Pileus Git - ~andy/linux/blob - 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 #include <trace/events/ext4.h>
  48
  49 static int ext4_split_extent(handle_t *handle,
  50                                 struct inode *inode,
  51                                 struct ext4_ext_path *path,
  52                                 struct ext4_map_blocks *map,
  53                                 int split_flag,
  54                                 int flags);
  55
  56 static int ext4_ext_truncate_extend_restart(handle_t *handle,
  57                                             struct inode *inode,
  58                                             int needed)
  59 {
  60         int err;
  61
  62         if (!ext4_handle_valid(handle))
  63                 return 0;
  64         if (handle->h_buffer_credits > needed)
  65                 return 0;
  66         err = ext4_journal_extend(handle, needed);
  67         if (err <= 0)
  68                 return err;
  69         err = ext4_truncate_restart_trans(handle, inode, needed);
  70         if (err == 0)
  71                 err = -EAGAIN;
  72
  73         return err;
  74 }
  75
  76 /*
  77  * could return:
  78  *  - EROFS
  79  *  - ENOMEM
  80  */
  81 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  82                                 struct ext4_ext_path *path)
  83 {
  84         if (path->p_bh) {
  85                 /* path points to block */
  86                 return ext4_journal_get_write_access(handle, path->p_bh);
  87         }
  88         /* path points to leaf/index in inode body */
  89         /* we use in-core data, no need to protect them */
  90         return 0;
  91 }
  92
  93 /*
  94  * could return:
  95  *  - EROFS
  96  *  - ENOMEM
  97  *  - EIO
  98  */
  99 #define ext4_ext_dirty(handle, inode, path) \
 100                 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 101 static int __ext4_ext_dirty(const char *where, unsigned int line,
 102                             handle_t *handle, struct inode *inode,
 103                             struct ext4_ext_path *path)
 104 {
 105         int err;
 106         if (path->p_bh) {
 107                 /* path points to block */
 108                 err = __ext4_handle_dirty_metadata(where, line, handle,
 109                                                    inode, path->p_bh);
 110         } else {
 111                 /* path points to leaf/index in inode body */
 112                 err = ext4_mark_inode_dirty(handle, inode);
 113         }
 114         return err;
 115 }
 116
 117 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 118                               struct ext4_ext_path *path,
 119                               ext4_lblk_t block)
 120 {
 121         int depth;
 122
 123         if (path) {
 124                 struct ext4_extent *ex;
 125                 depth = path->p_depth;
 126
 127                 /*
 128                  * Try to predict block placement assuming that we are
 129                  * filling in a file which will eventually be
 130                  * non-sparse --- i.e., in the case of libbfd writing
 131                  * an ELF object sections out-of-order but in a way
 132                  * the eventually results in a contiguous object or
 133                  * executable file, or some database extending a table
 134                  * space file.  However, this is actually somewhat
 135                  * non-ideal if we are writing a sparse file such as
 136                  * qemu or KVM writing a raw image file that is going
 137                  * to stay fairly sparse, since it will end up
 138                  * fragmenting the file system's free space.  Maybe we
 139                  * should have some hueristics or some way to allow
 140                  * userspace to pass a hint to file system,
 141                  * especially if the latter case turns out to be
 142                  * common.
 143                  */
 144                 ex = path[depth].p_ext;
 145                 if (ex) {
 146                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 147                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 148
 149                         if (block > ext_block)
 150                                 return ext_pblk + (block - ext_block);
 151                         else
 152                                 return ext_pblk - (ext_block - block);
 153                 }
 154
 155                 /* it looks like index is empty;
 156                  * try to find starting block from index itself */
 157                 if (path[depth].p_bh)
 158                         return path[depth].p_bh->b_blocknr;
 159         }
 160
 161         /* OK. use inode's group */
 162         return ext4_inode_to_goal_block(inode);
 163 }
 164
 165 /*
 166  * Allocation for a meta data block
 167  */
 168 static ext4_fsblk_t
 169 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 170                         struct ext4_ext_path *path,
 171                         struct ext4_extent *ex, int *err, unsigned int flags)
 172 {
 173         ext4_fsblk_t goal, newblock;
 174
 175         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 176         newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 177                                         NULL, err);
 178         return newblock;
 179 }
 180
 181 static inline int ext4_ext_space_block(struct inode *inode, int check)
 182 {
 183         int size;
 184
 185         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 186                         / sizeof(struct ext4_extent);
 187         if (!check) {
 188 #ifdef AGGRESSIVE_TEST
 189                 if (size > 6)
 190                         size = 6;
 191 #endif
 192         }
 193         return size;
 194 }
 195
 196 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 197 {
 198         int size;
 199
 200         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 201                         / sizeof(struct ext4_extent_idx);
 202         if (!check) {
 203 #ifdef AGGRESSIVE_TEST
 204                 if (size > 5)
 205                         size = 5;
 206 #endif
 207         }
 208         return size;
 209 }
 210
 211 static inline int ext4_ext_space_root(struct inode *inode, int check)
 212 {
 213         int size;
 214
 215         size = sizeof(EXT4_I(inode)->i_data);
 216         size -= sizeof(struct ext4_extent_header);
 217         size /= sizeof(struct ext4_extent);
 218         if (!check) {
 219 #ifdef AGGRESSIVE_TEST
 220                 if (size > 3)
 221                         size = 3;
 222 #endif
 223         }
 224         return size;
 225 }
 226
 227 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 228 {
 229         int size;
 230
 231         size = sizeof(EXT4_I(inode)->i_data);
 232         size -= sizeof(struct ext4_extent_header);
 233         size /= sizeof(struct ext4_extent_idx);
 234         if (!check) {
 235 #ifdef AGGRESSIVE_TEST
 236                 if (size > 4)
 237                         size = 4;
 238 #endif
 239         }
 240         return size;
 241 }
 242
 243 /*
 244  * Calculate the number of metadata blocks needed
 245  * to allocate @blocks
 246  * Worse case is one block per extent
 247  */
 248 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 249 {
 250         struct ext4_inode_info *ei = EXT4_I(inode);
 251         int idxs, num = 0;
 252
 253         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 254                 / sizeof(struct ext4_extent_idx));
 255
 256         /*
 257          * If the new delayed allocation block is contiguous with the
 258          * previous da block, it can share index blocks with the
 259          * previous block, so we only need to allocate a new index
 260          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 261          * an additional index block, and at ldxs**3 blocks, yet
 262          * another index blocks.
 263          */
 264         if (ei->i_da_metadata_calc_len &&
 265             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 266                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 267                         num++;
 268                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 269                         num++;
 270                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 271                         num++;
 272                         ei->i_da_metadata_calc_len = 0;
 273                 } else
 274                         ei->i_da_metadata_calc_len++;
 275                 ei->i_da_metadata_calc_last_lblock++;
 276                 return num;
 277         }
 278
 279         /*
 280          * In the worst case we need a new set of index blocks at
 281          * every level of the inode's extent tree.
 282          */
 283         ei->i_da_metadata_calc_len = 1;
 284         ei->i_da_metadata_calc_last_lblock = lblock;
 285         return ext_depth(inode) + 1;
 286 }
 287
 288 static int
 289 ext4_ext_max_entries(struct inode *inode, int depth)
 290 {
 291         int max;
 292
 293         if (depth == ext_depth(inode)) {
 294                 if (depth == 0)
 295                         max = ext4_ext_space_root(inode, 1);
 296                 else
 297                         max = ext4_ext_space_root_idx(inode, 1);
 298         } else {
 299                 if (depth == 0)
 300                         max = ext4_ext_space_block(inode, 1);
 301                 else
 302                         max = ext4_ext_space_block_idx(inode, 1);
 303         }
 304
 305         return max;
 306 }
 307
 308 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 309 {
 310         ext4_fsblk_t block = ext4_ext_pblock(ext);
 311         int len = ext4_ext_get_actual_len(ext);
 312
 313         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 314 }
 315
 316 static int ext4_valid_extent_idx(struct inode *inode,
 317                                 struct ext4_extent_idx *ext_idx)
 318 {
 319         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 320
 321         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 322 }
 323
 324 static int ext4_valid_extent_entries(struct inode *inode,
 325                                 struct ext4_extent_header *eh,
 326                                 int depth)
 327 {
 328         struct ext4_extent *ext;
 329         struct ext4_extent_idx *ext_idx;
 330         unsigned short entries;
 331         if (eh->eh_entries == 0)
 332                 return 1;
 333
 334         entries = le16_to_cpu(eh->eh_entries);
 335
 336         if (depth == 0) {
 337                 /* leaf entries */
 338                 ext = EXT_FIRST_EXTENT(eh);
 339                 while (entries) {
 340                         if (!ext4_valid_extent(inode, ext))
 341                                 return 0;
 342                         ext++;
 343                         entries--;
 344                 }
 345         } else {
 346                 ext_idx = EXT_FIRST_INDEX(eh);
 347                 while (entries) {
 348                         if (!ext4_valid_extent_idx(inode, ext_idx))
 349                                 return 0;
 350                         ext_idx++;
 351                         entries--;
 352                 }
 353         }
 354         return 1;
 355 }
 356
 357 static int __ext4_ext_check(const char *function, unsigned int line,
 358                             struct inode *inode, struct ext4_extent_header *eh,
 359                             int depth)
 360 {
 361         const char *error_msg;
 362         int max = 0;
 363
 364         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 365                 error_msg = "invalid magic";
 366                 goto corrupted;
 367         }
 368         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 369                 error_msg = "unexpected eh_depth";
 370                 goto corrupted;
 371         }
 372         if (unlikely(eh->eh_max == 0)) {
 373                 error_msg = "invalid eh_max";
 374                 goto corrupted;
 375         }
 376         max = ext4_ext_max_entries(inode, depth);
 377         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 378                 error_msg = "too large eh_max";
 379                 goto corrupted;
 380         }
 381         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 382                 error_msg = "invalid eh_entries";
 383                 goto corrupted;
 384         }
 385         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 386                 error_msg = "invalid extent entries";
 387                 goto corrupted;
 388         }
 389         return 0;
 390
 391 corrupted:
 392         ext4_error_inode(inode, function, line, 0,
 393                         "bad header/extent: %s - magic %x, "
 394                         "entries %u, max %u(%u), depth %u(%u)",
 395                         error_msg, le16_to_cpu(eh->eh_magic),
 396                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 397                         max, le16_to_cpu(eh->eh_depth), depth);
 398
 399         return -EIO;
 400 }
 401
 402 #define ext4_ext_check(inode, eh, depth)        \
 403         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 404
 405 int ext4_ext_check_inode(struct inode *inode)
 406 {
 407         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 408 }
 409
 410 #ifdef EXT_DEBUG
 411 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 412 {
 413         int k, l = path->p_depth;
 414
 415         ext_debug("path:");
 416         for (k = 0; k <= l; k++, path++) {
 417                 if (path->p_idx) {
 418                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 419                             ext4_idx_pblock(path->p_idx));
 420                 } else if (path->p_ext) {
 421                         ext_debug("  %d:[%d]%d:%llu ",
 422                                   le32_to_cpu(path->p_ext->ee_block),
 423                                   ext4_ext_is_uninitialized(path->p_ext),
 424                                   ext4_ext_get_actual_len(path->p_ext),
 425                                   ext4_ext_pblock(path->p_ext));
 426                 } else
 427                         ext_debug("  []");
 428         }
 429         ext_debug("\n");
 430 }
 431
 432 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 433 {
 434         int depth = ext_depth(inode);
 435         struct ext4_extent_header *eh;
 436         struct ext4_extent *ex;
 437         int i;
 438
 439         if (!path)
 440                 return;
 441
 442         eh = path[depth].p_hdr;
 443         ex = EXT_FIRST_EXTENT(eh);
 444
 445         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 446
 447         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 448                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 449                           ext4_ext_is_uninitialized(ex),
 450                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 451         }
 452         ext_debug("\n");
 453 }
 454
 455 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 456                         ext4_fsblk_t newblock, int level)
 457 {
 458         int depth = ext_depth(inode);
 459         struct ext4_extent *ex;
 460
 461         if (depth != level) {
 462                 struct ext4_extent_idx *idx;
 463                 idx = path[level].p_idx;
 464                 while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 465                         ext_debug("%d: move %d:%llu in new index %llu\n", level,
 466                                         le32_to_cpu(idx->ei_block),
 467                                         ext4_idx_pblock(idx),
 468                                         newblock);
 469                         idx++;
 470                 }
 471
 472                 return;
 473         }
 474
 475         ex = path[depth].p_ext;
 476         while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 477                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 478                                 le32_to_cpu(ex->ee_block),
 479                                 ext4_ext_pblock(ex),
 480                                 ext4_ext_is_uninitialized(ex),
 481                                 ext4_ext_get_actual_len(ex),
 482                                 newblock);
 483                 ex++;
 484         }
 485 }
 486
 487 #else
 488 #define ext4_ext_show_path(inode, path)
 489 #define ext4_ext_show_leaf(inode, path)
 490 #define ext4_ext_show_move(inode, path, newblock, level)
 491 #endif
 492
 493 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 494 {
 495         int depth = path->p_depth;
 496         int i;
 497
 498         for (i = 0; i <= depth; i++, path++)
 499                 if (path->p_bh) {
 500                         brelse(path->p_bh);
 501                         path->p_bh = NULL;
 502                 }
 503 }
 504
 505 /*
 506  * ext4_ext_binsearch_idx:
 507  * binary search for the closest index of the given block
 508  * the header must be checked before calling this
 509  */
 510 static void
 511 ext4_ext_binsearch_idx(struct inode *inode,
 512                         struct ext4_ext_path *path, ext4_lblk_t block)
 513 {
 514         struct ext4_extent_header *eh = path->p_hdr;
 515         struct ext4_extent_idx *r, *l, *m;
 516
 517
 518         ext_debug("binsearch for %u(idx):  ", block);
 519
 520         l = EXT_FIRST_INDEX(eh) + 1;
 521         r = EXT_LAST_INDEX(eh);
 522         while (l <= r) {
 523                 m = l + (r - l) / 2;
 524                 if (block < le32_to_cpu(m->ei_block))
 525                         r = m - 1;
 526                 else
 527                         l = m + 1;
 528                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 529                                 m, le32_to_cpu(m->ei_block),
 530                                 r, le32_to_cpu(r->ei_block));
 531         }
 532
 533         path->p_idx = l - 1;
 534         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 535                   ext4_idx_pblock(path->p_idx));
 536
 537 #ifdef CHECK_BINSEARCH
 538         {
 539                 struct ext4_extent_idx *chix, *ix;
 540                 int k;
 541
 542                 chix = ix = EXT_FIRST_INDEX(eh);
 543                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 544                   if (k != 0 &&
 545                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 546                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 547                                        "first=0x%p\n", k,
 548                                        ix, EXT_FIRST_INDEX(eh));
 549                                 printk(KERN_DEBUG "%u <= %u\n",
 550                                        le32_to_cpu(ix->ei_block),
 551                                        le32_to_cpu(ix[-1].ei_block));
 552                         }
 553                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 554                                            <= le32_to_cpu(ix[-1].ei_block));
 555                         if (block < le32_to_cpu(ix->ei_block))
 556                                 break;
 557                         chix = ix;
 558                 }
 559                 BUG_ON(chix != path->p_idx);
 560         }
 561 #endif
 562
 563 }
 564
 565 /*
 566  * ext4_ext_binsearch:
 567  * binary search for closest extent of the given block
 568  * the header must be checked before calling this
 569  */
 570 static void
 571 ext4_ext_binsearch(struct inode *inode,
 572                 struct ext4_ext_path *path, ext4_lblk_t block)
 573 {
 574         struct ext4_extent_header *eh = path->p_hdr;
 575         struct ext4_extent *r, *l, *m;
 576
 577         if (eh->eh_entries == 0) {
 578                 /*
 579                  * this leaf is empty:
 580                  * we get such a leaf in split/add case
 581                  */
 582                 return;
 583         }
 584
 585         ext_debug("binsearch for %u:  ", block);
 586
 587         l = EXT_FIRST_EXTENT(eh) + 1;
 588         r = EXT_LAST_EXTENT(eh);
 589
 590         while (l <= r) {
 591                 m = l + (r - l) / 2;
 592                 if (block < le32_to_cpu(m->ee_block))
 593                         r = m - 1;
 594                 else
 595                         l = m + 1;
 596                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 597                                 m, le32_to_cpu(m->ee_block),
 598                                 r, le32_to_cpu(r->ee_block));
 599         }
 600
 601         path->p_ext = l - 1;
 602         ext_debug("  -> %d:%llu:[%d]%d ",
 603                         le32_to_cpu(path->p_ext->ee_block),
 604                         ext4_ext_pblock(path->p_ext),
 605                         ext4_ext_is_uninitialized(path->p_ext),
 606                         ext4_ext_get_actual_len(path->p_ext));
