          * Cache the extent and update transaction to commit on fdatasync only
3556          * when it is _not_ an uninitialized extent.
3557          */
3558         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3559                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock,
3560                                                 EXT4_EXT_CACHE_EXTENT);
3561                 ext4_update_inode_fsync_trans(handle, inode, 1);
3562         } else
3563                 ext4_update_inode_fsync_trans(handle, inode, 0);
3564 out:
3565         if (allocated > map->m_len)
3566                 allocated = map->m_len;
3567         ext4_ext_show_leaf(inode, path);
3568         map->m_flags |= EXT4_MAP_MAPPED;
3569         map->m_pblk = newblock;
3570         map->m_len = allocated;
3571 out2:
3572         if (path) {
3573                 ext4_ext_drop_refs(path);
3574                 kfree(path);
3575         }
3576         return err ? err : allocated;
3577 }
3578
3579 void ext4_ext_truncate(struct inode *inode)
3580 {
3581         struct address_space *mapping = inode->i_mapping;
3582         struct super_block *sb = inode->i_sb;
3583         ext4_lblk_t last_block;
3584         handle_t *handle;
3585         int err = 0;
3586
3587         /*
3588          * probably first extent we're gonna free will be last in block
3589          */
3590         err = ext4_writepage_trans_blocks(inode);
3591         handle = ext4_journal_start(inode, err);
3592         if (IS_ERR(handle))
3593                 return;
3594
3595         if (inode->i_size & (sb->s_blocksize - 1))
3596                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3597
3598         if (ext4_orphan_add(handle, inode))
3599                 goto out_stop;
3600
3601         down_write(&EXT4_I(inode)->i_data_sem);
3602         ext4_ext_invalidate_cache(inode);
3603
3604         ext4_discard_preallocations(inode);
3605
3606         /*
3607          * TODO: optimization is possible here.
3608          * Probably we need not scan at all,
3609          * because page truncation is enough.
3610          */
3611
3612         /* we have to know where to truncate from in crash case */
3613         EXT4_I(inode)->i_disksize = inode->i_size;
3614         ext4_mark_inode_dirty(handle, inode);
3615
3616         last_block = (inode->i_size + sb->s_blocksize - 1)
3617                         >> EXT4_BLOCK_SIZE_BITS(sb);
3618         err = ext4_ext_remove_space(inode, last_block);
3619
3620         /* In a multi-transaction truncate, we only make the final
3621          * transaction synchronous.
3622          */
3623         if (IS_SYNC(inode))
3624                 ext4_handle_sync(handle);
3625
3626 out_stop:
3627         up_write(&EXT4_I(inode)->i_data_sem);
3628         /*
3629          * If this was a simple ftruncate() and the file will remain alive,
3630          * then we need to clear up the orphan record which we created above.
3631          * However, if this was a real unlink then we were called by
3632          * ext4_delete_inode(), and we allow that function to clean up the
3633          * orphan info for us.
3634          */
3635         if (inode->i_nlink)
3636                 ext4_orphan_del(handle, inode);
3637
3638         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3639         ext4_mark_inode_dirty(handle, inode);
3640         ext4_journal_stop(handle);
3641 }
3642
3643 static void ext4_falloc_update_inode(struct inode *inode,
3644                                 int mode, loff_t new_size, int update_ctime)
3645 {
3646         struct timespec now;
3647
3648         if (update_ctime) {
3649                 now = current_fs_time(inode->i_sb);
3650                 if (!timespec_equal(&inode->i_ctime, &now))
3651                         inode->i_ctime = now;
3652         }
3653         /*
3654          * Update only when preallocation was requested beyond
3655          * the file size.
3656          */
3657         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3658                 if (new_size > i_size_read(inode))
3659                         i_size_write(inode, new_size);
3660                 if (new_size > EXT4_I(inode)->i_disksize)
3661                         ext4_update_i_disksize(inode, new_size);
3662         } else {
3663                 /*
3664                  * Mark that we allocate beyond EOF so the subsequent truncate
3665                  * can proceed even if the new size is the same as i_size.
3666                  */
3667                 if (new_size > i_size_read(inode))
3668                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3669         }
3670
3671 }
3672
3673 /*
3674  * preallocate space for a file. This implements ext4's fallocate inode
3675  * operation, which gets called from sys_fallocate system call.
3676  * For block-mapped files, posix_fallocate should fall back to the method
3677  * of writing zeroes to the required new blocks (the same behavior which is
3678  * expected for file systems which do not support fallocate() system call).
