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_ext_truncate_extend_restart(handle_t *handle,
  50                                             struct inode *inode,
  51                                             int needed)
  52 {
  53         int err;
  54
  55         if (!ext4_handle_valid(handle))
  56                 return 0;
  57         if (handle->h_buffer_credits > needed)
  58                 return 0;
  59         err = ext4_journal_extend(handle, needed);
  60         if (err <= 0)
  61                 return err;
  62         err = ext4_truncate_restart_trans(handle, inode, needed);
  63         if (err == 0)
  64                 err = -EAGAIN;
  65
  66         return err;
  67 }
  68
  69 /*
  70  * could return:
  71  *  - EROFS
  72  *  - ENOMEM
  73  */
  74 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  75                                 struct ext4_ext_path *path)
  76 {
  77         if (path->p_bh) {
  78                 /* path points to block */
  79                 return ext4_journal_get_write_access(handle, path->p_bh);
  80         }
  81         /* path points to leaf/index in inode body */
  82         /* we use in-core data, no need to protect them */
  83         return 0;
  84 }
  85
  86 /*
  87  * could return:
  88  *  - EROFS
  89  *  - ENOMEM
  90  *  - EIO
  91  */
  92 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
  93                                 struct ext4_ext_path *path)
  94 {
  95         int err;
  96         if (path->p_bh) {
  97                 /* path points to block */
  98                 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
  99         } else {
 100                 /* path points to leaf/index in inode body */
 101                 err = ext4_mark_inode_dirty(handle, inode);
 102         }
 103         return err;
 104 }
 105
 106 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 107                               struct ext4_ext_path *path,
 108                               ext4_lblk_t block)
 109 {
 110         struct ext4_inode_info *ei = EXT4_I(inode);
 111         ext4_fsblk_t bg_start;
 112         ext4_fsblk_t last_block;
 113         ext4_grpblk_t colour;
 114         ext4_group_t block_group;
 115         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
 116         int depth;
 117
 118         if (path) {
 119                 struct ext4_extent *ex;
 120                 depth = path->p_depth;
 121
 122                 /*
 123                  * Try to predict block placement assuming that we are
 124                  * filling in a file which will eventually be
 125                  * non-sparse --- i.e., in the case of libbfd writing
 126                  * an ELF object sections out-of-order but in a way
 127                  * the eventually results in a contiguous object or
 128                  * executable file, or some database extending a table
 129                  * space file.  However, this is actually somewhat
 130                  * non-ideal if we are writing a sparse file such as
 131                  * qemu or KVM writing a raw image file that is going
 132                  * to stay fairly sparse, since it will end up
 133                  * fragmenting the file system's free space.  Maybe we
 134                  * should have some hueristics or some way to allow
 135                  * userspace to pass a hint to file system,
 136                  * especially if the latter case turns out to be
 137                  * common.
 138                  */
 139                 ex = path[depth].p_ext;
 140                 if (ex) {
 141                         ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 142                         ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 143
 144                         if (block > ext_block)
 145                                 return ext_pblk + (block - ext_block);
 146                         else
 147                                 return ext_pblk - (ext_block - block);
 148                 }
 149
 150                 /* it looks like index is empty;
 151                  * try to find starting block from index itself */
 152                 if (path[depth].p_bh)
 153                         return path[depth].p_bh->b_blocknr;
 154         }
 155
 156         /* OK. use inode's group */
 157         block_group = ei->i_block_group;
 158         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
 159                 /*
 160                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
 161                  * block groups per flexgroup, reserve the first block
 162                  * group for directories and special files.  Regular
 163                  * files will start at the second block group.  This
 164                  * tends to speed up directory access and improves
 165                  * fsck times.
 166                  */
 167                 block_group &= ~(flex_size-1);
 168                 if (S_ISREG(inode->i_mode))
 169                         block_group++;
 170         }
 171         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
 172         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
 173
 174         /*
 175          * If we are doing delayed allocation, we don't need take
 176          * colour into account.
 177          */
 178         if (test_opt(inode->i_sb, DELALLOC))
 179                 return bg_start;
 180
 181         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
 182                 colour = (current->pid % 16) *
 183                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 184         else
 185                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
 186         return bg_start + colour + block;
 187 }
 188
 189 /*
 190  * Allocation for a meta data block
 191  */
 192 static ext4_fsblk_t
 193 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 194                         struct ext4_ext_path *path,
 195                         struct ext4_extent *ex, int *err)
 196 {
 197         ext4_fsblk_t goal, newblock;
 198
 199         goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 200         newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
 201         return newblock;
 202 }
 203
 204 static inline int ext4_ext_space_block(struct inode *inode, int check)
 205 {
 206         int size;
 207
 208         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 209                         / sizeof(struct ext4_extent);
 210         if (!check) {
 211 #ifdef AGGRESSIVE_TEST
 212                 if (size > 6)
 213                         size = 6;
 214 #endif
 215         }
 216         return size;
 217 }
 218
 219 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 220 {
 221         int size;
 222
 223         size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 224                         / sizeof(struct ext4_extent_idx);
 225         if (!check) {
 226 #ifdef AGGRESSIVE_TEST
 227                 if (size > 5)
 228                         size = 5;
 229 #endif
 230         }
 231         return size;
 232 }
 233
 234 static inline int ext4_ext_space_root(struct inode *inode, int check)
 235 {
 236         int size;
 237
 238         size = sizeof(EXT4_I(inode)->i_data);
 239         size -= sizeof(struct ext4_extent_header);
 240         size /= sizeof(struct ext4_extent);
 241         if (!check) {
 242 #ifdef AGGRESSIVE_TEST
 243                 if (size > 3)
 244                         size = 3;
 245 #endif
 246         }
 247         return size;
 248 }
 249
 250 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 251 {
 252         int size;
 253
 254         size = sizeof(EXT4_I(inode)->i_data);
 255         size -= sizeof(struct ext4_extent_header);
 256         size /= sizeof(struct ext4_extent_idx);
 257         if (!check) {
 258 #ifdef AGGRESSIVE_TEST
 259                 if (size > 4)
 260                         size = 4;
 261 #endif
 262         }
 263         return size;
 264 }
 265
 266 /*
 267  * Calculate the number of metadata blocks needed
 268  * to allocate @blocks
 269  * Worse case is one block per extent
 270  */
 271 int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 272 {
 273         struct ext4_inode_info *ei = EXT4_I(inode);
 274         int idxs, num = 0;
 275
 276         idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 277                 / sizeof(struct ext4_extent_idx));
 278
 279         /*
 280          * If the new delayed allocation block is contiguous with the
 281          * previous da block, it can share index blocks with the
 282          * previous block, so we only need to allocate a new index
 283          * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 284          * an additional index block, and at ldxs**3 blocks, yet
 285          * another index blocks.
 286          */
 287         if (ei->i_da_metadata_calc_len &&
 288             ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 289                 if ((ei->i_da_metadata_calc_len % idxs) == 0)
 290                         num++;
 291                 if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 292                         num++;
 293                 if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 294                         num++;
 295                         ei->i_da_metadata_calc_len = 0;
 296                 } else
 297                         ei->i_da_metadata_calc_len++;
 298                 ei->i_da_metadata_calc_last_lblock++;
 299                 return num;
 300         }
 301
 302         /*
 303          * In the worst case we need a new set of index blocks at
 304          * every level of the inode's extent tree.
 305          */
 306         ei->i_da_metadata_calc_len = 1;
 307         ei->i_da_metadata_calc_last_lblock = lblock;
 308         return ext_depth(inode) + 1;
 309 }
 310
 311 static int
 312 ext4_ext_max_entries(struct inode *inode, int depth)
 313 {
 314         int max;
 315
 316         if (depth == ext_depth(inode)) {
 317                 if (depth == 0)
 318                         max = ext4_ext_space_root(inode, 1);
 319                 else
 320                         max = ext4_ext_space_root_idx(inode, 1);
 321         } else {
 322                 if (depth == 0)
 323                         max = ext4_ext_space_block(inode, 1);
 324                 else
 325                         max = ext4_ext_space_block_idx(inode, 1);
 326         }
 327
 328         return max;
 329 }
 330
 331 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 332 {
 333         ext4_fsblk_t block = ext4_ext_pblock(ext);
 334         int len = ext4_ext_get_actual_len(ext);
 335
 336         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 337 }
 338
 339 static int ext4_valid_extent_idx(struct inode *inode,
 340                                 struct ext4_extent_idx *ext_idx)
 341 {
 342         ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 343
 344         return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 345 }
 346
 347 static int ext4_valid_extent_entries(struct inode *inode,
 348                                 struct ext4_extent_header *eh,
 349                                 int depth)
 350 {
 351         struct ext4_extent *ext;
 352         struct ext4_extent_idx *ext_idx;
 353         unsigned short entries;
 354         if (eh->eh_entries == 0)
 355                 return 1;
 356
 357         entries = le16_to_cpu(eh->eh_entries);
 358
 359         if (depth == 0) {
 360                 /* leaf entries */
 361                 ext = EXT_FIRST_EXTENT(eh);
 362                 while (entries) {
 363                         if (!ext4_valid_extent(inode, ext))
 364                                 return 0;
 365                         ext++;
 366                         entries--;
 367                 }
 368         } else {
 369                 ext_idx = EXT_FIRST_INDEX(eh);
 370                 while (entries) {
 371                         if (!ext4_valid_extent_idx(inode, ext_idx))
 372                                 return 0;
 373                         ext_idx++;
 374                         entries--;
 375                 }
 376         }
 377         return 1;
 378 }
 379
 380 static int __ext4_ext_check(const char *function, unsigned int line,
 381                             struct inode *inode, struct ext4_extent_header *eh,
 382                             int depth)
 383 {
 384         const char *error_msg;
 385         int max = 0;
 386
 387         if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 388                 error_msg = "invalid magic";
 389                 goto corrupted;
 390         }
 391         if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 392                 error_msg = "unexpected eh_depth";
 393                 goto corrupted;
 394         }
 395         if (unlikely(eh->eh_max == 0)) {
 396                 error_msg = "invalid eh_max";
 397                 goto corrupted;
 398         }
 399         max = ext4_ext_max_entries(inode, depth);
 400         if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 401                 error_msg = "too large eh_max";
 402                 goto corrupted;
 403         }
 404         if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 405                 error_msg = "invalid eh_entries";
 406                 goto corrupted;
 407         }
 408         if (!ext4_valid_extent_entries(inode, eh, depth)) {
 409                 error_msg = "invalid extent entries";
 410                 goto corrupted;
 411         }
 412         return 0;
 413
 414 corrupted:
 415         ext4_error_inode(inode, function, line, 0,
 416                         "bad header/extent: %s - magic %x, "
 417                         "entries %u, max %u(%u), depth %u(%u)",
 418                         error_msg, le16_to_cpu(eh->eh_magic),
 419                         le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 420                         max, le16_to_cpu(eh->eh_depth), depth);
 421
 422         return -EIO;
 423 }
 424
 425 #define ext4_ext_check(inode, eh, depth)        \
 426         __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 427
 428 int ext4_ext_check_inode(struct inode *inode)
 429 {
 430         return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 431 }
 432
 433 #ifdef EXT_DEBUG
 434 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 435 {
 436         int k, l = path->p_depth;
 437
 438         ext_debug("path:");
 439         for (k = 0; k <= l; k++, path++) {
 440                 if (path->p_idx) {
 441                   ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 442                             ext4_idx_pblock(path->p_idx));
 443                 } else if (path->p_ext) {
 444                         ext_debug("  %d:[%d]%d:%llu ",
 445                                   le32_to_cpu(path->p_ext->ee_block),
 446                                   ext4_ext_is_uninitialized(path->p_ext),
 447                                   ext4_ext_get_actual_len(path->p_ext),
 448                                   ext4_ext_pblock(path->p_ext));
 449                 } else
 450                         ext_debug("  []");
 451         }
 452         ext_debug("\n");
 453 }
 454
 455 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 456 {
 457         int depth = ext_depth(inode);
 458         struct ext4_extent_header *eh;
 459         struct ext4_extent *ex;
 460         int i;
 461
 462         if (!path)
 463                 return;
 464
 465         eh = path[depth].p_hdr;
 466         ex = EXT_FIRST_EXTENT(eh);
 467
 468         ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 469
 470         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 471                 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 472                           ext4_ext_is_uninitialized(ex),
 473                           ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 474         }
 475         ext_debug("\n");
 476 }
 477 #else
 478 #define ext4_ext_show_path(inode, path)
 479 #define ext4_ext_show_leaf(inode, path)
 480 #endif
 481
 482 void ext4_ext_drop_refs(struct ext4_ext_path *path)
 483 {
 484         int depth = path->p_depth;
 485         int i;
 486
 487         for (i = 0; i <= depth; i++, path++)
 488                 if (path->p_bh) {
 489                         brelse(path->p_bh);
 490                         path->p_bh = NULL;
 491                 }
 492 }
 493
 494 /*
 495  * ext4_ext_binsearch_idx:
 496  * binary search for the closest index of the given block
 497  * the header must be checked before calling this
 498  */
 499 static void
 500 ext4_ext_binsearch_idx(struct inode *inode,
 501                         struct ext4_ext_path *path, ext4_lblk_t block)
 502 {
 503         struct ext4_extent_header *eh = path->p_hdr;
 504         struct ext4_extent_idx *r, *l, *m;
 505
 506
 507         ext_debug("binsearch for %u(idx):  ", block);
 508
 509         l = EXT_FIRST_INDEX(eh) + 1;
 510         r = EXT_LAST_INDEX(eh);
 511         while (l <= r) {
 512                 m = l + (r - l) / 2;
 513                 if (block < le32_to_cpu(m->ei_block))
 514                         r = m - 1;
 515                 else
 516                         l = m + 1;
 517                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 518                                 m, le32_to_cpu(m->ei_block),
 519                                 r, le32_to_cpu(r->ei_block));
 520         }
 521
 522         path->p_idx = l - 1;
 523         ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 524                   ext4_idx_pblock(path->p_idx));
 525
 526 #ifdef CHECK_BINSEARCH
 527         {
 528                 struct ext4_extent_idx *chix, *ix;
 529                 int k;
 530
 531                 chix = ix = EXT_FIRST_INDEX(eh);
 532                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 533                   if (k != 0 &&
 534                       le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 535                                 printk(KERN_DEBUG "k=%d, ix=0x%p, "
 536                                        "first=0x%p\n", k,
 537                                        ix, EXT_FIRST_INDEX(eh));
 538                                 printk(KERN_DEBUG "%u <= %u\n",
 539                                        le32_to_cpu(ix->ei_block),
 540                                        le32_to_cpu(ix[-1].ei_block));
 541                         }
 542                         BUG_ON(k && le32_to_cpu(ix->ei_block)
 543                                            <= le32_to_cpu(ix[-1].ei_block));
 544                         if (block < le32_to_cpu(ix->ei_block))
 545                                 break;
 546                         chix = ix;
 547                 }
 548                 BUG_ON(chix != path->p_idx);
 549         }
 550 #endif
 551
 552 }
 553
 554 /*
 555  * ext4_ext_binsearch:
 556  * binary search for closest extent of the given block
 557  * the header must be checked before calling this
 558  */
 559 static void
 560 ext4_ext_binsearch(struct inode *inode,
 561                 struct ext4_ext_path *path, ext4_lblk_t block)
 562 {
 563         struct ext4_extent_header *eh = path->p_hdr;
 564         struct ext4_extent *r, *l, *m;
 565
 566         if (eh->eh_entries == 0) {
 567                 /*
 568                  * this leaf is empty:
 569                  * we get such a leaf in split/add case
 570                  */
 571                 return;
 572         }
 573
 574         ext_debug("binsearch for %u:  ", block);
 575
 576         l = EXT_FIRST_EXTENT(eh) + 1;
 577         r = EXT_LAST_EXTENT(eh);
 578
 579         while (l <= r) {
 580                 m = l + (r - l) / 2;
 581                 if (block < le32_to_cpu(m->ee_block))
 582                         r = m - 1;
 583                 else
 584                         l = m + 1;
 585                 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 586                                 m, le32_to_cpu(m->ee_block),
 587                                 r, le32_to_cpu(r->ee_block));
 588         }
 589
 590         path->p_ext = l - 1;