 607
 608 #ifdef CHECK_BINSEARCH
 609         {
 610                 struct ext4_extent *chex, *ex;
 611                 int k;
 612
 613                 chex = ex = EXT_FIRST_EXTENT(eh);
 614                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 615                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 616                                           <= le32_to_cpu(ex[-1].ee_block));
 617                         if (block < le32_to_cpu(ex->ee_block))
 618                                 break;
 619                         chex = ex;
 620                 }
 621                 BUG_ON(chex != path->p_ext);
 622         }
 623 #endif
 624
 625 }
 626
 627 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 628 {
 629         struct ext4_extent_header *eh;
 630
 631         eh = ext_inode_hdr(inode);
 632         eh->eh_depth = 0;
 633         eh->eh_entries = 0;
 634         eh->eh_magic = EXT4_EXT_MAGIC;
 635         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 636         ext4_mark_inode_dirty(handle, inode);
 637         ext4_ext_invalidate_cache(inode);
 638         return 0;
 639 }
 640
 641 struct ext4_ext_path *
 642 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 643                                         struct ext4_ext_path *path)
 644 {
 645         struct ext4_extent_header *eh;
 646         struct buffer_head *bh;
 647         short int depth, i, ppos = 0, alloc = 0;
 648
 649         eh = ext_inode_hdr(inode);
 650         depth = ext_depth(inode);
 651
 652         /* account possible depth increase */
 653         if (!path) {
 654                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 655                                 GFP_NOFS);
 656                 if (!path)
 657                         return ERR_PTR(-ENOMEM);
 658                 alloc = 1;
 659         }
 660         path[0].p_hdr = eh;
 661         path[0].p_bh = NULL;
 662
 663         i = depth;
 664         /* walk through the tree */
 665         while (i) {
 666                 int need_to_validate = 0;
 667
 668                 ext_debug("depth %d: num %d, max %d\n",
 669                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 670
 671                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 672                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 673                 path[ppos].p_depth = i;
 674                 path[ppos].p_ext = NULL;
 675
 676                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 677                 if (unlikely(!bh))
 678                         goto err;
 679                 if (!bh_uptodate_or_lock(bh)) {
 680                         trace_ext4_ext_load_extent(inode, block,
 681                                                 path[ppos].p_block);
 682                         if (bh_submit_read(bh) < 0) {
 683                                 put_bh(bh);
 684                                 goto err;
 685                         }
 686                         /* validate the extent entries */
 687                         need_to_validate = 1;
 688                 }
 689                 eh = ext_block_hdr(bh);
 690                 ppos++;
 691                 if (unlikely(ppos > depth)) {
 692                         put_bh(bh);
 693                         EXT4_ERROR_INODE(inode,
 694                                          "ppos %d > depth %d", ppos, depth);
 695                         goto err;
 696                 }
 697                 path[ppos].p_bh = bh;
 698                 path[ppos].p_hdr = eh;
 699                 i--;
 700
 701                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 702                         goto err;
 703         }
 704
 705         path[ppos].p_depth = i;
 706         path[ppos].p_ext = NULL;
 707         path[ppos].p_idx = NULL;
 708
 709         /* find extent */
 710         ext4_ext_binsearch(inode, path + ppos, block);
 711         /* if not an empty leaf */
 712         if (path[ppos].p_ext)
 713                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 714
 715         ext4_ext_show_path(inode, path);
 716
 717         return path;
 718
 719 err:
 720         ext4_ext_drop_refs(path);
 721         if (alloc)
 722                 kfree(path);
 723         return ERR_PTR(-EIO);
 724 }
 725
 726 /*
 727  * ext4_ext_insert_index:
 728  * insert new index [@logical;@ptr] into the block at @curp;
 729  * check where to insert: before @curp or after @curp
 730  */
 731 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 732                                  struct ext4_ext_path *curp,
 733                                  int logical, ext4_fsblk_t ptr)
 734 {
 735         struct ext4_extent_idx *ix;
 736         int len, err;
 737
 738         err = ext4_ext_get_access(handle, inode, curp);
 739         if (err)
 740                 return err;
 741
 742         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 743                 EXT4_ERROR_INODE(inode,
 744                                  "logical %d == ei_block %d!",
 745                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 746                 return -EIO;
 747         }
 748
 749         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 750                              >= le16_to_cpu(curp->p_hdr->eh_max))) {
 751                 EXT4_ERROR_INODE(inode,
 752                                  "eh_entries %d >= eh_max %d!",
 753                                  le16_to_cpu(curp->p_hdr->eh_entries),
 754                                  le16_to_cpu(curp->p_hdr->eh_max));
 755                 return -EIO;
 756         }
 757
 758         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 759         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 760                 /* insert after */
 761                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 762                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 763                         len = len < 0 ? 0 : len;
 764                         ext_debug("insert new index %d after: %llu. "
 765                                         "move %d from 0x%p to 0x%p\n",
 766                                         logical, ptr, len,
 767                                         (curp->p_idx + 1), (curp->p_idx + 2));
 768                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 769                 }
 770                 ix = curp->p_idx + 1;
 771         } else {
 772                 /* insert before */
 773                 len = len * sizeof(struct ext4_extent_idx);
 774                 len = len < 0 ? 0 : len;
 775                 ext_debug("insert new index %d before: %llu. "
 776                                 "move %d from 0x%p to 0x%p\n",
 777                                 logical, ptr, len,
 778                                 curp->p_idx, (curp->p_idx + 1));
 779                 memmove(curp->p_idx + 1, curp->p_idx, len);
 780                 ix = curp->p_idx;
 781         }
 782
 783         ix->ei_block = cpu_to_le32(logical);
 784         ext4_idx_store_pblock(ix, ptr);
 785         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 786
 787         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 788                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 789                 return -EIO;
 790         }
 791
 792         err = ext4_ext_dirty(handle, inode, curp);
 793         ext4_std_error(inode->i_sb, err);
 794
 795         return err;
 796 }
 797
 798 /*
 799  * ext4_ext_split:
 800  * inserts new subtree into the path, using free index entry
 801  * at depth @at:
 802  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 803  * - makes decision where to split
 804  * - moves remaining extents and index entries (right to the split point)
 805  *   into the newly allocated blocks
 806  * - initializes subtree
 807  */
 808 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 809                           unsigned int flags,
 810                           struct ext4_ext_path *path,
 811                           struct ext4_extent *newext, int at)
 812 {
 813         struct buffer_head *bh = NULL;
 814         int depth = ext_depth(inode);
 815         struct ext4_extent_header *neh;
 816         struct ext4_extent_idx *fidx;
 817         int i = at, k, m, a;
 818         ext4_fsblk_t newblock, oldblock;
 819         __le32 border;
 820         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 821         int err = 0;
 822
 823         /* make decision: where to split? */
 824         /* FIXME: now decision is simplest: at current extent */
 825
 826         /* if current leaf will be split, then we should use
 827          * border from split point */
 828         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 829                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 830                 return -EIO;
 831         }
 832         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 833                 border = path[depth].p_ext[1].ee_block;
 834                 ext_debug("leaf will be split."
 835                                 " next leaf starts at %d\n",
 836                                   le32_to_cpu(border));
 837         } else {
 838                 border = newext->ee_block;
 839                 ext_debug("leaf will be added."
 840                                 " next leaf starts at %d\n",
 841                                 le32_to_cpu(border));
 842         }
 843
 844         /*
 845          * If error occurs, then we break processing
 846          * and mark filesystem read-only. index won't
 847          * be inserted and tree will be in consistent
 848          * state. Next mount will repair buffers too.
 849          */
 850
 851         /*
 852          * Get array to track all allocated blocks.
 853          * We need this to handle errors and free blocks
 854          * upon them.
 855          */
 856         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 857         if (!ablocks)
 858                 return -ENOMEM;
 859
 860         /* allocate all needed blocks */
 861         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 862         for (a = 0; a < depth - at; a++) {
 863                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 864                                                    newext, &err, flags);
 865                 if (newblock == 0)
 866                         goto cleanup;
 867                 ablocks[a] = newblock;
 868         }
 869
 870         /* initialize new leaf */
 871         newblock = ablocks[--a];
 872         if (unlikely(newblock == 0)) {
 873                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 874                 err = -EIO;
 875                 goto cleanup;
 876         }
 877         bh = sb_getblk(inode->i_sb, newblock);
 878         if (!bh) {
 879                 err = -EIO;
 880                 goto cleanup;
 881         }
 882         lock_buffer(bh);
 883
 884         err = ext4_journal_get_create_access(handle, bh);
 885         if (err)
 886                 goto cleanup;
 887
 888         neh = ext_block_hdr(bh);
 889         neh->eh_entries = 0;
 890         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 891         neh->eh_magic = EXT4_EXT_MAGIC;
 892         neh->eh_depth = 0;
 893
 894         /* move remainder of path[depth] to the new leaf */
 895         if (unlikely(path[depth].p_hdr->eh_entries !=
 896                      path[depth].p_hdr->eh_max)) {
 897                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 898                                  path[depth].p_hdr->eh_entries,
 899                                  path[depth].p_hdr->eh_max);
 900                 err = -EIO;
 901                 goto cleanup;
 902         }
 903         /* start copy from next extent */
 904         m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 905         ext4_ext_show_move(inode, path, newblock, depth);
 906         if (m) {
 907                 struct ext4_extent *ex;
 908                 ex = EXT_FIRST_EXTENT(neh);
 909                 memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 910                 le16_add_cpu(&neh->eh_entries, m);
 911         }
 912
 913         set_buffer_uptodate(bh);
 914         unlock_buffer(bh);
 915
 916         err = ext4_handle_dirty_metadata(handle, inode, bh);
 917         if (err)
 918                 goto cleanup;
 919         brelse(bh);
 920         bh = NULL;
 921
 922         /* correct old leaf */
 923         if (m) {
 924                 err = ext4_ext_get_access(handle, inode, path + depth);
 925                 if (err)
 926                         goto cleanup;
 927                 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 928                 err = ext4_ext_dirty(handle, inode, path + depth);
 929                 if (err)
 930                         goto cleanup;
 931
 932         }
 933
 934         /* create intermediate indexes */
 935         k = depth - at - 1;
 936         if (unlikely(k < 0)) {
 937                 EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 938                 err = -EIO;
 939                 goto cleanup;
 940         }
 941         if (k)
 942                 ext_debug("create %d intermediate indices\n", k);
 943         /* insert new index into current index block */
 944         /* current depth stored in i var */
 945         i = depth - 1;
 946         while (k--) {
 947                 oldblock = newblock;
 948                 newblock = ablocks[--a];
 949                 bh = sb_getblk(inode->i_sb, newblock);
 950                 if (!bh) {
 951                         err = -EIO;
 952                         goto cleanup;
 953                 }
 954                 lock_buffer(bh);
 955
 956                 err = ext4_journal_get_create_access(handle, bh);
 957                 if (err)
 958                         goto cleanup;
 959
 960                 neh = ext_block_hdr(bh);
 961                 neh->eh_entries = cpu_to_le16(1);
 962                 neh->eh_magic = EXT4_EXT_MAGIC;
 963                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 964                 neh->eh_depth = cpu_to_le16(depth - i);
 965                 fidx = EXT_FIRST_INDEX(neh);
 966                 fidx->ei_block = border;
 967                 ext4_idx_store_pblock(fidx, oldblock);
 968
 969                 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 970                                 i, newblock, le32_to_cpu(border), oldblock);
 971
 972                 /* move remainder of path[i] to the new index block */
 973                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 974                                         EXT_LAST_INDEX(path[i].p_hdr))) {
 975                         EXT4_ERROR_INODE(inode,
 976                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 977                                          le32_to_cpu(path[i].p_ext->ee_block));
 978                         err = -EIO;
 979                         goto cleanup;
 980                 }
 981                 /* start copy indexes */
 982                 m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
 983                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 984                                 EXT_MAX_INDEX(path[i].p_hdr));
 985                 ext4_ext_show_move(inode, path, newblock, i);
 986                 if (m) {
 987                         memmove(++fidx, path[i].p_idx,
 988                                 sizeof(struct ext4_extent_idx) * m);
 989                         le16_add_cpu(&neh->eh_entries, m);
 990                 }
 991                 set_buffer_uptodate(bh);
 992                 unlock_buffer(bh);
 993
 994                 err = ext4_handle_dirty_metadata(handle, inode, bh);
 995                 if (err)
 996                         goto cleanup;
 997                 brelse(bh);
 998                 bh = NULL;
 999
1000                 /* correct old index */
1001                 if (m) {
1002                         err = ext4_ext_get_access(handle, inode, path + i);
1003                         if (err)
1004                                 goto cleanup;
1005                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1006                         err = ext4_ext_dirty(handle, inode, path + i);
1007                         if (err)
1008                                 goto cleanup;
1009                 }
1010
1011                 i--;
1012         }
1013
1014         /* insert new index */
1015         err = ext4_ext_insert_index(handle, inode, path + at,
1016                                     le32_to_cpu(border), newblock);
1017
1018 cleanup:
1019         if (bh) {
1020                 if (buffer_locked(bh))
1021                         unlock_buffer(bh);
1022                 brelse(bh);
1023         }
1024
1025         if (err) {
1026                 /* free all allocated blocks in error case */
1027                 for (i = 0; i < depth; i++) {
1028                         if (!ablocks[i])
1029                                 continue;
1030                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1031                                          EXT4_FREE_BLOCKS_METADATA);
1032                 }
1033         }
1034         kfree(ablocks);
1035
1036         return err;
1037 }
1038
1039 /*
1040  * ext4_ext_grow_indepth:
1041  * implements tree growing procedure:
1042  * - allocates new block
1043  * - moves top-level data (index block or leaf) into the new block
1044  * - initializes new top-level, creating index that points to the
1045  *   just created block
1046  */
1047 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1048                                  unsigned int flags,
1049                                  struct ext4_ext_path *path,
1050                                  struct ext4_extent *newext)
1051 {
1052         struct ext4_ext_path *curp = path;
1053         struct ext4_extent_header *neh;
1054         struct buffer_head *bh;
1055         ext4_fsblk_t newblock;
1056         int err = 0;
1057
1058         newblock = ext4_ext_new_meta_block(handle, inode, path,
1059                 newext, &err, flags);
1060         if (newblock == 0)
1061                 return err;
1062
1063         bh = sb_getblk(inode->i_sb, newblock);
1064         if (!bh) {
1065                 err = -EIO;
1066                 ext4_std_error(inode->i_sb, err);
1067                 return err;
1068         }
1069         lock_buffer(bh);
1070
1071         err = ext4_journal_get_create_access(handle, bh);
1072         if (err) {
1073                 unlock_buffer(bh);
1074                 goto out;
1075         }
1076
1077         /* move top-level index/leaf into new block */
1078         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1079
1080         /* set size of new block */
1081         neh = ext_block_hdr(bh);
1082         /* old root could have indexes or leaves
1083          * so calculate e_max right way */
1084         if (ext_depth(inode))
1085                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1086         else
1087                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1088         neh->eh_magic = EXT4_EXT_MAGIC;
1089         set_buffer_uptodate(bh);
1090         unlock_buffer(bh);
1091
1092         err = ext4_handle_dirty_metadata(handle, inode, bh);
1093         if (err)
1094                 goto out;
1095
1096         /* create index in new top-level index: num,max,pointer */
1097         err = ext4_ext_get_access(handle, inode, curp);
1098         if (err)
1099                 goto out;
1100
1101         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1102         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1103         curp->p_hdr->eh_entries = cpu_to_le16(1);
1104         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1105
1106         if (path[0].p_hdr->eh_depth)
1107                 curp->p_idx->ei_block =
1108                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1109         else
1110                 curp->p_idx->ei_block =
1111                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1112         ext4_idx_store_pblock(curp->p_idx, newblock);
1113
1114         neh = ext_inode_hdr(inode);
1115         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1116                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1117                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1118                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1119
1120         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1121         err = ext4_ext_dirty(handle, inode, curp);
1122 out:
1123         brelse(bh);
1124
1125         return err;
1126 }
1127
1128 /*
1129  * ext4_ext_create_new_leaf:
1130  * finds empty index and adds new leaf.
1131  * if no free index is found, then it requests in-depth growing.