3679  */
3680 long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3681 {
3682         handle_t *handle;
3683         loff_t new_size;
3684         unsigned int max_blocks;
3685         int ret = 0;
3686         int ret2 = 0;
3687         int retries = 0;
3688         struct ext4_map_blocks map;
3689         unsigned int credits, blkbits = inode->i_blkbits;
3690
3691         /*
3692          * currently supporting (pre)allocate mode for extent-based
3693          * files _only_
3694          */
3695         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3696                 return -EOPNOTSUPP;
3697
3698         /* preallocation to directories is currently not supported */
3699         if (S_ISDIR(inode->i_mode))
3700                 return -ENODEV;
3701
3702         map.m_lblk = offset >> blkbits;
3703         /*
3704          * We can't just convert len to max_blocks because
3705          * If blocksize = 4096 offset = 3072 and len = 2048
3706          */
3707         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3708                 - map.m_lblk;
3709         /*
3710          * credits to insert 1 extent into extent tree
3711          */
3712         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3713         mutex_lock(&inode->i_mutex);
3714         ret = inode_newsize_ok(inode, (len + offset));
3715         if (ret) {
3716                 mutex_unlock(&inode->i_mutex);
3717                 return ret;
3718         }
3719 retry:
3720         while (ret >= 0 && ret < max_blocks) {
3721                 map.m_lblk = map.m_lblk + ret;
3722                 map.m_len = max_blocks = max_blocks - ret;
3723                 handle = ext4_journal_start(inode, credits);
3724                 if (IS_ERR(handle)) {
3725                         ret = PTR_ERR(handle);
3726                         break;
3727                 }
3728                 ret = ext4_map_blocks(handle, inode, &map,
3729                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3730                 if (ret <= 0) {
3731 #ifdef EXT4FS_DEBUG
3732                         WARN_ON(ret <= 0);
3733                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3734                                     "returned error inode#%lu, block=%u, "
3735                                     "max_blocks=%u", __func__,
3736                                     inode->i_ino, block, max_blocks);
3737 #endif
3738                         ext4_mark_inode_dirty(handle, inode);
3739                         ret2 = ext4_journal_stop(handle);
3740                         break;
3741                 }
3742                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3743                                                 blkbits) >> blkbits))
3744                         new_size = offset + len;
3745                 else
3746                         new_size = (map.m_lblk + ret) << blkbits;
3747
3748                 ext4_falloc_update_inode(inode, mode, new_size,
3749                                          (map.m_flags & EXT4_MAP_NEW));
3750                 ext4_mark_inode_dirty(handle, inode);
3751                 ret2 = ext4_journal_stop(handle);
3752                 if (ret2)
3753                         break;
3754         }
3755         if (ret == -ENOSPC &&
3756                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3757                 ret = 0;
3758                 goto retry;
3759         }
3760         mutex_unlock(&inode->i_mutex);
3761         return ret > 0 ? ret2 : ret;
3762 }
3763
3764 /*
3765  * This function convert a range of blocks to written extents
3766  * The caller of this function will pass the start offset and the size.
3767  * all unwritten extents within this range will be converted to
3768  * written extents.
3769  *
3770  * This function is called from the direct IO end io call back
3771  * function, to convert the fallocated extents after IO is completed.
3772  * Returns 0 on success.
3773  */
3774 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3775                                     ssize_t len)
3776 {
3777         handle_t *handle;
3778         unsigned int max_blocks;
3779         int ret = 0;
3780         int ret2 = 0;
3781         struct ext4_map_blocks map;
3782         unsigned int credits, blkbits = inode->i_blkbits;
3783
3784         map.m_lblk = offset >> blkbits;
3785         /*
3786          * We can't just convert len to max_blocks because
3787          * If blocksize = 4096 offset = 3072 and len = 2048
3788          */
3789         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3790                       map.m_lblk);
3791         /*
3792          * credits to insert 1 extent into extent tree
3793          */
3794         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3795         while (ret >= 0 && ret < max_blocks) {
3796                 map.m_lblk += ret;
3797                 map.m_len = (max_blocks -= ret);
3798                 handle = ext4_journal_start(inode, credits);
3799                 if (IS_ERR(handle)) {
3800                         ret = PTR_ERR(handle);
3801                         break;
3802                 }
3803                 ret = ext4_map_blocks(handle, inode, &map,
3804                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3805                 if (ret <= 0) {
3806                         WARN_ON(ret <= 0);
3807                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3808                                     "returned error inode#%lu, block=%u, "
3809                                     "max_blocks=%u", __func__,
3810                                     inode->i_ino, map.m_lblk, map.m_len);
3811                 }
3812                 ext4_mark_inode_dirty(handle, inode);
3813                 ret2 = ext4_journal_stop(handle);
3814                 if (ret <= 0 || ret2 )
3815                         break;
3816         }
3817         return ret > 0 ? ret2 : ret;
3818 }
3819 /*
3820  * Callback function called for each extent to gather FIEMAP information.
