linux/fs/ext4/extents.c
<<
>>
Prefs
   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/fs.h>
  33#include <linux/time.h>
  34#include <linux/jbd2.h>
  35#include <linux/highuid.h>
  36#include <linux/pagemap.h>
  37#include <linux/quotaops.h>
  38#include <linux/string.h>
  39#include <linux/slab.h>
  40#include <linux/falloc.h>
  41#include <asm/uaccess.h>
  42#include <linux/fiemap.h>
  43#include "ext4_jbd2.h"
  44
  45#include <trace/events/ext4.h>
  46
  47/*
  48 * used by extent splitting.
  49 */
  50#define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  51                                        due to ENOSPC */
  52#define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
  53#define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
  54
  55#define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  56#define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  57
  58static __le32 ext4_extent_block_csum(struct inode *inode,
  59                                     struct ext4_extent_header *eh)
  60{
  61        struct ext4_inode_info *ei = EXT4_I(inode);
  62        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  63        __u32 csum;
  64
  65        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  66                           EXT4_EXTENT_TAIL_OFFSET(eh));
  67        return cpu_to_le32(csum);
  68}
  69
  70static int ext4_extent_block_csum_verify(struct inode *inode,
  71                                         struct ext4_extent_header *eh)
  72{
  73        struct ext4_extent_tail *et;
  74
  75        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  76                EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  77                return 1;
  78
  79        et = find_ext4_extent_tail(eh);
  80        if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  81                return 0;
  82        return 1;
  83}
  84
  85static void ext4_extent_block_csum_set(struct inode *inode,
  86                                       struct ext4_extent_header *eh)
  87{
  88        struct ext4_extent_tail *et;
  89
  90        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
  91                EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
  92                return;
  93
  94        et = find_ext4_extent_tail(eh);
  95        et->et_checksum = ext4_extent_block_csum(inode, eh);
  96}
  97
  98static int ext4_split_extent(handle_t *handle,
  99                                struct inode *inode,
 100                                struct ext4_ext_path *path,
 101                                struct ext4_map_blocks *map,
 102                                int split_flag,
 103                                int flags);
 104
 105static int ext4_split_extent_at(handle_t *handle,
 106                             struct inode *inode,
 107                             struct ext4_ext_path *path,
 108                             ext4_lblk_t split,
 109                             int split_flag,
 110                             int flags);
 111
 112static int ext4_ext_truncate_extend_restart(handle_t *handle,
 113                                            struct inode *inode,
 114                                            int needed)
 115{
 116        int err;
 117
 118        if (!ext4_handle_valid(handle))
 119                return 0;
 120        if (handle->h_buffer_credits > needed)
 121                return 0;
 122        err = ext4_journal_extend(handle, needed);
 123        if (err <= 0)
 124                return err;
 125        err = ext4_truncate_restart_trans(handle, inode, needed);
 126        if (err == 0)
 127                err = -EAGAIN;
 128
 129        return err;
 130}
 131
 132/*
 133 * could return:
 134 *  - EROFS
 135 *  - ENOMEM
 136 */
 137static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 138                                struct ext4_ext_path *path)
 139{
 140        if (path->p_bh) {
 141                /* path points to block */
 142                return ext4_journal_get_write_access(handle, path->p_bh);
 143        }
 144        /* path points to leaf/index in inode body */
 145        /* we use in-core data, no need to protect them */
 146        return 0;
 147}
 148
 149/*
 150 * could return:
 151 *  - EROFS
 152 *  - ENOMEM
 153 *  - EIO
 154 */
 155#define ext4_ext_dirty(handle, inode, path) \
 156                __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 157static int __ext4_ext_dirty(const char *where, unsigned int line,
 158                            handle_t *handle, struct inode *inode,
 159                            struct ext4_ext_path *path)
 160{
 161        int err;
 162        if (path->p_bh) {
 163                ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
 164                /* path points to block */
 165                err = __ext4_handle_dirty_metadata(where, line, handle,
 166                                                   inode, path->p_bh);
 167        } else {
 168                /* path points to leaf/index in inode body */
 169                err = ext4_mark_inode_dirty(handle, inode);
 170        }
 171        return err;
 172}
 173
 174static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 175                              struct ext4_ext_path *path,
 176                              ext4_lblk_t block)
 177{
 178        if (path) {
 179                int depth = path->p_depth;
 180                struct ext4_extent *ex;
 181
 182                /*
 183                 * Try to predict block placement assuming that we are
 184                 * filling in a file which will eventually be
 185                 * non-sparse --- i.e., in the case of libbfd writing
 186                 * an ELF object sections out-of-order but in a way
 187                 * the eventually results in a contiguous object or
 188                 * executable file, or some database extending a table
 189                 * space file.  However, this is actually somewhat
 190                 * non-ideal if we are writing a sparse file such as
 191                 * qemu or KVM writing a raw image file that is going
 192                 * to stay fairly sparse, since it will end up
 193                 * fragmenting the file system's free space.  Maybe we
 194                 * should have some hueristics or some way to allow
 195                 * userspace to pass a hint to file system,
 196                 * especially if the latter case turns out to be
 197                 * common.
 198                 */
 199                ex = path[depth].p_ext;
 200                if (ex) {
 201                        ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 202                        ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 203
 204                        if (block > ext_block)
 205                                return ext_pblk + (block - ext_block);
 206                        else
 207                                return ext_pblk - (ext_block - block);
 208                }
 209
 210                /* it looks like index is empty;
 211                 * try to find starting block from index itself */
 212                if (path[depth].p_bh)
 213                        return path[depth].p_bh->b_blocknr;
 214        }
 215
 216        /* OK. use inode's group */
 217        return ext4_inode_to_goal_block(inode);
 218}
 219
 220/*
 221 * Allocation for a meta data block
 222 */
 223static ext4_fsblk_t
 224ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 225                        struct ext4_ext_path *path,
 226                        struct ext4_extent *ex, int *err, unsigned int flags)
 227{
 228        ext4_fsblk_t goal, newblock;
 229
 230        goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 231        newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 232                                        NULL, err);
 233        return newblock;
 234}
 235
 236static inline int ext4_ext_space_block(struct inode *inode, int check)
 237{
 238        int size;
 239
 240        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 241                        / sizeof(struct ext4_extent);
 242#ifdef AGGRESSIVE_TEST
 243        if (!check && size > 6)
 244                size = 6;
 245#endif
 246        return size;
 247}
 248
 249static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 250{
 251        int size;
 252
 253        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 254                        / sizeof(struct ext4_extent_idx);
 255#ifdef AGGRESSIVE_TEST
 256        if (!check && size > 5)
 257                size = 5;
 258#endif
 259        return size;
 260}
 261
 262static inline int ext4_ext_space_root(struct inode *inode, int check)
 263{
 264        int size;
 265
 266        size = sizeof(EXT4_I(inode)->i_data);
 267        size -= sizeof(struct ext4_extent_header);
 268        size /= sizeof(struct ext4_extent);
 269#ifdef AGGRESSIVE_TEST
 270        if (!check && size > 3)
 271                size = 3;
 272#endif
 273        return size;
 274}
 275
 276static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 277{
 278        int size;
 279
 280        size = sizeof(EXT4_I(inode)->i_data);
 281        size -= sizeof(struct ext4_extent_header);
 282        size /= sizeof(struct ext4_extent_idx);
 283#ifdef AGGRESSIVE_TEST
 284        if (!check && size > 4)
 285                size = 4;
 286#endif
 287        return size;
 288}
 289
 290/*
 291 * Calculate the number of metadata blocks needed
 292 * to allocate @blocks
 293 * Worse case is one block per extent
 294 */
 295int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 296{
 297        struct ext4_inode_info *ei = EXT4_I(inode);
 298        int idxs;
 299
 300        idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 301                / sizeof(struct ext4_extent_idx));
 302
 303        /*
 304         * If the new delayed allocation block is contiguous with the
 305         * previous da block, it can share index blocks with the
 306         * previous block, so we only need to allocate a new index
 307         * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 308         * an additional index block, and at ldxs**3 blocks, yet
 309         * another index blocks.
 310         */
 311        if (ei->i_da_metadata_calc_len &&
 312            ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 313                int num = 0;
 314
 315                if ((ei->i_da_metadata_calc_len % idxs) == 0)
 316                        num++;
 317                if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 318                        num++;
 319                if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 320                        num++;
 321                        ei->i_da_metadata_calc_len = 0;
 322                } else
 323                        ei->i_da_metadata_calc_len++;
 324                ei->i_da_metadata_calc_last_lblock++;
 325                return num;
 326        }
 327
 328        /*
 329         * In the worst case we need a new set of index blocks at
 330         * every level of the inode's extent tree.