1132  */
1133 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1134                                     unsigned int flags,
1135                                     struct ext4_ext_path *path,
1136                                     struct ext4_extent *newext)
1137 {
1138         struct ext4_ext_path *curp;
1139         int depth, i, err = 0;
1140
1141 repeat:
1142         i = depth = ext_depth(inode);
1143
1144         /* walk up to the tree and look for free index entry */
1145         curp = path + depth;
1146         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1147                 i--;
1148                 curp--;
1149         }
1150
1151         /* we use already allocated block for index block,
1152          * so subsequent data blocks should be contiguous */
1153         if (EXT_HAS_FREE_INDEX(curp)) {
1154                 /* if we found index with free entry, then use that
1155                  * entry: create all needed subtree and add new leaf */
1156                 err = ext4_ext_split(handle, inode, flags, path, newext, i);
1157                 if (err)
1158                         goto out;
1159
1160                 /* refill path */
1161                 ext4_ext_drop_refs(path);
1162                 path = ext4_ext_find_extent(inode,
1163                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1164                                     path);
1165                 if (IS_ERR(path))
1166                         err = PTR_ERR(path);
1167         } else {
1168                 /* tree is full, time to grow in depth */
1169                 err = ext4_ext_grow_indepth(handle, inode, flags,
1170                                             path, newext);
1171                 if (err)
1172                         goto out;
1173
1174                 /* refill path */
1175                 ext4_ext_drop_refs(path);
1176                 path = ext4_ext_find_extent(inode,
1177                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1178                                     path);
1179                 if (IS_ERR(path)) {
1180                         err = PTR_ERR(path);
1181                         goto out;
1182                 }
1183
1184                 /*
1185                  * only first (depth 0 -> 1) produces free space;
1186                  * in all other cases we have to split the grown tree
1187                  */
1188                 depth = ext_depth(inode);
1189                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1190                         /* now we need to split */
1191                         goto repeat;
1192                 }
1193         }
1194
1195 out:
1196         return err;
1197 }
1198
1199 /*
1200  * search the closest allocated block to the left for *logical
1201  * and returns it at @logical + it's physical address at @phys
1202  * if *logical is the smallest allocated block, the function
1203  * returns 0 at @phys
1204  * return value contains 0 (success) or error code
1205  */
1206 static int ext4_ext_search_left(struct inode *inode,
1207                                 struct ext4_ext_path *path,
1208                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1209 {
1210         struct ext4_extent_idx *ix;
1211         struct ext4_extent *ex;
1212         int depth, ee_len;
1213
1214         if (unlikely(path == NULL)) {
1215                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1216                 return -EIO;
1217         }
1218         depth = path->p_depth;
1219         *phys = 0;
1220
1221         if (depth == 0 && path->p_ext == NULL)
1222                 return 0;
1223
1224         /* usually extent in the path covers blocks smaller
1225          * then *logical, but it can be that extent is the
1226          * first one in the file */
1227
1228         ex = path[depth].p_ext;
1229         ee_len = ext4_ext_get_actual_len(ex);
1230         if (*logical < le32_to_cpu(ex->ee_block)) {
1231                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1232                         EXT4_ERROR_INODE(inode,
1233                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1234                                          *logical, le32_to_cpu(ex->ee_block));
1235                         return -EIO;
1236                 }
1237                 while (--depth >= 0) {
1238                         ix = path[depth].p_idx;
1239                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1240                                 EXT4_ERROR_INODE(inode,
1241                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1242                                   ix != NULL ? ix->ei_block : 0,
1243                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1244                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1245                                   depth);
1246                                 return -EIO;
1247                         }
1248                 }
1249                 return 0;
1250         }
1251
1252         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1253                 EXT4_ERROR_INODE(inode,
1254                                  "logical %d < ee_block %d + ee_len %d!",
1255                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1256                 return -EIO;
1257         }
1258
1259         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1260         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1261         return 0;
1262 }
1263
1264 /*
1265  * search the closest allocated block to the right for *logical
1266  * and returns it at @logical + it's physical address at @phys
1267  * if *logical is the smallest allocated block, the function
1268  * returns 0 at @phys
1269  * return value contains 0 (success) or error code
1270  */
1271 static int ext4_ext_search_right(struct inode *inode,
1272                                  struct ext4_ext_path *path,
1273                                  ext4_lblk_t *logical, ext4_fsblk_t *phys,
1274                                  struct ext4_extent **ret_ex)
1275 {
1276         struct buffer_head *bh = NULL;
1277         struct ext4_extent_header *eh;
1278         struct ext4_extent_idx *ix;
1279         struct ext4_extent *ex;
1280         ext4_fsblk_t block;
1281         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1282         int ee_len;
1283
1284         if (unlikely(path == NULL)) {
1285                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1286                 return -EIO;
1287         }
1288         depth = path->p_depth;
1289         *phys = 0;
1290
1291         if (depth == 0 && path->p_ext == NULL)
1292                 return 0;
1293
1294         /* usually extent in the path covers blocks smaller
1295          * then *logical, but it can be that extent is the
1296          * first one in the file */
1297
1298         ex = path[depth].p_ext;
1299         ee_len = ext4_ext_get_actual_len(ex);
1300         if (*logical < le32_to_cpu(ex->ee_block)) {
1301                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1302                         EXT4_ERROR_INODE(inode,
1303                                          "first_extent(path[%d].p_hdr) != ex",
1304                                          depth);
1305                         return -EIO;
1306                 }
1307                 while (--depth >= 0) {
1308                         ix = path[depth].p_idx;
1309                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1310                                 EXT4_ERROR_INODE(inode,
1311                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1312                                                  *logical);
1313                                 return -EIO;
1314                         }
1315                 }
1316                 goto found_extent;
1317         }
1318
1319         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1320                 EXT4_ERROR_INODE(inode,
1321                                  "logical %d < ee_block %d + ee_len %d!",
1322                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1323                 return -EIO;
1324         }
1325
1326         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1327                 /* next allocated block in this leaf */
1328                 ex++;
1329                 goto found_extent;
1330         }
1331
1332         /* go up and search for index to the right */
1333         while (--depth >= 0) {
1334                 ix = path[depth].p_idx;
1335                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1336                         goto got_index;
1337         }
1338
1339         /* we've gone up to the root and found no index to the right */
1340         return 0;
1341
1342 got_index:
1343         /* we've found index to the right, let's
1344          * follow it and find the closest allocated
1345          * block to the right */
1346         ix++;
1347         block = ext4_idx_pblock(ix);
1348         while (++depth < path->p_depth) {
1349                 bh = sb_bread(inode->i_sb, block);
1350                 if (bh == NULL)
1351                         return -EIO;
1352                 eh = ext_block_hdr(bh);
1353                 /* subtract from p_depth to get proper eh_depth */
1354                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1355                         put_bh(bh);
1356                         return -EIO;
1357                 }
1358                 ix = EXT_FIRST_INDEX(eh);
1359                 block = ext4_idx_pblock(ix);
1360                 put_bh(bh);
1361         }
1362
1363         bh = sb_bread(inode->i_sb, block);
1364         if (bh == NULL)
1365                 return -EIO;
1366         eh = ext_block_hdr(bh);
1367         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1368                 put_bh(bh);
1369                 return -EIO;
1370         }
1371         ex = EXT_FIRST_EXTENT(eh);
1372 found_extent:
1373         *logical = le32_to_cpu(ex->ee_block);
1374         *phys = ext4_ext_pblock(ex);
1375         *ret_ex = ex;
1376         if (bh)
1377                 put_bh(bh);
1378         return 0;
1379 }
1380
1381 /*
1382  * ext4_ext_next_allocated_block:
1383  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1384  * NOTE: it considers block number from index entry as
1385  * allocated block. Thus, index entries have to be consistent
1386  * with leaves.
1387  */
1388 static ext4_lblk_t
1389 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1390 {
1391         int depth;
1392
1393         BUG_ON(path == NULL);
1394         depth = path->p_depth;
1395
1396         if (depth == 0 && path->p_ext == NULL)
1397                 return EXT_MAX_BLOCKS;
1398
1399         while (depth >= 0) {
1400                 if (depth == path->p_depth) {
1401                         /* leaf */
1402                         if (path[depth].p_ext !=
1403                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1404                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1405                 } else {
1406                         /* index */
1407                         if (path[depth].p_idx !=
1408                                         EXT_LAST_INDEX(path[depth].p_hdr))
1409                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1410                 }
1411                 depth--;
1412         }
1413
1414         return EXT_MAX_BLOCKS;
1415 }
1416
1417 /*
1418  * ext4_ext_next_leaf_block:
1419  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1420  */
1421 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1422 {
1423         int depth;
1424
1425         BUG_ON(path == NULL);
1426         depth = path->p_depth;
1427
1428         /* zero-tree has no leaf blocks at all */
1429         if (depth == 0)
1430                 return EXT_MAX_BLOCKS;
1431
1432         /* go to index block */
1433         depth--;
1434
1435         while (depth >= 0) {
1436                 if (path[depth].p_idx !=
1437                                 EXT_LAST_INDEX(path[depth].p_hdr))
1438                         return (ext4_lblk_t)
1439                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1440                 depth--;
1441         }
1442
1443         return EXT_MAX_BLOCKS;
1444 }
1445
1446 /*
1447  * ext4_ext_correct_indexes:
1448  * if leaf gets modified and modified extent is first in the leaf,
1449  * then we have to correct all indexes above.
1450  * TODO: do we need to correct tree in all cases?
1451  */
1452 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1453                                 struct ext4_ext_path *path)
1454 {
1455         struct ext4_extent_header *eh;
1456         int depth = ext_depth(inode);
1457         struct ext4_extent *ex;
1458         __le32 border;
1459         int k, err = 0;
1460
1461         eh = path[depth].p_hdr;
1462         ex = path[depth].p_ext;
1463
1464         if (unlikely(ex == NULL || eh == NULL)) {
1465                 EXT4_ERROR_INODE(inode,
1466                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1467                 return -EIO;
1468         }
1469
1470         if (depth == 0) {
1471                 /* there is no tree at all */
1472                 return 0;
1473         }
1474
1475         if (ex != EXT_FIRST_EXTENT(eh)) {
1476                 /* we correct tree if first leaf got modified only */
1477                 return 0;
1478         }
1479
1480         /*
1481          * TODO: we need correction if border is smaller than current one
1482          */
1483         k = depth - 1;
1484         border = path[depth].p_ext->ee_block;
1485         err = ext4_ext_get_access(handle, inode, path + k);
1486         if (err)
1487                 return err;
1488         path[k].p_idx->ei_block = border;
1489         err = ext4_ext_dirty(handle, inode, path + k);
1490         if (err)
1491                 return err;
1492
1493         while (k--) {
1494                 /* change all left-side indexes */
1495                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1496                         break;
1497                 err = ext4_ext_get_access(handle, inode, path + k);
1498                 if (err)
1499                         break;
1500                 path[k].p_idx->ei_block = border;
1501                 err = ext4_ext_dirty(handle, inode, path + k);
1502                 if (err)
1503                         break;
1504         }
1505
1506         return err;
1507 }
1508
1509 int
1510 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1511                                 struct ext4_extent *ex2)
1512 {
1513         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1514
1515         /*
1516          * Make sure that either both extents are uninitialized, or
1517          * both are _not_.
1518          */
1519         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1520                 return 0;
1521
1522         if (ext4_ext_is_uninitialized(ex1))
1523                 max_len = EXT_UNINIT_MAX_LEN;
1524         else
1525                 max_len = EXT_INIT_MAX_LEN;
1526
1527         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1528         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1529
1530         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1531                         le32_to_cpu(ex2->ee_block))
1532                 return 0;
1533
1534         /*
1535          * To allow future support for preallocated extents to be added
1536          * as an RO_COMPAT feature, refuse to merge to extents if
1537          * this can result in the top bit of ee_len being set.
1538          */
1539         if (ext1_ee_len + ext2_ee_len > max_len)
1540                 return 0;
1541 #ifdef AGGRESSIVE_TEST
1542         if (ext1_ee_len >= 4)
1543                 return 0;
1544 #endif
1545
1546         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1547                 return 1;
1548         return 0;
1549 }
1550
1551 /*
1552  * This function tries to merge the "ex" extent to the next extent in the tree.
1553  * It always tries to merge towards right. If you want to merge towards
1554  * left, pass "ex - 1" as argument instead of "ex".
1555  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1556  * 1 if they got merged.
1557  */
1558 static int ext4_ext_try_to_merge_right(struct inode *inode,
1559                                  struct ext4_ext_path *path,
1560                                  struct ext4_extent *ex)
1561 {
1562         struct ext4_extent_header *eh;
1563         unsigned int depth, len;
1564         int merge_done = 0;
1565         int uninitialized = 0;
1566
1567         depth = ext_depth(inode);
1568         BUG_ON(path[depth].p_hdr == NULL);
1569         eh = path[depth].p_hdr;
1570
1571         while (ex < EXT_LAST_EXTENT(eh)) {
1572                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1573                         break;
1574                 /* merge with next extent! */
1575                 if (ext4_ext_is_uninitialized(ex))
1576                         uninitialized = 1;
1577                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1578                                 + ext4_ext_get_actual_len(ex + 1));
1579                 if (uninitialized)
1580                         ext4_ext_mark_uninitialized(ex);
1581
1582                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1583                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1584                                 * sizeof(struct ext4_extent);
1585                         memmove(ex + 1, ex + 2, len);
1586                 }
1587                 le16_add_cpu(&eh->eh_entries, -1);
1588                 merge_done = 1;
1589                 WARN_ON(eh->eh_entries == 0);
1590                 if (!eh->eh_entries)
1591                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1592         }
1593
1594         return merge_done;
1595 }
1596
1597 /*
1598  * This function tries to merge the @ex extent to neighbours in the tree.
1599  * return 1 if merge left else 0.
1600  */
1601 static int ext4_ext_try_to_merge(struct inode *inode,
1602                                   struct ext4_ext_path *path,
1603                                   struct ext4_extent *ex) {
1604         struct ext4_extent_header *eh;
1605         unsigned int depth;
1606         int merge_done = 0;
1607         int ret = 0;
1608
1609         depth = ext_depth(inode);
1610         BUG_ON(path[depth].p_hdr == NULL);
1611         eh = path[depth].p_hdr;
1612
1613         if (ex > EXT_FIRST_EXTENT(eh))
1614                 merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1615
1616         if (!merge_done)
1617                 ret = ext4_ext_try_to_merge_right(inode, path, ex);
1618
1619         return ret;
1620 }
1621
1622 /*
1623  * check if a portion of the "newext" extent overlaps with an
1624  * existing extent.
1625  *
1626  * If there is an overlap discovered, it updates the length of the newext
1627  * such that there will be no overlap, and then returns 1.
1628  * If there is no overlap found, it returns 0.
1629  */
1630 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1631                                            struct inode *inode,
1632                                            struct ext4_extent *newext,
1633                                            struct ext4_ext_path *path)
1634 {
1635         ext4_lblk_t b1, b2;
1636         unsigned int depth, len1;
1637         unsigned int ret = 0;
1638
1639         b1 = le32_to_cpu(newext->ee_block);
1640         len1 = ext4_ext_get_actual_len(newext);
1641         depth = ext_depth(inode);
1642         if (!path[depth].p_ext)
1643                 goto out;
1644         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1645         b2 &= ~(sbi->s_cluster_ratio - 1);
1646
1647         /*
1648          * get the next allocated block if the extent in the path
1649          * is before the requested block(s)
1650          */
1651         if (b2 < b1) {
1652                 b2 = ext4_ext_next_allocated_block(path);
1653                 if (b2 == EXT_MAX_BLOCKS)
1654                         goto out;
1655                 b2 &= ~(sbi->s_cluster_ratio - 1);
1656         }
1657
1658         /* check for wrap through zero on extent logical start block*/
1659         if (b1 + len1 < b1) {
1660                 len1 = EXT_MAX_BLOCKS - b1;
1661                 newext->ee_len = cpu_to_le16(len1);
1662                 ret = 1;
1663         }
1664
1665         /* check for overlap */
1666         if (b1 + len1 > b2) {
1667                 newext->ee_len = cpu_to_le16(b2 - b1);
1668                 ret = 1;
1669         }
1670 out:
1671         return ret;
1672 }
1673
1674 /*
1675  * ext4_ext_insert_extent:
1676  * tries to merge requsted extent into the existing extent or
1677  * inserts requested extent as new one into the tree,
1678  * creating new leaf in the no-space case.
1679  */
1680 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1681                                 struct ext4_ext_path *path,
1682                                 struct ext4_extent *newext, int flag)
1683 {
1684         struct ext4_extent_header *eh;
1685         struct ext4_extent *ex, *fex;
1686         struct ext4_extent *nearex; /* nearest extent */
1687         struct ext4_ext_path *npath = NULL;
1688         int depth, len, err;
1689         ext4_lblk_t next;
1690         unsigned uninitialized = 0;
1691         int flags = 0;
1692
1693         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1694                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1695                 return -EIO;
1696         }
1697         depth = ext_depth(inode);
1698         ex = path[depth].p_ext;
1699         if (unlikely(path[depth].p_hdr == NULL)) {
1700                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1701                 return -EIO;
1702         }
1703
1704         /* try to insert block into found extent and return */
1705         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1706                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1707                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1708                           ext4_ext_is_uninitialized(newext),
1709                           ext4_ext_get_actual_len(newext),
1710                           le32_to_cpu(ex->ee_block),
1711                           ext4_ext_is_uninitialized(ex),
1712                           ext4_ext_get_actual_len(ex),
1713                           ext4_ext_pblock(ex));
1714                 err = ext4_ext_get_access(handle, inode, path + depth);
1715                 if (err)
1716                         return err;
1717
1718                 /*
1719                  * ext4_can_extents_be_merged should have checked that either
1720                  * both extents are uninitialized, or both aren't. Thus we
1721                  * need to check only one of them here.
1722                  */
1723                 if (ext4_ext_is_uninitialized(ex))
1724                         uninitialized = 1;
1725                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1726                                         + ext4_ext_get_actual_len(newext));
1727                 if (uninitialized)
1728                         ext4_ext_mark_uninitialized(ex);
1729                 eh = path[depth].p_hdr;
1730                 nearex = ex;
1731                 goto merge;
1732         }
1733
1734         depth = ext_depth(inode);
1735         eh = path[depth].p_hdr;
1736         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1737                 goto has_space;
1738
1739         /* probably next leaf has space for us? */
1740         fex = EXT_LAST_EXTENT(eh);
1741         next = EXT_MAX_BLOCKS;
1742         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1743                 next = ext4_ext_next_leaf_block(path);
1744         if (next != EXT_MAX_BLOCKS) {
1745                 ext_debug("next leaf block - %d\n", next);
1746                 BUG_ON(npath != NULL);
1747                 npath = ext4_ext_find_extent(inode, next, NULL);
1748                 if (IS_ERR(npath))
1749                         return PTR_ERR(npath);
1750                 BUG_ON(npath->p_depth != path->p_depth);
1751                 eh = npath[depth].p_hdr;
1752                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1753                         ext_debug("next leaf isn't full(%d)\n",
1754                                   le16_to_cpu(eh->eh_entries));
1755                         path = npath;
1756                         goto has_space;
1757                 }
1758                 ext_debug("next leaf has no free space(%d,%d)\n",
1759                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1760         }
1761
1762         /*
1763          * There is no free space in the found leaf.
1764          * We're gonna add a new leaf in the tree.