3821  */
3822 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3823                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3824                        void *data)
3825 {
3826         struct fiemap_extent_info *fieinfo = data;
3827         unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
3828         __u64   logical;
3829         __u64   physical;
3830         __u64   length;
3831         __u32   flags = 0;
3832         int     error;
3833
3834         logical =  (__u64)newex->ec_block << blksize_bits;
3835
3836         if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3837                 pgoff_t offset;
3838                 struct page *page;
3839                 struct buffer_head *bh = NULL;
3840
3841                 offset = logical >> PAGE_SHIFT;
3842                 page = find_get_page(inode->i_mapping, offset);
3843                 if (!page || !page_has_buffers(page))
3844                         return EXT_CONTINUE;
3845
3846                 bh = page_buffers(page);
3847
3848                 if (!bh)
3849                         return EXT_CONTINUE;
3850
3851                 if (buffer_delay(bh)) {
3852                         flags |= FIEMAP_EXTENT_DELALLOC;
3853                         page_cache_release(page);
3854                 } else {
3855                         page_cache_release(page);
3856                         return EXT_CONTINUE;
3857                 }
3858         }
3859
3860         physical = (__u64)newex->ec_start << blksize_bits;
3861         length =   (__u64)newex->ec_len << blksize_bits;
3862
3863         if (ex && ext4_ext_is_uninitialized(ex))
3864                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3865
3866         /*
3867          * If this extent reaches EXT_MAX_BLOCK, it must be last.
3868          *
3869          * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3870          * this also indicates no more allocated blocks.
3871          *
3872          * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3873          */
3874         if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
3875             newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3876                 loff_t size = i_size_read(inode);
3877                 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3878
3879                 flags |= FIEMAP_EXTENT_LAST;
3880                 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3881                     logical+length > size)
3882                         length = (size - logical + bs - 1) & ~(bs-1);
3883         }
3884
3885         error = fiemap_fill_next_extent(fieinfo, logical, physical,
3886                                         length, flags);
3887         if (error < 0)
3888                 return error;
3889         if (error == 1)
3890                 return EXT_BREAK;
3891
3892         return EXT_CONTINUE;
3893 }
3894
3895 /* fiemap flags we can handle specified here */
3896 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3897
3898 static int ext4_xattr_fiemap(struct inode *inode,
3899                                 struct fiemap_extent_info *fieinfo)
3900 {
3901         __u64 physical = 0;
3902         __u64 length;
3903         __u32 flags = FIEMAP_EXTENT_LAST;
3904         int blockbits = inode->i_sb->s_blocksize_bits;
3905         int error = 0;
3906
3907         /* in-inode? */
3908         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3909                 struct ext4_iloc iloc;
3910                 int offset;     /* offset of xattr in inode */
3911
3912                 error = ext4_get_inode_loc(inode, &iloc);
3913                 if (error)
3914                         return error;
3915                 physical = iloc.bh->b_blocknr << blockbits;
3916                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3917                                 EXT4_I(inode)->i_extra_isize;
3918                 physical += offset;
3919                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3920                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3921                 brelse(iloc.bh);
3922         } else { /* external block */
3923                 physical = EXT4_I(inode)->i_file_acl << blockbits;
3924                 length = inode->i_sb->s_blocksize;
3925         }
3926
3927         if (physical)
3928                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3929                                                 length, flags);
3930         return (error < 0 ? error : 0);
3931 }
3932
3933 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3934                 __u64 start, __u64 len)
3935 {
3936         ext4_lblk_t start_blk;
3937         int error = 0;
3938
3939         /* fallback to generic here if not in extents fmt */
3940         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3941                 return generic_block_fiemap(inode, fieinfo, start, len,
3942                         ext4_get_block);
3943
3944         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3945                 return -EBADR;
3946
3947         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3948                 error = ext4_xattr_fiemap(inode, fieinfo);
3949         } else {
3950                 ext4_lblk_t len_blks;
3951                 __u64 last_blk;
3952
3953                 start_blk = start >> inode->i_sb->s_blocksize_bits;
3954                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
3955                 if (last_blk >= EXT_MAX_BLOCK)
3956                         last_blk = EXT_MAX_BLOCK-1;
3957                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
3958
3959                 /*
3960                  * Walk the extent tree gathering extent information.
3961                  * ext4_ext_fiemap_cb will push extents back to user.
3962                  */
3963                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3964                                           ext4_ext_fiemap_cb, fieinfo);
3965         }
3966
3967         return error;
3968 }
3969