 331         */
 332        ei->i_da_metadata_calc_len = 1;
 333        ei->i_da_metadata_calc_last_lblock = lblock;
 334        return ext_depth(inode) + 1;
 335}
 336
 337static int
 338ext4_ext_max_entries(struct inode *inode, int depth)
 339{
 340        int max;
 341
 342        if (depth == ext_depth(inode)) {
 343                if (depth == 0)
 344                        max = ext4_ext_space_root(inode, 1);
 345                else
 346                        max = ext4_ext_space_root_idx(inode, 1);
 347        } else {
 348                if (depth == 0)
 349                        max = ext4_ext_space_block(inode, 1);
 350                else
 351                        max = ext4_ext_space_block_idx(inode, 1);
 352        }
 353
 354        return max;
 355}
 356
 357static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 358{
 359        ext4_fsblk_t block = ext4_ext_pblock(ext);
 360        int len = ext4_ext_get_actual_len(ext);
 361
 362        if (len == 0)
 363                return 0;
 364        return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 365}
 366
 367static int ext4_valid_extent_idx(struct inode *inode,
 368                                struct ext4_extent_idx *ext_idx)
 369{
 370        ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 371
 372        return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 373}
 374
 375static int ext4_valid_extent_entries(struct inode *inode,
 376                                struct ext4_extent_header *eh,
 377                                int depth)
 378{
 379        unsigned short entries;
 380        if (eh->eh_entries == 0)
 381                return 1;
 382
 383        entries = le16_to_cpu(eh->eh_entries);
 384
 385        if (depth == 0) {
 386                /* leaf entries */
 387                struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 388                while (entries) {
 389                        if (!ext4_valid_extent(inode, ext))
 390                                return 0;
 391                        ext++;
 392                        entries--;
 393                }
 394        } else {
 395                struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 396                while (entries) {
 397                        if (!ext4_valid_extent_idx(inode, ext_idx))
 398                                return 0;
 399                        ext_idx++;
 400                        entries--;
 401                }
 402        }
 403        return 1;
 404}
 405
 406static int __ext4_ext_check(const char *function, unsigned int line,
 407                            struct inode *inode, struct ext4_extent_header *eh,
 408                            int depth)
 409{
 410        const char *error_msg;
 411        int max = 0;
 412
 413        if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 414                error_msg = "invalid magic";
 415                goto corrupted;
 416        }
 417        if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 418                error_msg = "unexpected eh_depth";
 419                goto corrupted;
 420        }
 421        if (unlikely(eh->eh_max == 0)) {
 422                error_msg = "invalid eh_max";
 423                goto corrupted;
 424        }
 425        max = ext4_ext_max_entries(inode, depth);
 426        if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 427                error_msg = "too large eh_max";
 428                goto corrupted;
 429        }
 430        if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 431                error_msg = "invalid eh_entries";
 432                goto corrupted;
 433        }
 434        if (!ext4_valid_extent_entries(inode, eh, depth)) {
 435                error_msg = "invalid extent entries";
 436                goto corrupted;
 437        }
 438        /* Verify checksum on non-root extent tree nodes */
 439        if (ext_depth(inode) != depth &&
 440            !ext4_extent_block_csum_verify(inode, eh)) {
 441                error_msg = "extent tree corrupted";
 442                goto corrupted;
 443        }
 444        return 0;
 445
 446corrupted:
 447        ext4_error_inode(inode, function, line, 0,
 448                        "bad header/extent: %s - magic %x, "
 449                        "entries %u, max %u(%u), depth %u(%u)",
 450                        error_msg, le16_to_cpu(eh->eh_magic),
 451                        le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 452                        max, le16_to_cpu(eh->eh_depth), depth);
 453
 454        return -EIO;
 455}
 456
 457#define ext4_ext_check(inode, eh, depth)        \
 458        __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 459
 460int ext4_ext_check_inode(struct inode *inode)
 461{
 462        return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 463}
 464
 465static int __ext4_ext_check_block(const char *function, unsigned int line,
 466                                  struct inode *inode,
 467                                  struct ext4_extent_header *eh,
 468                                  int depth,
 469                                  struct buffer_head *bh)
 470{
 471        int ret;
 472
 473        if (buffer_verified(bh))
 474                return 0;
 475        ret = ext4_ext_check(inode, eh, depth);
 476        if (ret)
 477                return ret;
 478        set_buffer_verified(bh);
 479        return ret;
 480}
 481
 482#define ext4_ext_check_block(inode, eh, depth, bh)      \
 483        __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
 484
 485#ifdef EXT_DEBUG
 486static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 487{
 488        int k, l = path->p_depth;
 489
 490        ext_debug("path:");
 491        for (k = 0; k <= l; k++, path++) {
 492                if (path->p_idx) {
 493                  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 494                            ext4_idx_pblock(path->p_idx));
 495                } else if (path->p_ext) {
 496                        ext_debug("  %d:[%d]%d:%llu ",
 497                                  le32_to_cpu(path->p_ext->ee_block),
 498                                  ext4_ext_is_uninitialized(path->p_ext),
 499                                  ext4_ext_get_actual_len(path->p_ext),
 500                                  ext4_ext_pblock(path->p_ext));
 501                } else
 502                        ext_debug("  []");
 503        }
 504        ext_debug("\n");
 505}
 506
 507static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 508{
 509        int depth = ext_depth(inode);
 510        struct ext4_extent_header *eh;
 511        struct ext4_extent *ex;
 512        int i;
 513
 514        if (!path)
 515                return;
 516
 517        eh = path[depth].p_hdr;
 518        ex = EXT_FIRST_EXTENT(eh);
 519
 520        ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 521
 522        for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 523                ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 524                          ext4_ext_is_uninitialized(ex),
 525                          ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 526        }
 527        ext_debug("\n");
 528}
 529
 530static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 531                        ext4_fsblk_t newblock, int level)
 532{
 533        int depth = ext_depth(inode);
 534        struct ext4_extent *ex;
 535
 536        if (depth != level) {
 537                struct ext4_extent_idx *idx;
 538                idx = path[level].p_idx;
 539                while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 540                        ext_debug("%d: move %d:%llu in new index %llu\n", level,
 541                                        le32_to_cpu(idx->ei_block),
 542                                        ext4_idx_pblock(idx),
 543                                        newblock);
 544                        idx++;
 545                }
 546
 547                return;
 548        }
 549
 550        ex = path[depth].p_ext;
 551        while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 552                ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 553                                le32_to_cpu(ex->ee_block),
 554                                ext4_ext_pblock(ex),
 555                                ext4_ext_is_uninitialized(ex),
 556                                ext4_ext_get_actual_len(ex),
 557                                newblock);
 558                ex++;
 559        }
 560}
 561
 562#else
 563#define ext4_ext_show_path(inode, path)
 564#define ext4_ext_show_leaf(inode, path)
 565#define ext4_ext_show_move(inode, path, newblock, level)
 566#endif
 567
 568void ext4_ext_drop_refs(struct ext4_ext_path *path)
 569{
 570        int depth = path->p_depth;
 571        int i;
 572
 573        for (i = 0; i <= depth; i++, path++)
 574                if (path->p_bh) {
 575                        brelse(path->p_bh);
 576                        path->p_bh = NULL;
 577                }
 578}
 579
 580/*
 581 * ext4_ext_binsearch_idx:
 582 * binary search for the closest index of the given block
 583 * the header must be checked before calling this
 584 */
 585static void
 586ext4_ext_binsearch_idx(struct inode *inode,
 587                        struct ext4_ext_path *path, ext4_lblk_t block)
 588{
 589        struct ext4_extent_header *eh = path->p_hdr;
 590        struct ext4_extent_idx *r, *l, *m;
 591
 592
 593        ext_debug("binsearch for %u(idx):  ", block);
 594
 595        l = EXT_FIRST_INDEX(eh) + 1;
 596        r = EXT_LAST_INDEX(eh);
 597        while (l <= r) {
 598                m = l + (r - l) / 2;
 599                if (block < le32_to_cpu(m->ei_block))
 600                        r = m - 1;
 601                else
 602                        l = m + 1;
 603                ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 604                                m, le32_to_cpu(m->ei_block),
 605                                r, le32_to_cpu(r->ei_block));
 606        }
 607
 608        path->p_idx = l - 1;
 609        ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 610                  ext4_idx_pblock(path->p_idx));
 611
 612#ifdef CHECK_BINSEARCH
 613        {
 614                struct ext4_extent_idx *chix, *ix;
 615                int k;
 616
 617                chix = ix = EXT_FIRST_INDEX(eh);
 618                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 619                  if (k != 0 &&
 620                      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 621                                printk(KERN_DEBUG "k=%d, ix=0x%p, "
 622                                       "first=0x%p\n", k,
 623                                       ix, EXT_FIRST_INDEX(eh));
 624                                printk(KERN_DEBUG "%u <= %u\n",
 625                                       le32_to_cpu(ix->ei_block),
 626                                       le32_to_cpu(ix[-1].ei_block));
 627                        }
 628                        BUG_ON(k && le32_to_cpu(ix->ei_block)
 629                                           <= le32_to_cpu(ix[-1].ei_block));