 591         ext_debug("  -> %d:%llu:[%d]%d ",
 592                         le32_to_cpu(path->p_ext->ee_block),
 593                         ext4_ext_pblock(path->p_ext),
 594                         ext4_ext_is_uninitialized(path->p_ext),
 595                         ext4_ext_get_actual_len(path->p_ext));
 596
 597 #ifdef CHECK_BINSEARCH
 598         {
 599                 struct ext4_extent *chex, *ex;
 600                 int k;
 601
 602                 chex = ex = EXT_FIRST_EXTENT(eh);
 603                 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 604                         BUG_ON(k && le32_to_cpu(ex->ee_block)
 605                                           <= le32_to_cpu(ex[-1].ee_block));
 606                         if (block < le32_to_cpu(ex->ee_block))
 607                                 break;
 608                         chex = ex;
 609                 }
 610                 BUG_ON(chex != path->p_ext);
 611         }
 612 #endif
 613
 614 }
 615
 616 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 617 {
 618         struct ext4_extent_header *eh;
 619
 620         eh = ext_inode_hdr(inode);
 621         eh->eh_depth = 0;
 622         eh->eh_entries = 0;
 623         eh->eh_magic = EXT4_EXT_MAGIC;
 624         eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 625         ext4_mark_inode_dirty(handle, inode);
 626         ext4_ext_invalidate_cache(inode);
 627         return 0;
 628 }
 629
 630 struct ext4_ext_path *
 631 ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 632                                         struct ext4_ext_path *path)
 633 {
 634         struct ext4_extent_header *eh;
 635         struct buffer_head *bh;
 636         short int depth, i, ppos = 0, alloc = 0;
 637
 638         eh = ext_inode_hdr(inode);
 639         depth = ext_depth(inode);
 640
 641         /* account possible depth increase */
 642         if (!path) {
 643                 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 644                                 GFP_NOFS);
 645                 if (!path)
 646                         return ERR_PTR(-ENOMEM);
 647                 alloc = 1;
 648         }
 649         path[0].p_hdr = eh;
 650         path[0].p_bh = NULL;
 651
 652         i = depth;
 653         /* walk through the tree */
 654         while (i) {
 655                 int need_to_validate = 0;
 656
 657                 ext_debug("depth %d: num %d, max %d\n",
 658                           ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 659
 660                 ext4_ext_binsearch_idx(inode, path + ppos, block);
 661                 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 662                 path[ppos].p_depth = i;
 663                 path[ppos].p_ext = NULL;
 664
 665                 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 666                 if (unlikely(!bh))
 667                         goto err;
 668                 if (!bh_uptodate_or_lock(bh)) {
 669                         trace_ext4_ext_load_extent(inode, block,
 670                                                 path[ppos].p_block);
 671                         if (bh_submit_read(bh) < 0) {
 672                                 put_bh(bh);
 673                                 goto err;
 674                         }
 675                         /* validate the extent entries */
 676                         need_to_validate = 1;
 677                 }
 678                 eh = ext_block_hdr(bh);
 679                 ppos++;
 680                 if (unlikely(ppos > depth)) {
 681                         put_bh(bh);
 682                         EXT4_ERROR_INODE(inode,
 683                                          "ppos %d > depth %d", ppos, depth);
 684                         goto err;
 685                 }
 686                 path[ppos].p_bh = bh;
 687                 path[ppos].p_hdr = eh;
 688                 i--;
 689
 690                 if (need_to_validate && ext4_ext_check(inode, eh, i))
 691                         goto err;
 692         }
 693
 694         path[ppos].p_depth = i;
 695         path[ppos].p_ext = NULL;
 696         path[ppos].p_idx = NULL;
 697
 698         /* find extent */
 699         ext4_ext_binsearch(inode, path + ppos, block);
 700         /* if not an empty leaf */
 701         if (path[ppos].p_ext)
 702                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 703
 704         ext4_ext_show_path(inode, path);
 705
 706         return path;
 707
 708 err:
 709         ext4_ext_drop_refs(path);
 710         if (alloc)
 711                 kfree(path);
 712         return ERR_PTR(-EIO);
 713 }
 714
 715 /*
 716  * ext4_ext_insert_index:
 717  * insert new index [@logical;@ptr] into the block at @curp;
 718  * check where to insert: before @curp or after @curp
 719  */
 720 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 721                                  struct ext4_ext_path *curp,
 722                                  int logical, ext4_fsblk_t ptr)
 723 {
 724         struct ext4_extent_idx *ix;
 725         int len, err;
 726
 727         err = ext4_ext_get_access(handle, inode, curp);
 728         if (err)
 729                 return err;
 730
 731         if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 732                 EXT4_ERROR_INODE(inode,
 733                                  "logical %d == ei_block %d!",
 734                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 735                 return -EIO;
 736         }
 737         len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
 738         if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 739                 /* insert after */
 740                 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
 741                         len = (len - 1) * sizeof(struct ext4_extent_idx);
 742                         len = len < 0 ? 0 : len;
 743                         ext_debug("insert new index %d after: %llu. "
 744                                         "move %d from 0x%p to 0x%p\n",
 745                                         logical, ptr, len,
 746                                         (curp->p_idx + 1), (curp->p_idx + 2));
 747                         memmove(curp->p_idx + 2, curp->p_idx + 1, len);
 748                 }
 749                 ix = curp->p_idx + 1;
 750         } else {
 751                 /* insert before */
 752                 len = len * sizeof(struct ext4_extent_idx);
 753                 len = len < 0 ? 0 : len;
 754                 ext_debug("insert new index %d before: %llu. "
 755                                 "move %d from 0x%p to 0x%p\n",
 756                                 logical, ptr, len,
 757                                 curp->p_idx, (curp->p_idx + 1));
 758                 memmove(curp->p_idx + 1, curp->p_idx, len);
 759                 ix = curp->p_idx;
 760         }
 761
 762         ix->ei_block = cpu_to_le32(logical);
 763         ext4_idx_store_pblock(ix, ptr);
 764         le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 765
 766         if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 767                              > le16_to_cpu(curp->p_hdr->eh_max))) {
 768                 EXT4_ERROR_INODE(inode,
 769                                  "logical %d == ei_block %d!",
 770                                  logical, le32_to_cpu(curp->p_idx->ei_block));
 771                 return -EIO;
 772         }
 773         if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 774                 EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 775                 return -EIO;
 776         }
 777
 778         err = ext4_ext_dirty(handle, inode, curp);
 779         ext4_std_error(inode->i_sb, err);
 780
 781         return err;
 782 }
 783
 784 /*
 785  * ext4_ext_split:
 786  * inserts new subtree into the path, using free index entry
 787  * at depth @at:
 788  * - allocates all needed blocks (new leaf and all intermediate index blocks)
 789  * - makes decision where to split
 790  * - moves remaining extents and index entries (right to the split point)
 791  *   into the newly allocated blocks
 792  * - initializes subtree
 793  */
 794 static int ext4_ext_split(handle_t *handle, struct inode *inode,
 795                                 struct ext4_ext_path *path,
 796                                 struct ext4_extent *newext, int at)
 797 {
 798         struct buffer_head *bh = NULL;
 799         int depth = ext_depth(inode);
 800         struct ext4_extent_header *neh;
 801         struct ext4_extent_idx *fidx;
 802         struct ext4_extent *ex;
 803         int i = at, k, m, a;
 804         ext4_fsblk_t newblock, oldblock;
 805         __le32 border;
 806         ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 807         int err = 0;
 808
 809         /* make decision: where to split? */
 810         /* FIXME: now decision is simplest: at current extent */
 811
 812         /* if current leaf will be split, then we should use
 813          * border from split point */
 814         if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 815                 EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 816                 return -EIO;
 817         }
 818         if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 819                 border = path[depth].p_ext[1].ee_block;
 820                 ext_debug("leaf will be split."
 821                                 " next leaf starts at %d\n",
 822                                   le32_to_cpu(border));
 823         } else {
 824                 border = newext->ee_block;
 825                 ext_debug("leaf will be added."
 826                                 " next leaf starts at %d\n",
 827                                 le32_to_cpu(border));
 828         }
 829
 830         /*
 831          * If error occurs, then we break processing
 832          * and mark filesystem read-only. index won't
 833          * be inserted and tree will be in consistent
 834          * state. Next mount will repair buffers too.
 835          */
 836
 837         /*
 838          * Get array to track all allocated blocks.
 839          * We need this to handle errors and free blocks
 840          * upon them.
 841          */
 842         ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 843         if (!ablocks)
 844                 return -ENOMEM;
 845
 846         /* allocate all needed blocks */
 847         ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 848         for (a = 0; a < depth - at; a++) {
 849                 newblock = ext4_ext_new_meta_block(handle, inode, path,
 850                                                    newext, &err);
 851                 if (newblock == 0)
 852                         goto cleanup;
 853                 ablocks[a] = newblock;
 854         }
 855
 856         /* initialize new leaf */
 857         newblock = ablocks[--a];
 858         if (unlikely(newblock == 0)) {
 859                 EXT4_ERROR_INODE(inode, "newblock == 0!");
 860                 err = -EIO;
 861                 goto cleanup;
 862         }
 863         bh = sb_getblk(inode->i_sb, newblock);
 864         if (!bh) {
 865                 err = -EIO;
 866                 goto cleanup;
 867         }
 868         lock_buffer(bh);
 869
 870         err = ext4_journal_get_create_access(handle, bh);
 871         if (err)
 872                 goto cleanup;
 873
 874         neh = ext_block_hdr(bh);
 875         neh->eh_entries = 0;
 876         neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 877         neh->eh_magic = EXT4_EXT_MAGIC;
 878         neh->eh_depth = 0;
 879         ex = EXT_FIRST_EXTENT(neh);
 880
 881         /* move remainder of path[depth] to the new leaf */
 882         if (unlikely(path[depth].p_hdr->eh_entries !=
 883                      path[depth].p_hdr->eh_max)) {
 884                 EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 885                                  path[depth].p_hdr->eh_entries,
 886                                  path[depth].p_hdr->eh_max);
 887                 err = -EIO;
 888                 goto cleanup;
 889         }
 890         /* start copy from next extent */
 891         /* TODO: we could do it by single memmove */
 892         m = 0;
 893         path[depth].p_ext++;
 894         while (path[depth].p_ext <=
 895                         EXT_MAX_EXTENT(path[depth].p_hdr)) {
 896                 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 897                                 le32_to_cpu(path[depth].p_ext->ee_block),
 898                                 ext4_ext_pblock(path[depth].p_ext),
 899                                 ext4_ext_is_uninitialized(path[depth].p_ext),
 900                                 ext4_ext_get_actual_len(path[depth].p_ext),
 901                                 newblock);
 902                 /*memmove(ex++, path[depth].p_ext++,
 903                                 sizeof(struct ext4_extent));
 904                 neh->eh_entries++;*/
 905                 path[depth].p_ext++;
 906                 m++;
 907         }
 908         if (m) {
 909                 memmove(ex, path[depth].p_ext-m, 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                 /* copy indexes */
 972                 m = 0;
 973                 path[i].p_idx++;
 974
 975                 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 976                                 EXT_MAX_INDEX(path[i].p_hdr));
 977                 if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 978                                         EXT_LAST_INDEX(path[i].p_hdr))) {
 979                         EXT4_ERROR_INODE(inode,
 980                                          "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 981                                          le32_to_cpu(path[i].p_ext->ee_block));
 982                         err = -EIO;
 983                         goto cleanup;
 984                 }
 985                 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
 986                         ext_debug("%d: move %d:%llu in new index %llu\n", i,
 987                                         le32_to_cpu(path[i].p_idx->ei_block),
 988                                         ext4_idx_pblock(path[i].p_idx),
 989                                         newblock);
 990                         /*memmove(++fidx, path[i].p_idx++,
 991                                         sizeof(struct ext4_extent_idx));
 992                         neh->eh_entries++;
 993                         BUG_ON(neh->eh_entries > neh->eh_max);*/
 994                         path[i].p_idx++;
 995                         m++;
 996                 }
 997                 if (m) {
 998                         memmove(++fidx, path[i].p_idx - m,
 999                                 sizeof(struct ext4_extent_idx) * m);
1000                         le16_add_cpu(&neh->eh_entries, m);
1001                 }
1002                 set_buffer_uptodate(bh);
1003                 unlock_buffer(bh);
1004
1005                 err = ext4_handle_dirty_metadata(handle, inode, bh);
1006                 if (err)
1007                         goto cleanup;
1008                 brelse(bh);
1009                 bh = NULL;
1010
1011                 /* correct old index */
1012                 if (m) {
1013                         err = ext4_ext_get_access(handle, inode, path + i);
1014                         if (err)
1015                                 goto cleanup;
1016                         le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1017                         err = ext4_ext_dirty(handle, inode, path + i);
1018                         if (err)
1019                                 goto cleanup;
1020                 }
1021
1022                 i--;
1023         }
1024
1025         /* insert new index */
1026         err = ext4_ext_insert_index(handle, inode, path + at,
1027                                     le32_to_cpu(border), newblock);
1028
1029 cleanup:
1030         if (bh) {
1031                 if (buffer_locked(bh))
1032                         unlock_buffer(bh);
1033                 brelse(bh);
1034         }
1035
1036         if (err) {
1037                 /* free all allocated blocks in error case */
1038                 for (i = 0; i < depth; i++) {
1039                         if (!ablocks[i])
1040                                 continue;
1041                         ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1042                                          EXT4_FREE_BLOCKS_METADATA);
1043                 }
1044         }
1045         kfree(ablocks);
1046
1047         return err;
1048 }
1049
1050 /*
1051  * ext4_ext_grow_indepth:
1052  * implements tree growing procedure:
1053  * - allocates new block
1054  * - moves top-level data (index block or leaf) into the new block
1055  * - initializes new top-level, creating index that points to the
1056  *   just created block
1057  */
1058 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1059                                         struct ext4_ext_path *path,
1060                                         struct ext4_extent *newext)
1061 {
1062         struct ext4_ext_path *curp = path;
1063         struct ext4_extent_header *neh;
1064         struct buffer_head *bh;
1065         ext4_fsblk_t newblock;
1066         int err = 0;
1067
1068         newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
1069         if (newblock == 0)
1070                 return err;
1071
1072         bh = sb_getblk(inode->i_sb, newblock);
1073         if (!bh) {
1074                 err = -EIO;
1075                 ext4_std_error(inode->i_sb, err);
1076                 return err;
1077         }
1078         lock_buffer(bh);
1079
1080         err = ext4_journal_get_create_access(handle, bh);
1081         if (err) {
1082                 unlock_buffer(bh);
1083                 goto out;
1084         }
1085
1086         /* move top-level index/leaf into new block */
1087         memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1088
1089         /* set size of new block */
1090         neh = ext_block_hdr(bh);
1091         /* old root could have indexes or leaves
1092          * so calculate e_max right way */
1093         if (ext_depth(inode))
1094                 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1095         else
1096                 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1097         neh->eh_magic = EXT4_EXT_MAGIC;
1098         set_buffer_uptodate(bh);
1099         unlock_buffer(bh);
1100
1101         err = ext4_handle_dirty_metadata(handle, inode, bh);
1102         if (err)
1103                 goto out;
1104
1105         /* create index in new top-level index: num,max,pointer */
1106         err = ext4_ext_get_access(handle, inode, curp);
1107         if (err)
1108                 goto out;
1109
1110         curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1111         curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1112         curp->p_hdr->eh_entries = cpu_to_le16(1);
1113         curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
1114
1115         if (path[0].p_hdr->eh_depth)
1116                 curp->p_idx->ei_block =
1117                         EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1118         else
1119                 curp->p_idx->ei_block =
1120                         EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
1121         ext4_idx_store_pblock(curp->p_idx, newblock);
1122
1123         neh = ext_inode_hdr(inode);
1124         ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1125                   le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1126                   le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1127                   ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1128
1129         neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1130         err = ext4_ext_dirty(handle, inode, curp);
1131 out:
1132         brelse(bh);
1133
1134         return err;
1135 }
1136
1137 /*
1138  * ext4_ext_create_new_leaf:
1139  * finds empty index and adds new leaf.
1140  * if no free index is found, then it requests in-depth growing.