1765          */
1766         if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1767                 flags = EXT4_MB_USE_ROOT_BLOCKS;
1768         err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1769         if (err)
1770                 goto cleanup;
1771         depth = ext_depth(inode);
1772         eh = path[depth].p_hdr;
1773
1774 has_space:
1775         nearex = path[depth].p_ext;
1776
1777         err = ext4_ext_get_access(handle, inode, path + depth);
1778         if (err)
1779                 goto cleanup;
1780
1781         if (!nearex) {
1782                 /* there is no extent in this leaf, create first one */
1783                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1784                                 le32_to_cpu(newext->ee_block),
1785                                 ext4_ext_pblock(newext),
1786                                 ext4_ext_is_uninitialized(newext),
1787                                 ext4_ext_get_actual_len(newext));
1788                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1789         } else if (le32_to_cpu(newext->ee_block)
1790                            > le32_to_cpu(nearex->ee_block)) {
1791 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1792                 if (nearex != EXT_LAST_EXTENT(eh)) {
1793                         len = EXT_MAX_EXTENT(eh) - nearex;
1794                         len = (len - 1) * sizeof(struct ext4_extent);
1795                         len = len < 0 ? 0 : len;
1796                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1797                                         "move %d from 0x%p to 0x%p\n",
1798                                         le32_to_cpu(newext->ee_block),
1799                                         ext4_ext_pblock(newext),
1800                                         ext4_ext_is_uninitialized(newext),
1801                                         ext4_ext_get_actual_len(newext),
1802                                         nearex, len, nearex + 1, nearex + 2);
1803                         memmove(nearex + 2, nearex + 1, len);
1804                 }
1805                 path[depth].p_ext = nearex + 1;
1806         } else {
1807                 BUG_ON(newext->ee_block == nearex->ee_block);
1808                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1809                 len = len < 0 ? 0 : len;
1810                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1811                                 "move %d from 0x%p to 0x%p\n",
1812                                 le32_to_cpu(newext->ee_block),
1813                                 ext4_ext_pblock(newext),
1814                                 ext4_ext_is_uninitialized(newext),
1815                                 ext4_ext_get_actual_len(newext),
1816                                 nearex, len, nearex, nearex + 1);
1817                 memmove(nearex + 1, nearex, len);
1818                 path[depth].p_ext = nearex;
1819         }
1820
1821         le16_add_cpu(&eh->eh_entries, 1);
1822         nearex = path[depth].p_ext;
1823         nearex->ee_block = newext->ee_block;
1824         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1825         nearex->ee_len = newext->ee_len;
1826
1827 merge:
1828         /* try to merge extents to the right */
1829         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1830                 ext4_ext_try_to_merge(inode, path, nearex);
1831
1832         /* try to merge extents to the left */
1833
1834         /* time to correct all indexes above */
1835         err = ext4_ext_correct_indexes(handle, inode, path);
1836         if (err)
1837                 goto cleanup;
1838
1839         err = ext4_ext_dirty(handle, inode, path + depth);
1840
1841 cleanup:
1842         if (npath) {
1843                 ext4_ext_drop_refs(npath);
1844                 kfree(npath);
1845         }
1846         ext4_ext_invalidate_cache(inode);
1847         return err;
1848 }
1849
1850 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1851                                ext4_lblk_t num, ext_prepare_callback func,
1852                                void *cbdata)
1853 {
1854         struct ext4_ext_path *path = NULL;
1855         struct ext4_ext_cache cbex;
1856         struct ext4_extent *ex;
1857         ext4_lblk_t next, start = 0, end = 0;
1858         ext4_lblk_t last = block + num;
1859         int depth, exists, err = 0;
1860
1861         BUG_ON(func == NULL);
1862         BUG_ON(inode == NULL);
1863
1864         while (block < last && block != EXT_MAX_BLOCKS) {
1865                 num = last - block;
1866                 /* find extent for this block */
1867                 down_read(&EXT4_I(inode)->i_data_sem);
1868                 path = ext4_ext_find_extent(inode, block, path);
1869                 up_read(&EXT4_I(inode)->i_data_sem);
1870                 if (IS_ERR(path)) {
1871                         err = PTR_ERR(path);
1872                         path = NULL;
1873                         break;
1874                 }
1875
1876                 depth = ext_depth(inode);
1877                 if (unlikely(path[depth].p_hdr == NULL)) {
1878                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1879                         err = -EIO;
1880                         break;
1881                 }
1882                 ex = path[depth].p_ext;
1883                 next = ext4_ext_next_allocated_block(path);
1884
1885                 exists = 0;
1886                 if (!ex) {
1887                         /* there is no extent yet, so try to allocate
1888                          * all requested space */
1889                         start = block;
1890                         end = block + num;
1891                 } else if (le32_to_cpu(ex->ee_block) > block) {
1892                         /* need to allocate space before found extent */
1893                         start = block;
1894                         end = le32_to_cpu(ex->ee_block);
1895                         if (block + num < end)
1896                                 end = block + num;
1897                 } else if (block >= le32_to_cpu(ex->ee_block)
1898                                         + ext4_ext_get_actual_len(ex)) {
1899                         /* need to allocate space after found extent */
1900                         start = block;
1901                         end = block + num;
1902                         if (end >= next)
1903                                 end = next;
1904                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1905                         /*
1906                          * some part of requested space is covered
1907                          * by found extent
1908                          */
1909                         start = block;
1910                         end = le32_to_cpu(ex->ee_block)
1911                                 + ext4_ext_get_actual_len(ex);
1912                         if (block + num < end)
1913                                 end = block + num;
1914                         exists = 1;
1915                 } else {
1916                         BUG();
1917                 }
1918                 BUG_ON(end <= start);
1919
1920                 if (!exists) {
1921                         cbex.ec_block = start;
1922                         cbex.ec_len = end - start;
1923                         cbex.ec_start = 0;
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 = ext4_ext_pblock(ex);
1928                 }
1929
1930                 if (unlikely(cbex.ec_len == 0)) {
1931                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1932                         err = -EIO;
1933                         break;
1934                 }
1935                 err = func(inode, next, &cbex, ex, cbdata);
1936                 ext4_ext_drop_refs(path);
1937
1938                 if (err < 0)
1939                         break;
1940
1941                 if (err == EXT_REPEAT)
1942                         continue;
1943                 else if (err == EXT_BREAK) {
1944                         err = 0;
1945                         break;
1946                 }
1947
1948                 if (ext_depth(inode) != depth) {
1949                         /* depth was changed. we have to realloc path */
1950                         kfree(path);
1951                         path = NULL;
1952                 }
1953
1954                 block = cbex.ec_block + cbex.ec_len;
1955         }
1956
1957         if (path) {
1958                 ext4_ext_drop_refs(path);
1959                 kfree(path);
1960         }
1961
1962         return err;
1963 }
1964
1965 static void
1966 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1967                         __u32 len, ext4_fsblk_t start)
1968 {
1969         struct ext4_ext_cache *cex;
1970         BUG_ON(len == 0);
1971         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1972         cex = &EXT4_I(inode)->i_cached_extent;
1973         cex->ec_block = block;
1974         cex->ec_len = len;
1975         cex->ec_start = start;
1976         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1977 }
1978
1979 /*
1980  * ext4_ext_put_gap_in_cache:
1981  * calculate boundaries of the gap that the requested block fits into
1982  * and cache this gap
1983  */
1984 static void
1985 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1986                                 ext4_lblk_t block)
1987 {
1988         int depth = ext_depth(inode);
1989         unsigned long len;
1990         ext4_lblk_t lblock;
1991         struct ext4_extent *ex;
1992
1993         ex = path[depth].p_ext;
1994         if (ex == NULL) {
1995                 /* there is no extent yet, so gap is [0;-] */
1996                 lblock = 0;
1997                 len = EXT_MAX_BLOCKS;
1998                 ext_debug("cache gap(whole file):");
1999         } else if (block < le32_to_cpu(ex->ee_block)) {
2000                 lblock = block;
2001                 len = le32_to_cpu(ex->ee_block) - block;
2002                 ext_debug("cache gap(before): %u [%u:%u]",
2003                                 block,
2004                                 le32_to_cpu(ex->ee_block),
2005                                  ext4_ext_get_actual_len(ex));
2006         } else if (block >= le32_to_cpu(ex->ee_block)
2007                         + ext4_ext_get_actual_len(ex)) {
2008                 ext4_lblk_t next;
2009                 lblock = le32_to_cpu(ex->ee_block)
2010                         + ext4_ext_get_actual_len(ex);
2011
2012                 next = ext4_ext_next_allocated_block(path);
2013                 ext_debug("cache gap(after): [%u:%u] %u",
2014                                 le32_to_cpu(ex->ee_block),
2015                                 ext4_ext_get_actual_len(ex),
2016                                 block);
2017                 BUG_ON(next == lblock);
2018                 len = next - lblock;
2019         } else {
2020                 lblock = len = 0;
2021                 BUG();
2022         }
2023
2024         ext_debug(" -> %u:%lu\n", lblock, len);
2025         ext4_ext_put_in_cache(inode, lblock, len, 0);
2026 }
2027
2028 /*
2029  * ext4_ext_check_cache()
2030  * Checks to see if the given block is in the cache.
2031  * If it is, the cached extent is stored in the given
2032  * cache extent pointer.  If the cached extent is a hole,
2033  * this routine should be used instead of
2034  * ext4_ext_in_cache if the calling function needs to
2035  * know the size of the hole.
2036  *
2037  * @inode: The files inode
2038  * @block: The block to look for in the cache
2039  * @ex:    Pointer where the cached extent will be stored
2040  *         if it contains block
2041  *
2042  * Return 0 if cache is invalid; 1 if the cache is valid
2043  */
2044 static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2045         struct ext4_ext_cache *ex){
2046         struct ext4_ext_cache *cex;
2047         struct ext4_sb_info *sbi;
2048         int ret = 0;
2049
2050         /*
2051          * We borrow i_block_reservation_lock to protect i_cached_extent
2052          */
2053         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2054         cex = &EXT4_I(inode)->i_cached_extent;
2055         sbi = EXT4_SB(inode->i_sb);
2056
2057         /* has cache valid data? */
2058         if (cex->ec_len == 0)
2059                 goto errout;
2060
2061         if (in_range(block, cex->ec_block, cex->ec_len)) {
2062                 memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2063                 ext_debug("%u cached by %u:%u:%llu\n",
2064                                 block,
2065                                 cex->ec_block, cex->ec_len, cex->ec_start);
2066                 ret = 1;
2067         }
2068 errout:
2069         if (!ret)
2070                 sbi->extent_cache_misses++;
2071         else
2072                 sbi->extent_cache_hits++;
2073         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2074         return ret;
2075 }
2076
2077 /*
2078  * ext4_ext_in_cache()
2079  * Checks to see if the given block is in the cache.
2080  * If it is, the cached extent is stored in the given
2081  * extent pointer.
2082  *
2083  * @inode: The files inode
2084  * @block: The block to look for in the cache
2085  * @ex:    Pointer where the cached extent will be stored
2086  *         if it contains block
2087  *
2088  * Return 0 if cache is invalid; 1 if the cache is valid
2089  */
2090 static int
2091 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2092                         struct ext4_extent *ex)
2093 {
2094         struct ext4_ext_cache cex;
2095         int ret = 0;
2096
2097         if (ext4_ext_check_cache(inode, block, &cex)) {
2098                 ex->ee_block = cpu_to_le32(cex.ec_block);
2099                 ext4_ext_store_pblock(ex, cex.ec_start);
2100                 ex->ee_len = cpu_to_le16(cex.ec_len);
2101                 ret = 1;
2102         }
2103
2104         return ret;
2105 }
2106
2107
2108 /*
2109  * ext4_ext_rm_idx:
2110  * removes index from the index block.
2111  */
2112 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2113                         struct ext4_ext_path *path)
2114 {
2115         int err;
2116         ext4_fsblk_t leaf;
2117
2118         /* free index block */
2119         path--;
2120         leaf = ext4_idx_pblock(path->p_idx);
2121         if (unlikely(path->p_hdr->eh_entries == 0)) {
2122                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2123                 return -EIO;
2124         }
2125         err = ext4_ext_get_access(handle, inode, path);
2126         if (err)
2127                 return err;
2128
2129         if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2130                 int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2131                 len *= sizeof(struct ext4_extent_idx);
2132                 memmove(path->p_idx, path->p_idx + 1, len);
2133         }
2134
2135         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2136         err = ext4_ext_dirty(handle, inode, path);
2137         if (err)
2138                 return err;
2139         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2140         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2141                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2142         return err;
2143 }
2144
2145 /*
2146  * ext4_ext_calc_credits_for_single_extent:
2147  * This routine returns max. credits that needed to insert an extent
2148  * to the extent tree.
2149  * When pass the actual path, the caller should calculate credits
2150  * under i_data_sem.
2151  */
2152 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2153                                                 struct ext4_ext_path *path)
2154 {
2155         if (path) {
2156                 int depth = ext_depth(inode);
2157                 int ret = 0;
2158
2159                 /* probably there is space in leaf? */
2160                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2161                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2162
2163                         /*
2164                          *  There are some space in the leaf tree, no
2165                          *  need to account for leaf block credit
2166                          *
2167                          *  bitmaps and block group descriptor blocks
2168                          *  and other metadat blocks still need to be
2169                          *  accounted.
2170                          */
2171                         /* 1 bitmap, 1 block group descriptor */
2172                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2173                         return ret;
2174                 }
2175         }
2176
2177         return ext4_chunk_trans_blocks(inode, nrblocks);
2178 }
2179
2180 /*
2181  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2182  *
2183  * if nrblocks are fit in a single extent (chunk flag is 1), then
2184  * in the worse case, each tree level index/leaf need to be changed
2185  * if the tree split due to insert a new extent, then the old tree
2186  * index/leaf need to be updated too
2187  *
2188  * If the nrblocks are discontiguous, they could cause
2189  * the whole tree split more than once, but this is really rare.
2190  */
2191 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2192 {
2193         int index;
2194         int depth = ext_depth(inode);
2195
2196         if (chunk)
2197                 index = depth * 2;
2198         else
2199                 index = depth * 3;
2200
2201         return index;
2202 }
2203
2204 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2205                                 struct ext4_extent *ex,
2206                                 ext4_lblk_t from, ext4_lblk_t to)
2207 {
2208         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2209         int flags = EXT4_FREE_BLOCKS_FORGET;
2210
2211         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2212                 flags |= EXT4_FREE_BLOCKS_METADATA;
2213 #ifdef EXTENTS_STATS
2214         {
2215                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2216                 spin_lock(&sbi->s_ext_stats_lock);
2217                 sbi->s_ext_blocks += ee_len;
2218                 sbi->s_ext_extents++;
2219                 if (ee_len < sbi->s_ext_min)
2220                         sbi->s_ext_min = ee_len;
2221                 if (ee_len > sbi->s_ext_max)
2222                         sbi->s_ext_max = ee_len;
2223                 if (ext_depth(inode) > sbi->s_depth_max)
2224                         sbi->s_depth_max = ext_depth(inode);
2225                 spin_unlock(&sbi->s_ext_stats_lock);
2226         }
2227 #endif
2228         if (from >= le32_to_cpu(ex->ee_block)
2229             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2230                 /* tail removal */
2231                 ext4_lblk_t num;
2232                 ext4_fsblk_t start;
2233
2234                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2235                 start = ext4_ext_pblock(ex) + ee_len - num;
2236                 ext_debug("free last %u blocks starting %llu\n", num, start);
2237                 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2238         } else if (from == le32_to_cpu(ex->ee_block)
2239                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2240                 /* head removal */
2241                 ext4_lblk_t num;
2242                 ext4_fsblk_t start;
2243
2244                 num = to - from;
2245                 start = ext4_ext_pblock(ex);
2246
2247                 ext_debug("free first %u blocks starting %llu\n", num, start);
2248                 ext4_free_blocks(handle, inode, 0, start, num, flags);
2249
2250         } else {
2251                 printk(KERN_INFO "strange request: removal(2) "
2252                                 "%u-%u from %u:%u\n",
2253                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2254         }
2255         return 0;
2256 }
2257
2258
2259 /*
2260  * ext4_ext_rm_leaf() Removes the extents associated with the
2261  * blocks appearing between "start" and "end", and splits the extents
2262  * if "start" and "end" appear in the same extent
2263  *
2264  * @handle: The journal handle
2265  * @inode:  The files inode
2266  * @path:   The path to the leaf
2267  * @start:  The first block to remove
2268  * @end:   The last block to remove
2269  */
2270 static int
2271 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2272                 struct ext4_ext_path *path, ext4_lblk_t start,
2273                 ext4_lblk_t end)
2274 {
2275         int err = 0, correct_index = 0;
2276         int depth = ext_depth(inode), credits;
2277         struct ext4_extent_header *eh;
2278         ext4_lblk_t a, b, block;
2279         unsigned num;
2280         ext4_lblk_t ex_ee_block;
2281         unsigned short ex_ee_len;
2282         unsigned uninitialized = 0;
2283         struct ext4_extent *ex;
2284         struct ext4_map_blocks map;
2285
2286         /* the header must be checked already in ext4_ext_remove_space() */
2287         ext_debug("truncate since %u in leaf\n", start);
2288         if (!path[depth].p_hdr)
2289                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2290         eh = path[depth].p_hdr;
2291         if (unlikely(path[depth].p_hdr == NULL)) {
2292                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2293                 return -EIO;
2294         }
2295         /* find where to start removing */
2296         ex = EXT_LAST_EXTENT(eh);
2297
2298         ex_ee_block = le32_to_cpu(ex->ee_block);
2299         ex_ee_len = ext4_ext_get_actual_len(ex);
2300
2301         while (ex >= EXT_FIRST_EXTENT(eh) &&
2302                         ex_ee_block + ex_ee_len > start) {
2303
2304                 if (ext4_ext_is_uninitialized(ex))
2305                         uninitialized = 1;
2306                 else
2307                         uninitialized = 0;
2308
2309                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2310                          uninitialized, ex_ee_len);
2311                 path[depth].p_ext = ex;
2312
2313                 a = ex_ee_block > start ? ex_ee_block : start;
2314                 b = ex_ee_block+ex_ee_len - 1 < end ?
2315                         ex_ee_block+ex_ee_len - 1 : end;
2316
2317                 ext_debug("  border %u:%u\n", a, b);
2318
2319                 /* If this extent is beyond the end of the hole, skip it */
2320                 if (end <= ex_ee_block) {
2321                         ex--;
2322                         ex_ee_block = le32_to_cpu(ex->ee_block);
2323                         ex_ee_len = ext4_ext_get_actual_len(ex);
2324                         continue;
2325                 } else if (a != ex_ee_block &&
2326                         b != ex_ee_block + ex_ee_len - 1) {
2327                         /*
2328                          * If this is a truncate, then this condition should
2329                          * never happen because at least one of the end points
2330                          * needs to be on the edge of the extent.
2331                          */
2332                         if (end == EXT_MAX_BLOCKS - 1) {
2333                                 ext_debug("  bad truncate %u:%u\n",
2334                                                 start, end);
2335                                 block = 0;
2336                                 num = 0;
2337                                 err = -EIO;
2338                                 goto out;
2339                         }
2340                         /*
2341                          * else this is a hole punch, so the extent needs to
2342                          * be split since neither edge of the hole is on the
2343                          * extent edge
2344                          */
2345                         else{
2346                                 map.m_pblk = ext4_ext_pblock(ex);
2347                                 map.m_lblk = ex_ee_block;
2348                                 map.m_len = b - ex_ee_block;
2349
2350                                 err = ext4_split_extent(handle,
2351                                         inode, path, &map, 0,
2352                                         EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
2353                                         EXT4_GET_BLOCKS_PRE_IO);
2354
2355                                 if (err < 0)
2356                                         goto out;
2357
2358                                 ex_ee_len = ext4_ext_get_actual_len(ex);
2359
2360                                 b = ex_ee_block+ex_ee_len - 1 < end ?