 630                        if (block < le32_to_cpu(ix->ei_block))
 631                                break;
 632                        chix = ix;
 633                }
 634                BUG_ON(chix != path->p_idx);
 635        }
 636#endif
 637
 638}
 639
 640/*
 641 * ext4_ext_binsearch:
 642 * binary search for closest extent of the given block
 643 * the header must be checked before calling this
 644 */
 645static void
 646ext4_ext_binsearch(struct inode *inode,
 647                struct ext4_ext_path *path, ext4_lblk_t block)
 648{
 649        struct ext4_extent_header *eh = path->p_hdr;
 650        struct ext4_extent *r, *l, *m;
 651
 652        if (eh->eh_entries == 0) {
 653                /*
 654                 * this leaf is empty:
 655                 * we get such a leaf in split/add case
 656                 */
 657                return;
 658        }
 659
 660        ext_debug("binsearch for %u:  ", block);
 661
 662        l = EXT_FIRST_EXTENT(eh) + 1;
 663        r = EXT_LAST_EXTENT(eh);
 664
 665        while (l <= r) {
 666                m = l + (r - l) / 2;
 667                if (block < le32_to_cpu(m->ee_block))
 668                        r = m - 1;
 669                else
 670                        l = m + 1;
 671                ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 672                                m, le32_to_cpu(m->ee_block),
 673                                r, le32_to_cpu(r->ee_block));
 674        }
 675
 676        path->p_ext = l - 1;
 677        ext_debug("  -> %d:%llu:[%d]%d ",
 678                        le32_to_cpu(path->p_ext->ee_block),
 679                        ext4_ext_pblock(path->p_ext),
 680                        ext4_ext_is_uninitialized(path->p_ext),
 681                        ext4_ext_get_actual_len(path->p_ext));
 682
 683#ifdef CHECK_BINSEARCH
 684        {
 685                struct ext4_extent *chex, *ex;
 686                int k;
 687
 688                chex = ex = EXT_FIRST_EXTENT(eh);
 689                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 690                        BUG_ON(k && le32_to_cpu(ex->ee_block)
 691                                          <= le32_to_cpu(ex[-1].ee_block));
 692                        if (block < le32_to_cpu(ex->ee_block))
 693                                break;
 694                        chex = ex;
 695                }
 696                BUG_ON(chex != path->p_ext);
 697        }
 698#endif
 699
 700}
 701
 702int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 703{
 704        struct ext4_extent_header *eh;
 705
 706        eh = ext_inode_hdr(inode);
 707        eh->eh_depth = 0;
 708        eh->eh_entries = 0;
 709        eh->eh_magic = EXT4_EXT_MAGIC;
 710        eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 711        ext4_mark_inode_dirty(handle, inode);
 712        ext4_ext_invalidate_cache(inode);
 713        return 0;
 714}
 715
 716struct ext4_ext_path *
 717ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 718                                        struct ext4_ext_path *path)
 719{
 720        struct ext4_extent_header *eh;
 721        struct buffer_head *bh;
 722        short int depth, i, ppos = 0, alloc = 0;
 723
 724        eh = ext_inode_hdr(inode);
 725        depth = ext_depth(inode);
 726
 727        /* account possible depth increase */
 728        if (!path) {
 729                path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 730                                GFP_NOFS);
 731                if (!path)
 732                        return ERR_PTR(-ENOMEM);
 733                alloc = 1;
 734        }
 735        path[0].p_hdr = eh;
 736        path[0].p_bh = NULL;
 737
 738        i = depth;
 739        /* walk through the tree */
 740        while (i) {
 741                ext_debug("depth %d: num %d, max %d\n",
 742                          ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 743
 744                ext4_ext_binsearch_idx(inode, path + ppos, block);
 745                path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 746                path[ppos].p_depth = i;
 747                path[ppos].p_ext = NULL;
 748
 749                bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 750                if (unlikely(!bh))
 751                        goto err;
 752                if (!bh_uptodate_or_lock(bh)) {
 753                        trace_ext4_ext_load_extent(inode, block,
 754                                                path[ppos].p_block);
 755                        if (bh_submit_read(bh) < 0) {
 756                                put_bh(bh);
 757                                goto err;
 758                        }
 759                }
 760                eh = ext_block_hdr(bh);
 761                ppos++;
 762                if (unlikely(ppos > depth)) {
 763                        put_bh(bh);
 764                        EXT4_ERROR_INODE(inode,
 765                                         "ppos %d > depth %d", ppos, depth);
 766                        goto err;
 767                }
 768                path[ppos].p_bh = bh;
 769                path[ppos].p_hdr = eh;
 770                i--;
 771
 772                if (ext4_ext_check_block(inode, eh, i, bh))
 773                        goto err;
 774        }
 775
 776        path[ppos].p_depth = i;
 777        path[ppos].p_ext = NULL;
 778        path[ppos].p_idx = NULL;
 779
 780        /* find extent */
 781        ext4_ext_binsearch(inode, path + ppos, block);
 782        /* if not an empty leaf */
 783        if (path[ppos].p_ext)
 784                path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 785
 786        ext4_ext_show_path(inode, path);
 787
 788        return path;
 789
 790err:
 791        ext4_ext_drop_refs(path);
 792        if (alloc)
 793                kfree(path);
 794        return ERR_PTR(-EIO);
 795}
 796
 797/*
 798 * ext4_ext_insert_index:
 799 * insert new index [@logical;@ptr] into the block at @curp;
 800 * check where to insert: before @curp or after @curp
 801 */
 802static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 803                                 struct ext4_ext_path *curp,
 804                                 int logical, ext4_fsblk_t ptr)
 805{
 806        struct ext4_extent_idx *ix;
 807        int len, err;
 808
 809        err = ext4_ext_get_access(handle, inode, curp);
 810        if (err)
 811                return err;
 812
 813        if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 814                EXT4_ERROR_INODE(inode,
 815                                 "logical %d == ei_block %d!",
 816                                 logical, le32_to_cpu(curp->p_idx->ei_block));
 817                return -EIO;
 818        }
 819
 820        if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 821                             >= le16_to_cpu(curp->p_hdr->eh_max))) {
 822                EXT4_ERROR_INODE(inode,
 823                                 "eh_entries %d >= eh_max %d!",
 824                                 le16_to_cpu(curp->p_hdr->eh_entries),
 825                                 le16_to_cpu(curp->p_hdr->eh_max));
 826                return -EIO;
 827        }
 828
 829        if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 830                /* insert after */
 831                ext_debug("insert new index %d after: %llu\n", logical, ptr);
 832                ix = curp->p_idx + 1;
 833        } else {
 834                /* insert before */
 835                ext_debug("insert new index %d before: %llu\n", logical, ptr);
 836                ix = curp->p_idx;
 837        }
 838
 839        len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 840        BUG_ON(len < 0);
 841        if (len > 0) {
 842                ext_debug("insert new index %d: "
 843                                "move %d indices from 0x%p to 0x%p\n",
 844                                logical, len, ix, ix + 1);
 845                memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 846        }
 847
 848        if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 849                EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 850                return -EIO;
 851        }
 852
 853        ix->ei_block = cpu_to_le32(logical);
 854        ext4_idx_store_pblock(ix, ptr);
 855        le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 856
 857        if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 858                EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 859                return -EIO;
 860        }
 861
 862        err = ext4_ext_dirty(handle, inode, curp);
 863        ext4_std_error(inode->i_sb, err);
 864
 865        return err;
 866}
 867
 868/*
 869 * ext4_ext_split:
 870 * inserts new subtree into the path, using free index entry
 871 * at depth @at:
 872 * - allocates all needed blocks (new leaf and all intermediate index blocks)
 873 * - makes decision where to split
 874 * - moves remaining extents and index entries (right to the split point)
 875 *   into the newly allocated blocks
 876 * - initializes subtree
 877 */
 878static int ext4_ext_split(handle_t *handle, struct inode *inode,
 879                          unsigned int flags,
 880                          struct ext4_ext_path *path,
 881                          struct ext4_extent *newext, int at)
 882{
 883        struct buffer_head *bh = NULL;
 884        int depth = ext_depth(inode);
 885        struct ext4_extent_header *neh;
 886        struct ext4_extent_idx *fidx;
 887        int i = at, k, m, a;
 888        ext4_fsblk_t newblock, oldblock;
 889        __le32 border;
 890        ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 891        int err = 0;
 892
 893        /* make decision: where to split? */
 894        /* FIXME: now decision is simplest: at current extent */
 895
 896        /* if current leaf will be split, then we should use
 897         * border from split point */
 898        if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 899                EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 900                return -EIO;
 901        }
 902        if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 903                border = path[depth].p_ext[1].ee_block;
 904                ext_debug("leaf will be split."
 905                                " next leaf starts at %d\n",
 906                                  le32_to_cpu(border));
 907        } else {
 908                border = newext->ee_block;
 909                ext_debug("leaf will be added."
 910                                " next leaf starts at %d\n",
 911                                le32_to_cpu(border));
 912        }
 913
 914        /*
 915         * If error occurs, then we break processing
 916         * and mark filesystem read-only. index won't
 917         * be inserted and tree will be in consistent
 918         * state. Next mount will repair buffers too.
 919         */
 920
 921        /*
 922         * Get array to track all allocated blocks.
 923         * We need this to handle errors and free blocks
 924         * upon them.