1141  */
1142 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1143                                         struct ext4_ext_path *path,
1144                                         struct ext4_extent *newext)
1145 {
1146         struct ext4_ext_path *curp;
1147         int depth, i, err = 0;
1148
1149 repeat:
1150         i = depth = ext_depth(inode);
1151
1152         /* walk up to the tree and look for free index entry */
1153         curp = path + depth;
1154         while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1155                 i--;
1156                 curp--;
1157         }
1158
1159         /* we use already allocated block for index block,
1160          * so subsequent data blocks should be contiguous */
1161         if (EXT_HAS_FREE_INDEX(curp)) {
1162                 /* if we found index with free entry, then use that
1163                  * entry: create all needed subtree and add new leaf */
1164                 err = ext4_ext_split(handle, inode, path, newext, i);
1165                 if (err)
1166                         goto out;
1167
1168                 /* refill path */
1169                 ext4_ext_drop_refs(path);
1170                 path = ext4_ext_find_extent(inode,
1171                                     (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1172                                     path);
1173                 if (IS_ERR(path))
1174                         err = PTR_ERR(path);
1175         } else {
1176                 /* tree is full, time to grow in depth */
1177                 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1178                 if (err)
1179                         goto out;
1180
1181                 /* refill path */
1182                 ext4_ext_drop_refs(path);
1183                 path = ext4_ext_find_extent(inode,
1184                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1185                                     path);
1186                 if (IS_ERR(path)) {
1187                         err = PTR_ERR(path);
1188                         goto out;
1189                 }
1190
1191                 /*
1192                  * only first (depth 0 -> 1) produces free space;
1193                  * in all other cases we have to split the grown tree
1194                  */
1195                 depth = ext_depth(inode);
1196                 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1197                         /* now we need to split */
1198                         goto repeat;
1199                 }
1200         }
1201
1202 out:
1203         return err;
1204 }
1205
1206 /*
1207  * search the closest allocated block to the left for *logical
1208  * and returns it at @logical + it's physical address at @phys
1209  * if *logical is the smallest allocated block, the function
1210  * returns 0 at @phys
1211  * return value contains 0 (success) or error code
1212  */
1213 static int ext4_ext_search_left(struct inode *inode,
1214                                 struct ext4_ext_path *path,
1215                                 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1216 {
1217         struct ext4_extent_idx *ix;
1218         struct ext4_extent *ex;
1219         int depth, ee_len;
1220
1221         if (unlikely(path == NULL)) {
1222                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1223                 return -EIO;
1224         }
1225         depth = path->p_depth;
1226         *phys = 0;
1227
1228         if (depth == 0 && path->p_ext == NULL)
1229                 return 0;
1230
1231         /* usually extent in the path covers blocks smaller
1232          * then *logical, but it can be that extent is the
1233          * first one in the file */
1234
1235         ex = path[depth].p_ext;
1236         ee_len = ext4_ext_get_actual_len(ex);
1237         if (*logical < le32_to_cpu(ex->ee_block)) {
1238                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1239                         EXT4_ERROR_INODE(inode,
1240                                          "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1241                                          *logical, le32_to_cpu(ex->ee_block));
1242                         return -EIO;
1243                 }
1244                 while (--depth >= 0) {
1245                         ix = path[depth].p_idx;
1246                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1247                                 EXT4_ERROR_INODE(inode,
1248                                   "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1249                                   ix != NULL ? ix->ei_block : 0,
1250                                   EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1251                                     EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block : 0,
1252                                   depth);
1253                                 return -EIO;
1254                         }
1255                 }
1256                 return 0;
1257         }
1258
1259         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1260                 EXT4_ERROR_INODE(inode,
1261                                  "logical %d < ee_block %d + ee_len %d!",
1262                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1263                 return -EIO;
1264         }
1265
1266         *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1267         *phys = ext4_ext_pblock(ex) + ee_len - 1;
1268         return 0;
1269 }
1270
1271 /*
1272  * search the closest allocated block to the right for *logical
1273  * and returns it at @logical + it's physical address at @phys
1274  * if *logical is the smallest allocated block, the function
1275  * returns 0 at @phys
1276  * return value contains 0 (success) or error code
1277  */
1278 static int ext4_ext_search_right(struct inode *inode,
1279                                  struct ext4_ext_path *path,
1280                                  ext4_lblk_t *logical, ext4_fsblk_t *phys)
1281 {
1282         struct buffer_head *bh = NULL;
1283         struct ext4_extent_header *eh;
1284         struct ext4_extent_idx *ix;
1285         struct ext4_extent *ex;
1286         ext4_fsblk_t block;
1287         int depth;      /* Note, NOT eh_depth; depth from top of tree */
1288         int ee_len;
1289
1290         if (unlikely(path == NULL)) {
1291                 EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1292                 return -EIO;
1293         }
1294         depth = path->p_depth;
1295         *phys = 0;
1296
1297         if (depth == 0 && path->p_ext == NULL)
1298                 return 0;
1299
1300         /* usually extent in the path covers blocks smaller
1301          * then *logical, but it can be that extent is the
1302          * first one in the file */
1303
1304         ex = path[depth].p_ext;
1305         ee_len = ext4_ext_get_actual_len(ex);
1306         if (*logical < le32_to_cpu(ex->ee_block)) {
1307                 if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1308                         EXT4_ERROR_INODE(inode,
1309                                          "first_extent(path[%d].p_hdr) != ex",
1310                                          depth);
1311                         return -EIO;
1312                 }
1313                 while (--depth >= 0) {
1314                         ix = path[depth].p_idx;
1315                         if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1316                                 EXT4_ERROR_INODE(inode,
1317                                                  "ix != EXT_FIRST_INDEX *logical %d!",
1318                                                  *logical);
1319                                 return -EIO;
1320                         }
1321                 }
1322                 *logical = le32_to_cpu(ex->ee_block);
1323                 *phys = ext4_ext_pblock(ex);
1324                 return 0;
1325         }
1326
1327         if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1328                 EXT4_ERROR_INODE(inode,
1329                                  "logical %d < ee_block %d + ee_len %d!",
1330                                  *logical, le32_to_cpu(ex->ee_block), ee_len);
1331                 return -EIO;
1332         }
1333
1334         if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1335                 /* next allocated block in this leaf */
1336                 ex++;
1337                 *logical = le32_to_cpu(ex->ee_block);
1338                 *phys = ext4_ext_pblock(ex);
1339                 return 0;
1340         }
1341
1342         /* go up and search for index to the right */
1343         while (--depth >= 0) {
1344                 ix = path[depth].p_idx;
1345                 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1346                         goto got_index;
1347         }
1348
1349         /* we've gone up to the root and found no index to the right */
1350         return 0;
1351
1352 got_index:
1353         /* we've found index to the right, let's
1354          * follow it and find the closest allocated
1355          * block to the right */
1356         ix++;
1357         block = ext4_idx_pblock(ix);
1358         while (++depth < path->p_depth) {
1359                 bh = sb_bread(inode->i_sb, block);
1360                 if (bh == NULL)
1361                         return -EIO;
1362                 eh = ext_block_hdr(bh);
1363                 /* subtract from p_depth to get proper eh_depth */
1364                 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1365                         put_bh(bh);
1366                         return -EIO;
1367                 }
1368                 ix = EXT_FIRST_INDEX(eh);
1369                 block = ext4_idx_pblock(ix);
1370                 put_bh(bh);
1371         }
1372
1373         bh = sb_bread(inode->i_sb, block);
1374         if (bh == NULL)
1375                 return -EIO;
1376         eh = ext_block_hdr(bh);
1377         if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1378                 put_bh(bh);
1379                 return -EIO;
1380         }
1381         ex = EXT_FIRST_EXTENT(eh);
1382         *logical = le32_to_cpu(ex->ee_block);
1383         *phys = ext4_ext_pblock(ex);
1384         put_bh(bh);
1385         return 0;
1386 }
1387
1388 /*
1389  * ext4_ext_next_allocated_block:
1390  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1391  * NOTE: it considers block number from index entry as
1392  * allocated block. Thus, index entries have to be consistent
1393  * with leaves.
1394  */
1395 static ext4_lblk_t
1396 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1397 {
1398         int depth;
1399
1400         BUG_ON(path == NULL);
1401         depth = path->p_depth;
1402
1403         if (depth == 0 && path->p_ext == NULL)
1404                 return EXT_MAX_BLOCK;
1405
1406         while (depth >= 0) {
1407                 if (depth == path->p_depth) {
1408                         /* leaf */
1409                         if (path[depth].p_ext !=
1410                                         EXT_LAST_EXTENT(path[depth].p_hdr))
1411                           return le32_to_cpu(path[depth].p_ext[1].ee_block);
1412                 } else {
1413                         /* index */
1414                         if (path[depth].p_idx !=
1415                                         EXT_LAST_INDEX(path[depth].p_hdr))
1416                           return le32_to_cpu(path[depth].p_idx[1].ei_block);
1417                 }
1418                 depth--;
1419         }
1420
1421         return EXT_MAX_BLOCK;
1422 }
1423
1424 /*
1425  * ext4_ext_next_leaf_block:
1426  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1427  */
1428 static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
1429                                         struct ext4_ext_path *path)
1430 {
1431         int depth;
1432
1433         BUG_ON(path == NULL);
1434         depth = path->p_depth;
1435
1436         /* zero-tree has no leaf blocks at all */
1437         if (depth == 0)
1438                 return EXT_MAX_BLOCK;
1439
1440         /* go to index block */
1441         depth--;
1442
1443         while (depth >= 0) {
1444                 if (path[depth].p_idx !=
1445                                 EXT_LAST_INDEX(path[depth].p_hdr))
1446                         return (ext4_lblk_t)
1447                                 le32_to_cpu(path[depth].p_idx[1].ei_block);
1448                 depth--;
1449         }
1450
1451         return EXT_MAX_BLOCK;
1452 }
1453
1454 /*
1455  * ext4_ext_correct_indexes:
1456  * if leaf gets modified and modified extent is first in the leaf,
1457  * then we have to correct all indexes above.
1458  * TODO: do we need to correct tree in all cases?
1459  */
1460 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1461                                 struct ext4_ext_path *path)
1462 {
1463         struct ext4_extent_header *eh;
1464         int depth = ext_depth(inode);
1465         struct ext4_extent *ex;
1466         __le32 border;
1467         int k, err = 0;
1468
1469         eh = path[depth].p_hdr;
1470         ex = path[depth].p_ext;
1471
1472         if (unlikely(ex == NULL || eh == NULL)) {
1473                 EXT4_ERROR_INODE(inode,
1474                                  "ex %p == NULL or eh %p == NULL", ex, eh);
1475                 return -EIO;
1476         }
1477
1478         if (depth == 0) {
1479                 /* there is no tree at all */
1480                 return 0;
1481         }
1482
1483         if (ex != EXT_FIRST_EXTENT(eh)) {
1484                 /* we correct tree if first leaf got modified only */
1485                 return 0;
1486         }
1487
1488         /*
1489          * TODO: we need correction if border is smaller than current one
1490          */
1491         k = depth - 1;
1492         border = path[depth].p_ext->ee_block;
1493         err = ext4_ext_get_access(handle, inode, path + k);
1494         if (err)
1495                 return err;
1496         path[k].p_idx->ei_block = border;
1497         err = ext4_ext_dirty(handle, inode, path + k);
1498         if (err)
1499                 return err;
1500
1501         while (k--) {
1502                 /* change all left-side indexes */
1503                 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1504                         break;
1505                 err = ext4_ext_get_access(handle, inode, path + k);
1506                 if (err)
1507                         break;
1508                 path[k].p_idx->ei_block = border;
1509                 err = ext4_ext_dirty(handle, inode, path + k);
1510                 if (err)
1511                         break;
1512         }
1513
1514         return err;
1515 }
1516
1517 int
1518 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1519                                 struct ext4_extent *ex2)
1520 {
1521         unsigned short ext1_ee_len, ext2_ee_len, max_len;
1522
1523         /*
1524          * Make sure that either both extents are uninitialized, or
1525          * both are _not_.
1526          */
1527         if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1528                 return 0;
1529
1530         if (ext4_ext_is_uninitialized(ex1))
1531                 max_len = EXT_UNINIT_MAX_LEN;
1532         else
1533                 max_len = EXT_INIT_MAX_LEN;
1534
1535         ext1_ee_len = ext4_ext_get_actual_len(ex1);
1536         ext2_ee_len = ext4_ext_get_actual_len(ex2);
1537
1538         if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1539                         le32_to_cpu(ex2->ee_block))
1540                 return 0;
1541
1542         /*
1543          * To allow future support for preallocated extents to be added
1544          * as an RO_COMPAT feature, refuse to merge to extents if
1545          * this can result in the top bit of ee_len being set.
1546          */
1547         if (ext1_ee_len + ext2_ee_len > max_len)
1548                 return 0;
1549 #ifdef AGGRESSIVE_TEST
1550         if (ext1_ee_len >= 4)
1551                 return 0;
1552 #endif
1553
1554         if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1555                 return 1;
1556         return 0;
1557 }
1558
1559 /*
1560  * This function tries to merge the "ex" extent to the next extent in the tree.
1561  * It always tries to merge towards right. If you want to merge towards
1562  * left, pass "ex - 1" as argument instead of "ex".
1563  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1564  * 1 if they got merged.
1565  */
1566 static int ext4_ext_try_to_merge(struct inode *inode,
1567                                  struct ext4_ext_path *path,
1568                                  struct ext4_extent *ex)
1569 {
1570         struct ext4_extent_header *eh;
1571         unsigned int depth, len;
1572         int merge_done = 0;
1573         int uninitialized = 0;
1574
1575         depth = ext_depth(inode);
1576         BUG_ON(path[depth].p_hdr == NULL);
1577         eh = path[depth].p_hdr;
1578
1579         while (ex < EXT_LAST_EXTENT(eh)) {
1580                 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1581                         break;
1582                 /* merge with next extent! */
1583                 if (ext4_ext_is_uninitialized(ex))
1584                         uninitialized = 1;
1585                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1586                                 + ext4_ext_get_actual_len(ex + 1));
1587                 if (uninitialized)
1588                         ext4_ext_mark_uninitialized(ex);
1589
1590                 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1591                         len = (EXT_LAST_EXTENT(eh) - ex - 1)
1592                                 * sizeof(struct ext4_extent);
1593                         memmove(ex + 1, ex + 2, len);
1594                 }
1595                 le16_add_cpu(&eh->eh_entries, -1);
1596                 merge_done = 1;
1597                 WARN_ON(eh->eh_entries == 0);
1598                 if (!eh->eh_entries)
1599                         EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1600         }
1601
1602         return merge_done;
1603 }
1604
1605 /*
1606  * check if a portion of the "newext" extent overlaps with an
1607  * existing extent.
1608  *
1609  * If there is an overlap discovered, it updates the length of the newext
1610  * such that there will be no overlap, and then returns 1.
1611  * If there is no overlap found, it returns 0.
1612  */
1613 static unsigned int ext4_ext_check_overlap(struct inode *inode,
1614                                            struct ext4_extent *newext,
1615                                            struct ext4_ext_path *path)
1616 {
1617         ext4_lblk_t b1, b2;
1618         unsigned int depth, len1;
1619         unsigned int ret = 0;
1620
1621         b1 = le32_to_cpu(newext->ee_block);
1622         len1 = ext4_ext_get_actual_len(newext);
1623         depth = ext_depth(inode);
1624         if (!path[depth].p_ext)
1625                 goto out;
1626         b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1627
1628         /*
1629          * get the next allocated block if the extent in the path
1630          * is before the requested block(s)
1631          */
1632         if (b2 < b1) {
1633                 b2 = ext4_ext_next_allocated_block(path);
1634                 if (b2 == EXT_MAX_BLOCK)
1635                         goto out;
1636         }
1637
1638         /* check for wrap through zero on extent logical start block*/
1639         if (b1 + len1 < b1) {
1640                 len1 = EXT_MAX_BLOCK - b1;
1641                 newext->ee_len = cpu_to_le16(len1);
1642                 ret = 1;
1643         }
1644
1645         /* check for overlap */
1646         if (b1 + len1 > b2) {
1647                 newext->ee_len = cpu_to_le16(b2 - b1);
1648                 ret = 1;
1649         }
1650 out:
1651         return ret;
1652 }
1653
1654 /*
1655  * ext4_ext_insert_extent:
1656  * tries to merge requsted extent into the existing extent or
1657  * inserts requested extent as new one into the tree,
1658  * creating new leaf in the no-space case.
1659  */
1660 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1661                                 struct ext4_ext_path *path,
1662                                 struct ext4_extent *newext, int flag)
1663 {
1664         struct ext4_extent_header *eh;
1665         struct ext4_extent *ex, *fex;
1666         struct ext4_extent *nearex; /* nearest extent */
1667         struct ext4_ext_path *npath = NULL;
1668         int depth, len, err;
1669         ext4_lblk_t next;
1670         unsigned uninitialized = 0;
1671
1672         if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1673                 EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1674                 return -EIO;
1675         }
1676         depth = ext_depth(inode);
1677         ex = path[depth].p_ext;
1678         if (unlikely(path[depth].p_hdr == NULL)) {
1679                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1680                 return -EIO;
1681         }
1682
1683         /* try to insert block into found extent and return */
1684         if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1685                 && ext4_can_extents_be_merged(inode, ex, newext)) {
1686                 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1687                           ext4_ext_is_uninitialized(newext),
1688                           ext4_ext_get_actual_len(newext),
1689                           le32_to_cpu(ex->ee_block),
1690                           ext4_ext_is_uninitialized(ex),
1691                           ext4_ext_get_actual_len(ex),
1692                           ext4_ext_pblock(ex));
1693                 err = ext4_ext_get_access(handle, inode, path + depth);
1694                 if (err)
1695                         return err;
1696
1697                 /*
1698                  * ext4_can_extents_be_merged should have checked that either
1699                  * both extents are uninitialized, or both aren't. Thus we
1700                  * need to check only one of them here.