2361                                         ex_ee_block+ex_ee_len - 1 : end;
2362
2363                                 /* Then remove tail of this extent */
2364                                 block = ex_ee_block;
2365                                 num = a - block;
2366                         }
2367                 } else if (a != ex_ee_block) {
2368                         /* remove tail of the extent */
2369                         block = ex_ee_block;
2370                         num = a - block;
2371                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2372                         /* remove head of the extent */
2373                         block = b;
2374                         num =  ex_ee_block + ex_ee_len - b;
2375
2376                         /*
2377                          * If this is a truncate, this condition
2378                          * should never happen
2379                          */
2380                         if (end == EXT_MAX_BLOCKS - 1) {
2381                                 ext_debug("  bad truncate %u:%u\n",
2382                                         start, end);
2383                                 err = -EIO;
2384                                 goto out;
2385                         }
2386                 } else {
2387                         /* remove whole extent: excellent! */
2388                         block = ex_ee_block;
2389                         num = 0;
2390                         if (a != ex_ee_block) {
2391                                 ext_debug("  bad truncate %u:%u\n",
2392                                         start, end);
2393                                 err = -EIO;
2394                                 goto out;
2395                         }
2396
2397                         if (b != ex_ee_block + ex_ee_len - 1) {
2398                                 ext_debug("  bad truncate %u:%u\n",
2399                                         start, end);
2400                                 err = -EIO;
2401                                 goto out;
2402                         }
2403                 }
2404
2405                 /*
2406                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2407                  * descriptor) for each block group; assume two block
2408                  * groups plus ex_ee_len/blocks_per_block_group for
2409                  * the worst case
2410                  */
2411                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2412                 if (ex == EXT_FIRST_EXTENT(eh)) {
2413                         correct_index = 1;
2414                         credits += (ext_depth(inode)) + 1;
2415                 }
2416                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2417
2418                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2419                 if (err)
2420                         goto out;
2421
2422                 err = ext4_ext_get_access(handle, inode, path + depth);
2423                 if (err)
2424                         goto out;
2425
2426                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2427                 if (err)
2428                         goto out;
2429
2430                 if (num == 0) {
2431                         /* this extent is removed; mark slot entirely unused */
2432                         ext4_ext_store_pblock(ex, 0);
2433                 } else if (block != ex_ee_block) {
2434                         /*
2435                          * If this was a head removal, then we need to update
2436                          * the physical block since it is now at a different
2437                          * location
2438                          */
2439                         ext4_ext_store_pblock(ex, ext4_ext_pblock(ex) + (b-a));
2440                 }
2441
2442                 ex->ee_block = cpu_to_le32(block);
2443                 ex->ee_len = cpu_to_le16(num);
2444                 /*
2445                  * Do not mark uninitialized if all the blocks in the
2446                  * extent have been removed.
2447                  */
2448                 if (uninitialized && num)
2449                         ext4_ext_mark_uninitialized(ex);
2450
2451                 err = ext4_ext_dirty(handle, inode, path + depth);
2452                 if (err)
2453                         goto out;
2454
2455                 /*
2456                  * If the extent was completely released,
2457                  * we need to remove it from the leaf
2458                  */
2459                 if (num == 0) {
2460                         if (end != EXT_MAX_BLOCKS - 1) {
2461                                 /*
2462                                  * For hole punching, we need to scoot all the
2463                                  * extents up when an extent is removed so that
2464                                  * we dont have blank extents in the middle
2465                                  */
2466                                 memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2467                                         sizeof(struct ext4_extent));
2468
2469                                 /* Now get rid of the one at the end */
2470                                 memset(EXT_LAST_EXTENT(eh), 0,
2471                                         sizeof(struct ext4_extent));
2472                         }
2473                         le16_add_cpu(&eh->eh_entries, -1);
2474                 }
2475
2476                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2477                                 ext4_ext_pblock(ex));
2478                 ex--;
2479                 ex_ee_block = le32_to_cpu(ex->ee_block);
2480                 ex_ee_len = ext4_ext_get_actual_len(ex);
2481         }
2482
2483         if (correct_index && eh->eh_entries)
2484                 err = ext4_ext_correct_indexes(handle, inode, path);
2485
2486         /* if this leaf is free, then we should
2487          * remove it from index block above */
2488         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2489                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2490
2491 out:
2492         return err;
2493 }
2494
2495 /*
2496  * ext4_ext_more_to_rm:
2497  * returns 1 if current index has to be freed (even partial)
2498  */
2499 static int
2500 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2501 {
2502         BUG_ON(path->p_idx == NULL);
2503
2504         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2505                 return 0;
2506
2507         /*
2508          * if truncate on deeper level happened, it wasn't partial,
2509          * so we have to consider current index for truncation
2510          */
2511         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2512                 return 0;
2513         return 1;
2514 }
2515
2516 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2517 {
2518         struct super_block *sb = inode->i_sb;
2519         int depth = ext_depth(inode);
2520         struct ext4_ext_path *path;
2521         handle_t *handle;
2522         int i, err;
2523
2524         ext_debug("truncate since %u\n", start);
2525
2526         /* probably first extent we're gonna free will be last in block */
2527         handle = ext4_journal_start(inode, depth + 1);
2528         if (IS_ERR(handle))
2529                 return PTR_ERR(handle);
2530
2531 again:
2532         ext4_ext_invalidate_cache(inode);
2533
2534         /*
2535          * We start scanning from right side, freeing all the blocks
2536          * after i_size and walking into the tree depth-wise.
2537          */
2538         depth = ext_depth(inode);
2539         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2540         if (path == NULL) {
2541                 ext4_journal_stop(handle);
2542                 return -ENOMEM;
2543         }
2544         path[0].p_depth = depth;
2545         path[0].p_hdr = ext_inode_hdr(inode);
2546         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2547                 err = -EIO;
2548                 goto out;
2549         }
2550         i = err = 0;
2551
2552         while (i >= 0 && err == 0) {
2553                 if (i == depth) {
2554                         /* this is leaf block */
2555                         err = ext4_ext_rm_leaf(handle, inode, path,
2556                                         start, EXT_MAX_BLOCKS - 1);
2557                         /* root level has p_bh == NULL, brelse() eats this */
2558                         brelse(path[i].p_bh);
2559                         path[i].p_bh = NULL;
2560                         i--;
2561                         continue;
2562                 }
2563
2564                 /* this is index block */
2565                 if (!path[i].p_hdr) {
2566                         ext_debug("initialize header\n");
2567                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2568                 }
2569
2570                 if (!path[i].p_idx) {
2571                         /* this level hasn't been touched yet */
2572                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2573                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2574                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2575                                   path[i].p_hdr,
2576                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2577                 } else {
2578                         /* we were already here, see at next index */
2579                         path[i].p_idx--;
2580                 }
2581
2582                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2583                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2584                                 path[i].p_idx);
2585                 if (ext4_ext_more_to_rm(path + i)) {
2586                         struct buffer_head *bh;
2587                         /* go to the next level */
2588                         ext_debug("move to level %d (block %llu)\n",
2589                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2590                         memset(path + i + 1, 0, sizeof(*path));
2591                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2592                         if (!bh) {
2593                                 /* should we reset i_size? */
2594                                 err = -EIO;
2595                                 break;
2596                         }
2597                         if (WARN_ON(i + 1 > depth)) {
2598                                 err = -EIO;
2599                                 break;
2600                         }
2601                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2602                                                         depth - i - 1)) {
2603                                 err = -EIO;
2604                                 break;
2605                         }
2606                         path[i + 1].p_bh = bh;
2607
2608                         /* save actual number of indexes since this
2609                          * number is changed at the next iteration */
2610                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2611                         i++;
2612                 } else {
2613                         /* we finished processing this index, go up */
2614                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2615                                 /* index is empty, remove it;
2616                                  * handle must be already prepared by the
2617                                  * truncatei_leaf() */
2618                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2619                         }
2620                         /* root level has p_bh == NULL, brelse() eats this */
2621                         brelse(path[i].p_bh);
2622                         path[i].p_bh = NULL;
2623                         i--;
2624                         ext_debug("return to level %d\n", i);
2625                 }
2626         }
2627
2628         /* TODO: flexible tree reduction should be here */
2629         if (path->p_hdr->eh_entries == 0) {
2630                 /*
2631                  * truncate to zero freed all the tree,
2632                  * so we need to correct eh_depth
2633                  */
2634                 err = ext4_ext_get_access(handle, inode, path);
2635                 if (err == 0) {
2636                         ext_inode_hdr(inode)->eh_depth = 0;
2637                         ext_inode_hdr(inode)->eh_max =
2638                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2639                         err = ext4_ext_dirty(handle, inode, path);
2640                 }
2641         }
2642 out:
2643         ext4_ext_drop_refs(path);
2644         kfree(path);
2645         if (err == -EAGAIN)
2646                 goto again;
2647         ext4_journal_stop(handle);
2648
2649         return err;
2650 }
2651
2652 /*
2653  * called at mount time
2654  */
2655 void ext4_ext_init(struct super_block *sb)
2656 {
2657         /*
2658          * possible initialization would be here
2659          */
2660
2661         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2662 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2663                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2664 #ifdef AGGRESSIVE_TEST
2665                 printk(", aggressive tests");
2666 #endif
2667 #ifdef CHECK_BINSEARCH
2668                 printk(", check binsearch");
2669 #endif
2670 #ifdef EXTENTS_STATS
2671                 printk(", stats");
2672 #endif
2673                 printk("\n");
2674 #endif
2675 #ifdef EXTENTS_STATS
2676                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2677                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2678                 EXT4_SB(sb)->s_ext_max = 0;
2679 #endif
2680         }
2681 }
2682
2683 /*
2684  * called at umount time
2685  */
2686 void ext4_ext_release(struct super_block *sb)
2687 {
2688         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2689                 return;
2690
2691 #ifdef EXTENTS_STATS
2692         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2693                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2694                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2695                         sbi->s_ext_blocks, sbi->s_ext_extents,
2696                         sbi->s_ext_blocks / sbi->s_ext_extents);
2697                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2698                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2699         }
2700 #endif
2701 }
2702
2703 /* FIXME!! we need to try to merge to left or right after zero-out  */
2704 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2705 {
2706         ext4_fsblk_t ee_pblock;
2707         unsigned int ee_len;
2708         int ret;
2709
2710         ee_len    = ext4_ext_get_actual_len(ex);
2711         ee_pblock = ext4_ext_pblock(ex);
2712
2713         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2714         if (ret > 0)
2715                 ret = 0;
2716
2717         return ret;
2718 }
2719
2720 /*
2721  * used by extent splitting.
2722  */
2723 #define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
2724                                         due to ENOSPC */
2725 #define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
2726 #define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
2727
2728 /*
2729  * ext4_split_extent_at() splits an extent at given block.
2730  *
2731  * @handle: the journal handle
2732  * @inode: the file inode
2733  * @path: the path to the extent
2734  * @split: the logical block where the extent is splitted.
2735  * @split_flags: indicates if the extent could be zeroout if split fails, and
2736  *               the states(init or uninit) of new extents.
2737  * @flags: flags used to insert new extent to extent tree.
2738  *
2739  *
2740  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2741  * of which are deterimined by split_flag.
2742  *
2743  * There are two cases:
2744  *  a> the extent are splitted into two extent.
2745  *  b> split is not needed, and just mark the extent.
2746  *
2747  * return 0 on success.
2748  */
2749 static int ext4_split_extent_at(handle_t *handle,
2750                              struct inode *inode,
2751                              struct ext4_ext_path *path,
2752                              ext4_lblk_t split,
2753                              int split_flag,
2754                              int flags)
2755 {
2756         ext4_fsblk_t newblock;
2757         ext4_lblk_t ee_block;
2758         struct ext4_extent *ex, newex, orig_ex;
2759         struct ext4_extent *ex2 = NULL;
2760         unsigned int ee_len, depth;
2761         int err = 0;
2762
2763         ext_debug("ext4_split_extents_at: inode %lu, logical"
2764                 "block %llu\n", inode->i_ino, (unsigned long long)split);
2765
2766         ext4_ext_show_leaf(inode, path);
2767
2768         depth = ext_depth(inode);
2769         ex = path[depth].p_ext;
2770         ee_block = le32_to_cpu(ex->ee_block);
2771         ee_len = ext4_ext_get_actual_len(ex);
2772         newblock = split - ee_block + ext4_ext_pblock(ex);
2773
2774         BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2775
2776         err = ext4_ext_get_access(handle, inode, path + depth);
2777         if (err)
2778                 goto out;
2779
2780         if (split == ee_block) {
2781                 /*
2782                  * case b: block @split is the block that the extent begins with
2783                  * then we just change the state of the extent, and splitting
2784                  * is not needed.
2785                  */
2786                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2787                         ext4_ext_mark_uninitialized(ex);
2788                 else
2789                         ext4_ext_mark_initialized(ex);
2790
2791                 if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2792                         ext4_ext_try_to_merge(inode, path, ex);
2793
2794                 err = ext4_ext_dirty(handle, inode, path + depth);
2795                 goto out;
2796         }
2797
2798         /* case a */
2799         memcpy(&orig_ex, ex, sizeof(orig_ex));
2800         ex->ee_len = cpu_to_le16(split - ee_block);
2801         if (split_flag & EXT4_EXT_MARK_UNINIT1)
2802                 ext4_ext_mark_uninitialized(ex);
2803
2804         /*
2805          * path may lead to new leaf, not to original leaf any more
2806          * after ext4_ext_insert_extent() returns,
2807          */
2808         err = ext4_ext_dirty(handle, inode, path + depth);
2809         if (err)
2810                 goto fix_extent_len;
2811
2812         ex2 = &newex;
2813         ex2->ee_block = cpu_to_le32(split);
2814         ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2815         ext4_ext_store_pblock(ex2, newblock);
2816         if (split_flag & EXT4_EXT_MARK_UNINIT2)
2817                 ext4_ext_mark_uninitialized(ex2);
2818
2819         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2820         if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2821                 err = ext4_ext_zeroout(inode, &orig_ex);
2822                 if (err)
2823                         goto fix_extent_len;
2824                 /* update the extent length and mark as initialized */
2825                 ex->ee_len = cpu_to_le32(ee_len);
2826                 ext4_ext_try_to_merge(inode, path, ex);
2827                 err = ext4_ext_dirty(handle, inode, path + depth);
2828                 goto out;
2829         } else if (err)
2830                 goto fix_extent_len;
2831
2832 out:
2833         ext4_ext_show_leaf(inode, path);
2834         return err;
2835
2836 fix_extent_len:
2837         ex->ee_len = orig_ex.ee_len;
2838         ext4_ext_dirty(handle, inode, path + depth);
2839         return err;
2840 }
2841
2842 /*
2843  * ext4_split_extents() splits an extent and mark extent which is covered
2844  * by @map as split_flags indicates
2845  *
2846  * It may result in splitting the extent into multiple extents (upto three)
2847  * There are three possibilities:
2848  *   a> There is no split required
2849  *   b> Splits in two extents: Split is happening at either end of the extent
2850  *   c> Splits in three extents: Somone is splitting in middle of the extent
2851  *
2852  */
2853 static int ext4_split_extent(handle_t *handle,
2854                               struct inode *inode,
2855                               struct ext4_ext_path *path,
2856                               struct ext4_map_blocks *map,
2857                               int split_flag,
2858                               int flags)
2859 {
2860         ext4_lblk_t ee_block;
2861         struct ext4_extent *ex;
2862         unsigned int ee_len, depth;
2863         int err = 0;
2864         int uninitialized;
2865         int split_flag1, flags1;
2866
2867         depth = ext_depth(inode);
2868         ex = path[depth].p_ext;
2869         ee_block = le32_to_cpu(ex->ee_block);
2870         ee_len = ext4_ext_get_actual_len(ex);
2871         uninitialized = ext4_ext_is_uninitialized(ex);
2872
2873         if (map->m_lblk + map->m_len < ee_block + ee_len) {
2874                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2875                               EXT4_EXT_MAY_ZEROOUT : 0;
2876                 flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2877                 if (uninitialized)
2878                         split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2879                                        EXT4_EXT_MARK_UNINIT2;
2880                 err = ext4_split_extent_at(handle, inode, path,
2881                                 map->m_lblk + map->m_len, split_flag1, flags1);
2882                 if (err)
2883                         goto out;
2884         }
2885
2886         ext4_ext_drop_refs(path);
2887         path = ext4_ext_find_extent(inode, map->m_lblk, path);
2888         if (IS_ERR(path))
2889                 return PTR_ERR(path);
2890
2891         if (map->m_lblk >= ee_block) {
2892                 split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2893                               EXT4_EXT_MAY_ZEROOUT : 0;
2894                 if (uninitialized)
2895                         split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2896                 if (split_flag & EXT4_EXT_MARK_UNINIT2)
2897                         split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2898                 err = ext4_split_extent_at(handle, inode, path,
2899                                 map->m_lblk, split_flag1, flags);
2900                 if (err)
2901                         goto out;
2902         }
2903
2904         ext4_ext_show_leaf(inode, path);
2905 out:
2906         return err ? err : map->m_len;
2907 }
2908
2909 #define EXT4_EXT_ZERO_LEN 7
2910 /*
2911  * This function is called by ext4_ext_map_blocks() if someone tries to write
2912  * to an uninitialized extent. It may result in splitting the uninitialized
2913  * extent into multiple extents (up to three - one initialized and two
2914  * uninitialized).
2915  * There are three possibilities:
2916  *   a> There is no split required: Entire extent should be initialized
2917  *   b> Splits in two extents: Write is happening at either end of the extent
2918  *   c> Splits in three extents: Somone is writing in middle of the extent
2919  */
2920 static int ext4_ext_convert_to_initialized(handle_t *handle,
2921                                            struct inode *inode,
2922                                            struct ext4_map_blocks *map,
2923                                            struct ext4_ext_path *path)
2924 {
2925         struct ext4_map_blocks split_map;
2926         struct ext4_extent zero_ex;
2927         struct ext4_extent *ex;
2928         ext4_lblk_t ee_block, eof_block;
2929         unsigned int allocated, ee_len, depth;
2930         int err = 0;
2931         int split_flag = 0;
2932
2933         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2934                 "block %llu, max_blocks %u\n", inode->i_ino,
2935                 (unsigned long long)map->m_lblk, map->m_len);
2936
2937         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2938                 inode->i_sb->s_blocksize_bits;
2939         if (eof_block < map->m_lblk + map->m_len)
2940                 eof_block = map->m_lblk + map->m_len;
2941
2942         depth = ext_depth(inode);
2943         ex = path[depth].p_ext;
2944         ee_block = le32_to_cpu(ex->ee_block);
2945         ee_len = ext4_ext_get_actual_len(ex);
2946         allocated = ee_len - (map->m_lblk - ee_block);
2947
2948         WARN_ON(map->m_lblk < ee_block);
2949         /*
2950          * It is safe to convert extent to initialized via explicit
2951          * zeroout only if extent is fully insde i_size or new_size.