 925         */
 926        ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 927        if (!ablocks)
 928                return -ENOMEM;
 929
 930        /* allocate all needed blocks */
 931        ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 932        for (a = 0; a < depth - at; a++) {
 933                newblock = ext4_ext_new_meta_block(handle, inode, path,
 934                                                   newext, &err, flags);
 935                if (newblock == 0)
 936                        goto cleanup;
 937                ablocks[a] = newblock;
 938        }
 939
 940        /* initialize new leaf */
 941        newblock = ablocks[--a];
 942        if (unlikely(newblock == 0)) {
 943                EXT4_ERROR_INODE(inode, "newblock == 0!");
 944                err = -EIO;
 945                goto cleanup;
 946        }
 947        bh = sb_getblk(inode->i_sb, newblock);
 948        if (!bh) {
 949                err = -EIO;
 950                goto cleanup;
 951        }
 952        lock_buffer(bh);
 953
 954        err = ext4_journal_get_create_access(handle, bh);
 955        if (err)
 956                goto cleanup;
 957
 958        neh = ext_block_hdr(bh);
 959        neh->eh_entries = 0;
 960        neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 961        neh->eh_magic = EXT4_EXT_MAGIC;
 962        neh->eh_depth = 0;
 963
 964        /* move remainder of path[depth] to the new leaf */
 965        if (unlikely(path[depth].p_hdr->eh_entries !=
 966                     path[depth].p_hdr->eh_max)) {
 967                EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 968                                 path[depth].p_hdr->eh_entries,
 969                                 path[depth].p_hdr->eh_max);
 970                err = -EIO;
 971                goto cleanup;
 972        }
 973        /* start copy from next extent */
 974        m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 975        ext4_ext_show_move(inode, path, newblock, depth);
 976        if (m) {
 977                struct ext4_extent *ex;
 978                ex = EXT_FIRST_EXTENT(neh);
 979                memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 980                le16_add_cpu(&neh->eh_entries, m);
 981        }
 982
 983        ext4_extent_block_csum_set(inode, neh);
 984        set_buffer_uptodate(bh);
 985        unlock_buffer(bh);
 986
 987        err = ext4_handle_dirty_metadata(handle, inode, bh);
 988        if (err)
 989                goto cleanup;
 990        brelse(bh);
 991        bh = NULL;
 992
 993        /* correct old leaf */
 994        if (m) {
 995                err = ext4_ext_get_access(handle, inode, path + depth);
 996                if (err)
 997                        goto cleanup;
 998                le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 999                err = ext4_ext_dirty(handle, inode, path + depth);
1000                if (err)
1001                        goto cleanup;
1002
1003        }
1004
1005        /* create intermediate indexes */
1006        k = depth - at - 1;
1007        if (unlikely(k < 0)) {
1008                EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1009                err = -EIO;
1010                goto cleanup;
1011        }
1012        if (k)
1013                ext_debug("create %d intermediate indices\n", k);
1014        /* insert new index into current index block */
1015        /* current depth stored in i var */
1016        i = depth - 1;
1017        while (k--) {
1018                oldblock = newblock;
1019                newblock = ablocks[--a];
1020                bh = sb_getblk(inode->i_sb, newblock);
1021                if (!bh) {
1022                        err = -EIO;
1023                        goto cleanup;
1024                }
1025                lock_buffer(bh);
1026
1027                err = ext4_journal_get_create_access(handle, bh);
1028                if (err)
1029                        goto cleanup;
1030
1031                neh = ext_block_hdr(bh);
1032                neh->eh_entries = cpu_to_le16(1);
1033                neh->eh_magic = EXT4_EXT_MAGIC;
1034                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1035                neh->eh_depth = cpu_to_le16(depth - i);
1036                fidx = EXT_FIRST_INDEX(neh);
1037                fidx->ei_block = border;
1038                ext4_idx_store_pblock(fidx, oldblock);
1039
1040                ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1041                                i, newblock, le32_to_cpu(border), oldblock);
1042
1043                /* move remainder of path[i] to the new index block */
1044                if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1045                                        EXT_LAST_INDEX(path[i].p_hdr))) {
1046                        EXT4_ERROR_INODE(inode,
1047                                         "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1048                                         le32_to_cpu(path[i].p_ext->ee_block));
1049                        err = -EIO;
1050                        goto cleanup;
1051                }
1052                /* start copy indexes */
1053                m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1054                ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1055                                EXT_MAX_INDEX(path[i].p_hdr));
1056                ext4_ext_show_move(inode, path, newblock, i);
1057                if (m) {
1058                        memmove(++fidx, path[i].p_idx,
1059                                sizeof(struct ext4_extent_idx) * m);
1060                        le16_add_cpu(&neh->eh_entries, m);
1061                }
1062                ext4_extent_block_csum_set(inode, neh);
1063                set_buffer_uptodate(bh);
1064                unlock_buffer(bh);
1065
1066                err = ext4_handle_dirty_metadata(handle, inode, bh);
1067                if (err)
1068                        goto cleanup;
1069                brelse(bh);
1070                bh = NULL;
1071
1072                /* correct old index */
1073                if (m) {
1074                        err = ext4_ext_get_access(handle, inode, path + i);
1075                        if (err)
1076                                goto cleanup;
1077                        le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1078                        err = ext4_ext_dirty(handle, inode, path + i);
1079                        if (err)
1080                                goto cleanup;
1081                }
1082
1083                i--;
1084        }
1085
1086        /* insert new index */
1087        err = ext4_ext_insert_index(handle, inode, path + at,
1088                                    le32_to_cpu(border), newblock);
1089
1090cleanup:
1091        if (bh) {
1092                if (buffer_locked(bh))
1093                        unlock_buffer(bh);
1094                brelse(bh);
1095        }
1096
1097        if (err) {
1098                /* free all allocated blocks in error case */
1099                for (i = 0; i < depth; i++) {
1100                        if (!ablocks[i])
1101                                continue;
1102                        ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1103                                         EXT4_FREE_BLOCKS_METADATA);
1104                }
1105        }
1106        kfree(ablocks);
1107
1108        return err;
1109}
1110
1111/*
1112 * ext4_ext_grow_indepth:
1113 * implements tree growing procedure:
1114 * - allocates new block
1115 * - moves top-level data (index block or leaf) into the new block
1116 * - initializes new top-level, creating index that points to the
1117 *   just created block
1118 */
1119static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1120                                 unsigned int flags,
1121                                 struct ext4_extent *newext)
1122{
1123        struct ext4_extent_header *neh;
1124        struct buffer_head *bh;
1125        ext4_fsblk_t newblock;
1126        int err = 0;
1127
1128        newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1129                newext, &err, flags);
1130        if (newblock == 0)
1131                return err;
1132
1133        bh = sb_getblk(inode->i_sb, newblock);
1134        if (!bh) {
1135                err = -EIO;
1136                ext4_std_error(inode->i_sb, err);
1137                return err;
1138        }
1139        lock_buffer(bh);
1140
1141        err = ext4_journal_get_create_access(handle, bh);
1142        if (err) {
1143                unlock_buffer(bh);
1144                goto out;
1145        }
1146
1147        /* move top-level index/leaf into new block */
1148        memmove(bh->b_data, EXT4_I(inode)->i_data,
1149                sizeof(EXT4_I(inode)->i_data));
1150
1151        /* set size of new block */
1152        neh = ext_block_hdr(bh);
1153        /* old root could have indexes or leaves
1154         * so calculate e_max right way */
1155        if (ext_depth(inode))
1156                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1157        else
1158                neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1159        neh->eh_magic = EXT4_EXT_MAGIC;
1160        ext4_extent_block_csum_set(inode, neh);
1161        set_buffer_uptodate(bh);
1162        unlock_buffer(bh);
1163
1164        err = ext4_handle_dirty_metadata(handle, inode, bh);
1165        if (err)
1166                goto out;
1167
1168        /* Update top-level index: num,max,pointer */
1169        neh = ext_inode_hdr(inode);
1170        neh->eh_entries = cpu_to_le16(1);
1171        ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1172        if (neh->eh_depth == 0) {
1173                /* Root extent block becomes index block */
1174                neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1175                EXT_FIRST_INDEX(neh)->ei_block =
1176                        EXT_FIRST_EXTENT(neh)->ee_block;
1177        }
1178        ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1179                  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1180                  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1181                  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1182
1183        le16_add_cpu(&neh->eh_depth, 1);
1184        ext4_mark_inode_dirty(handle, inode);
1185out:
1186        brelse(bh);
1187
1188        return err;
1189}
1190
1191/*
1192 * ext4_ext_create_new_leaf:
1193 * finds empty index and adds new leaf.
1194 * if no free index is found, then it requests in-depth growing.
1195 */
1196static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1197                                    unsigned int flags,
1198                                    struct ext4_ext_path *path,
1199                                    struct ext4_extent *newext)
1200{
1201        struct ext4_ext_path *curp;
1202        int depth, i, err = 0;
1203
1204repeat:
1205        i = depth = ext_depth(inode);
1206
1207        /* walk up to the tree and look for free index entry */
1208        curp = path + depth;
1209        while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1210                i--;
1211                curp--;
1212        }
1213
1214        /* we use already allocated block for index block,
1215         * so subsequent data blocks should be contiguous */
1216        if (EXT_HAS_FREE_INDEX(curp)) {
1217                /* if we found index with free entry, then use that
1218                 * entry: create all needed subtree and add new leaf */
1219                err = ext4_ext_split(handle, inode, flags, path, newext, i);
1220                if (err)
1221                        goto out;
1222
1223                /* refill path */
1224                ext4_ext_drop_refs(path);
1225                path = ext4_ext_find_extent(inode,
1226                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1227                                    path);
1228                if (IS_ERR(path))
1229                        err = PTR_ERR(path);
1230        } else {
1231                /* tree is full, time to grow in depth */
1232                err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1233                if (err)
1234                        goto out;
1235
1236                /* refill path */
1237                ext4_ext_drop_refs(path);
1238                path = ext4_ext_find_extent(inode,
1239                                   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1240                                    path);
1241                if (IS_ERR(path)) {
1242                        err = PTR_ERR(path);
1243                        goto out;
1244                }
1245
1246                /*
1247                 * only first (depth 0 -> 1) produces free space;
1248                 * in all other cases we have to split the grown tree
1249                 */
1250                depth = ext_depth(inode);
1251                if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1252                        /* now we need to split */
1253                        goto repeat;
1254                }
1255        }
1256
1257out:
1258        return err;
1259}
1260
1261/*
1262 * search the closest allocated block to the left for *logical
1263 * and returns it at @logical + it's physical address at @phys