1701                  */
1702                 if (ext4_ext_is_uninitialized(ex))
1703                         uninitialized = 1;
1704                 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1705                                         + ext4_ext_get_actual_len(newext));
1706                 if (uninitialized)
1707                         ext4_ext_mark_uninitialized(ex);
1708                 eh = path[depth].p_hdr;
1709                 nearex = ex;
1710                 goto merge;
1711         }
1712
1713 repeat:
1714         depth = ext_depth(inode);
1715         eh = path[depth].p_hdr;
1716         if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1717                 goto has_space;
1718
1719         /* probably next leaf has space for us? */
1720         fex = EXT_LAST_EXTENT(eh);
1721         next = ext4_ext_next_leaf_block(inode, path);
1722         if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1723             && next != EXT_MAX_BLOCK) {
1724                 ext_debug("next leaf block - %d\n", next);
1725                 BUG_ON(npath != NULL);
1726                 npath = ext4_ext_find_extent(inode, next, NULL);
1727                 if (IS_ERR(npath))
1728                         return PTR_ERR(npath);
1729                 BUG_ON(npath->p_depth != path->p_depth);
1730                 eh = npath[depth].p_hdr;
1731                 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1732                         ext_debug("next leaf isnt full(%d)\n",
1733                                   le16_to_cpu(eh->eh_entries));
1734                         path = npath;
1735                         goto repeat;
1736                 }
1737                 ext_debug("next leaf has no free space(%d,%d)\n",
1738                           le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1739         }
1740
1741         /*
1742          * There is no free space in the found leaf.
1743          * We're gonna add a new leaf in the tree.
1744          */
1745         err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1746         if (err)
1747                 goto cleanup;
1748         depth = ext_depth(inode);
1749         eh = path[depth].p_hdr;
1750
1751 has_space:
1752         nearex = path[depth].p_ext;
1753
1754         err = ext4_ext_get_access(handle, inode, path + depth);
1755         if (err)
1756                 goto cleanup;
1757
1758         if (!nearex) {
1759                 /* there is no extent in this leaf, create first one */
1760                 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1761                                 le32_to_cpu(newext->ee_block),
1762                                 ext4_ext_pblock(newext),
1763                                 ext4_ext_is_uninitialized(newext),
1764                                 ext4_ext_get_actual_len(newext));
1765                 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1766         } else if (le32_to_cpu(newext->ee_block)
1767                            > le32_to_cpu(nearex->ee_block)) {
1768 /*              BUG_ON(newext->ee_block == nearex->ee_block); */
1769                 if (nearex != EXT_LAST_EXTENT(eh)) {
1770                         len = EXT_MAX_EXTENT(eh) - nearex;
1771                         len = (len - 1) * sizeof(struct ext4_extent);
1772                         len = len < 0 ? 0 : len;
1773                         ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1774                                         "move %d from 0x%p to 0x%p\n",
1775                                         le32_to_cpu(newext->ee_block),
1776                                         ext4_ext_pblock(newext),
1777                                         ext4_ext_is_uninitialized(newext),
1778                                         ext4_ext_get_actual_len(newext),
1779                                         nearex, len, nearex + 1, nearex + 2);
1780                         memmove(nearex + 2, nearex + 1, len);
1781                 }
1782                 path[depth].p_ext = nearex + 1;
1783         } else {
1784                 BUG_ON(newext->ee_block == nearex->ee_block);
1785                 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1786                 len = len < 0 ? 0 : len;
1787                 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1788                                 "move %d from 0x%p to 0x%p\n",
1789                                 le32_to_cpu(newext->ee_block),
1790                                 ext4_ext_pblock(newext),
1791                                 ext4_ext_is_uninitialized(newext),
1792                                 ext4_ext_get_actual_len(newext),
1793                                 nearex, len, nearex + 1, nearex + 2);
1794                 memmove(nearex + 1, nearex, len);
1795                 path[depth].p_ext = nearex;
1796         }
1797
1798         le16_add_cpu(&eh->eh_entries, 1);
1799         nearex = path[depth].p_ext;
1800         nearex->ee_block = newext->ee_block;
1801         ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1802         nearex->ee_len = newext->ee_len;
1803
1804 merge:
1805         /* try to merge extents to the right */
1806         if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1807                 ext4_ext_try_to_merge(inode, path, nearex);
1808
1809         /* try to merge extents to the left */
1810
1811         /* time to correct all indexes above */
1812         err = ext4_ext_correct_indexes(handle, inode, path);
1813         if (err)
1814                 goto cleanup;
1815
1816         err = ext4_ext_dirty(handle, inode, path + depth);
1817
1818 cleanup:
1819         if (npath) {
1820                 ext4_ext_drop_refs(npath);
1821                 kfree(npath);
1822         }
1823         ext4_ext_invalidate_cache(inode);
1824         return err;
1825 }
1826
1827 static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1828                                ext4_lblk_t num, ext_prepare_callback func,
1829                                void *cbdata)
1830 {
1831         struct ext4_ext_path *path = NULL;
1832         struct ext4_ext_cache cbex;
1833         struct ext4_extent *ex;
1834         ext4_lblk_t next, start = 0, end = 0;
1835         ext4_lblk_t last = block + num;
1836         int depth, exists, err = 0;
1837
1838         BUG_ON(func == NULL);
1839         BUG_ON(inode == NULL);
1840
1841         while (block < last && block != EXT_MAX_BLOCK) {
1842                 num = last - block;
1843                 /* find extent for this block */
1844                 down_read(&EXT4_I(inode)->i_data_sem);
1845                 path = ext4_ext_find_extent(inode, block, path);
1846                 up_read(&EXT4_I(inode)->i_data_sem);
1847                 if (IS_ERR(path)) {
1848                         err = PTR_ERR(path);
1849                         path = NULL;
1850                         break;
1851                 }
1852
1853                 depth = ext_depth(inode);
1854                 if (unlikely(path[depth].p_hdr == NULL)) {
1855                         EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1856                         err = -EIO;
1857                         break;
1858                 }
1859                 ex = path[depth].p_ext;
1860                 next = ext4_ext_next_allocated_block(path);
1861
1862                 exists = 0;
1863                 if (!ex) {
1864                         /* there is no extent yet, so try to allocate
1865                          * all requested space */
1866                         start = block;
1867                         end = block + num;
1868                 } else if (le32_to_cpu(ex->ee_block) > block) {
1869                         /* need to allocate space before found extent */
1870                         start = block;
1871                         end = le32_to_cpu(ex->ee_block);
1872                         if (block + num < end)
1873                                 end = block + num;
1874                 } else if (block >= le32_to_cpu(ex->ee_block)
1875                                         + ext4_ext_get_actual_len(ex)) {
1876                         /* need to allocate space after found extent */
1877                         start = block;
1878                         end = block + num;
1879                         if (end >= next)
1880                                 end = next;
1881                 } else if (block >= le32_to_cpu(ex->ee_block)) {
1882                         /*
1883                          * some part of requested space is covered
1884                          * by found extent
1885                          */
1886                         start = block;
1887                         end = le32_to_cpu(ex->ee_block)
1888                                 + ext4_ext_get_actual_len(ex);
1889                         if (block + num < end)
1890                                 end = block + num;
1891                         exists = 1;
1892                 } else {
1893                         BUG();
1894                 }
1895                 BUG_ON(end <= start);
1896
1897                 if (!exists) {
1898                         cbex.ec_block = start;
1899                         cbex.ec_len = end - start;
1900                         cbex.ec_start = 0;
1901                 } else {
1902                         cbex.ec_block = le32_to_cpu(ex->ee_block);
1903                         cbex.ec_len = ext4_ext_get_actual_len(ex);
1904                         cbex.ec_start = ext4_ext_pblock(ex);
1905                 }
1906
1907                 if (unlikely(cbex.ec_len == 0)) {
1908                         EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1909                         err = -EIO;
1910                         break;
1911                 }
1912                 err = func(inode, path, &cbex, ex, cbdata);
1913                 ext4_ext_drop_refs(path);
1914
1915                 if (err < 0)
1916                         break;
1917
1918                 if (err == EXT_REPEAT)
1919                         continue;
1920                 else if (err == EXT_BREAK) {
1921                         err = 0;
1922                         break;
1923                 }
1924
1925                 if (ext_depth(inode) != depth) {
1926                         /* depth was changed. we have to realloc path */
1927                         kfree(path);
1928                         path = NULL;
1929                 }
1930
1931                 block = cbex.ec_block + cbex.ec_len;
1932         }
1933
1934         if (path) {
1935                 ext4_ext_drop_refs(path);
1936                 kfree(path);
1937         }
1938
1939         return err;
1940 }
1941
1942 static void
1943 ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1944                         __u32 len, ext4_fsblk_t start)
1945 {
1946         struct ext4_ext_cache *cex;
1947         BUG_ON(len == 0);
1948         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1949         cex = &EXT4_I(inode)->i_cached_extent;
1950         cex->ec_block = block;
1951         cex->ec_len = len;
1952         cex->ec_start = start;
1953         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1954 }
1955
1956 /*
1957  * ext4_ext_put_gap_in_cache:
1958  * calculate boundaries of the gap that the requested block fits into
1959  * and cache this gap
1960  */
1961 static void
1962 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1963                                 ext4_lblk_t block)
1964 {
1965         int depth = ext_depth(inode);
1966         unsigned long len;
1967         ext4_lblk_t lblock;
1968         struct ext4_extent *ex;
1969
1970         ex = path[depth].p_ext;
1971         if (ex == NULL) {
1972                 /* there is no extent yet, so gap is [0;-] */
1973                 lblock = 0;
1974                 len = EXT_MAX_BLOCK;
1975                 ext_debug("cache gap(whole file):");
1976         } else if (block < le32_to_cpu(ex->ee_block)) {
1977                 lblock = block;
1978                 len = le32_to_cpu(ex->ee_block) - block;
1979                 ext_debug("cache gap(before): %u [%u:%u]",
1980                                 block,
1981                                 le32_to_cpu(ex->ee_block),
1982                                  ext4_ext_get_actual_len(ex));
1983         } else if (block >= le32_to_cpu(ex->ee_block)
1984                         + ext4_ext_get_actual_len(ex)) {
1985                 ext4_lblk_t next;
1986                 lblock = le32_to_cpu(ex->ee_block)
1987                         + ext4_ext_get_actual_len(ex);
1988
1989                 next = ext4_ext_next_allocated_block(path);
1990                 ext_debug("cache gap(after): [%u:%u] %u",
1991                                 le32_to_cpu(ex->ee_block),
1992                                 ext4_ext_get_actual_len(ex),
1993                                 block);
1994                 BUG_ON(next == lblock);
1995                 len = next - lblock;
1996         } else {
1997                 lblock = len = 0;
1998                 BUG();
1999         }
2000
2001         ext_debug(" -> %u:%lu\n", lblock, len);
2002         ext4_ext_put_in_cache(inode, lblock, len, 0);
2003 }
2004
2005 /*
2006  * Return 0 if cache is invalid; 1 if the cache is valid
2007  */
2008 static int
2009 ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2010                         struct ext4_extent *ex)
2011 {
2012         struct ext4_ext_cache *cex;
2013         int ret = 0;
2014
2015         /*
2016          * We borrow i_block_reservation_lock to protect i_cached_extent
2017          */
2018         spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2019         cex = &EXT4_I(inode)->i_cached_extent;
2020
2021         /* has cache valid data? */
2022         if (cex->ec_len == 0)
2023                 goto errout;
2024
2025         if (in_range(block, cex->ec_block, cex->ec_len)) {
2026                 ex->ee_block = cpu_to_le32(cex->ec_block);
2027                 ext4_ext_store_pblock(ex, cex->ec_start);
2028                 ex->ee_len = cpu_to_le16(cex->ec_len);
2029                 ext_debug("%u cached by %u:%u:%llu\n",
2030                                 block,
2031                                 cex->ec_block, cex->ec_len, cex->ec_start);
2032                 ret = 1;
2033         }
2034 errout:
2035         spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2036         return ret;
2037 }
2038
2039 /*
2040  * ext4_ext_rm_idx:
2041  * removes index from the index block.
2042  * It's used in truncate case only, thus all requests are for
2043  * last index in the block only.
2044  */
2045 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2046                         struct ext4_ext_path *path)
2047 {
2048         int err;
2049         ext4_fsblk_t leaf;
2050
2051         /* free index block */
2052         path--;
2053         leaf = ext4_idx_pblock(path->p_idx);
2054         if (unlikely(path->p_hdr->eh_entries == 0)) {
2055                 EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2056                 return -EIO;
2057         }
2058         err = ext4_ext_get_access(handle, inode, path);
2059         if (err)
2060                 return err;
2061         le16_add_cpu(&path->p_hdr->eh_entries, -1);
2062         err = ext4_ext_dirty(handle, inode, path);
2063         if (err)
2064                 return err;
2065         ext_debug("index is empty, remove it, free block %llu\n", leaf);
2066         ext4_free_blocks(handle, inode, NULL, leaf, 1,
2067                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2068         return err;
2069 }
2070
2071 /*
2072  * ext4_ext_calc_credits_for_single_extent:
2073  * This routine returns max. credits that needed to insert an extent
2074  * to the extent tree.
2075  * When pass the actual path, the caller should calculate credits
2076  * under i_data_sem.
2077  */
2078 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2079                                                 struct ext4_ext_path *path)
2080 {
2081         if (path) {
2082                 int depth = ext_depth(inode);
2083                 int ret = 0;
2084
2085                 /* probably there is space in leaf? */
2086                 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2087                                 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2088
2089                         /*
2090                          *  There are some space in the leaf tree, no
2091                          *  need to account for leaf block credit
2092                          *
2093                          *  bitmaps and block group descriptor blocks
2094                          *  and other metadat blocks still need to be
2095                          *  accounted.
2096                          */
2097                         /* 1 bitmap, 1 block group descriptor */
2098                         ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2099                         return ret;
2100                 }
2101         }
2102
2103         return ext4_chunk_trans_blocks(inode, nrblocks);
2104 }
2105
2106 /*
2107  * How many index/leaf blocks need to change/allocate to modify nrblocks?
2108  *
2109  * if nrblocks are fit in a single extent (chunk flag is 1), then
2110  * in the worse case, each tree level index/leaf need to be changed
2111  * if the tree split due to insert a new extent, then the old tree
2112  * index/leaf need to be updated too
2113  *
2114  * If the nrblocks are discontiguous, they could cause
2115  * the whole tree split more than once, but this is really rare.