2952          */
2953         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
2954
2955         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2956         if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
2957             (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2958                 err = ext4_ext_zeroout(inode, ex);
2959                 if (err)
2960                         goto out;
2961
2962                 err = ext4_ext_get_access(handle, inode, path + depth);
2963                 if (err)
2964                         goto out;
2965                 ext4_ext_mark_initialized(ex);
2966                 ext4_ext_try_to_merge(inode, path, ex);
2967                 err = ext4_ext_dirty(handle, inode, path + depth);
2968                 goto out;
2969         }
2970
2971         /*
2972          * four cases:
2973          * 1. split the extent into three extents.
2974          * 2. split the extent into two extents, zeroout the first half.
2975          * 3. split the extent into two extents, zeroout the second half.
2976          * 4. split the extent into two extents with out zeroout.
2977          */
2978         split_map.m_lblk = map->m_lblk;
2979         split_map.m_len = map->m_len;
2980
2981         if (allocated > map->m_len) {
2982                 if (allocated <= EXT4_EXT_ZERO_LEN &&
2983                     (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2984                         /* case 3 */
2985                         zero_ex.ee_block =
2986                                          cpu_to_le32(map->m_lblk);
2987                         zero_ex.ee_len = cpu_to_le16(allocated);
2988                         ext4_ext_store_pblock(&zero_ex,
2989                                 ext4_ext_pblock(ex) + map->m_lblk - ee_block);
2990                         err = ext4_ext_zeroout(inode, &zero_ex);
2991                         if (err)
2992                                 goto out;
2993                         split_map.m_lblk = map->m_lblk;
2994                         split_map.m_len = allocated;
2995                 } else if ((map->m_lblk - ee_block + map->m_len <
2996                            EXT4_EXT_ZERO_LEN) &&
2997                            (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2998                         /* case 2 */
2999                         if (map->m_lblk != ee_block) {
3000                                 zero_ex.ee_block = ex->ee_block;
3001                                 zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3002                                                         ee_block);
3003                                 ext4_ext_store_pblock(&zero_ex,
3004                                                       ext4_ext_pblock(ex));
3005                                 err = ext4_ext_zeroout(inode, &zero_ex);
3006                                 if (err)
3007                                         goto out;
3008                         }
3009
3010                         split_map.m_lblk = ee_block;
3011                         split_map.m_len = map->m_lblk - ee_block + map->m_len;
3012                         allocated = map->m_len;
3013                 }
3014         }
3015
3016         allocated = ext4_split_extent(handle, inode, path,
3017                                        &split_map, split_flag, 0);
3018         if (allocated < 0)
3019                 err = allocated;
3020
3021 out:
3022         return err ? err : allocated;
3023 }
3024
3025 /*
3026  * This function is called by ext4_ext_map_blocks() from
3027  * ext4_get_blocks_dio_write() when DIO to write
3028  * to an uninitialized extent.
3029  *
3030  * Writing to an uninitialized extent may result in splitting the uninitialized
3031  * extent into multiple /initialized uninitialized extents (up to three)
3032  * There are three possibilities:
3033  *   a> There is no split required: Entire extent should be uninitialized
3034  *   b> Splits in two extents: Write is happening at either end of the extent
3035  *   c> Splits in three extents: Somone is writing in middle of the extent
3036  *
3037  * One of more index blocks maybe needed if the extent tree grow after
3038  * the uninitialized extent split. To prevent ENOSPC occur at the IO
3039  * complete, we need to split the uninitialized extent before DIO submit
3040  * the IO. The uninitialized extent called at this time will be split
3041  * into three uninitialized extent(at most). After IO complete, the part
3042  * being filled will be convert to initialized by the end_io callback function
3043  * via ext4_convert_unwritten_extents().
3044  *
3045  * Returns the size of uninitialized extent to be written on success.
3046  */
3047 static int ext4_split_unwritten_extents(handle_t *handle,
3048                                         struct inode *inode,
3049                                         struct ext4_map_blocks *map,
3050                                         struct ext4_ext_path *path,
3051                                         int flags)
3052 {
3053         ext4_lblk_t eof_block;
3054         ext4_lblk_t ee_block;
3055         struct ext4_extent *ex;
3056         unsigned int ee_len;
3057         int split_flag = 0, depth;
3058
3059         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3060                 "block %llu, max_blocks %u\n", inode->i_ino,
3061                 (unsigned long long)map->m_lblk, map->m_len);
3062
3063         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3064                 inode->i_sb->s_blocksize_bits;
3065         if (eof_block < map->m_lblk + map->m_len)
3066                 eof_block = map->m_lblk + map->m_len;
3067         /*
3068          * It is safe to convert extent to initialized via explicit
3069          * zeroout only if extent is fully insde i_size or new_size.
3070          */
3071         depth = ext_depth(inode);
3072         ex = path[depth].p_ext;
3073         ee_block = le32_to_cpu(ex->ee_block);
3074         ee_len = ext4_ext_get_actual_len(ex);
3075
3076         split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3077         split_flag |= EXT4_EXT_MARK_UNINIT2;
3078
3079         flags |= EXT4_GET_BLOCKS_PRE_IO;
3080         return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3081 }
3082
3083 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3084                                               struct inode *inode,
3085                                               struct ext4_ext_path *path)
3086 {
3087         struct ext4_extent *ex;
3088         int depth;
3089         int err = 0;
3090
3091         depth = ext_depth(inode);
3092         ex = path[depth].p_ext;
3093
3094         ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3095                 "block %llu, max_blocks %u\n", inode->i_ino,
3096                 (unsigned long long)le32_to_cpu(ex->ee_block),
3097                 ext4_ext_get_actual_len(ex));
3098
3099         err = ext4_ext_get_access(handle, inode, path + depth);
3100         if (err)
3101                 goto out;
3102         /* first mark the extent as initialized */
3103         ext4_ext_mark_initialized(ex);
3104
3105         /* note: ext4_ext_correct_indexes() isn't needed here because
3106          * borders are not changed
3107          */
3108         ext4_ext_try_to_merge(inode, path, ex);
3109
3110         /* Mark modified extent as dirty */
3111         err = ext4_ext_dirty(handle, inode, path + depth);
3112 out:
3113         ext4_ext_show_leaf(inode, path);
3114         return err;
3115 }
3116
3117 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3118                         sector_t block, int count)
3119 {
3120         int i;
3121         for (i = 0; i < count; i++)
3122                 unmap_underlying_metadata(bdev, block + i);
3123 }
3124
3125 /*
3126  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3127  */
3128 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3129                               ext4_lblk_t lblk,
3130                               struct ext4_ext_path *path,
3131                               unsigned int len)
3132 {
3133         int i, depth;
3134         struct ext4_extent_header *eh;
3135         struct ext4_extent *last_ex;
3136
3137         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3138                 return 0;
3139
3140         depth = ext_depth(inode);
3141         eh = path[depth].p_hdr;
3142
3143         if (unlikely(!eh->eh_entries)) {
3144                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3145                                  "EOFBLOCKS_FL set");
3146                 return -EIO;
3147         }
3148         last_ex = EXT_LAST_EXTENT(eh);
3149         /*
3150          * We should clear the EOFBLOCKS_FL flag if we are writing the
3151          * last block in the last extent in the file.  We test this by
3152          * first checking to see if the caller to
3153          * ext4_ext_get_blocks() was interested in the last block (or
3154          * a block beyond the last block) in the current extent.  If
3155          * this turns out to be false, we can bail out from this
3156          * function immediately.
3157          */
3158         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3159             ext4_ext_get_actual_len(last_ex))
3160                 return 0;
3161         /*
3162          * If the caller does appear to be planning to write at or
3163          * beyond the end of the current extent, we then test to see
3164          * if the current extent is the last extent in the file, by
3165          * checking to make sure it was reached via the rightmost node
3166          * at each level of the tree.
3167          */
3168         for (i = depth-1; i >= 0; i--)
3169                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3170                         return 0;
3171         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3172         return ext4_mark_inode_dirty(handle, inode);
3173 }
3174
3175 static int
3176 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3177                         struct ext4_map_blocks *map,
3178                         struct ext4_ext_path *path, int flags,
3179                         unsigned int allocated, ext4_fsblk_t newblock)
3180 {
3181         int ret = 0;
3182         int err = 0;
3183         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3184
3185         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3186                   "block %llu, max_blocks %u, flags %d, allocated %u",
3187                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3188                   flags, allocated);
3189         ext4_ext_show_leaf(inode, path);
3190
3191         /* get_block() before submit the IO, split the extent */
3192         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3193                 ret = ext4_split_unwritten_extents(handle, inode, map,
3194                                                    path, flags);
3195                 /*
3196                  * Flag the inode(non aio case) or end_io struct (aio case)
3197                  * that this IO needs to conversion to written when IO is
3198                  * completed
3199                  */
3200                 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3201                         io->flag = EXT4_IO_END_UNWRITTEN;
3202                         atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3203                 } else
3204                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3205                 if (ext4_should_dioread_nolock(inode))
3206                         map->m_flags |= EXT4_MAP_UNINIT;
3207                 goto out;
3208         }
3209         /* IO end_io complete, convert the filled extent to written */
3210         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3211                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3212                                                         path);
3213                 if (ret >= 0) {
3214                         ext4_update_inode_fsync_trans(handle, inode, 1);
3215                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3216                                                  path, map->m_len);
3217                 } else
3218                         err = ret;
3219                 goto out2;
3220         }
3221         /* buffered IO case */
3222         /*
3223          * repeat fallocate creation request
3224          * we already have an unwritten extent
3225          */
3226         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3227                 goto map_out;
3228
3229         /* buffered READ or buffered write_begin() lookup */
3230         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3231                 /*
3232                  * We have blocks reserved already.  We
3233                  * return allocated blocks so that delalloc
3234                  * won't do block reservation for us.  But
3235                  * the buffer head will be unmapped so that
3236                  * a read from the block returns 0s.
3237                  */
3238                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3239                 goto out1;
3240         }
3241
3242         /* buffered write, writepage time, convert*/
3243         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3244         if (ret >= 0) {
3245                 ext4_update_inode_fsync_trans(handle, inode, 1);
3246                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3247                                          map->m_len);
3248                 if (err < 0)
3249                         goto out2;
3250         }
3251
3252 out:
3253         if (ret <= 0) {
3254                 err = ret;
3255                 goto out2;
3256         } else
3257                 allocated = ret;
3258         map->m_flags |= EXT4_MAP_NEW;
3259         /*
3260          * if we allocated more blocks than requested
3261          * we need to make sure we unmap the extra block
3262          * allocated. The actual needed block will get
3263          * unmapped later when we find the buffer_head marked
3264          * new.
3265          */
3266         if (allocated > map->m_len) {
3267                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3268                                         newblock + map->m_len,
3269                                         allocated - map->m_len);
3270                 allocated = map->m_len;
3271         }
3272
3273         /*
3274          * If we have done fallocate with the offset that is already
3275          * delayed allocated, we would have block reservation
3276          * and quota reservation done in the delayed write path.
3277          * But fallocate would have already updated quota and block
3278          * count for this offset. So cancel these reservation
3279          */
3280         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3281                 ext4_da_update_reserve_space(inode, allocated, 0);
3282
3283 map_out:
3284         map->m_flags |= EXT4_MAP_MAPPED;
3285 out1:
3286         if (allocated > map->m_len)
3287                 allocated = map->m_len;
3288         ext4_ext_show_leaf(inode, path);
3289         map->m_pblk = newblock;
3290         map->m_len = allocated;
3291 out2:
3292         if (path) {
3293                 ext4_ext_drop_refs(path);
3294                 kfree(path);
3295         }
3296         return err ? err : allocated;
3297 }
3298
3299 /*
3300  * get_implied_cluster_alloc - check to see if the requested
3301  * allocation (in the map structure) overlaps with a cluster already
3302  * allocated in an extent.
3303  *      @sbi    The ext4-specific superblock structure
3304  *      @map    The requested lblk->pblk mapping
3305  *      @ex     The extent structure which might contain an implied
3306  *                      cluster allocation
3307  *
3308  * This function is called by ext4_ext_map_blocks() after we failed to
3309  * find blocks that were already in the inode's extent tree.  Hence,
3310  * we know that the beginning of the requested region cannot overlap
3311  * the extent from the inode's extent tree.  There are three cases we
3312  * want to catch.  The first is this case:
3313  *
3314  *               |--- cluster # N--|
3315  *    |--- extent ---|  |---- requested region ---|
3316  *                      |==========|
3317  *
3318  * The second case that we need to test for is this one:
3319  *
3320  *   |--------- cluster # N ----------------|
3321  *         |--- requested region --|   |------- extent ----|
3322  *         |=======================|
3323  *
3324  * The third case is when the requested region lies between two extents
3325  * within the same cluster:
3326  *          |------------- cluster # N-------------|
3327  * |----- ex -----|                  |---- ex_right ----|
3328  *                  |------ requested region ------|
3329  *                  |================|
3330  *
3331  * In each of the above cases, we need to set the map->m_pblk and
3332  * map->m_len so it corresponds to the return the extent labelled as
3333  * "|====|" from cluster #N, since it is already in use for data in
3334  * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3335  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3336  * as a new "allocated" block region.  Otherwise, we will return 0 and
3337  * ext4_ext_map_blocks() will then allocate one or more new clusters
3338  * by calling ext4_mb_new_blocks().
3339  */
3340 static int get_implied_cluster_alloc(struct ext4_sb_info *sbi,
3341                                      struct ext4_map_blocks *map,
3342                                      struct ext4_extent *ex,
3343                                      struct ext4_ext_path *path)
3344 {
3345         ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3346         ext4_lblk_t ex_cluster_start, ex_cluster_end;
3347         ext4_lblk_t rr_cluster_start, rr_cluster_end;
3348         ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3349         ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3350         unsigned short ee_len = ext4_ext_get_actual_len(ex);
3351
3352         /* The extent passed in that we are trying to match */
3353         ex_cluster_start = EXT4_B2C(sbi, ee_block);
3354         ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3355
3356         /* The requested region passed into ext4_map_blocks() */
3357         rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3358         rr_cluster_end = EXT4_B2C(sbi, map->m_lblk + map->m_len - 1);
3359
3360         if ((rr_cluster_start == ex_cluster_end) ||
3361             (rr_cluster_start == ex_cluster_start)) {
3362                 if (rr_cluster_start == ex_cluster_end)
3363                         ee_start += ee_len - 1;
3364                 map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3365                         c_offset;
3366                 map->m_len = min(map->m_len,
3367                                  (unsigned) sbi->s_cluster_ratio - c_offset);
3368                 /*
3369                  * Check for and handle this case:
3370                  *
3371                  *   |--------- cluster # N-------------|
3372                  *                     |------- extent ----|
3373                  *         |--- requested region ---|
3374                  *         |===========|
3375                  */
3376
3377                 if (map->m_lblk < ee_block)
3378                         map->m_len = min(map->m_len, ee_block - map->m_lblk);
3379
3380                 /*
3381                  * Check for the case where there is already another allocated
3382                  * block to the right of 'ex' but before the end of the cluster.
3383                  *
3384                  *          |------------- cluster # N-------------|
3385                  * |----- ex -----|                  |---- ex_right ----|
3386                  *                  |------ requested region ------|
3387                  *                  |================|
3388                  */
3389                 if (map->m_lblk > ee_block) {
3390                         ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3391                         map->m_len = min(map->m_len, next - map->m_lblk);
3392                 }
3393                 return 1;
3394         }
3395         return 0;
3396 }
3397
3398
3399 /*
3400  * Block allocation/map/preallocation routine for extents based files
3401  *
3402  *
3403  * Need to be called with
3404  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3405  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3406  *
3407  * return > 0, number of of blocks already mapped/allocated
3408  *          if create == 0 and these are pre-allocated blocks
3409  *              buffer head is unmapped
3410  *          otherwise blocks are mapped
3411  *
3412  * return = 0, if plain look up failed (blocks have not been allocated)
3413  *          buffer head is unmapped
3414  *
3415  * return < 0, error case.
3416  */
3417 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3418                         struct ext4_map_blocks *map, int flags)
3419 {
3420         struct ext4_ext_path *path = NULL;
3421         struct ext4_extent newex, *ex, *ex2;
3422         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3423         ext4_fsblk_t newblock = 0;
3424         int free_on_err = 0, err = 0, depth, ret;
3425         unsigned int allocated = 0, offset = 0;
3426         unsigned int punched_out = 0;
3427         unsigned int result = 0;
3428         struct ext4_allocation_request ar;
3429         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3430         ext4_lblk_t cluster_offset;
3431         struct ext4_map_blocks punch_map;
3432
3433         ext_debug("blocks %u/%u requested for inode %lu\n",
3434                   map->m_lblk, map->m_len, inode->i_ino);
3435         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3436
3437         /* check in cache */
3438         if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3439                 ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3440                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3441                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3442                                 /*
3443                                  * block isn't allocated yet and
3444                                  * user doesn't want to allocate it
3445                                  */
3446                                 goto out2;
3447                         }
3448                         /* we should allocate requested block */
3449                 } else {
3450                         /* block is already allocated */
3451                         newblock = map->m_lblk
3452                                    - le32_to_cpu(newex.ee_block)
3453                                    + ext4_ext_pblock(&newex);
3454                         /* number of remaining blocks in the extent */
3455                         allocated = ext4_ext_get_actual_len(&newex) -
3456                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3457                         goto out;
3458                 }
3459         }
3460
3461         /* find extent for this block */
3462         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3463         if (IS_ERR(path)) {
3464                 err = PTR_ERR(path);
3465                 path = NULL;
3466                 goto out2;
3467         }
3468
3469         depth = ext_depth(inode);
3470
3471         /*
3472          * consistent leaf must not be empty;
3473          * this situation is possible, though, _during_ tree modification;
3474          * this is why assert can't be put in ext4_ext_find_extent()
3475          */
3476         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3477                 EXT4_ERROR_INODE(inode, "bad extent address "
3478                                  "lblock: %lu, depth: %d pblock %lld",
3479                                  (unsigned long) map->m_lblk, depth,
3480                                  path[depth].p_block);
3481                 err = -EIO;
3482                 goto out2;
3483         }
3484
3485         ex = path[depth].p_ext;
3486         if (ex) {
3487                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3488                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3489                 unsigned short ee_len;
3490
3491                 /*
3492                  * Uninitialized extents are treated as holes, except that
3493                  * we split out initialized portions during a write.