1264 * if *logical is the smallest allocated block, the function
1265 * returns 0 at @phys
1266 * return value contains 0 (success) or error code
1267 */
1268static int ext4_ext_search_left(struct inode *inode,
1269                                struct ext4_ext_path *path,
1270                                ext4_lblk_t *logical, ext4_fsblk_t *phys)
1271{
1272        struct ext4_extent_idx *ix;
1273        struct ext4_extent *ex;
1274        int depth, ee_len;
1275
1276        if (unlikely(path == NULL)) {
1277                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1278                return -EIO;
1279        }
1280        depth = path->p_depth;
1281        *phys = 0;
1282
1283        if (depth == 0 && path->p_ext == NULL)
1284                return 0;
1285
1286        /* usually extent in the path covers blocks smaller
1287         * then *logical, but it can be that extent is the
1288         * first one in the file */
1289
1290        ex = path[depth].p_ext;
1291        ee_len = ext4_ext_get_actual_len(ex);
1292        if (*logical < le32_to_cpu(ex->ee_block)) {
1293                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1294                        EXT4_ERROR_INODE(inode,
1295                                         "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1296                                         *logical, le32_to_cpu(ex->ee_block));
1297                        return -EIO;
1298                }
1299                while (--depth >= 0) {
1300                        ix = path[depth].p_idx;
1301                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1302                                EXT4_ERROR_INODE(inode,
1303                                  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1304                                  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1305                                  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1306                le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1307                                  depth);
1308                                return -EIO;
1309                        }
1310                }
1311                return 0;
1312        }
1313
1314        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1315                EXT4_ERROR_INODE(inode,
1316                                 "logical %d < ee_block %d + ee_len %d!",
1317                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1318                return -EIO;
1319        }
1320
1321        *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1322        *phys = ext4_ext_pblock(ex) + ee_len - 1;
1323        return 0;
1324}
1325
1326/*
1327 * search the closest allocated block to the right for *logical
1328 * and returns it at @logical + it's physical address at @phys
1329 * if *logical is the largest allocated block, the function
1330 * returns 0 at @phys
1331 * return value contains 0 (success) or error code
1332 */
1333static int ext4_ext_search_right(struct inode *inode,
1334                                 struct ext4_ext_path *path,
1335                                 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1336                                 struct ext4_extent **ret_ex)
1337{
1338        struct buffer_head *bh = NULL;
1339        struct ext4_extent_header *eh;
1340        struct ext4_extent_idx *ix;
1341        struct ext4_extent *ex;
1342        ext4_fsblk_t block;
1343        int depth;      /* Note, NOT eh_depth; depth from top of tree */
1344        int ee_len;
1345
1346        if (unlikely(path == NULL)) {
1347                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1348                return -EIO;
1349        }
1350        depth = path->p_depth;
1351        *phys = 0;
1352
1353        if (depth == 0 && path->p_ext == NULL)
1354                return 0;
1355
1356        /* usually extent in the path covers blocks smaller
1357         * then *logical, but it can be that extent is the
1358         * first one in the file */
1359
1360        ex = path[depth].p_ext;
1361        ee_len = ext4_ext_get_actual_len(ex);
1362        if (*logical < le32_to_cpu(ex->ee_block)) {
1363                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1364                        EXT4_ERROR_INODE(inode,
1365                                         "first_extent(path[%d].p_hdr) != ex",
1366                                         depth);
1367                        return -EIO;
1368                }
1369                while (--depth >= 0) {
1370                        ix = path[depth].p_idx;
1371                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1372                                EXT4_ERROR_INODE(inode,
1373                                                 "ix != EXT_FIRST_INDEX *logical %d!",
1374                                                 *logical);
1375                                return -EIO;
1376                        }
1377                }
1378                goto found_extent;
1379        }
1380
1381        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1382                EXT4_ERROR_INODE(inode,
1383                                 "logical %d < ee_block %d + ee_len %d!",
1384                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1385                return -EIO;
1386        }
1387
1388        if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1389                /* next allocated block in this leaf */
1390                ex++;
1391                goto found_extent;
1392        }
1393
1394        /* go up and search for index to the right */
1395        while (--depth >= 0) {
1396                ix = path[depth].p_idx;
1397                if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1398                        goto got_index;
1399        }
1400
1401        /* we've gone up to the root and found no index to the right */
1402        return 0;
1403
1404got_index:
1405        /* we've found index to the right, let's
1406         * follow it and find the closest allocated
1407         * block to the right */
1408        ix++;
1409        block = ext4_idx_pblock(ix);
1410        while (++depth < path->p_depth) {
1411                bh = sb_bread(inode->i_sb, block);
1412                if (bh == NULL)
1413                        return -EIO;
1414                eh = ext_block_hdr(bh);
1415                /* subtract from p_depth to get proper eh_depth */
1416                if (ext4_ext_check_block(inode, eh,
1417                                         path->p_depth - depth, bh)) {
1418                        put_bh(bh);
1419                        return -EIO;
1420                }
1421                ix = EXT_FIRST_INDEX(eh);
1422                block = ext4_idx_pblock(ix);
1423                put_bh(bh);
1424        }
1425
1426        bh = sb_bread(inode->i_sb, block);
1427        if (bh == NULL)
1428                return -EIO;
1429        eh = ext_block_hdr(bh);
1430        if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) {
1431                put_bh(bh);
1432                return -EIO;
1433        }
1434        ex = EXT_FIRST_EXTENT(eh);
1435found_extent:
1436        *logical = le32_to_cpu(ex->ee_block);
1437        *phys = ext4_ext_pblock(ex);
1438        *ret_ex = ex;
1439        if (bh)
1440                put_bh(bh);
1441        return 0;
1442}
1443
1444/*
1445 * ext4_ext_next_allocated_block:
1446 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1447 * NOTE: it considers block number from index entry as
1448 * allocated block. Thus, index entries have to be consistent
1449 * with leaves.
1450 */
1451static ext4_lblk_t
1452ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1453{
1454        int depth;
1455
1456        BUG_ON(path == NULL);
1457        depth = path->p_depth;
1458
1459        if (depth == 0 && path->p_ext == NULL)
1460                return EXT_MAX_BLOCKS;
1461
1462        while (depth >= 0) {
1463                if (depth == path->p_depth) {
1464                        /* leaf */
1465                        if (path[depth].p_ext &&
1466                                path[depth].p_ext !=
1467                                        EXT_LAST_EXTENT(path[depth].p_hdr))
1468                          return le32_to_cpu(path[depth].p_ext[1].ee_block);
1469                } else {
1470                        /* index */
1471                        if (path[depth].p_idx !=
1472                                        EXT_LAST_INDEX(path[depth].p_hdr))
1473                          return le32_to_cpu(path[depth].p_idx[1].ei_block);
1474                }
1475                depth--;
1476        }
1477
1478        return EXT_MAX_BLOCKS;
1479}
1480
1481/*
1482 * ext4_ext_next_leaf_block:
1483 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1484 */
1485static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1486{
1487        int depth;
1488
1489        BUG_ON(path == NULL);
1490        depth = path->p_depth;
1491
1492        /* zero-tree has no leaf blocks at all */
1493        if (depth == 0)
1494                return EXT_MAX_BLOCKS;
1495
1496        /* go to index block */
1497        depth--;
1498
1499        while (depth >= 0) {
1500                if (path[depth].p_idx !=
1501                                EXT_LAST_INDEX(path[depth].p_hdr))
1502                        return (ext4_lblk_t)
1503                                le32_to_cpu(path[depth].p_idx[1].ei_block);
1504                depth--;
1505        }
1506
1507        return EXT_MAX_BLOCKS;
1508}
1509
1510/*
1511 * ext4_ext_correct_indexes:
1512 * if leaf gets modified and modified extent is first in the leaf,
1513 * then we have to correct all indexes above.
1514 * TODO: do we need to correct tree in all cases?
1515 */
1516static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1517                                struct ext4_ext_path *path)
1518{
1519        struct ext4_extent_header *eh;
1520        int depth = ext_depth(inode);
1521        struct ext4_extent *ex;
1522        __le32 border;
1523        int k, err = 0;
1524
1525        eh = path[depth].p_hdr;
1526        ex = path[depth].p_ext;
1527
1528        if (unlikely(ex == NULL || eh == NULL)) {
1529                EXT4_ERROR_INODE(inode,
1530                                 "ex %p == NULL or eh %p == NULL", ex, eh);
1531                return -EIO;
1532        }
1533
1534        if (depth == 0) {
1535                /* there is no tree at all */
1536                return 0;
1537        }
1538
1539        if (ex != EXT_FIRST_EXTENT(eh)) {
1540                /* we correct tree if first leaf got modified only */
1541                return 0;
1542        }
1543
1544        /*
1545         * TODO: we need correction if border is smaller than current one
1546         */
1547        k = depth - 1;
1548        border = path[depth].p_ext->ee_block;
1549        err = ext4_ext_get_access(handle, inode, path + k);
1550        if (err)
1551                return err;
1552        path[k].p_idx->ei_block = border;
1553        err = ext4_ext_dirty(handle, inode, path + k);
1554        if (err)
1555                return err;
1556
1557        while (k--) {
1558                /* change all left-side indexes */
1559                if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1560                        break;
1561                err = ext4_ext_get_access(handle, inode, path + k);
1562                if (err)
1563                        break;
1564                path[k].p_idx->ei_block = border;
1565                err = ext4_ext_dirty(handle, inode, path + k);
1566                if (err)
1567                        break;
1568        }
1569
1570        return err;
1571}
1572
1573int
1574ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1575                                struct ext4_extent *ex2)
1576{
1577        unsigned short ext1_ee_len, ext2_ee_len, max_len;
1578
1579        /*
1580         * Make sure that either both extents are uninitialized, or
1581         * both are _not_.
1582         */
1583        if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1584                return 0;
1585
1586        if (ext4_ext_is_uninitialized(ex1))
1587                max_len = EXT_UNINIT_MAX_LEN;
1588        else
1589                max_len = EXT_INIT_MAX_LEN;
1590
1591        ext1_ee_len = ext4_ext_get_actual_len(ex1);
1592        ext2_ee_len = ext4_ext_get_actual_len(ex2);
1593
1594        if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1595                        le32_to_cpu(ex2->ee_block))
1596                return 0;
1597
1598        /*
1599         * To allow future support for preallocated extents to be added
1600         * as an RO_COMPAT feature, refuse to merge to extents if
1601         * this can result in the top bit of ee_len being set.
1602         */
1603        if (ext1_ee_len + ext2_ee_len > max_len)
1604                return 0;
1605#ifdef AGGRESSIVE_TEST
1606        if (ext1_ee_len >= 4)
1607                return 0;
1608#endif
1609
1610        if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1611                return 1;
1612        return 0;
1613}
1614
1615/*
1616 * This function tries to merge the "ex" extent to the next extent in the tree.
1617 * It always tries to merge towards right. If you want to merge towards
1618 * left, pass "ex - 1" as argument instead of "ex".
1619 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1620 * 1 if they got merged.