2116  */
2117 int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2118 {
2119         int index;
2120         int depth = ext_depth(inode);
2121
2122         if (chunk)
2123                 index = depth * 2;
2124         else
2125                 index = depth * 3;
2126
2127         return index;
2128 }
2129
2130 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2131                                 struct ext4_extent *ex,
2132                                 ext4_lblk_t from, ext4_lblk_t to)
2133 {
2134         unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2135         int flags = EXT4_FREE_BLOCKS_FORGET;
2136
2137         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2138                 flags |= EXT4_FREE_BLOCKS_METADATA;
2139 #ifdef EXTENTS_STATS
2140         {
2141                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2142                 spin_lock(&sbi->s_ext_stats_lock);
2143                 sbi->s_ext_blocks += ee_len;
2144                 sbi->s_ext_extents++;
2145                 if (ee_len < sbi->s_ext_min)
2146                         sbi->s_ext_min = ee_len;
2147                 if (ee_len > sbi->s_ext_max)
2148                         sbi->s_ext_max = ee_len;
2149                 if (ext_depth(inode) > sbi->s_depth_max)
2150                         sbi->s_depth_max = ext_depth(inode);
2151                 spin_unlock(&sbi->s_ext_stats_lock);
2152         }
2153 #endif
2154         if (from >= le32_to_cpu(ex->ee_block)
2155             && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2156                 /* tail removal */
2157                 ext4_lblk_t num;
2158                 ext4_fsblk_t start;
2159
2160                 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2161                 start = ext4_ext_pblock(ex) + ee_len - num;
2162                 ext_debug("free last %u blocks starting %llu\n", num, start);
2163                 ext4_free_blocks(handle, inode, NULL, start, num, flags);
2164         } else if (from == le32_to_cpu(ex->ee_block)
2165                    && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2166                 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
2167                         from, to, le32_to_cpu(ex->ee_block), ee_len);
2168         } else {
2169                 printk(KERN_INFO "strange request: removal(2) "
2170                                 "%u-%u from %u:%u\n",
2171                                 from, to, le32_to_cpu(ex->ee_block), ee_len);
2172         }
2173         return 0;
2174 }
2175
2176 static int
2177 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2178                 struct ext4_ext_path *path, ext4_lblk_t start)
2179 {
2180         int err = 0, correct_index = 0;
2181         int depth = ext_depth(inode), credits;
2182         struct ext4_extent_header *eh;
2183         ext4_lblk_t a, b, block;
2184         unsigned num;
2185         ext4_lblk_t ex_ee_block;
2186         unsigned short ex_ee_len;
2187         unsigned uninitialized = 0;
2188         struct ext4_extent *ex;
2189
2190         /* the header must be checked already in ext4_ext_remove_space() */
2191         ext_debug("truncate since %u in leaf\n", start);
2192         if (!path[depth].p_hdr)
2193                 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2194         eh = path[depth].p_hdr;
2195         if (unlikely(path[depth].p_hdr == NULL)) {
2196                 EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2197                 return -EIO;
2198         }
2199         /* find where to start removing */
2200         ex = EXT_LAST_EXTENT(eh);
2201
2202         ex_ee_block = le32_to_cpu(ex->ee_block);
2203         ex_ee_len = ext4_ext_get_actual_len(ex);
2204
2205         while (ex >= EXT_FIRST_EXTENT(eh) &&
2206                         ex_ee_block + ex_ee_len > start) {
2207
2208                 if (ext4_ext_is_uninitialized(ex))
2209                         uninitialized = 1;
2210                 else
2211                         uninitialized = 0;
2212
2213                 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2214                          uninitialized, ex_ee_len);
2215                 path[depth].p_ext = ex;
2216
2217                 a = ex_ee_block > start ? ex_ee_block : start;
2218                 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2219                         ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2220
2221                 ext_debug("  border %u:%u\n", a, b);
2222
2223                 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2224                         block = 0;
2225                         num = 0;
2226                         BUG();
2227                 } else if (a != ex_ee_block) {
2228                         /* remove tail of the extent */
2229                         block = ex_ee_block;
2230                         num = a - block;
2231                 } else if (b != ex_ee_block + ex_ee_len - 1) {
2232                         /* remove head of the extent */
2233                         block = a;
2234                         num = b - a;
2235                         /* there is no "make a hole" API yet */
2236                         BUG();
2237                 } else {
2238                         /* remove whole extent: excellent! */
2239                         block = ex_ee_block;
2240                         num = 0;
2241                         BUG_ON(a != ex_ee_block);
2242                         BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2243                 }
2244
2245                 /*
2246                  * 3 for leaf, sb, and inode plus 2 (bmap and group
2247                  * descriptor) for each block group; assume two block
2248                  * groups plus ex_ee_len/blocks_per_block_group for
2249                  * the worst case
2250                  */
2251                 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2252                 if (ex == EXT_FIRST_EXTENT(eh)) {
2253                         correct_index = 1;
2254                         credits += (ext_depth(inode)) + 1;
2255                 }
2256                 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2257
2258                 err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2259                 if (err)
2260                         goto out;
2261
2262                 err = ext4_ext_get_access(handle, inode, path + depth);
2263                 if (err)
2264                         goto out;
2265
2266                 err = ext4_remove_blocks(handle, inode, ex, a, b);
2267                 if (err)
2268                         goto out;
2269
2270                 if (num == 0) {
2271                         /* this extent is removed; mark slot entirely unused */
2272                         ext4_ext_store_pblock(ex, 0);
2273                         le16_add_cpu(&eh->eh_entries, -1);
2274                 }
2275
2276                 ex->ee_block = cpu_to_le32(block);
2277                 ex->ee_len = cpu_to_le16(num);
2278                 /*
2279                  * Do not mark uninitialized if all the blocks in the
2280                  * extent have been removed.
2281                  */
2282                 if (uninitialized && num)
2283                         ext4_ext_mark_uninitialized(ex);
2284
2285                 err = ext4_ext_dirty(handle, inode, path + depth);
2286                 if (err)
2287                         goto out;
2288
2289                 ext_debug("new extent: %u:%u:%llu\n", block, num,
2290                                 ext4_ext_pblock(ex));
2291                 ex--;
2292                 ex_ee_block = le32_to_cpu(ex->ee_block);
2293                 ex_ee_len = ext4_ext_get_actual_len(ex);
2294         }
2295
2296         if (correct_index && eh->eh_entries)
2297                 err = ext4_ext_correct_indexes(handle, inode, path);
2298
2299         /* if this leaf is free, then we should
2300          * remove it from index block above */
2301         if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2302                 err = ext4_ext_rm_idx(handle, inode, path + depth);
2303
2304 out:
2305         return err;
2306 }
2307
2308 /*
2309  * ext4_ext_more_to_rm:
2310  * returns 1 if current index has to be freed (even partial)
2311  */
2312 static int
2313 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2314 {
2315         BUG_ON(path->p_idx == NULL);
2316
2317         if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2318                 return 0;
2319
2320         /*
2321          * if truncate on deeper level happened, it wasn't partial,
2322          * so we have to consider current index for truncation
2323          */
2324         if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2325                 return 0;
2326         return 1;
2327 }
2328
2329 static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2330 {
2331         struct super_block *sb = inode->i_sb;
2332         int depth = ext_depth(inode);
2333         struct ext4_ext_path *path;
2334         handle_t *handle;
2335         int i, err;
2336
2337         ext_debug("truncate since %u\n", start);
2338
2339         /* probably first extent we're gonna free will be last in block */
2340         handle = ext4_journal_start(inode, depth + 1);
2341         if (IS_ERR(handle))
2342                 return PTR_ERR(handle);
2343
2344 again:
2345         ext4_ext_invalidate_cache(inode);
2346
2347         /*
2348          * We start scanning from right side, freeing all the blocks
2349          * after i_size and walking into the tree depth-wise.
2350          */
2351         depth = ext_depth(inode);
2352         path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2353         if (path == NULL) {
2354                 ext4_journal_stop(handle);
2355                 return -ENOMEM;
2356         }
2357         path[0].p_depth = depth;
2358         path[0].p_hdr = ext_inode_hdr(inode);
2359         if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2360                 err = -EIO;
2361                 goto out;
2362         }
2363         i = err = 0;
2364
2365         while (i >= 0 && err == 0) {
2366                 if (i == depth) {
2367                         /* this is leaf block */
2368                         err = ext4_ext_rm_leaf(handle, inode, path, start);
2369                         /* root level has p_bh == NULL, brelse() eats this */
2370                         brelse(path[i].p_bh);
2371                         path[i].p_bh = NULL;
2372                         i--;
2373                         continue;
2374                 }
2375
2376                 /* this is index block */
2377                 if (!path[i].p_hdr) {
2378                         ext_debug("initialize header\n");
2379                         path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2380                 }
2381
2382                 if (!path[i].p_idx) {
2383                         /* this level hasn't been touched yet */
2384                         path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2385                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2386                         ext_debug("init index ptr: hdr 0x%p, num %d\n",
2387                                   path[i].p_hdr,
2388                                   le16_to_cpu(path[i].p_hdr->eh_entries));
2389                 } else {
2390                         /* we were already here, see at next index */
2391                         path[i].p_idx--;
2392                 }
2393
2394                 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2395                                 i, EXT_FIRST_INDEX(path[i].p_hdr),
2396                                 path[i].p_idx);
2397                 if (ext4_ext_more_to_rm(path + i)) {
2398                         struct buffer_head *bh;
2399                         /* go to the next level */
2400                         ext_debug("move to level %d (block %llu)\n",
2401                                   i + 1, ext4_idx_pblock(path[i].p_idx));
2402                         memset(path + i + 1, 0, sizeof(*path));
2403                         bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2404                         if (!bh) {
2405                                 /* should we reset i_size? */
2406                                 err = -EIO;
2407                                 break;
2408                         }
2409                         if (WARN_ON(i + 1 > depth)) {
2410                                 err = -EIO;
2411                                 break;
2412                         }
2413                         if (ext4_ext_check(inode, ext_block_hdr(bh),
2414                                                         depth - i - 1)) {
2415                                 err = -EIO;
2416                                 break;
2417                         }
2418                         path[i + 1].p_bh = bh;
2419
2420                         /* save actual number of indexes since this
2421                          * number is changed at the next iteration */
2422                         path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2423                         i++;
2424                 } else {
2425                         /* we finished processing this index, go up */
2426                         if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2427                                 /* index is empty, remove it;
2428                                  * handle must be already prepared by the
2429                                  * truncatei_leaf() */
2430                                 err = ext4_ext_rm_idx(handle, inode, path + i);
2431                         }
2432                         /* root level has p_bh == NULL, brelse() eats this */
2433                         brelse(path[i].p_bh);
2434                         path[i].p_bh = NULL;
2435                         i--;
2436                         ext_debug("return to level %d\n", i);
2437                 }
2438         }
2439
2440         /* TODO: flexible tree reduction should be here */
2441         if (path->p_hdr->eh_entries == 0) {
2442                 /*
2443                  * truncate to zero freed all the tree,
2444                  * so we need to correct eh_depth
2445                  */
2446                 err = ext4_ext_get_access(handle, inode, path);
2447                 if (err == 0) {
2448                         ext_inode_hdr(inode)->eh_depth = 0;
2449                         ext_inode_hdr(inode)->eh_max =
2450                                 cpu_to_le16(ext4_ext_space_root(inode, 0));
2451                         err = ext4_ext_dirty(handle, inode, path);
2452                 }
2453         }
2454 out:
2455         ext4_ext_drop_refs(path);
2456         kfree(path);
2457         if (err == -EAGAIN)
2458                 goto again;
2459         ext4_journal_stop(handle);
2460
2461         return err;
2462 }
2463
2464 /*
2465  * called at mount time
2466  */
2467 void ext4_ext_init(struct super_block *sb)
2468 {
2469         /*
2470          * possible initialization would be here
2471          */
2472
2473         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2474 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2475                 printk(KERN_INFO "EXT4-fs: file extents enabled");
2476 #ifdef AGGRESSIVE_TEST
2477                 printk(", aggressive tests");
2478 #endif
2479 #ifdef CHECK_BINSEARCH
2480                 printk(", check binsearch");
2481 #endif
2482 #ifdef EXTENTS_STATS
2483                 printk(", stats");
2484 #endif
2485                 printk("\n");
2486 #endif
2487 #ifdef EXTENTS_STATS
2488                 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2489                 EXT4_SB(sb)->s_ext_min = 1 << 30;
2490                 EXT4_SB(sb)->s_ext_max = 0;
2491 #endif
2492         }
2493 }
2494
2495 /*
2496  * called at umount time
2497  */
2498 void ext4_ext_release(struct super_block *sb)
2499 {
2500         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2501                 return;
2502
2503 #ifdef EXTENTS_STATS
2504         if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2505                 struct ext4_sb_info *sbi = EXT4_SB(sb);
2506                 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2507                         sbi->s_ext_blocks, sbi->s_ext_extents,
2508                         sbi->s_ext_blocks / sbi->s_ext_extents);
2509                 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2510                         sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2511         }
2512 #endif
2513 }
2514
2515 /* FIXME!! we need to try to merge to left or right after zero-out  */
2516 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2517 {
2518         ext4_fsblk_t ee_pblock;
2519         unsigned int ee_len;
2520         int ret;
2521
2522         ee_len    = ext4_ext_get_actual_len(ex);
2523         ee_pblock = ext4_ext_pblock(ex);
2524
2525         ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2526         if (ret > 0)
2527                 ret = 0;
2528
2529         return ret;
2530 }
2531
2532 #define EXT4_EXT_ZERO_LEN 7
2533 /*
2534  * This function is called by ext4_ext_map_blocks() if someone tries to write
2535  * to an uninitialized extent. It may result in splitting the uninitialized
2536  * extent into multiple extents (upto three - one initialized and two
2537  * uninitialized).
2538  * There are three possibilities:
2539  *   a> There is no split required: Entire extent should be initialized
2540  *   b> Splits in two extents: Write is happening at either end of the extent
2541  *   c> Splits in three extents: Somone is writing in middle of the extent
2542  */
2543 static int ext4_ext_convert_to_initialized(handle_t *handle,
2544                                            struct inode *inode,
2545                                            struct ext4_map_blocks *map,
2546                                            struct ext4_ext_path *path)
2547 {
2548         struct ext4_extent *ex, newex, orig_ex;
2549         struct ext4_extent *ex1 = NULL;
2550         struct ext4_extent *ex2 = NULL;
2551         struct ext4_extent *ex3 = NULL;
2552         struct ext4_extent_header *eh;
2553         ext4_lblk_t ee_block, eof_block;
2554         unsigned int allocated, ee_len, depth;
2555         ext4_fsblk_t newblock;
2556         int err = 0;
2557         int ret = 0;
2558         int may_zeroout;
2559
2560         ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2561                 "block %llu, max_blocks %u\n", inode->i_ino,
2562                 (unsigned long long)map->m_lblk, map->m_len);
2563
2564         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2565                 inode->i_sb->s_blocksize_bits;
2566         if (eof_block < map->m_lblk + map->m_len)
2567                 eof_block = map->m_lblk + map->m_len;
2568
2569         depth = ext_depth(inode);
2570         eh = path[depth].p_hdr;
2571         ex = path[depth].p_ext;
2572         ee_block = le32_to_cpu(ex->ee_block);
2573         ee_len = ext4_ext_get_actual_len(ex);
2574         allocated = ee_len - (map->m_lblk - ee_block);
2575         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2576
2577         ex2 = ex;
2578         orig_ex.ee_block = ex->ee_block;
2579         orig_ex.ee_len   = cpu_to_le16(ee_len);
2580         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2581
2582         /*
2583          * It is safe to convert extent to initialized via explicit
2584          * zeroout only if extent is fully insde i_size or new_size.
2585          */
2586         may_zeroout = ee_block + ee_len <= eof_block;
2587
2588         err = ext4_ext_get_access(handle, inode, path + depth);
2589         if (err)
2590                 goto out;
2591         /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2592         if (ee_len <= 2*EXT4_EXT_ZERO_LEN && may_zeroout) {
2593                 err =  ext4_ext_zeroout(inode, &orig_ex);
2594                 if (err)
2595                         goto fix_extent_len;
2596                 /* update the extent length and mark as initialized */
2597                 ex->ee_block = orig_ex.ee_block;
2598                 ex->ee_len   = orig_ex.ee_len;
2599                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2600                 ext4_ext_dirty(handle, inode, path + depth);
2601                 /* zeroed the full extent */
2602                 return allocated;
2603         }
2604
2605         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2606         if (map->m_lblk > ee_block) {
2607                 ex1 = ex;
2608                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2609                 ext4_ext_mark_uninitialized(ex1);
2610                 ex2 = &newex;
2611         }
2612         /*
2613          * for sanity, update the length of the ex2 extent before
2614          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2615          * overlap of blocks.
2616          */
2617         if (!ex1 && allocated > map->m_len)
2618                 ex2->ee_len = cpu_to_le16(map->m_len);
2619         /* ex3: to ee_block + ee_len : uninitialised */
2620         if (allocated > map->m_len) {
2621                 unsigned int newdepth;
2622                 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2623                 if (allocated <= EXT4_EXT_ZERO_LEN && may_zeroout) {
2624                         /*
2625                          * map->m_lblk == ee_block is handled by the zerouout
2626                          * at the beginning.
2627                          * Mark first half uninitialized.
2628                          * Mark second half initialized and zero out the
2629                          * initialized extent
2630                          */
2631                         ex->ee_block = orig_ex.ee_block;
2632                         ex->ee_len   = cpu_to_le16(ee_len - allocated);
2633                         ext4_ext_mark_uninitialized(ex);
2634                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2635                         ext4_ext_dirty(handle, inode, path + depth);
2636
2637                         ex3 = &newex;
2638                         ex3->ee_block = cpu_to_le32(map->m_lblk);
2639                         ext4_ext_store_pblock(ex3, newblock);
2640                         ex3->ee_len = cpu_to_le16(allocated);
2641                         err = ext4_ext_insert_extent(handle, inode, path,
2642                                                         ex3, 0);
2643                         if (err == -ENOSPC) {
2644                                 err =  ext4_ext_zeroout(inode, &orig_ex);
2645                                 if (err)
2646                                         goto fix_extent_len;
2647                                 ex->ee_block = orig_ex.ee_block;
2648                                 ex->ee_len   = orig_ex.ee_len;
2649                                 ext4_ext_store_pblock(ex,
2650                                         ext4_ext_pblock(&orig_ex));
2651                                 ext4_ext_dirty(handle, inode, path + depth);
2652                                 /* blocks available from map->m_lblk */
2653                                 return allocated;
2654
2655                         } else if (err)
2656                                 goto fix_extent_len;
2657
2658                         /*
2659                          * We need to zero out the second half because
2660                          * an fallocate request can update file size and
2661                          * converting the second half to initialized extent
2662                          * implies that we can leak some junk data to user
2663                          * space.