3494                  */
3495                 ee_len = ext4_ext_get_actual_len(ex);
3496                 /* if found extent covers block, simply return it */
3497                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3498                         newblock = map->m_lblk - ee_block + ee_start;
3499                         /* number of remaining blocks in the extent */
3500                         allocated = ee_len - (map->m_lblk - ee_block);
3501                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3502                                   ee_block, ee_len, newblock);
3503
3504                         if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3505                                 /*
3506                                  * Do not put uninitialized extent
3507                                  * in the cache
3508                                  */
3509                                 if (!ext4_ext_is_uninitialized(ex)) {
3510                                         ext4_ext_put_in_cache(inode, ee_block,
3511                                                 ee_len, ee_start);
3512                                         goto out;
3513                                 }
3514                                 ret = ext4_ext_handle_uninitialized_extents(
3515                                         handle, inode, map, path, flags,
3516                                         allocated, newblock);
3517                                 return ret;
3518                         }
3519
3520                         /*
3521                          * Punch out the map length, but only to the
3522                          * end of the extent
3523                          */
3524                         punched_out = allocated < map->m_len ?
3525                                 allocated : map->m_len;
3526
3527                         /*
3528                          * Sense extents need to be converted to
3529                          * uninitialized, they must fit in an
3530                          * uninitialized extent
3531                          */
3532                         if (punched_out > EXT_UNINIT_MAX_LEN)
3533                                 punched_out = EXT_UNINIT_MAX_LEN;
3534
3535                         punch_map.m_lblk = map->m_lblk;
3536                         punch_map.m_pblk = newblock;
3537                         punch_map.m_len = punched_out;
3538                         punch_map.m_flags = 0;
3539
3540                         /* Check to see if the extent needs to be split */
3541                         if (punch_map.m_len != ee_len ||
3542                                 punch_map.m_lblk != ee_block) {
3543
3544                                 ret = ext4_split_extent(handle, inode,
3545                                 path, &punch_map, 0,
3546                                 EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3547                                 EXT4_GET_BLOCKS_PRE_IO);
3548
3549                                 if (ret < 0) {
3550                                         err = ret;
3551                                         goto out2;
3552                                 }
3553                                 /*
3554                                  * find extent for the block at
3555                                  * the start of the hole
3556                                  */
3557                                 ext4_ext_drop_refs(path);
3558                                 kfree(path);
3559
3560                                 path = ext4_ext_find_extent(inode,
3561                                 map->m_lblk, NULL);
3562                                 if (IS_ERR(path)) {
3563                                         err = PTR_ERR(path);
3564                                         path = NULL;
3565                                         goto out2;
3566                                 }
3567
3568                                 depth = ext_depth(inode);
3569                                 ex = path[depth].p_ext;
3570                                 ee_len = ext4_ext_get_actual_len(ex);
3571                                 ee_block = le32_to_cpu(ex->ee_block);
3572                                 ee_start = ext4_ext_pblock(ex);
3573
3574                         }
3575
3576                         ext4_ext_mark_uninitialized(ex);
3577
3578                         ext4_ext_invalidate_cache(inode);
3579
3580                         err = ext4_ext_rm_leaf(handle, inode, path,
3581                                 map->m_lblk, map->m_lblk + punched_out);
3582
3583                         if (!err && path->p_hdr->eh_entries == 0) {
3584                                 /*
3585                                  * Punch hole freed all of this sub tree,
3586                                  * so we need to correct eh_depth
3587                                  */
3588                                 err = ext4_ext_get_access(handle, inode, path);
3589                                 if (err == 0) {
3590                                         ext_inode_hdr(inode)->eh_depth = 0;
3591                                         ext_inode_hdr(inode)->eh_max =
3592                                         cpu_to_le16(ext4_ext_space_root(
3593                                                 inode, 0));
3594
3595                                         err = ext4_ext_dirty(
3596                                                 handle, inode, path);
3597                                 }
3598                         }
3599
3600                         goto out2;
3601                 }
3602         }
3603
3604         /*
3605          * requested block isn't allocated yet;
3606          * we couldn't try to create block if create flag is zero
3607          */
3608         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3609                 /*
3610                  * put just found gap into cache to speed up
3611                  * subsequent requests
3612                  */
3613                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3614                 goto out2;
3615         }
3616
3617         /*
3618          * Okay, we need to do block allocation.
3619          */
3620         newex.ee_block = cpu_to_le32(map->m_lblk);
3621         cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3622
3623         /*
3624          * If we are doing bigalloc, check to see if the extent returned
3625          * by ext4_ext_find_extent() implies a cluster we can use.
3626          */
3627         if (cluster_offset && ex &&
3628             get_implied_cluster_alloc(sbi, map, ex, path)) {
3629                 ar.len = allocated = map->m_len;
3630                 newblock = map->m_pblk;
3631                 goto got_allocated_blocks;
3632         }
3633
3634         /* find neighbour allocated blocks */
3635         ar.lleft = map->m_lblk;
3636         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3637         if (err)
3638                 goto out2;
3639         ar.lright = map->m_lblk;
3640         ex2 = NULL;
3641         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3642         if (err)
3643                 goto out2;
3644
3645         /* Check if the extent after searching to the right implies a
3646          * cluster we can use. */
3647         if ((sbi->s_cluster_ratio > 1) && ex2 &&
3648             get_implied_cluster_alloc(sbi, map, ex2, path)) {
3649                 ar.len = allocated = map->m_len;
3650                 newblock = map->m_pblk;
3651                 goto got_allocated_blocks;
3652         }
3653
3654         /*
3655          * See if request is beyond maximum number of blocks we can have in
3656          * a single extent. For an initialized extent this limit is
3657          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3658          * EXT_UNINIT_MAX_LEN.
3659          */
3660         if (map->m_len > EXT_INIT_MAX_LEN &&
3661             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3662                 map->m_len = EXT_INIT_MAX_LEN;
3663         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3664                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3665                 map->m_len = EXT_UNINIT_MAX_LEN;
3666
3667         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3668         newex.ee_len = cpu_to_le16(map->m_len);
3669         err = ext4_ext_check_overlap(sbi, inode, &newex, path);
3670         if (err)
3671                 allocated = ext4_ext_get_actual_len(&newex);
3672         else
3673                 allocated = map->m_len;
3674
3675         /* allocate new block */
3676         ar.inode = inode;
3677         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3678         ar.logical = map->m_lblk;
3679         /*
3680          * We calculate the offset from the beginning of the cluster
3681          * for the logical block number, since when we allocate a
3682          * physical cluster, the physical block should start at the
3683          * same offset from the beginning of the cluster.  This is
3684          * needed so that future calls to get_implied_cluster_alloc()
3685          * work correctly.
3686          */
3687         offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3688         ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3689         ar.goal -= offset;
3690         ar.logical -= offset;
3691         if (S_ISREG(inode->i_mode))
3692                 ar.flags = EXT4_MB_HINT_DATA;
3693         else
3694                 /* disable in-core preallocation for non-regular files */
3695                 ar.flags = 0;
3696         if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
3697                 ar.flags |= EXT4_MB_HINT_NOPREALLOC;
3698         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3699         if (!newblock)
3700                 goto out2;
3701         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3702                   ar.goal, newblock, allocated);
3703         free_on_err = 1;
3704         ar.len = EXT4_C2B(sbi, ar.len) - offset;
3705         if (ar.len > allocated)
3706                 ar.len = allocated;
3707
3708 got_allocated_blocks:
3709         /* try to insert new extent into found leaf and return */
3710         ext4_ext_store_pblock(&newex, newblock + offset);
3711         newex.ee_len = cpu_to_le16(ar.len);
3712         /* Mark uninitialized */
3713         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3714                 ext4_ext_mark_uninitialized(&newex);
3715                 /*
3716                  * io_end structure was created for every IO write to an
3717                  * uninitialized extent. To avoid unnecessary conversion,
3718                  * here we flag the IO that really needs the conversion.
3719                  * For non asycn direct IO case, flag the inode state
3720                  * that we need to perform conversion when IO is done.
3721                  */
3722                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3723                         if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3724                                 io->flag = EXT4_IO_END_UNWRITTEN;
3725                                 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3726                         } else
3727                                 ext4_set_inode_state(inode,
3728                                                      EXT4_STATE_DIO_UNWRITTEN);
3729                 }
3730                 if (ext4_should_dioread_nolock(inode))
3731                         map->m_flags |= EXT4_MAP_UNINIT;
3732         }
3733
3734         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3735         if (!err)
3736                 err = ext4_ext_insert_extent(handle, inode, path,
3737                                              &newex, flags);
3738         if (err && free_on_err) {
3739                 int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
3740                         EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
3741                 /* free data blocks we just allocated */
3742                 /* not a good idea to call discard here directly,
3743                  * but otherwise we'd need to call it every free() */
3744                 ext4_discard_preallocations(inode);
3745                 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3746                                  ext4_ext_get_actual_len(&newex), fb_flags);
3747                 goto out2;
3748         }
3749
3750         /* previous routine could use block we allocated */
3751         newblock = ext4_ext_pblock(&newex);
3752         allocated = ext4_ext_get_actual_len(&newex);
3753         if (allocated > map->m_len)
3754                 allocated = map->m_len;
3755         map->m_flags |= EXT4_MAP_NEW;
3756
3757         /*
3758          * Update reserved blocks/metadata blocks after successful
3759          * block allocation which had been deferred till now.
3760          */
3761         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3762                 ext4_da_update_reserve_space(inode, allocated, 1);
3763
3764         /*
3765          * Cache the extent and update transaction to commit on fdatasync only
3766          * when it is _not_ an uninitialized extent.
3767          */
3768         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3769                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3770                 ext4_update_inode_fsync_trans(handle, inode, 1);
3771         } else
3772                 ext4_update_inode_fsync_trans(handle, inode, 0);
3773 out:
3774         if (allocated > map->m_len)
3775                 allocated = map->m_len;
3776         ext4_ext_show_leaf(inode, path);
3777         map->m_flags |= EXT4_MAP_MAPPED;
3778         map->m_pblk = newblock;
3779         map->m_len = allocated;
3780 out2:
3781         if (path) {
3782                 ext4_ext_drop_refs(path);
3783                 kfree(path);
3784         }
3785         trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3786                 newblock, map->m_len, err ? err : allocated);
3787
3788         result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
3789                         punched_out : allocated;
3790
3791         return err ? err : result;
3792 }
3793
3794 void ext4_ext_truncate(struct inode *inode)
3795 {
3796         struct address_space *mapping = inode->i_mapping;
3797         struct super_block *sb = inode->i_sb;
3798         ext4_lblk_t last_block;
3799         handle_t *handle;
3800         loff_t page_len;
3801         int err = 0;
3802
3803         /*
3804          * finish any pending end_io work so we won't run the risk of
3805          * converting any truncated blocks to initialized later
3806          */
3807         ext4_flush_completed_IO(inode);
3808
3809         /*
3810          * probably first extent we're gonna free will be last in block
3811          */
3812         err = ext4_writepage_trans_blocks(inode);
3813         handle = ext4_journal_start(inode, err);
3814         if (IS_ERR(handle))
3815                 return;
3816
3817         if (inode->i_size % PAGE_CACHE_SIZE != 0) {
3818                 page_len = PAGE_CACHE_SIZE -
3819                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
3820
3821                 err = ext4_discard_partial_page_buffers(handle,
3822                         mapping, inode->i_size, page_len, 0);
3823
3824                 if (err)
3825                         goto out_stop;
3826         }
3827
3828         if (ext4_orphan_add(handle, inode))
3829                 goto out_stop;
3830
3831         down_write(&EXT4_I(inode)->i_data_sem);
3832         ext4_ext_invalidate_cache(inode);
3833
3834         ext4_discard_preallocations(inode);
3835
3836         /*
3837          * TODO: optimization is possible here.
3838          * Probably we need not scan at all,
3839          * because page truncation is enough.
3840          */
3841
3842         /* we have to know where to truncate from in crash case */
3843         EXT4_I(inode)->i_disksize = inode->i_size;
3844         ext4_mark_inode_dirty(handle, inode);
3845
3846         last_block = (inode->i_size + sb->s_blocksize - 1)
3847                         >> EXT4_BLOCK_SIZE_BITS(sb);
3848         err = ext4_ext_remove_space(inode, last_block);
3849
3850         /* In a multi-transaction truncate, we only make the final
3851          * transaction synchronous.
3852          */
3853         if (IS_SYNC(inode))
3854                 ext4_handle_sync(handle);
3855
3856         up_write(&EXT4_I(inode)->i_data_sem);
3857
3858 out_stop:
3859         /*
3860          * If this was a simple ftruncate() and the file will remain alive,
3861          * then we need to clear up the orphan record which we created above.
3862          * However, if this was a real unlink then we were called by
3863          * ext4_delete_inode(), and we allow that function to clean up the
3864          * orphan info for us.
3865          */
3866         if (inode->i_nlink)
3867                 ext4_orphan_del(handle, inode);
3868
3869         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3870         ext4_mark_inode_dirty(handle, inode);
3871         ext4_journal_stop(handle);
3872 }
3873
3874 static void ext4_falloc_update_inode(struct inode *inode,
3875                                 int mode, loff_t new_size, int update_ctime)
3876 {
3877         struct timespec now;
3878
3879         if (update_ctime) {
3880                 now = current_fs_time(inode->i_sb);
3881                 if (!timespec_equal(&inode->i_ctime, &now))
3882                         inode->i_ctime = now;
3883         }
3884         /*
3885          * Update only when preallocation was requested beyond
3886          * the file size.
3887          */
3888         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3889                 if (new_size > i_size_read(inode))
3890                         i_size_write(inode, new_size);
3891                 if (new_size > EXT4_I(inode)->i_disksize)
3892                         ext4_update_i_disksize(inode, new_size);
3893         } else {
3894                 /*
3895                  * Mark that we allocate beyond EOF so the subsequent truncate
3896                  * can proceed even if the new size is the same as i_size.
3897                  */
3898                 if (new_size > i_size_read(inode))
3899                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3900         }
3901
3902 }
3903
3904 /*
3905  * preallocate space for a file. This implements ext4's fallocate file
3906  * operation, which gets called from sys_fallocate system call.
3907  * For block-mapped files, posix_fallocate should fall back to the method
3908  * of writing zeroes to the required new blocks (the same behavior which is
3909  * expected for file systems which do not support fallocate() system call).
3910  */
3911 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3912 {
3913         struct inode *inode = file->f_path.dentry->d_inode;
3914         handle_t *handle;
3915         loff_t new_size;
3916         unsigned int max_blocks;
3917         int ret = 0;
3918         int ret2 = 0;
3919         int retries = 0;
3920         struct ext4_map_blocks map;
3921         unsigned int credits, blkbits = inode->i_blkbits;
3922
3923         /*
3924          * currently supporting (pre)allocate mode for extent-based
3925          * files _only_
3926          */
3927         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3928                 return -EOPNOTSUPP;
3929
3930         /* Return error if mode is not supported */
3931         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3932                 return -EOPNOTSUPP;
3933
3934         if (mode & FALLOC_FL_PUNCH_HOLE)
3935                 return ext4_punch_hole(file, offset, len);
3936
3937         trace_ext4_fallocate_enter(inode, offset, len, mode);
3938         map.m_lblk = offset >> blkbits;
3939         /*
3940          * We can't just convert len to max_blocks because
3941          * If blocksize = 4096 offset = 3072 and len = 2048
3942          */
3943         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3944                 - map.m_lblk;
3945         /*
3946          * credits to insert 1 extent into extent tree
3947          */
3948         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3949         mutex_lock(&inode->i_mutex);
3950         ret = inode_newsize_ok(inode, (len + offset));
3951         if (ret) {
3952                 mutex_unlock(&inode->i_mutex);
3953                 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3954                 return ret;
3955         }
3956 retry:
3957         while (ret >= 0 && ret < max_blocks) {
3958                 map.m_lblk = map.m_lblk + ret;
3959                 map.m_len = max_blocks = max_blocks - ret;
3960                 handle = ext4_journal_start(inode, credits);
3961                 if (IS_ERR(handle)) {
3962                         ret = PTR_ERR(handle);
3963                         break;
3964                 }
3965                 ret = ext4_map_blocks(handle, inode, &map,
3966                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT |
3967                                       EXT4_GET_BLOCKS_NO_NORMALIZE);
3968                 if (ret <= 0) {
3969 #ifdef EXT4FS_DEBUG
3970                         WARN_ON(ret <= 0);
3971                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3972                                     "returned error inode#%lu, block=%u, "
3973                                     "max_blocks=%u", __func__,
3974                                     inode->i_ino, map.m_lblk, max_blocks);
3975 #endif
3976                         ext4_mark_inode_dirty(handle, inode);
3977                         ret2 = ext4_journal_stop(handle);
3978                         break;
3979                 }
3980                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3981                                                 blkbits) >> blkbits))
3982                         new_size = offset + len;
3983                 else
3984                         new_size = ((loff_t) map.m_lblk + ret) << blkbits;
3985
3986                 ext4_falloc_update_inode(inode, mode, new_size,
3987                                          (map.m_flags & EXT4_MAP_NEW));
3988                 ext4_mark_inode_dirty(handle, inode);
3989                 ret2 = ext4_journal_stop(handle);
3990                 if (ret2)
3991                         break;
3992         }
3993         if (ret == -ENOSPC &&
3994                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3995                 ret = 0;
3996                 goto retry;
3997         }
3998         mutex_unlock(&inode->i_mutex);
3999         trace_ext4_fallocate_exit(inode, offset, max_blocks,
4000                                 ret > 0 ? ret2 : ret);
4001         return ret > 0 ? ret2 : ret;
4002 }
4003
4004 /*
4005  * This function convert a range of blocks to written extents
4006  * The caller of this function will pass the start offset and the size.