1621 */
1622static int ext4_ext_try_to_merge_right(struct inode *inode,
1623                                 struct ext4_ext_path *path,
1624                                 struct ext4_extent *ex)
1625{
1626        struct ext4_extent_header *eh;
1627        unsigned int depth, len;
1628        int merge_done = 0;
1629        int uninitialized = 0;
1630
1631        depth = ext_depth(inode);
1632        BUG_ON(path[depth].p_hdr == NULL);
1633        eh = path[depth].p_hdr;
1634
1635        while (ex < EXT_LAST_EXTENT(eh)) {
1636                if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1637                        break;
1638                /* merge with next extent! */
1639                if (ext4_ext_is_uninitialized(ex))
1640                        uninitialized = 1;
1641                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1642                                + ext4_ext_get_actual_len(ex + 1));
1643                if (uninitialized)
1644                        ext4_ext_mark_uninitialized(ex);
1645
1646                if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1647                        len = (EXT_LAST_EXTENT(eh) - ex - 1)
1648                                * sizeof(struct ext4_extent);
1649                        memmove(ex + 1, ex + 2, len);
1650                }
1651                le16_add_cpu(&eh->eh_entries, -1);
1652                merge_done = 1;
1653                WARN_ON(eh->eh_entries == 0);
1654                if (!eh->eh_entries)
1655                        EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1656        }
1657
1658        return merge_done;
1659}
1660
1661/*
1662 * This function does a very simple check to see if we can collapse
1663 * an extent tree with a single extent tree leaf block into the inode.
1664 */
1665static void ext4_ext_try_to_merge_up(handle_t *handle,
1666                                     struct inode *inode,
1667                                     struct ext4_ext_path *path)
1668{
1669        size_t s;
1670        unsigned max_root = ext4_ext_space_root(inode, 0);
1671        ext4_fsblk_t blk;
1672
1673        if ((path[0].p_depth != 1) ||
1674            (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1675            (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1676                return;
1677
1678        /*
1679         * We need to modify the block allocation bitmap and the block
1680         * group descriptor to release the extent tree block.  If we
1681         * can't get the journal credits, give up.
1682         */
1683        if (ext4_journal_extend(handle, 2))
1684                return;
1685
1686        /*
1687         * Copy the extent data up to the inode
1688         */
1689        blk = ext4_idx_pblock(path[0].p_idx);
1690        s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1691                sizeof(struct ext4_extent_idx);
1692        s += sizeof(struct ext4_extent_header);
1693
1694        memcpy(path[0].p_hdr, path[1].p_hdr, s);
1695        path[0].p_depth = 0;
1696        path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1697                (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1698        path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1699
1700        brelse(path[1].p_bh);
1701        ext4_free_blocks(handle, inode, NULL, blk, 1,
1702                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1703}
1704
1705/*
1706 * This function tries to merge the @ex extent to neighbours in the tree.
1707 * return 1 if merge left else 0.
1708 */
1709static void ext4_ext_try_to_merge(handle_t *handle,
1710                                  struct inode *inode,
1711                                  struct ext4_ext_path *path,
1712                                  struct ext4_extent *ex) {
1713        struct ext4_extent_header *eh;
1714        unsigned int depth;
1715        int merge_done = 0;
1716
1717        depth = ext_depth(inode);
1718        BUG_ON(path[depth].p_hdr == NULL);
1719        eh = path[depth].p_hdr;
1720
1721        if (ex > EXT_FIRST_EXTENT(eh))
1722                merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1723
1724        if (!merge_done)
1725                (void) ext4_ext_try_to_merge_right(inode, path, ex);
1726
1727        ext4_ext_try_to_merge_up(handle, inode, path);
1728}
1729
1730/*
1731 * check if a portion of the "newext" extent overlaps with an
1732 * existing extent.
1733 *
1734 * If there is an overlap discovered, it updates the length of the newext
1735 * such that there will be no overlap, and then returns 1.
1736 * If there is no overlap found, it returns 0.
1737 */
1738static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1739                                           struct inode *inode,
1740                                           struct ext4_extent *newext,
1741                                           struct ext4_ext_path *path)
1742{
1743        ext4_lblk_t b1, b2;
1744        unsigned int depth, len1;
1745        unsigned int ret = 0;
1746
1747        b1 = le32_to_cpu(newext->ee_block);
1748        len1 = ext4_ext_get_actual_len(newext);
1749        depth = ext_depth(inode);
1750        if (!path[depth].p_ext)
1751                goto out;
1752        b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1753        b2 &= ~(sbi->s_cluster_ratio - 1);
1754
1755        /*
1756         * get the next allocated block if the extent in the path
1757         * is before the requested block(s)
1758         */
1759        if (b2 < b1) {
1760                b2 = ext4_ext_next_allocated_block(path);
1761                if (b2 == EXT_MAX_BLOCKS)
1762                        goto out;
1763                b2 &= ~(sbi->s_cluster_ratio - 1);
1764        }
1765
1766        /* check for wrap through zero on extent logical start block*/
1767        if (b1 + len1 < b1) {
1768                len1 = EXT_MAX_BLOCKS - b1;
1769                newext->ee_len = cpu_to_le16(len1);
1770                ret = 1;
1771        }
1772
1773        /* check for overlap */
1774        if (b1 + len1 > b2) {
1775                newext->ee_len = cpu_to_le16(b2 - b1);
1776                ret = 1;
1777        }
1778out:
1779        return ret;
1780}
1781
1782/*
1783 * ext4_ext_insert_extent:
1784 * tries to merge requsted extent into the existing extent or
1785 * inserts requested extent as new one into the tree,
1786 * creating new leaf in the no-space case.
1787 */
1788int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1789                                struct ext4_ext_path *path,
1790                                struct ext4_extent *newext, int flag)
1791{
1792        struct ext4_extent_header *eh;
1793        struct ext4_extent *ex, *fex;
1794        struct ext4_extent *nearex; /* nearest extent */
1795        struct ext4_ext_path *npath = NULL;
1796        int depth, len, err;
1797        ext4_lblk_t next;
1798        unsigned uninitialized = 0;
1799        int flags = 0;
1800
1801        if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1802                EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1803                return -EIO;
1804        }
1805        depth = ext_depth(inode);
1806        ex = path[depth].p_ext;
1807        if (unlikely(path[depth].p_hdr == NULL)) {
1808                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1809                return -EIO;
1810        }
1811
1812        /* try to insert block into found extent and return */
1813        if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1814                && ext4_can_extents_be_merged(inode, ex, newext)) {
1815                ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1816                          ext4_ext_is_uninitialized(newext),
1817                          ext4_ext_get_actual_len(newext),
1818                          le32_to_cpu(ex->ee_block),
1819                          ext4_ext_is_uninitialized(ex),
1820                          ext4_ext_get_actual_len(ex),
1821                          ext4_ext_pblock(ex));
1822                err = ext4_ext_get_access(handle, inode, path + depth);
1823                if (err)
1824                        return err;
1825
1826                /*
1827                 * ext4_can_extents_be_merged should have checked that either
1828                 * both extents are uninitialized, or both aren't. Thus we
1829                 * need to check only one of them here.
1830                 */
1831                if (ext4_ext_is_uninitialized(ex))
1832                        uninitialized = 1;
1833                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1834                                        + ext4_ext_get_actual_len(newext));
1835                if (uninitialized)
1836                        ext4_ext_mark_uninitialized(ex);
1837                eh = path[depth].p_hdr;
1838                nearex = ex;
1839                goto merge;
1840        }
1841
1842        depth = ext_depth(inode);
1843        eh = path[depth].p_hdr;
1844        if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1845                goto has_space;
1846
1847        /* probably next leaf has space for us? */
1848        fex = EXT_LAST_EXTENT(eh);
1849        next = EXT_MAX_BLOCKS;
1850        if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1851                next = ext4_ext_next_leaf_block(path);
1852        if (next != EXT_MAX_BLOCKS) {
1853                ext_debug("next leaf block - %u\n", next);
1854                BUG_ON(npath != NULL);
1855                npath = ext4_ext_find_extent(inode, next, NULL);
1856                if (IS_ERR(npath))
1857                        return PTR_ERR(npath);
1858                BUG_ON(npath->p_depth != path->p_depth);
1859                eh = npath[depth].p_hdr;
1860                if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1861                        ext_debug("next leaf isn't full(%d)\n",
1862                                  le16_to_cpu(eh->eh_entries));
1863                        path = npath;
1864                        goto has_space;
1865                }
1866                ext_debug("next leaf has no free space(%d,%d)\n",
1867                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1868        }
1869
1870        /*
1871         * There is no free space in the found leaf.
1872         * We're gonna add a new leaf in the tree.