2664                          */
2665                         err =  ext4_ext_zeroout(inode, ex3);
2666                         if (err) {
2667                                 /*
2668                                  * We should actually mark the
2669                                  * second half as uninit and return error
2670                                  * Insert would have changed the extent
2671                                  */
2672                                 depth = ext_depth(inode);
2673                                 ext4_ext_drop_refs(path);
2674                                 path = ext4_ext_find_extent(inode, map->m_lblk,
2675                                                             path);
2676                                 if (IS_ERR(path)) {
2677                                         err = PTR_ERR(path);
2678                                         return err;
2679                                 }
2680                                 /* get the second half extent details */
2681                                 ex = path[depth].p_ext;
2682                                 err = ext4_ext_get_access(handle, inode,
2683                                                                 path + depth);
2684                                 if (err)
2685                                         return err;
2686                                 ext4_ext_mark_uninitialized(ex);
2687                                 ext4_ext_dirty(handle, inode, path + depth);
2688                                 return err;
2689                         }
2690
2691                         /* zeroed the second half */
2692                         return allocated;
2693                 }
2694                 ex3 = &newex;
2695                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2696                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2697                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2698                 ext4_ext_mark_uninitialized(ex3);
2699                 err = ext4_ext_insert_extent(handle, inode, path, ex3, 0);
2700                 if (err == -ENOSPC && may_zeroout) {
2701                         err =  ext4_ext_zeroout(inode, &orig_ex);
2702                         if (err)
2703                                 goto fix_extent_len;
2704                         /* update the extent length and mark as initialized */
2705                         ex->ee_block = orig_ex.ee_block;
2706                         ex->ee_len   = orig_ex.ee_len;
2707                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2708                         ext4_ext_dirty(handle, inode, path + depth);
2709                         /* zeroed the full extent */
2710                         /* blocks available from map->m_lblk */
2711                         return allocated;
2712
2713                 } else if (err)
2714                         goto fix_extent_len;
2715                 /*
2716                  * The depth, and hence eh & ex might change
2717                  * as part of the insert above.
2718                  */
2719                 newdepth = ext_depth(inode);
2720                 /*
2721                  * update the extent length after successful insert of the
2722                  * split extent
2723                  */
2724                 ee_len -= ext4_ext_get_actual_len(ex3);
2725                 orig_ex.ee_len = cpu_to_le16(ee_len);
2726                 may_zeroout = ee_block + ee_len <= eof_block;
2727
2728                 depth = newdepth;
2729                 ext4_ext_drop_refs(path);
2730                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2731                 if (IS_ERR(path)) {
2732                         err = PTR_ERR(path);
2733                         goto out;
2734                 }
2735                 eh = path[depth].p_hdr;
2736                 ex = path[depth].p_ext;
2737                 if (ex2 != &newex)
2738                         ex2 = ex;
2739
2740                 err = ext4_ext_get_access(handle, inode, path + depth);
2741                 if (err)
2742                         goto out;
2743
2744                 allocated = map->m_len;
2745
2746                 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2747                  * to insert a extent in the middle zerout directly
2748                  * otherwise give the extent a chance to merge to left
2749                  */
2750                 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2751                         map->m_lblk != ee_block && may_zeroout) {
2752                         err =  ext4_ext_zeroout(inode, &orig_ex);
2753                         if (err)
2754                                 goto fix_extent_len;
2755                         /* update the extent length and mark as initialized */
2756                         ex->ee_block = orig_ex.ee_block;
2757                         ex->ee_len   = orig_ex.ee_len;
2758                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2759                         ext4_ext_dirty(handle, inode, path + depth);
2760                         /* zero out the first half */
2761                         /* blocks available from map->m_lblk */
2762                         return allocated;
2763                 }
2764         }
2765         /*
2766          * If there was a change of depth as part of the
2767          * insertion of ex3 above, we need to update the length
2768          * of the ex1 extent again here
2769          */
2770         if (ex1 && ex1 != ex) {
2771                 ex1 = ex;
2772                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2773                 ext4_ext_mark_uninitialized(ex1);
2774                 ex2 = &newex;
2775         }
2776         /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2777         ex2->ee_block = cpu_to_le32(map->m_lblk);
2778         ext4_ext_store_pblock(ex2, newblock);
2779         ex2->ee_len = cpu_to_le16(allocated);
2780         if (ex2 != ex)
2781                 goto insert;
2782         /*
2783          * New (initialized) extent starts from the first block
2784          * in the current extent. i.e., ex2 == ex
2785          * We have to see if it can be merged with the extent
2786          * on the left.
2787          */
2788         if (ex2 > EXT_FIRST_EXTENT(eh)) {
2789                 /*
2790                  * To merge left, pass "ex2 - 1" to try_to_merge(),
2791                  * since it merges towards right _only_.
2792                  */
2793                 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2794                 if (ret) {
2795                         err = ext4_ext_correct_indexes(handle, inode, path);
2796                         if (err)
2797                                 goto out;
2798                         depth = ext_depth(inode);
2799                         ex2--;
2800                 }
2801         }
2802         /*
2803          * Try to Merge towards right. This might be required
2804          * only when the whole extent is being written to.
2805          * i.e. ex2 == ex and ex3 == NULL.
2806          */
2807         if (!ex3) {
2808                 ret = ext4_ext_try_to_merge(inode, path, ex2);
2809                 if (ret) {
2810                         err = ext4_ext_correct_indexes(handle, inode, path);
2811                         if (err)
2812                                 goto out;
2813                 }
2814         }
2815         /* Mark modified extent as dirty */
2816         err = ext4_ext_dirty(handle, inode, path + depth);
2817         goto out;
2818 insert:
2819         err = ext4_ext_insert_extent(handle, inode, path, &newex, 0);
2820         if (err == -ENOSPC && may_zeroout) {
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_block = orig_ex.ee_block;
2826                 ex->ee_len   = orig_ex.ee_len;
2827                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2828                 ext4_ext_dirty(handle, inode, path + depth);
2829                 /* zero out the first half */
2830                 return allocated;
2831         } else if (err)
2832                 goto fix_extent_len;
2833 out:
2834         ext4_ext_show_leaf(inode, path);
2835         return err ? err : allocated;
2836
2837 fix_extent_len:
2838         ex->ee_block = orig_ex.ee_block;
2839         ex->ee_len   = orig_ex.ee_len;
2840         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2841         ext4_ext_mark_uninitialized(ex);
2842         ext4_ext_dirty(handle, inode, path + depth);
2843         return err;
2844 }
2845
2846 /*
2847  * This function is called by ext4_ext_map_blocks() from
2848  * ext4_get_blocks_dio_write() when DIO to write
2849  * to an uninitialized extent.
2850  *
2851  * Writing to an uninitialized extent may result in splitting the uninitialized
2852  * extent into multiple /initialized uninitialized extents (up to three)
2853  * There are three possibilities:
2854  *   a> There is no split required: Entire extent should be uninitialized
2855  *   b> Splits in two extents: Write is happening at either end of the extent
2856  *   c> Splits in three extents: Somone is writing in middle of the extent
2857  *
2858  * One of more index blocks maybe needed if the extent tree grow after
2859  * the uninitialized extent split. To prevent ENOSPC occur at the IO
2860  * complete, we need to split the uninitialized extent before DIO submit
2861  * the IO. The uninitialized extent called at this time will be split
2862  * into three uninitialized extent(at most). After IO complete, the part
2863  * being filled will be convert to initialized by the end_io callback function
2864  * via ext4_convert_unwritten_extents().
2865  *
2866  * Returns the size of uninitialized extent to be written on success.
2867  */
2868 static int ext4_split_unwritten_extents(handle_t *handle,
2869                                         struct inode *inode,
2870                                         struct ext4_map_blocks *map,
2871                                         struct ext4_ext_path *path,
2872                                         int flags)
2873 {
2874         struct ext4_extent *ex, newex, orig_ex;
2875         struct ext4_extent *ex1 = NULL;
2876         struct ext4_extent *ex2 = NULL;
2877         struct ext4_extent *ex3 = NULL;
2878         ext4_lblk_t ee_block, eof_block;
2879         unsigned int allocated, ee_len, depth;
2880         ext4_fsblk_t newblock;
2881         int err = 0;
2882         int may_zeroout;
2883
2884         ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2885                 "block %llu, max_blocks %u\n", inode->i_ino,
2886                 (unsigned long long)map->m_lblk, map->m_len);
2887
2888         eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2889                 inode->i_sb->s_blocksize_bits;
2890         if (eof_block < map->m_lblk + map->m_len)
2891                 eof_block = map->m_lblk + map->m_len;
2892
2893         depth = ext_depth(inode);
2894         ex = path[depth].p_ext;
2895         ee_block = le32_to_cpu(ex->ee_block);
2896         ee_len = ext4_ext_get_actual_len(ex);
2897         allocated = ee_len - (map->m_lblk - ee_block);
2898         newblock = map->m_lblk - ee_block + ext4_ext_pblock(ex);
2899
2900         ex2 = ex;
2901         orig_ex.ee_block = ex->ee_block;
2902         orig_ex.ee_len   = cpu_to_le16(ee_len);
2903         ext4_ext_store_pblock(&orig_ex, ext4_ext_pblock(ex));
2904
2905         /*
2906          * It is safe to convert extent to initialized via explicit
2907          * zeroout only if extent is fully insde i_size or new_size.
2908          */
2909         may_zeroout = ee_block + ee_len <= eof_block;
2910
2911         /*
2912          * If the uninitialized extent begins at the same logical
2913          * block where the write begins, and the write completely
2914          * covers the extent, then we don't need to split it.
2915          */
2916         if ((map->m_lblk == ee_block) && (allocated <= map->m_len))
2917                 return allocated;
2918
2919         err = ext4_ext_get_access(handle, inode, path + depth);
2920         if (err)
2921                 goto out;
2922         /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2923         if (map->m_lblk > ee_block) {
2924                 ex1 = ex;
2925                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2926                 ext4_ext_mark_uninitialized(ex1);
2927                 ex2 = &newex;
2928         }
2929         /*
2930          * for sanity, update the length of the ex2 extent before
2931          * we insert ex3, if ex1 is NULL. This is to avoid temporary
2932          * overlap of blocks.
2933          */
2934         if (!ex1 && allocated > map->m_len)
2935                 ex2->ee_len = cpu_to_le16(map->m_len);
2936         /* ex3: to ee_block + ee_len : uninitialised */
2937         if (allocated > map->m_len) {
2938                 unsigned int newdepth;
2939                 ex3 = &newex;
2940                 ex3->ee_block = cpu_to_le32(map->m_lblk + map->m_len);
2941                 ext4_ext_store_pblock(ex3, newblock + map->m_len);
2942                 ex3->ee_len = cpu_to_le16(allocated - map->m_len);
2943                 ext4_ext_mark_uninitialized(ex3);
2944                 err = ext4_ext_insert_extent(handle, inode, path, ex3, flags);
2945                 if (err == -ENOSPC && may_zeroout) {
2946                         err =  ext4_ext_zeroout(inode, &orig_ex);
2947                         if (err)
2948                                 goto fix_extent_len;
2949                         /* update the extent length and mark as initialized */
2950                         ex->ee_block = orig_ex.ee_block;
2951                         ex->ee_len   = orig_ex.ee_len;
2952                         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
2953                         ext4_ext_dirty(handle, inode, path + depth);
2954                         /* zeroed the full extent */
2955                         /* blocks available from map->m_lblk */
2956                         return allocated;
2957
2958                 } else if (err)
2959                         goto fix_extent_len;
2960                 /*
2961                  * The depth, and hence eh & ex might change
2962                  * as part of the insert above.
2963                  */
2964                 newdepth = ext_depth(inode);
2965                 /*
2966                  * update the extent length after successful insert of the
2967                  * split extent
2968                  */
2969                 ee_len -= ext4_ext_get_actual_len(ex3);
2970                 orig_ex.ee_len = cpu_to_le16(ee_len);
2971                 may_zeroout = ee_block + ee_len <= eof_block;
2972
2973                 depth = newdepth;
2974                 ext4_ext_drop_refs(path);
2975                 path = ext4_ext_find_extent(inode, map->m_lblk, path);
2976                 if (IS_ERR(path)) {
2977                         err = PTR_ERR(path);
2978                         goto out;
2979                 }
2980                 ex = path[depth].p_ext;
2981                 if (ex2 != &newex)
2982                         ex2 = ex;
2983
2984                 err = ext4_ext_get_access(handle, inode, path + depth);
2985                 if (err)
2986                         goto out;
2987
2988                 allocated = map->m_len;
2989         }
2990         /*
2991          * If there was a change of depth as part of the
2992          * insertion of ex3 above, we need to update the length
2993          * of the ex1 extent again here
2994          */
2995         if (ex1 && ex1 != ex) {
2996                 ex1 = ex;
2997                 ex1->ee_len = cpu_to_le16(map->m_lblk - ee_block);
2998                 ext4_ext_mark_uninitialized(ex1);
2999                 ex2 = &newex;
3000         }
3001         /*
3002          * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
3003          * using direct I/O, uninitialised still.
3004          */
3005         ex2->ee_block = cpu_to_le32(map->m_lblk);
3006         ext4_ext_store_pblock(ex2, newblock);
3007         ex2->ee_len = cpu_to_le16(allocated);
3008         ext4_ext_mark_uninitialized(ex2);
3009         if (ex2 != ex)
3010                 goto insert;
3011         /* Mark modified extent as dirty */
3012         err = ext4_ext_dirty(handle, inode, path + depth);
3013         ext_debug("out here\n");
3014         goto out;
3015 insert:
3016         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3017         if (err == -ENOSPC && may_zeroout) {
3018                 err =  ext4_ext_zeroout(inode, &orig_ex);
3019                 if (err)
3020                         goto fix_extent_len;
3021                 /* update the extent length and mark as initialized */
3022                 ex->ee_block = orig_ex.ee_block;
3023                 ex->ee_len   = orig_ex.ee_len;
3024                 ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3025                 ext4_ext_dirty(handle, inode, path + depth);
3026                 /* zero out the first half */
3027                 return allocated;
3028         } else if (err)
3029                 goto fix_extent_len;
3030 out:
3031         ext4_ext_show_leaf(inode, path);
3032         return err ? err : allocated;
3033
3034 fix_extent_len:
3035         ex->ee_block = orig_ex.ee_block;
3036         ex->ee_len   = orig_ex.ee_len;
3037         ext4_ext_store_pblock(ex, ext4_ext_pblock(&orig_ex));
3038         ext4_ext_mark_uninitialized(ex);
3039         ext4_ext_dirty(handle, inode, path + depth);
3040         return err;
3041 }
3042 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3043                                               struct inode *inode,
3044                                               struct ext4_ext_path *path)
3045 {
3046         struct ext4_extent *ex;
3047         struct ext4_extent_header *eh;
3048         int depth;
3049         int err = 0;
3050         int ret = 0;
3051
3052         depth = ext_depth(inode);
3053         eh = path[depth].p_hdr;
3054         ex = path[depth].p_ext;
3055
3056         err = ext4_ext_get_access(handle, inode, path + depth);
3057         if (err)
3058                 goto out;
3059         /* first mark the extent as initialized */
3060         ext4_ext_mark_initialized(ex);
3061
3062         /*
3063          * We have to see if it can be merged with the extent
3064          * on the left.
3065          */
3066         if (ex > EXT_FIRST_EXTENT(eh)) {
3067                 /*
3068                  * To merge left, pass "ex - 1" to try_to_merge(),
3069                  * since it merges towards right _only_.
3070                  */
3071                 ret = ext4_ext_try_to_merge(inode, path, ex - 1);
3072                 if (ret) {
3073                         err = ext4_ext_correct_indexes(handle, inode, path);
3074                         if (err)
3075                                 goto out;
3076                         depth = ext_depth(inode);
3077                         ex--;
3078                 }
3079         }
3080         /*
3081          * Try to Merge towards right.