4007  * all unwritten extents within this range will be converted to
4008  * written extents.
4009  *
4010  * This function is called from the direct IO end io call back
4011  * function, to convert the fallocated extents after IO is completed.
4012  * Returns 0 on success.
4013  */
4014 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4015                                     ssize_t len)
4016 {
4017         handle_t *handle;
4018         unsigned int max_blocks;
4019         int ret = 0;
4020         int ret2 = 0;
4021         struct ext4_map_blocks map;
4022         unsigned int credits, blkbits = inode->i_blkbits;
4023
4024         map.m_lblk = offset >> blkbits;
4025         /*
4026          * We can't just convert len to max_blocks because
4027          * If blocksize = 4096 offset = 3072 and len = 2048
4028          */
4029         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4030                       map.m_lblk);
4031         /*
4032          * credits to insert 1 extent into extent tree
4033          */
4034         credits = ext4_chunk_trans_blocks(inode, max_blocks);
4035         while (ret >= 0 && ret < max_blocks) {
4036                 map.m_lblk += ret;
4037                 map.m_len = (max_blocks -= ret);
4038                 handle = ext4_journal_start(inode, credits);
4039                 if (IS_ERR(handle)) {
4040                         ret = PTR_ERR(handle);
4041                         break;
4042                 }
4043                 ret = ext4_map_blocks(handle, inode, &map,
4044                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4045                 if (ret <= 0) {
4046                         WARN_ON(ret <= 0);
4047                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
4048                                     "returned error inode#%lu, block=%u, "
4049                                     "max_blocks=%u", __func__,
4050                                     inode->i_ino, map.m_lblk, map.m_len);
4051                 }
4052                 ext4_mark_inode_dirty(handle, inode);
4053                 ret2 = ext4_journal_stop(handle);
4054                 if (ret <= 0 || ret2 )
4055                         break;
4056         }
4057         return ret > 0 ? ret2 : ret;
4058 }
4059
4060 /*
4061  * Callback function called for each extent to gather FIEMAP information.
4062  */
4063 static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4064                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
4065                        void *data)
4066 {
4067         __u64   logical;
4068         __u64   physical;
4069         __u64   length;
4070         __u32   flags = 0;
4071         int             ret = 0;
4072         struct fiemap_extent_info *fieinfo = data;
4073         unsigned char blksize_bits;
4074
4075         blksize_bits = inode->i_sb->s_blocksize_bits;
4076         logical = (__u64)newex->ec_block << blksize_bits;
4077
4078         if (newex->ec_start == 0) {
4079                 /*
4080                  * No extent in extent-tree contains block @newex->ec_start,
4081                  * then the block may stay in 1)a hole or 2)delayed-extent.
4082                  *
4083                  * Holes or delayed-extents are processed as follows.
4084                  * 1. lookup dirty pages with specified range in pagecache.
4085                  *    If no page is got, then there is no delayed-extent and
4086                  *    return with EXT_CONTINUE.
4087                  * 2. find the 1st mapped buffer,
4088                  * 3. check if the mapped buffer is both in the request range
4089                  *    and a delayed buffer. If not, there is no delayed-extent,
4090                  *    then return.
4091                  * 4. a delayed-extent is found, the extent will be collected.
4092                  */
4093                 ext4_lblk_t     end = 0;
4094                 pgoff_t         last_offset;
4095                 pgoff_t         offset;
4096                 pgoff_t         index;
4097                 pgoff_t         start_index = 0;
4098                 struct page     **pages = NULL;
4099                 struct buffer_head *bh = NULL;
4100                 struct buffer_head *head = NULL;
4101                 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4102
4103                 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4104                 if (pages == NULL)
4105                         return -ENOMEM;
4106
4107                 offset = logical >> PAGE_SHIFT;
4108 repeat:
4109                 last_offset = offset;
4110                 head = NULL;
4111                 ret = find_get_pages_tag(inode->i_mapping, &offset,
4112                                         PAGECACHE_TAG_DIRTY, nr_pages, pages);
4113
4114                 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4115                         /* First time, try to find a mapped buffer. */
4116                         if (ret == 0) {
4117 out:
4118                                 for (index = 0; index < ret; index++)
4119                                         page_cache_release(pages[index]);
4120                                 /* just a hole. */
4121                                 kfree(pages);
4122                                 return EXT_CONTINUE;
4123                         }
4124                         index = 0;
4125
4126 next_page:
4127                         /* Try to find the 1st mapped buffer. */
4128                         end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4129                                   blksize_bits;
4130                         if (!page_has_buffers(pages[index]))
4131                                 goto out;
4132                         head = page_buffers(pages[index]);
4133                         if (!head)
4134                                 goto out;
4135
4136                         index++;
4137                         bh = head;
4138                         do {
4139                                 if (end >= newex->ec_block +
4140                                         newex->ec_len)
4141                                         /* The buffer is out of
4142                                          * the request range.
4143                                          */
4144                                         goto out;
4145
4146                                 if (buffer_mapped(bh) &&
4147                                     end >= newex->ec_block) {
4148                                         start_index = index - 1;
4149                                         /* get the 1st mapped buffer. */
4150                                         goto found_mapped_buffer;
4151                                 }
4152
4153                                 bh = bh->b_this_page;
4154                                 end++;
4155                         } while (bh != head);
4156
4157                         /* No mapped buffer in the range found in this page,
4158                          * We need to look up next page.
4159                          */
4160                         if (index >= ret) {
4161                                 /* There is no page left, but we need to limit
4162                                  * newex->ec_len.
4163                                  */
4164                                 newex->ec_len = end - newex->ec_block;
4165                                 goto out;
4166                         }
4167                         goto next_page;
4168                 } else {
4169                         /*Find contiguous delayed buffers. */
4170                         if (ret > 0 && pages[0]->index == last_offset)
4171                                 head = page_buffers(pages[0]);
4172                         bh = head;
4173                         index = 1;
4174                         start_index = 0;
4175                 }
4176
4177 found_mapped_buffer:
4178                 if (bh != NULL && buffer_delay(bh)) {
4179                         /* 1st or contiguous delayed buffer found. */
4180                         if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4181                                 /*
4182                                  * 1st delayed buffer found, record
4183                                  * the start of extent.
4184                                  */
4185                                 flags |= FIEMAP_EXTENT_DELALLOC;
4186                                 newex->ec_block = end;
4187                                 logical = (__u64)end << blksize_bits;
4188                         }
4189                         /* Find contiguous delayed buffers. */
4190                         do {
4191                                 if (!buffer_delay(bh))
4192                                         goto found_delayed_extent;
4193                                 bh = bh->b_this_page;
4194                                 end++;
4195                         } while (bh != head);
4196
4197                         for (; index < ret; index++) {
4198                                 if (!page_has_buffers(pages[index])) {
4199                                         bh = NULL;
4200                                         break;
4201                                 }
4202                                 head = page_buffers(pages[index]);
4203                                 if (!head) {
4204                                         bh = NULL;
4205                                         break;
4206                                 }
4207
4208                                 if (pages[index]->index !=
4209                                     pages[start_index]->index + index
4210                                     - start_index) {
4211                                         /* Blocks are not contiguous. */
4212                                         bh = NULL;
4213                                         break;
4214                                 }
4215                                 bh = head;
4216                                 do {
4217                                         if (!buffer_delay(bh))
4218                                                 /* Delayed-extent ends. */
4219                                                 goto found_delayed_extent;
4220                                         bh = bh->b_this_page;
4221                                         end++;
4222                                 } while (bh != head);
4223                         }
4224                 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4225                         /* a hole found. */
4226                         goto out;
4227
4228 found_delayed_extent:
4229                 newex->ec_len = min(end - newex->ec_block,
4230                                                 (ext4_lblk_t)EXT_INIT_MAX_LEN);
4231                 if (ret == nr_pages && bh != NULL &&
4232                         newex->ec_len < EXT_INIT_MAX_LEN &&
4233                         buffer_delay(bh)) {
4234                         /* Have not collected an extent and continue. */
4235                         for (index = 0; index < ret; index++)
4236                                 page_cache_release(pages[index]);
4237                         goto repeat;
4238                 }
4239
4240                 for (index = 0; index < ret; index++)
4241                         page_cache_release(pages[index]);
4242                 kfree(pages);
4243         }
4244
4245         physical = (__u64)newex->ec_start << blksize_bits;
4246         length =   (__u64)newex->ec_len << blksize_bits;
4247
4248         if (ex && ext4_ext_is_uninitialized(ex))
4249                 flags |= FIEMAP_EXTENT_UNWRITTEN;
4250
4251         if (next == EXT_MAX_BLOCKS)
4252                 flags |= FIEMAP_EXTENT_LAST;
4253
4254         ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4255                                         length, flags);
4256         if (ret < 0)
4257                 return ret;
4258         if (ret == 1)
4259                 return EXT_BREAK;
4260         return EXT_CONTINUE;
4261 }
4262 /* fiemap flags we can handle specified here */
4263 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4264
4265 static int ext4_xattr_fiemap(struct inode *inode,
4266                                 struct fiemap_extent_info *fieinfo)
4267 {
4268         __u64 physical = 0;
4269         __u64 length;
4270         __u32 flags = FIEMAP_EXTENT_LAST;
4271         int blockbits = inode->i_sb->s_blocksize_bits;
4272         int error = 0;
4273
4274         /* in-inode? */
4275         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4276                 struct ext4_iloc iloc;
4277                 int offset;     /* offset of xattr in inode */
4278
4279                 error = ext4_get_inode_loc(inode, &iloc);
4280                 if (error)
4281                         return error;
4282                 physical = iloc.bh->b_blocknr << blockbits;
4283                 offset = EXT4_GOOD_OLD_INODE_SIZE +
4284                                 EXT4_I(inode)->i_extra_isize;
4285                 physical += offset;
4286                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4287                 flags |= FIEMAP_EXTENT_DATA_INLINE;
4288                 brelse(iloc.bh);
4289         } else { /* external block */
4290                 physical = EXT4_I(inode)->i_file_acl << blockbits;
4291                 length = inode->i_sb->s_blocksize;
4292         }
4293
4294         if (physical)
4295                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4296                                                 length, flags);
4297         return (error < 0 ? error : 0);
4298 }
4299
4300 /*
4301  * ext4_ext_punch_hole
4302  *
4303  * Punches a hole of "length" bytes in a file starting
4304  * at byte "offset"
4305  *
4306  * @inode:  The inode of the file to punch a hole in
4307  * @offset: The starting byte offset of the hole
4308  * @length: The length of the hole
4309  *
4310  * Returns the number of blocks removed or negative on err
4311  */
4312 int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4313 {
4314         struct inode *inode = file->f_path.dentry->d_inode;
4315         struct super_block *sb = inode->i_sb;
4316         struct ext4_ext_cache cache_ex;
4317         ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4318         struct address_space *mapping = inode->i_mapping;
4319         struct ext4_map_blocks map;
4320         handle_t *handle;
4321         loff_t first_page, last_page, page_len;
4322         loff_t first_page_offset, last_page_offset;
4323         int ret, credits, blocks_released, err = 0;
4324
4325         /* No need to punch hole beyond i_size */
4326         if (offset >= inode->i_size)
4327                 return 0;
4328
4329         /*
4330          * If the hole extends beyond i_size, set the hole
4331          * to end after the page that contains i_size
4332          */
4333         if (offset + length > inode->i_size) {
4334                 length = inode->i_size +
4335                    PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4336                    offset;
4337         }
4338
4339         first_block = (offset + sb->s_blocksize - 1) >>
4340                 EXT4_BLOCK_SIZE_BITS(sb);
4341         last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4342
4343         first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4344         last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4345
4346         first_page_offset = first_page << PAGE_CACHE_SHIFT;
4347         last_page_offset = last_page << PAGE_CACHE_SHIFT;
4348
4349         /*
4350          * Write out all dirty pages to avoid race conditions
4351          * Then release them.
4352          */
4353         if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4354                 err = filemap_write_and_wait_range(mapping,
4355                         offset, offset + length - 1);
4356
4357                 if (err)
4358                         return err;
4359         }
4360
4361         /* Now release the pages */
4362         if (last_page_offset > first_page_offset) {
4363                 truncate_inode_pages_range(mapping, first_page_offset,
4364                                            last_page_offset-1);
4365         }
4366
4367         /* finish any pending end_io work */
4368         ext4_flush_completed_IO(inode);
4369
4370         credits = ext4_writepage_trans_blocks(inode);
4371         handle = ext4_journal_start(inode, credits);
4372         if (IS_ERR(handle))
4373                 return PTR_ERR(handle);
4374
4375         err = ext4_orphan_add(handle, inode);
4376         if (err)
4377                 goto out;
4378
4379         /*
4380          * Now we need to zero out the non-page-aligned data in the
4381          * pages at the start and tail of the hole, and unmap the buffer
4382          * heads for the block aligned regions of the page that were
4383          * completely zeroed.
4384          */
4385         if (first_page > last_page) {
4386                 /*
4387                  * If the file space being truncated is contained within a page
4388                  * just zero out and unmap the middle of that page
4389                  */
4390                 err = ext4_discard_partial_page_buffers(handle,
4391                         mapping, offset, length, 0);
4392
4393                 if (err)
4394                         goto out;
4395         } else {
4396                 /*
4397                  * zero out and unmap the partial page that contains
4398                  * the start of the hole
4399                  */
4400                 page_len  = first_page_offset - offset;
4401                 if (page_len > 0) {
4402                         err = ext4_discard_partial_page_buffers(handle, mapping,
4403                                                    offset, page_len, 0);
4404                         if (err)
4405                                 goto out;
4406                 }
4407
4408                 /*
4409                  * zero out and unmap the partial page that contains
4410                  * the end of the hole
4411                  */
4412                 page_len = offset + length - last_page_offset;
4413                 if (page_len > 0) {
4414                         err = ext4_discard_partial_page_buffers(handle, mapping,
4415                                         last_page_offset, page_len, 0);
4416                         if (err)
4417                                 goto out;
4418                 }
4419         }
4420
4421
4422         /*
4423          * If i_size is contained in the last page, we need to
4424          * unmap and zero the partial page after i_size
4425          */
4426         if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4427            inode->i_size % PAGE_CACHE_SIZE != 0) {
4428
4429                 page_len = PAGE_CACHE_SIZE -
4430                         (inode->i_size & (PAGE_CACHE_SIZE - 1));
4431
4432                 if (page_len > 0) {
4433                         err = ext4_discard_partial_page_buffers(handle,
4434                           mapping, inode->i_size, page_len, 0);
4435
4436                         if (err)
4437                                 goto out;
4438                 }
4439         }
4440
4441         /* If there are no blocks to remove, return now */
4442         if (first_block >= last_block)
4443                 goto out;
4444
4445         down_write(&EXT4_I(inode)->i_data_sem);
4446         ext4_ext_invalidate_cache(inode);
4447         ext4_discard_preallocations(inode);
4448
4449         /*
4450          * Loop over all the blocks and identify blocks
4451          * that need to be punched out
4452          */
4453         iblock = first_block;
4454         blocks_released = 0;
4455         while (iblock < last_block) {
4456                 max_blocks = last_block - iblock;
4457                 num_blocks = 1;
4458                 memset(&map, 0, sizeof(map));
4459                 map.m_lblk = iblock;
4460                 map.m_len = max_blocks;
4461                 ret = ext4_ext_map_blocks(handle, inode, &map,
4462                         EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4463
4464                 if (ret > 0) {
4465                         blocks_released += ret;
4466                         num_blocks = ret;
4467                 } else if (ret == 0) {
4468                         /*
4469                          * If map blocks could not find the block,
4470                          * then it is in a hole.  If the hole was
4471                          * not already cached, then map blocks should
4472                          * put it in the cache.  So we can get the hole
4473                          * out of the cache
4474                          */
4475                         memset(&cache_ex, 0, sizeof(cache_ex));
4476                         if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4477                                 !cache_ex.ec_start) {
4478
4479                                 /* The hole is cached */
4480                                 num_blocks = cache_ex.ec_block +
4481                                 cache_ex.ec_len - iblock;
4482
4483                         } else {
4484                                 /* The block could not be identified */
4485                                 err = -EIO;
4486                                 break;
4487                         }
4488                 } else {
4489                         /* Map blocks error */
4490                         err = ret;
4491                         break;
4492                 }
4493
4494                 if (num_blocks == 0) {
4495                         /* This condition should never happen */
4496                         ext_debug("Block lookup failed");
4497                         err = -EIO;
4498                         break;
4499                 }
4500
4501                 iblock += num_blocks;
4502         }
4503
4504         if (blocks_released > 0) {
4505                 ext4_ext_invalidate_cache(inode);
4506                 ext4_discard_preallocations(inode);
4507         }
4508
4509         if (IS_SYNC(inode))
4510                 ext4_handle_sync(handle);
4511
4512         up_write(&EXT4_I(inode)->i_data_sem);
4513
4514 out:
4515         ext4_orphan_del(handle, inode);
4516         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4517         ext4_mark_inode_dirty(handle, inode);
4518         ext4_journal_stop(handle);
4519         return err;
4520 }
4521 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4522                 __u64 start, __u64 len)
4523 {
4524         ext4_lblk_t start_blk;
4525         int error = 0;
4526
4527         /* fallback to generic here if not in extents fmt */
4528         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4529                 return generic_block_fiemap(inode, fieinfo, start, len,
4530                         ext4_get_block);
4531
4532         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4533                 return -EBADR;
4534
4535         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4536                 error = ext4_xattr_fiemap(inode, fieinfo);
4537         } else {
4538                 ext4_lblk_t len_blks;
4539                 __u64 last_blk;
4540
4541                 start_blk = start >> inode->i_sb->s_blocksize_bits;
4542                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4543                 if (last_blk >= EXT_MAX_BLOCKS)
4544                         last_blk = EXT_MAX_BLOCKS-1;
4545                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4546
4547                 /*
4548                  * Walk the extent tree gathering extent information.
4549                  * ext4_ext_fiemap_cb will push extents back to user.
4550                  */
4551                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4552                                           ext4_ext_fiemap_cb, fieinfo);
4553         }
4554
4555         return error;
4556 }