1873         */
1874        if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1875                flags = EXT4_MB_USE_ROOT_BLOCKS;
1876        err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1877        if (err)
1878                goto cleanup;
1879        depth = ext_depth(inode);
1880        eh = path[depth].p_hdr;
1881
1882has_space:
1883        nearex = path[depth].p_ext;
1884
1885        err = ext4_ext_get_access(handle, inode, path + depth);
1886        if (err)
1887                goto cleanup;
1888
1889        if (!nearex) {
1890                /* there is no extent in this leaf, create first one */
1891                ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1892                                le32_to_cpu(newext->ee_block),
1893                                ext4_ext_pblock(newext),
1894                                ext4_ext_is_uninitialized(newext),
1895                                ext4_ext_get_actual_len(newext));
1896                nearex = EXT_FIRST_EXTENT(eh);
1897        } else {
1898                if (le32_to_cpu(newext->ee_block)
1899                           > le32_to_cpu(nearex->ee_block)) {
1900                        /* Insert after */
1901                        ext_debug("insert %u:%llu:[%d]%d before: "
1902                                        "nearest %p\n",
1903                                        le32_to_cpu(newext->ee_block),
1904                                        ext4_ext_pblock(newext),
1905                                        ext4_ext_is_uninitialized(newext),
1906                                        ext4_ext_get_actual_len(newext),
1907                                        nearex);
1908                        nearex++;
1909                } else {
1910                        /* Insert before */
1911                        BUG_ON(newext->ee_block == nearex->ee_block);
1912                        ext_debug("insert %u:%llu:[%d]%d after: "
1913                                        "nearest %p\n",
1914                                        le32_to_cpu(newext->ee_block),
1915                                        ext4_ext_pblock(newext),
1916                                        ext4_ext_is_uninitialized(newext),
1917                                        ext4_ext_get_actual_len(newext),
1918                                        nearex);
1919                }
1920                len = EXT_LAST_EXTENT(eh) - nearex + 1;
1921                if (len > 0) {
1922                        ext_debug("insert %u:%llu:[%d]%d: "
1923                                        "move %d extents from 0x%p to 0x%p\n",
1924                                        le32_to_cpu(newext->ee_block),
1925                                        ext4_ext_pblock(newext),
1926                                        ext4_ext_is_uninitialized(newext),
1927                                        ext4_ext_get_actual_len(newext),
1928                                        len, nearex, nearex + 1);
1929                        memmove(nearex + 1, nearex,
1930                                len * sizeof(struct ext4_extent));
1931                }
1932        }
1933
1934        le16_add_cpu(&eh->eh_entries, 1);
1935        path[depth].p_ext = nearex;
1936        nearex->ee_block = newext->ee_block;
1937        ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1938        nearex->ee_len = newext->ee_len;
1939
1940merge:
1941        /* try to merge extents */
1942        if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1943                ext4_ext_try_to_merge(handle, inode, path, nearex);
1944
1945
1946        /* time to correct all indexes above */
1947        err = ext4_ext_correct_indexes(handle, inode, path);
1948        if (err)
1949                goto cleanup;
1950
1951        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
1952
1953cleanup:
1954        if (npath) {
1955                ext4_ext_drop_refs(npath);
1956                kfree(npath);
1957        }
1958        ext4_ext_invalidate_cache(inode);
1959        return err;
1960}
1961
1962static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1963                               ext4_lblk_t num, ext_prepare_callback func,
1964                               void *cbdata)
1965{
1966        struct ext4_ext_path *path = NULL;
1967        struct ext4_ext_cache cbex;
1968        struct ext4_extent *ex;
1969        ext4_lblk_t next, start = 0, end = 0;
1970        ext4_lblk_t last = block + num;
1971        int depth, exists, err = 0;
1972
1973        BUG_ON(func == NULL);
1974        BUG_ON(inode == NULL);
1975
1976        while (block < last && block != EXT_MAX_BLOCKS) {
1977                num = last - block;
1978                /* find extent for this block */
1979                down_read(&EXT4_I(inode)->i_data_sem);
1980                path = ext4_ext_find_extent(inode, block, path);
1981                up_read(&EXT4_I(inode)->i_data_sem);
1982                if (IS_ERR(path)) {
1983                        err = PTR_ERR(path);
1984                        path = NULL;
1985                        break;
1986                }
1987
1988                depth = ext_depth(inode);
1989                if (unlikely(path[depth].p_hdr == NULL)) {
1990                        EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1991                        err = -EIO;
1992                        break;
1993                }
1994                ex = path[depth].p_ext;
1995                next = ext4_ext_next_allocated_block(path);
1996
1997                exists = 0;
1998                if (!ex) {
1999                        /* there is no extent yet, so try to allocate
2000                         * all requested space */
2001                        start = block;
2002                        end = block + num;
2003                } else if (le32_to_cpu(ex->ee_block) > block) {
2004                        /* need to allocate space before found extent */
2005                        start = block;
2006                        end = le32_to_cpu(ex->ee_block);
2007                        if (block + num < end)
2008                                end = block + num;
2009                } else if (block >= le32_to_cpu(ex->ee_block)
2010                                        + ext4_ext_get_actual_len(ex)) {
2011                        /* need to allocate space after found extent */
2012                        start = block;
2013                        end = block + num;
2014                        if (end >= next)
2015                                end = next;
2016                } else if (block >= le32_to_cpu(ex->ee_block)) {
2017                        /*
2018                         * some part of requested space is covered
2019                         * by found extent
2020                         */
2021                        start = block;
2022                        end = le32_to_cpu(ex->ee_block)
2023                                + ext4_ext_get_actual_len(ex);
2024                        if (block + num < end)
2025                                end = block + num;
2026                        exists = 1;
2027                } else {
2028                        BUG();
2029                }
2030                BUG_ON(end <= start);
2031
2032                if (!exists) {
2033                        cbex.ec_block = start;
2034                        cbex.ec_len = end - start;
2035                        cbex.ec_start = 0;
2036                } else {
2037                        cbex.ec_block = le32_to_cpu(ex->ee_block);
2038                        cbex.ec_len = ext4_ext_get_actual_len(ex);
2039                        cbex.ec_start = ext4_ext_pblock(ex);
2040                }
2041
2042                if (unlikely(cbex.ec_len == 0)) {
2043                        EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
2044                        err = -EIO;
2045                        break;
2046                }
2047                err = func(inode, next, &cbex, ex, cbdata);
2048                ext4_ext_drop_refs(path);
2049
2050                if (err < 0)
2051                        break;
2052
2053                if (err == EXT_REPEAT)
2054                        continue;
2055                else if (err == EXT_BREAK) {
2056                        err = 0;
2057                        break;
2058                }
2059
2060                if (ext_depth(inode) != depth) {
2061                        /* depth was changed. we have to realloc path */
2062                        kfree(path);
2063                        path = NULL;
2064                }
2065
2066                block = cbex.ec_block + cbex.ec_len;
2067        }
2068
2069        if (path) {
2070                ext4_ext_drop_refs(path);
2071                kfree(path);
2072        }
2073
2074        return err;
2075}
2076
2077static void
2078ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
2079                        __u32 len, ext4_fsblk_t start)
2080{
2081        struct ext4_ext_cache *cex;
2082        BUG_ON(len == 0);
2083        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2084        trace_ext4_ext_put_in_cache(inode, block, len, start);
2085        cex = &EXT4_I(inode)->i_cached_extent;
2086        cex->ec_block = block;
2087        cex->ec_len = len;
2088        cex->ec_start = start;
2089        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2090}
2091
2092/*
2093 * ext4_ext_put_gap_in_cache:
2094 * calculate boundaries of the gap that the requested block fits into
2095 * and cache this gap
2096 */
2097static void
2098ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
2099                                ext4_lblk_t block)
2100{
2101        int depth = ext_depth(inode);
2102        unsigned long len;
2103        ext4_lblk_t lblock;
2104        struct ext4_extent *ex;
2105
2106        ex = path[depth].p_ext;
2107        if (ex == NULL) {
2108                /* there is no extent yet, so gap is [0;-] */
2109                lblock = 0;
2110                len = EXT_MAX_BLOCKS;
2111                ext_debug("cache gap(whole file):");
2112        } else if (block < le32_to_cpu(ex->ee_block)) {
2113                lblock = block;
2114                len = le32_to_cpu(ex->ee_block) - block;
2115                ext_debug("cache gap(before): %u [%u:%u]",
2116                                block,
2117                                le32_to_cpu(ex->ee_block),
2118                                 ext4_ext_get_actual_len(ex));
2119        } else if (block >= le32_to_cpu(ex->ee_block)
2120                        + ext4_ext_get_actual_len(ex)) {
2121                ext4_lblk_t next;
2122                lblock = le32_to_cpu(ex->ee_block)
2123                        + ext4_ext_get_actual_len(ex);
2124
2125                next = ext4_ext_next_allocated_block(path);
2126                ext_debug("cache gap(after): [%u:%u] %u",
2127                                le32_to_cpu(ex->ee_block),
2128                                ext4_ext_get_actual_len(ex),
2129                                block);
2130                BUG_ON(next == lblock);
2131                len = next - lblock;
2132        } else {
2133                lblock = len = 0;
2134                BUG();
2135        }
2136
2137        ext_debug(" -> %u:%lu\n", lblock, len);
2138        ext4_ext_put_in_cache(inode, lblock, len, 0);
2139}
2140
2141/*
2142 * ext4_ext_in_cache()
2143 * Checks to see if the given block is in the cache.
2144 * If it is, the cached extent is stored in the given
2145 * cache extent pointer.
2146 *
2147 * @inode: The files inode
2148 * @block: The block to look for in the cache
2149 * @ex:    Pointer where the cached extent will be stored
2150 *         if it contains block
2151 *
2152 * Return 0 if cache is invalid; 1 if the cache is valid
2153 */
2154static int
2155ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2156                  struct ext4_extent *ex)
2157{
2158        struct ext4_ext_cache *cex;
2159        struct ext4_sb_info *sbi;
2160        int ret = 0;
2161
2162        /*
2163         * We borrow i_block_reservation_lock to protect i_cached_extent
2164         */
2165        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2166        cex = &EXT4_I(inode)->i_cached_extent;
2167        sbi = EXT4_SB(inode->i_sb);
2168
2169        /* has cache valid data? */
2170        if (cex->ec_len == 0)
2171                goto errout;
2172
2173        if (in_range(block, cex->ec_block, cex->ec_len)) {
2174                ex->ee_block = cpu_to_le32(cex->ec_block);
2175                ext4_ext_store_pblock(ex, cex->ec_start);
2176                ex->ee_len = cpu_to_le16(cex->ec_len);
2177                ext_debug("%u cached by %u:%u:%llu\n",
2178                                block,
2179                                cex->ec_block, cex->ec_len, cex->ec_start);
2180                ret = 1;
2181        }
2182errout:
2183        trace_ext4_ext_in_cache(inode, block, ret);
2184        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2185        return ret;
2186}
2187
2188/*
2189 * ext4_ext_rm_idx:
2190 * removes index from the index block.