3082          */
3083         ret = ext4_ext_try_to_merge(inode, path, ex);
3084         if (ret) {
3085                 err = ext4_ext_correct_indexes(handle, inode, path);
3086                 if (err)
3087                         goto out;
3088                 depth = ext_depth(inode);
3089         }
3090         /* Mark modified extent as dirty */
3091         err = ext4_ext_dirty(handle, inode, path + depth);
3092 out:
3093         ext4_ext_show_leaf(inode, path);
3094         return err;
3095 }
3096
3097 static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3098                         sector_t block, int count)
3099 {
3100         int i;
3101         for (i = 0; i < count; i++)
3102                 unmap_underlying_metadata(bdev, block + i);
3103 }
3104
3105 /*
3106  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3107  */
3108 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3109                               ext4_lblk_t lblk,
3110                               struct ext4_ext_path *path,
3111                               unsigned int len)
3112 {
3113         int i, depth;
3114         struct ext4_extent_header *eh;
3115         struct ext4_extent *last_ex;
3116
3117         if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3118                 return 0;
3119
3120         depth = ext_depth(inode);
3121         eh = path[depth].p_hdr;
3122
3123         if (unlikely(!eh->eh_entries)) {
3124                 EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3125                                  "EOFBLOCKS_FL set");
3126                 return -EIO;
3127         }
3128         last_ex = EXT_LAST_EXTENT(eh);
3129         /*
3130          * We should clear the EOFBLOCKS_FL flag if we are writing the
3131          * last block in the last extent in the file.  We test this by
3132          * first checking to see if the caller to
3133          * ext4_ext_get_blocks() was interested in the last block (or
3134          * a block beyond the last block) in the current extent.  If
3135          * this turns out to be false, we can bail out from this
3136          * function immediately.
3137          */
3138         if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3139             ext4_ext_get_actual_len(last_ex))
3140                 return 0;
3141         /*
3142          * If the caller does appear to be planning to write at or
3143          * beyond the end of the current extent, we then test to see
3144          * if the current extent is the last extent in the file, by
3145          * checking to make sure it was reached via the rightmost node
3146          * at each level of the tree.
3147          */
3148         for (i = depth-1; i >= 0; i--)
3149                 if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3150                         return 0;
3151         ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3152         return ext4_mark_inode_dirty(handle, inode);
3153 }
3154
3155 static int
3156 ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3157                         struct ext4_map_blocks *map,
3158                         struct ext4_ext_path *path, int flags,
3159                         unsigned int allocated, ext4_fsblk_t newblock)
3160 {
3161         int ret = 0;
3162         int err = 0;
3163         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3164
3165         ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3166                   "block %llu, max_blocks %u, flags %d, allocated %u",
3167                   inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3168                   flags, allocated);
3169         ext4_ext_show_leaf(inode, path);
3170
3171         /* get_block() before submit the IO, split the extent */
3172         if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3173                 ret = ext4_split_unwritten_extents(handle, inode, map,
3174                                                    path, flags);
3175                 /*
3176                  * Flag the inode(non aio case) or end_io struct (aio case)
3177                  * that this IO needs to convertion to written when IO is
3178                  * completed
3179                  */
3180                 if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3181                         io->flag = EXT4_IO_END_UNWRITTEN;
3182                         atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3183                 } else
3184                         ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3185                 if (ext4_should_dioread_nolock(inode))
3186                         map->m_flags |= EXT4_MAP_UNINIT;
3187                 goto out;
3188         }
3189         /* IO end_io complete, convert the filled extent to written */
3190         if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3191                 ret = ext4_convert_unwritten_extents_endio(handle, inode,
3192                                                         path);
3193                 if (ret >= 0) {
3194                         ext4_update_inode_fsync_trans(handle, inode, 1);
3195                         err = check_eofblocks_fl(handle, inode, map->m_lblk,
3196                                                  path, map->m_len);
3197                 } else
3198                         err = ret;
3199                 goto out2;
3200         }
3201         /* buffered IO case */
3202         /*
3203          * repeat fallocate creation request
3204          * we already have an unwritten extent
3205          */
3206         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3207                 goto map_out;
3208
3209         /* buffered READ or buffered write_begin() lookup */
3210         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3211                 /*
3212                  * We have blocks reserved already.  We
3213                  * return allocated blocks so that delalloc
3214                  * won't do block reservation for us.  But
3215                  * the buffer head will be unmapped so that
3216                  * a read from the block returns 0s.
3217                  */
3218                 map->m_flags |= EXT4_MAP_UNWRITTEN;
3219                 goto out1;
3220         }
3221
3222         /* buffered write, writepage time, convert*/
3223         ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3224         if (ret >= 0) {
3225                 ext4_update_inode_fsync_trans(handle, inode, 1);
3226                 err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3227                                          map->m_len);
3228                 if (err < 0)
3229                         goto out2;
3230         }
3231
3232 out:
3233         if (ret <= 0) {
3234                 err = ret;
3235                 goto out2;
3236         } else
3237                 allocated = ret;
3238         map->m_flags |= EXT4_MAP_NEW;
3239         /*
3240          * if we allocated more blocks than requested
3241          * we need to make sure we unmap the extra block
3242          * allocated. The actual needed block will get
3243          * unmapped later when we find the buffer_head marked
3244          * new.
3245          */
3246         if (allocated > map->m_len) {
3247                 unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3248                                         newblock + map->m_len,
3249                                         allocated - map->m_len);
3250                 allocated = map->m_len;
3251         }
3252
3253         /*
3254          * If we have done fallocate with the offset that is already
3255          * delayed allocated, we would have block reservation
3256          * and quota reservation done in the delayed write path.
3257          * But fallocate would have already updated quota and block
3258          * count for this offset. So cancel these reservation
3259          */
3260         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3261                 ext4_da_update_reserve_space(inode, allocated, 0);
3262
3263 map_out:
3264         map->m_flags |= EXT4_MAP_MAPPED;
3265 out1:
3266         if (allocated > map->m_len)
3267                 allocated = map->m_len;
3268         ext4_ext_show_leaf(inode, path);
3269         map->m_pblk = newblock;
3270         map->m_len = allocated;
3271 out2:
3272         if (path) {
3273                 ext4_ext_drop_refs(path);
3274                 kfree(path);
3275         }
3276         return err ? err : allocated;
3277 }
3278
3279 /*
3280  * Block allocation/map/preallocation routine for extents based files
3281  *
3282  *
3283  * Need to be called with
3284  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3285  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3286  *
3287  * return > 0, number of of blocks already mapped/allocated
3288  *          if create == 0 and these are pre-allocated blocks
3289  *              buffer head is unmapped
3290  *          otherwise blocks are mapped
3291  *
3292  * return = 0, if plain look up failed (blocks have not been allocated)
3293  *          buffer head is unmapped
3294  *
3295  * return < 0, error case.
3296  */
3297 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3298                         struct ext4_map_blocks *map, int flags)
3299 {
3300         struct ext4_ext_path *path = NULL;
3301         struct ext4_extent newex, *ex;
3302         ext4_fsblk_t newblock = 0;
3303         int err = 0, depth, ret;
3304         unsigned int allocated = 0;
3305         struct ext4_allocation_request ar;
3306         ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3307
3308         ext_debug("blocks %u/%u requested for inode %lu\n",
3309                   map->m_lblk, map->m_len, inode->i_ino);
3310         trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3311
3312         /* check in cache */
3313         if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3314                 if (!newex.ee_start_lo && !newex.ee_start_hi) {
3315                         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3316                                 /*
3317                                  * block isn't allocated yet and
3318                                  * user doesn't want to allocate it
3319                                  */
3320                                 goto out2;
3321                         }
3322                         /* we should allocate requested block */
3323                 } else {
3324                         /* block is already allocated */
3325                         newblock = map->m_lblk
3326                                    - le32_to_cpu(newex.ee_block)
3327                                    + ext4_ext_pblock(&newex);
3328                         /* number of remaining blocks in the extent */
3329                         allocated = ext4_ext_get_actual_len(&newex) -
3330                                 (map->m_lblk - le32_to_cpu(newex.ee_block));
3331                         goto out;
3332                 }
3333         }
3334
3335         /* find extent for this block */
3336         path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3337         if (IS_ERR(path)) {
3338                 err = PTR_ERR(path);
3339                 path = NULL;
3340                 goto out2;
3341         }
3342
3343         depth = ext_depth(inode);
3344
3345         /*
3346          * consistent leaf must not be empty;
3347          * this situation is possible, though, _during_ tree modification;
3348          * this is why assert can't be put in ext4_ext_find_extent()
3349          */
3350         if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3351                 EXT4_ERROR_INODE(inode, "bad extent address "
3352                                  "lblock: %lu, depth: %d pblock %lld",
3353                                  (unsigned long) map->m_lblk, depth,
3354                                  path[depth].p_block);
3355                 err = -EIO;
3356                 goto out2;
3357         }
3358
3359         ex = path[depth].p_ext;
3360         if (ex) {
3361                 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3362                 ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3363                 unsigned short ee_len;
3364
3365                 /*
3366                  * Uninitialized extents are treated as holes, except that
3367                  * we split out initialized portions during a write.
3368                  */
3369                 ee_len = ext4_ext_get_actual_len(ex);
3370                 /* if found extent covers block, simply return it */
3371                 if (in_range(map->m_lblk, ee_block, ee_len)) {
3372                         newblock = map->m_lblk - ee_block + ee_start;
3373                         /* number of remaining blocks in the extent */
3374                         allocated = ee_len - (map->m_lblk - ee_block);
3375                         ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3376                                   ee_block, ee_len, newblock);
3377
3378                         /* Do not put uninitialized extent in the cache */
3379                         if (!ext4_ext_is_uninitialized(ex)) {
3380                                 ext4_ext_put_in_cache(inode, ee_block,
3381                                                         ee_len, ee_start);
3382                                 goto out;
3383                         }
3384                         ret = ext4_ext_handle_uninitialized_extents(handle,
3385                                         inode, map, path, flags, allocated,
3386                                         newblock);
3387                         return ret;
3388                 }
3389         }
3390
3391         /*
3392          * requested block isn't allocated yet;
3393          * we couldn't try to create block if create flag is zero
3394          */
3395         if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3396                 /*
3397                  * put just found gap into cache to speed up
3398                  * subsequent requests
3399                  */
3400                 ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3401                 goto out2;
3402         }
3403         /*
3404          * Okay, we need to do block allocation.
3405          */
3406
3407         /* find neighbour allocated blocks */
3408         ar.lleft = map->m_lblk;
3409         err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3410         if (err)
3411                 goto out2;
3412         ar.lright = map->m_lblk;
3413         err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
3414         if (err)
3415                 goto out2;
3416
3417         /*
3418          * See if request is beyond maximum number of blocks we can have in
3419          * a single extent. For an initialized extent this limit is
3420          * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3421          * EXT_UNINIT_MAX_LEN.
3422          */
3423         if (map->m_len > EXT_INIT_MAX_LEN &&
3424             !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3425                 map->m_len = EXT_INIT_MAX_LEN;
3426         else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3427                  (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3428                 map->m_len = EXT_UNINIT_MAX_LEN;
3429
3430         /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3431         newex.ee_block = cpu_to_le32(map->m_lblk);
3432         newex.ee_len = cpu_to_le16(map->m_len);
3433         err = ext4_ext_check_overlap(inode, &newex, path);
3434         if (err)
3435                 allocated = ext4_ext_get_actual_len(&newex);
3436         else
3437                 allocated = map->m_len;
3438
3439         /* allocate new block */
3440         ar.inode = inode;
3441         ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3442         ar.logical = map->m_lblk;
3443         ar.len = allocated;
3444         if (S_ISREG(inode->i_mode))
3445                 ar.flags = EXT4_MB_HINT_DATA;
3446         else
3447                 /* disable in-core preallocation for non-regular files */
3448                 ar.flags = 0;
3449         newblock = ext4_mb_new_blocks(handle, &ar, &err);
3450         if (!newblock)
3451                 goto out2;
3452         ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3453                   ar.goal, newblock, allocated);
3454
3455         /* try to insert new extent into found leaf and return */
3456         ext4_ext_store_pblock(&newex, newblock);
3457         newex.ee_len = cpu_to_le16(ar.len);
3458         /* Mark uninitialized */
3459         if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
3460                 ext4_ext_mark_uninitialized(&newex);
3461                 /*
3462                  * io_end structure was created for every IO write to an
3463                  * uninitialized extent. To avoid unecessary conversion,
3464                  * here we flag the IO that really needs the conversion.
3465                  * For non asycn direct IO case, flag the inode state
3466                  * that we need to perform convertion when IO is done.
3467                  */
3468                 if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3469                         if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
3470                                 io->flag = EXT4_IO_END_UNWRITTEN;
3471                                 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3472                         } else
3473                                 ext4_set_inode_state(inode,
3474                                                      EXT4_STATE_DIO_UNWRITTEN);
3475                 }
3476                 if (ext4_should_dioread_nolock(inode))
3477                         map->m_flags |= EXT4_MAP_UNINIT;
3478         }
3479
3480         err = check_eofblocks_fl(handle, inode, map->m_lblk, path, ar.len);
3481         if (err)
3482                 goto out2;
3483
3484         err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
3485         if (err) {
3486                 /* free data blocks we just allocated */
3487                 /* not a good idea to call discard here directly,
3488                  * but otherwise we'd need to call it every free() */
3489                 ext4_discard_preallocations(inode);
3490                 ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
3491                                  ext4_ext_get_actual_len(&newex), 0);
3492                 goto out2;
3493         }
3494
3495         /* previous routine could use block we allocated */
3496         newblock = ext4_ext_pblock(&newex);
3497         allocated = ext4_ext_get_actual_len(&newex);
3498         if (allocated > map->m_len)
3499                 allocated = map->m_len;
3500         map->m_flags |= EXT4_MAP_NEW;
3501
3502         /*
3503          * Update reserved blocks/metadata blocks after successful
3504          * block allocation which had been deferred till now.
3505          */
3506         if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
3507                 ext4_da_update_reserve_space(inode, allocated, 1);
3508
3509         /*
3510          * Cache the extent and update transaction to commit on fdatasync only
3511          * when it is _not_ an uninitialized extent.
3512          */
3513         if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
3514                 ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
3515                 ext4_update_inode_fsync_trans(handle, inode, 1);
3516         } else
3517                 ext4_update_inode_fsync_trans(handle, inode, 0);
3518 out:
3519         if (allocated > map->m_len)
3520                 allocated = map->m_len;
3521         ext4_ext_show_leaf(inode, path);
3522         map->m_flags |= EXT4_MAP_MAPPED;
3523         map->m_pblk = newblock;
3524         map->m_len = allocated;
3525 out2:
3526         if (path) {
3527                 ext4_ext_drop_refs(path);
3528                 kfree(path);
3529         }
3530         trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
3531                 newblock, map->m_len, err ? err : allocated);
3532         return err ? err : allocated;
3533 }
3534
3535 void ext4_ext_truncate(struct inode *inode)
3536 {
3537         struct address_space *mapping = inode->i_mapping;
3538         struct super_block *sb = inode->i_sb;
3539         ext4_lblk_t last_block;
3540         handle_t *handle;
3541         int err = 0;
3542
3543         /*
3544          * finish any pending end_io work so we won't run the risk of
3545          * converting any truncated blocks to initialized later
3546          */
3547         ext4_flush_completed_IO(inode);
3548
3549         /*
3550          * probably first extent we're gonna free will be last in block
3551          */
3552         err = ext4_writepage_trans_blocks(inode);
3553         handle = ext4_journal_start(inode, err);
3554         if (IS_ERR(handle))
3555                 return;
3556
3557         if (inode->i_size & (sb->s_blocksize - 1))
3558                 ext4_block_truncate_page(handle, mapping, inode->i_size);
3559
3560         if (ext4_orphan_add(handle, inode))
3561                 goto out_stop;
3562
3563         down_write(&EXT4_I(inode)->i_data_sem);
3564         ext4_ext_invalidate_cache(inode);
3565
3566         ext4_discard_preallocations(inode);
3567
3568         /*
3569          * TODO: optimization is possible here.
3570          * Probably we need not scan at all,
3571          * because page truncation is enough.
3572          */
3573
3574         /* we have to know where to truncate from in crash case */
3575         EXT4_I(inode)->i_disksize = inode->i_size;
3576         ext4_mark_inode_dirty(handle, inode);
3577
3578         last_block = (inode->i_size + sb->s_blocksize - 1)
3579                         >> EXT4_BLOCK_SIZE_BITS(sb);
3580         err = ext4_ext_remove_space(inode, last_block);
3581
3582         /* In a multi-transaction truncate, we only make the final
3583          * transaction synchronous.
3584          */
3585         if (IS_SYNC(inode))
3586                 ext4_handle_sync(handle);
3587
3588 out_stop:
3589         up_write(&EXT4_I(inode)->i_data_sem);
3590         /*
3591          * If this was a simple ftruncate() and the file will remain alive,
3592          * then we need to clear up the orphan record which we created above.
3593          * However, if this was a real unlink then we were called by
3594          * ext4_delete_inode(), and we allow that function to clean up the
3595          * orphan info for us.
3596          */
3597         if (inode->i_nlink)
3598                 ext4_orphan_del(handle, inode);
3599
3600         inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3601         ext4_mark_inode_dirty(handle, inode);
3602         ext4_journal_stop(handle);
3603 }
3604
3605 static void ext4_falloc_update_inode(struct inode *inode,
3606                                 int mode, loff_t new_size, int update_ctime)
3607 {
3608         struct timespec now;
3609
3610         if (update_ctime) {
3611                 now = current_fs_time(inode->i_sb);
3612                 if (!timespec_equal(&inode->i_ctime, &now))
3613                         inode->i_ctime = now;
3614         }
3615         /*
3616          * Update only when preallocation was requested beyond
3617          * the file size.
3618          */
3619         if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3620                 if (new_size > i_size_read(inode))
3621                         i_size_write(inode, new_size);
3622                 if (new_size > EXT4_I(inode)->i_disksize)
3623                         ext4_update_i_disksize(inode, new_size);
3624         } else {
3625                 /*
3626                  * Mark that we allocate beyond EOF so the subsequent truncate
3627                  * can proceed even if the new size is the same as i_size.
3628                  */
3629                 if (new_size > i_size_read(inode))
3630                         ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3631         }
3632
3633 }
3634
3635 /*
3636  * preallocate space for a file. This implements ext4's fallocate file
3637  * operation, which gets called from sys_fallocate system call.
3638  * For block-mapped files, posix_fallocate should fall back to the method
3639  * of writing zeroes to the required new blocks (the same behavior which is
3640  * expected for file systems which do not support fallocate() system call).
3641  */
3642 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
3643 {
3644         struct inode *inode = file->f_path.dentry->d_inode;
3645         handle_t *handle;
3646         loff_t new_size;
3647         unsigned int max_blocks;
3648         int ret = 0;
3649         int ret2 = 0;
3650         int retries = 0;
3651         struct ext4_map_blocks map;
3652         unsigned int credits, blkbits = inode->i_blkbits;
3653
3654         /* We only support the FALLOC_FL_KEEP_SIZE mode */
3655         if (mode & ~FALLOC_FL_KEEP_SIZE)
3656                 return -EOPNOTSUPP;
3657
3658         /*
3659          * currently supporting (pre)allocate mode for extent-based
3660          * files _only_
3661          */
3662         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3663                 return -EOPNOTSUPP;
3664
3665         trace_ext4_fallocate_enter(inode, offset, len, mode);
3666         map.m_lblk = offset >> blkbits;
3667         /*
3668          * We can't just convert len to max_blocks because
3669          * If blocksize = 4096 offset = 3072 and len = 2048
3670          */
3671         max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
3672                 - map.m_lblk;
3673         /*
3674          * credits to insert 1 extent into extent tree
3675          */
3676         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3677         mutex_lock(&inode->i_mutex);
3678         ret = inode_newsize_ok(inode, (len + offset));
3679         if (ret) {
3680                 mutex_unlock(&inode->i_mutex);
3681                 trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
3682                 return ret;
3683         }
3684 retry:
3685         while (ret >= 0 && ret < max_blocks) {
3686                 map.m_lblk = map.m_lblk + ret;
3687                 map.m_len = max_blocks = max_blocks - ret;
3688                 handle = ext4_journal_start(inode, credits);
3689                 if (IS_ERR(handle)) {
3690                         ret = PTR_ERR(handle);
3691                         break;
3692                 }
3693                 ret = ext4_map_blocks(handle, inode, &map,
3694                                       EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
3695                 if (ret <= 0) {
3696 #ifdef EXT4FS_DEBUG
3697                         WARN_ON(ret <= 0);
3698                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3699                                     "returned error inode#%lu, block=%u, "
3700                                     "max_blocks=%u", __func__,
3701                                     inode->i_ino, map.m_lblk, max_blocks);
3702 #endif
3703                         ext4_mark_inode_dirty(handle, inode);
3704                         ret2 = ext4_journal_stop(handle);
3705                         break;
3706                 }
3707                 if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3708                                                 blkbits) >> blkbits))
3709                         new_size = offset + len;
3710                 else
3711                         new_size = (map.m_lblk + ret) << blkbits;
3712
3713                 ext4_falloc_update_inode(inode, mode, new_size,
3714                                          (map.m_flags & EXT4_MAP_NEW));
3715                 ext4_mark_inode_dirty(handle, inode);
3716                 ret2 = ext4_journal_stop(handle);
3717                 if (ret2)
3718                         break;
3719         }
3720         if (ret == -ENOSPC &&
3721                         ext4_should_retry_alloc(inode->i_sb, &retries)) {
3722                 ret = 0;
3723                 goto retry;
3724         }
3725         mutex_unlock(&inode->i_mutex);
3726         trace_ext4_fallocate_exit(inode, offset, max_blocks,
3727                                 ret > 0 ? ret2 : ret);
3728         return ret > 0 ? ret2 : ret;
3729 }
3730
3731 /*
3732  * This function convert a range of blocks to written extents
3733  * The caller of this function will pass the start offset and the size.
3734  * all unwritten extents within this range will be converted to
3735  * written extents.
3736  *
3737  * This function is called from the direct IO end io call back
3738  * function, to convert the fallocated extents after IO is completed.
3739  * Returns 0 on success.
3740  */
3741 int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
3742                                     ssize_t len)
3743 {
3744         handle_t *handle;
3745         unsigned int max_blocks;
3746         int ret = 0;
3747         int ret2 = 0;
3748         struct ext4_map_blocks map;
3749         unsigned int credits, blkbits = inode->i_blkbits;
3750
3751         map.m_lblk = offset >> blkbits;
3752         /*
3753          * We can't just convert len to max_blocks because
3754          * If blocksize = 4096 offset = 3072 and len = 2048
3755          */
3756         max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
3757                       map.m_lblk);
3758         /*
3759          * credits to insert 1 extent into extent tree
3760          */
3761         credits = ext4_chunk_trans_blocks(inode, max_blocks);
3762         while (ret >= 0 && ret < max_blocks) {
3763                 map.m_lblk += ret;
3764                 map.m_len = (max_blocks -= ret);
3765                 handle = ext4_journal_start(inode, credits);
3766                 if (IS_ERR(handle)) {
3767                         ret = PTR_ERR(handle);
3768                         break;
3769                 }
3770                 ret = ext4_map_blocks(handle, inode, &map,
3771                                       EXT4_GET_BLOCKS_IO_CONVERT_EXT);
3772                 if (ret <= 0) {
3773                         WARN_ON(ret <= 0);
3774                         printk(KERN_ERR "%s: ext4_ext_map_blocks "
3775                                     "returned error inode#%lu, block=%u, "
3776                                     "max_blocks=%u", __func__,
3777                                     inode->i_ino, map.m_lblk, map.m_len);
3778                 }
3779                 ext4_mark_inode_dirty(handle, inode);
3780                 ret2 = ext4_journal_stop(handle);
3781                 if (ret <= 0 || ret2 )
3782                         break;
3783         }
3784         return ret > 0 ? ret2 : ret;
3785 }
3786
3787 /*
3788  * Callback function called for each extent to gather FIEMAP information.
3789  */
3790 static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
3791                        struct ext4_ext_cache *newex, struct ext4_extent *ex,
3792                        void *data)
3793 {
3794         __u64   logical;
3795         __u64   physical;
3796         __u64   length;
3797         loff_t  size;
3798         __u32   flags = 0;
3799         int             ret = 0;
3800         struct fiemap_extent_info *fieinfo = data;
3801         unsigned char blksize_bits;
3802
3803         blksize_bits = inode->i_sb->s_blocksize_bits;
3804         logical = (__u64)newex->ec_block << blksize_bits;
3805
3806         if (newex->ec_start == 0) {
3807                 /*
3808                  * No extent in extent-tree contains block @newex->ec_start,
3809                  * then the block may stay in 1)a hole or 2)delayed-extent.
3810                  *
3811                  * Holes or delayed-extents are processed as follows.
3812                  * 1. lookup dirty pages with specified range in pagecache.
3813                  *    If no page is got, then there is no delayed-extent and
3814                  *    return with EXT_CONTINUE.
3815                  * 2. find the 1st mapped buffer,
3816                  * 3. check if the mapped buffer is both in the request range
3817                  *    and a delayed buffer. If not, there is no delayed-extent,
3818                  *    then return.
3819                  * 4. a delayed-extent is found, the extent will be collected.
3820                  */
3821                 ext4_lblk_t     end = 0;
3822                 pgoff_t         last_offset;
3823                 pgoff_t         offset;
3824                 pgoff_t         index;
3825                 struct page     **pages = NULL;
3826                 struct buffer_head *bh = NULL;
3827                 struct buffer_head *head = NULL;
3828                 unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
3829
3830                 pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
3831                 if (pages == NULL)
3832                         return -ENOMEM;
3833
3834                 offset = logical >> PAGE_SHIFT;
3835 repeat:
3836                 last_offset = offset;
3837                 head = NULL;
3838                 ret = find_get_pages_tag(inode->i_mapping, &offset,
3839                                         PAGECACHE_TAG_DIRTY, nr_pages, pages);
3840
3841                 if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3842                         /* First time, try to find a mapped buffer. */
3843                         if (ret == 0) {
3844 out:
3845                                 for (index = 0; index < ret; index++)
3846                                         page_cache_release(pages[index]);
3847                                 /* just a hole. */
3848                                 kfree(pages);
3849                                 return EXT_CONTINUE;
3850                         }
3851
3852                         /* Try to find the 1st mapped buffer. */
3853                         end = ((__u64)pages[0]->index << PAGE_SHIFT) >>
3854                                   blksize_bits;
3855                         if (!page_has_buffers(pages[0]))
3856                                 goto out;
3857                         head = page_buffers(pages[0]);
3858                         if (!head)
3859                                 goto out;
3860
3861                         bh = head;
3862                         do {
3863                                 if (buffer_mapped(bh)) {
3864                                         /* get the 1st mapped buffer. */
3865                                         if (end > newex->ec_block +
3866                                                 newex->ec_len)
3867                                                 /* The buffer is out of
3868                                                  * the request range.
3869                                                  */
3870                                                 goto out;
3871                                         goto found_mapped_buffer;
3872                                 }
3873                                 bh = bh->b_this_page;
3874                                 end++;
3875                         } while (bh != head);
3876
3877                         /* No mapped buffer found. */
3878                         goto out;
3879                 } else {
3880                         /*Find contiguous delayed buffers. */
3881                         if (ret > 0 && pages[0]->index == last_offset)
3882                                 head = page_buffers(pages[0]);
3883                         bh = head;
3884                 }
3885
3886 found_mapped_buffer:
3887                 if (bh != NULL && buffer_delay(bh)) {
3888                         /* 1st or contiguous delayed buffer found. */
3889                         if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
3890                                 /*
3891                                  * 1st delayed buffer found, record
3892                                  * the start of extent.
3893                                  */
3894                                 flags |= FIEMAP_EXTENT_DELALLOC;
3895                                 newex->ec_block = end;
3896                                 logical = (__u64)end << blksize_bits;
3897                         }
3898                         /* Find contiguous delayed buffers. */
3899                         do {
3900                                 if (!buffer_delay(bh))
3901                                         goto found_delayed_extent;
3902                                 bh = bh->b_this_page;
3903                                 end++;
3904                         } while (bh != head);
3905
3906                         for (index = 1; index < ret; index++) {
3907                                 if (!page_has_buffers(pages[index])) {
3908                                         bh = NULL;
3909                                         break;
3910                                 }
3911                                 head = page_buffers(pages[index]);
3912                                 if (!head) {
3913                                         bh = NULL;
3914                                         break;
3915                                 }
3916                                 if (pages[index]->index !=
3917                                         pages[0]->index + index) {
3918                                         /* Blocks are not contiguous. */
3919                                         bh = NULL;
3920                                         break;
3921                                 }
3922                                 bh = head;
3923                                 do {
3924                                         if (!buffer_delay(bh))
3925                                                 /* Delayed-extent ends. */
3926                                                 goto found_delayed_extent;
3927                                         bh = bh->b_this_page;
3928                                         end++;
3929                                 } while (bh != head);
3930                         }
3931                 } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
3932                         /* a hole found. */
3933                         goto out;
3934
3935 found_delayed_extent:
3936                 newex->ec_len = min(end - newex->ec_block,
3937                                                 (ext4_lblk_t)EXT_INIT_MAX_LEN);
3938                 if (ret == nr_pages && bh != NULL &&
3939                         newex->ec_len < EXT_INIT_MAX_LEN &&
3940                         buffer_delay(bh)) {
3941                         /* Have not collected an extent and continue. */
3942                         for (index = 0; index < ret; index++)
3943                                 page_cache_release(pages[index]);
3944                         goto repeat;
3945                 }
3946
3947                 for (index = 0; index < ret; index++)
3948                         page_cache_release(pages[index]);
3949                 kfree(pages);
3950         }
3951
3952         physical = (__u64)newex->ec_start << blksize_bits;
3953         length =   (__u64)newex->ec_len << blksize_bits;
3954
3955         if (ex && ext4_ext_is_uninitialized(ex))
3956                 flags |= FIEMAP_EXTENT_UNWRITTEN;
3957
3958         size = i_size_read(inode);
3959         if (logical + length >= size)
3960                 flags |= FIEMAP_EXTENT_LAST;
3961
3962         ret = fiemap_fill_next_extent(fieinfo, logical, physical,
3963                                         length, flags);
3964         if (ret < 0)
3965                 return ret;
3966         if (ret == 1)
3967                 return EXT_BREAK;
3968         return EXT_CONTINUE;
3969 }
3970
3971 /* fiemap flags we can handle specified here */
3972 #define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3973
3974 static int ext4_xattr_fiemap(struct inode *inode,
3975                                 struct fiemap_extent_info *fieinfo)
3976 {
3977         __u64 physical = 0;
3978         __u64 length;
3979         __u32 flags = FIEMAP_EXTENT_LAST;
3980         int blockbits = inode->i_sb->s_blocksize_bits;
3981         int error = 0;
3982
3983         /* in-inode? */
3984         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
3985                 struct ext4_iloc iloc;
3986                 int offset;     /* offset of xattr in inode */
3987
3988                 error = ext4_get_inode_loc(inode, &iloc);
3989                 if (error)
3990                         return error;
3991                 physical = iloc.bh->b_blocknr << blockbits;
3992                 offset = EXT4_GOOD_OLD_INODE_SIZE +
3993                                 EXT4_I(inode)->i_extra_isize;
3994                 physical += offset;
3995                 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3996                 flags |= FIEMAP_EXTENT_DATA_INLINE;
3997                 brelse(iloc.bh);
3998         } else { /* external block */
3999                 physical = EXT4_I(inode)->i_file_acl << blockbits;
4000                 length = inode->i_sb->s_blocksize;
4001         }
4002
4003         if (physical)
4004                 error = fiemap_fill_next_extent(fieinfo, 0, physical,
4005                                                 length, flags);
4006         return (error < 0 ? error : 0);
4007 }
4008
4009 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4010                 __u64 start, __u64 len)
4011 {
4012         ext4_lblk_t start_blk;
4013         int error = 0;
4014
4015         /* fallback to generic here if not in extents fmt */
4016         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4017                 return generic_block_fiemap(inode, fieinfo, start, len,
4018                         ext4_get_block);
4019
4020         if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4021                 return -EBADR;
4022
4023         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4024                 error = ext4_xattr_fiemap(inode, fieinfo);
4025         } else {
4026                 ext4_lblk_t len_blks;
4027                 __u64 last_blk;
4028
4029                 start_blk = start >> inode->i_sb->s_blocksize_bits;
4030                 last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4031                 if (last_blk >= EXT_MAX_BLOCK)
4032                         last_blk = EXT_MAX_BLOCK-1;
4033                 len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4034
4035                 /*
4036                  * Walk the extent tree gathering extent information.
4037                  * ext4_ext_fiemap_cb will push extents back to user.
4038                  */
4039                 error = ext4_ext_walk_space(inode, start_blk, len_blks,
4040                                           ext4_ext_fiemap_cb, fieinfo);
4041         }
4042
4043         return error;
4044 }
4045