2191 */
2192static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2193                        struct ext4_ext_path *path, int depth)
2194{
2195        int err;
2196        ext4_fsblk_t leaf;
2197
2198        /* free index block */
2199        depth--;
2200        path = path + depth;
2201        leaf = ext4_idx_pblock(path->p_idx);
2202        if (unlikely(path->p_hdr->eh_entries == 0)) {
2203                EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2204                return -EIO;
2205        }
2206        err = ext4_ext_get_access(handle, inode, path);
2207        if (err)
2208                return err;
2209
2210        if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2211                int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2212                len *= sizeof(struct ext4_extent_idx);
2213                memmove(path->p_idx, path->p_idx + 1, len);
2214        }
2215
2216        le16_add_cpu(&path->p_hdr->eh_entries, -1);
2217        err = ext4_ext_dirty(handle, inode, path);
2218        if (err)
2219                return err;
2220        ext_debug("index is empty, remove it, free block %llu\n", leaf);
2221        trace_ext4_ext_rm_idx(inode, leaf);
2222
2223        ext4_free_blocks(handle, inode, NULL, leaf, 1,
2224                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2225
2226        while (--depth >= 0) {
2227                if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2228                        break;
2229                path--;
2230                err = ext4_ext_get_access(handle, inode, path);
2231                if (err)
2232                        break;
2233                path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2234                err = ext4_ext_dirty(handle, inode, path);
2235                if (err)
2236                        break;
2237        }
2238        return err;
2239}
2240
2241/*
2242 * ext4_ext_calc_credits_for_single_extent:
2243 * This routine returns max. credits that needed to insert an extent
2244 * to the extent tree.
2245 * When pass the actual path, the caller should calculate credits
2246 * under i_data_sem.
2247 */
2248int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2249                                                struct ext4_ext_path *path)
2250{
2251        if (path) {
2252                int depth = ext_depth(inode);
2253                int ret = 0;
2254
2255                /* probably there is space in leaf? */
2256                if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2257                                < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2258
2259                        /*
2260                         *  There are some space in the leaf tree, no
2261                         *  need to account for leaf block credit
2262                         *
2263                         *  bitmaps and block group descriptor blocks
2264                         *  and other metadata blocks still need to be
2265                         *  accounted.
2266                         */
2267                        /* 1 bitmap, 1 block group descriptor */
2268                        ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2269                        return ret;
2270                }
2271        }
2272
2273        return ext4_chunk_trans_blocks(inode, nrblocks);
2274}
2275
2276/*
2277 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2278 *
2279 * if nrblocks are fit in a single extent (chunk flag is 1), then
2280 * in the worse case, each tree level index/leaf need to be changed
2281 * if the tree split due to insert a new extent, then the old tree
2282 * index/leaf need to be updated too
2283 *
2284 * If the nrblocks are discontiguous, they could cause
2285 * the whole tree split more than once, but this is really rare.
2286 */
2287int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2288{
2289        int index;
2290        int depth = ext_depth(inode);
2291
2292        if (chunk)
2293                index = depth * 2;
2294        else
2295                index = depth * 3;
2296
2297        return index;
2298}
2299
2300static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2301                              struct ext4_extent *ex,
2302                              ext4_fsblk_t *partial_cluster,
2303                              ext4_lblk_t from, ext4_lblk_t to)
2304{
2305        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2306        unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2307        ext4_fsblk_t pblk;
2308        int flags = 0;
2309
2310        if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2311                flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2312        else if (ext4_should_journal_data(inode))
2313                flags |= EXT4_FREE_BLOCKS_FORGET;
2314
2315        /*
2316         * For bigalloc file systems, we never free a partial cluster
2317         * at the beginning of the extent.  Instead, we make a note
2318         * that we tried freeing the cluster, and check to see if we
2319         * need to free it on a subsequent call to ext4_remove_blocks,
2320         * or at the end of the ext4_truncate() operation.
2321         */
2322        flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2323
2324        trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2325        /*
2326         * If we have a partial cluster, and it's different from the
2327         * cluster of the last block, we need to explicitly free the
2328         * partial cluster here.
2329         */
2330        pblk = ext4_ext_pblock(ex) + ee_len - 1;
2331        if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2332                ext4_free_blocks(handle, inode, NULL,
2333                                 EXT4_C2B(sbi, *partial_cluster),
2334                                 sbi->s_cluster_ratio, flags);
2335                *partial_cluster = 0;
2336        }
2337
2338#ifdef EXTENTS_STATS
2339        {
2340                struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2341                spin_lock(&sbi->s_ext_stats_lock);
2342                sbi->s_ext_blocks += ee_len;
2343                sbi->s_ext_extents++;
2344                if (ee_len < sbi->s_ext_min)
2345                        sbi->s_ext_min = ee_len;
2346                if (ee_len > sbi->s_ext_max)
2347                        sbi->s_ext_max = ee_len;
2348                if (ext_depth(inode) > sbi->s_depth_max)
2349                        sbi->s_depth_max = ext_depth(inode);
2350                spin_unlock(&sbi->s_ext_stats_lock);
2351        }
2352#endif
2353        if (from >= le32_to_cpu(ex->ee_block)
2354            && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2355                /* tail removal */
2356                ext4_lblk_t num;
2357
2358                num = le32_to_cpu(ex->ee_block) + ee_len - from;
2359                pblk = ext4_ext_pblock(ex) + ee_len - num;
2360                ext_debug("free last %u blocks starting %llu\n", num, pblk);
2361                ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2362                /*
2363                 * If the block range to be freed didn't start at the
2364                 * beginning of a cluster, and we removed the entire
2365                 * extent, save the partial cluster here, since we
2366                 * might need to delete if we determine that the
2367                 * truncate operation has removed all of the blocks in
2368                 * the cluster.
2369                 */
2370                if (pblk & (sbi->s_cluster_ratio - 1) &&
2371                    (ee_len == num))
2372                        *partial_cluster = EXT4_B2C(sbi, pblk);
2373                else
2374                        *partial_cluster = 0;
2375        } else if (from == le32_to_cpu(ex->ee_block)
2376                   && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2377                /* head removal */
2378                ext4_lblk_t num;
2379                ext4_fsblk_t start;
2380
2381                num = to - from;
2382                start = ext4_ext_pblock(ex);
2383
2384                ext_debug("free first %u blocks starting %llu\n", num, start);
2385                ext4_free_blocks(handle, inode, NULL, start, num, flags);
2386
2387        } else {
2388                printk(KERN_INFO "strange request: removal(2) "
2389                                "%u-%u from %u:%u\n",
2390                                from, to, le32_to_cpu(ex->ee_block), ee_len);
2391        }
2392        return 0;
2393}
2394
2395
2396/*
2397 * ext4_ext_rm_leaf() Removes the extents associated with the
2398 * blocks appearing between "start" and "end", and splits the extents
2399 * if "start" and "end" appear in the same extent
2400 *
2401 * @handle: The journal handle
2402 * @inode:  The files inode
2403 * @path:   The path to the leaf
2404 * @start:  The first block to remove
2405 * @end:   The last block to remove
2406 */
2407static int
2408ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2409                 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2410                 ext4_lblk_t start, ext4_lblk_t end)
2411{
2412        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2413        int err = 0, correct_index = 0;
2414        int depth = ext_depth(inode), credits;
2415        struct ext4_extent_header *eh;
2416        ext4_lblk_t a, b;
2417        unsigned num;
2418        ext4_lblk_t ex_ee_block;
2419        unsigned short ex_ee_len;
2420        unsigned uninitialized = 0;
2421        struct ext4_extent *ex;
2422
2423        /* the header must be checked already in ext4_ext_remove_space() */
2424        ext_debug("truncate since %u in leaf to %u\n", start, end);
2425        if (!path[depth].p_hdr)
2426                path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2427        eh = path[depth].p_hdr;
2428        if (unlikely(path[depth].p_hdr == NULL)) {
2429                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2430                return -EIO;
2431        }
2432        /* find where to start removing */
2433        ex = EXT_LAST_EXTENT(eh);
2434
2435        ex_ee_block = le32_to_cpu(ex->ee_block);
2436        ex_ee_len = ext4_ext_get_actual_len(ex);
2437
2438        trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2439
2440        while (ex >= EXT_FIRST_EXTENT(eh) &&
2441                        ex_ee_block + ex_ee_len > start) {
2442
2443                if (ext4_ext_is_uninitialized(ex))
2444                        uninitialized = 1;
2445                else
2446                        uninitialized = 0;
2447
2448                ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2449                         uninitialized, ex_ee_len);
2450                path[depth].p_ext = ex;
2451
2452                a = ex_ee_block > start ? ex_ee_block : start;
2453                b = ex_ee_block+ex_ee_len - 1 < end ?
2454                        ex_ee_block+ex_ee_len - 1 : end;
2455
2456                ext_debug("  border %u:%u\n", a, b);
2457
2458                /* If this extent is beyond the end of the hole, skip it */
2459                if (end < ex_ee_block) {
2460                        ex--;
2461                        ex_ee_block = le32_to_cpu(ex->ee_block);
2462                        ex_ee_len = ext4_ext_get_actual_len(ex);
2463                        continue;
2464                } else if (b != ex_ee_block + ex_ee_len - 1) {
2465                        EXT4_ERROR_INODE(inode,
2466                                         "can not handle truncate %u:%u "
2467                                         "on extent %u:%u",
2468                                         start, end, ex_ee_block,
2469                                         ex_ee_block + ex_ee_len - 1);
2470                        err = -EIO;
2471                        goto out;
2472                } else if (a != ex_ee_block) {
2473                        /* remove tail of the extent */
2474                        num = a - ex_ee_block;
2475                } else {
2476                        /* remove whole extent: excellent! */
2477                        num = 0;
2478                }
2479                /*
2480                 * 3 for leaf, sb, and inode plus 2 (bmap and group
2481                 * descriptor) for each block group; assume two block
2482                 * groups plus ex_ee_len/blocks_per_block_group for
2483                 * the worst case
2484                 */
2485                credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2486                if (ex == EXT_FIRST_EXTENT(eh)) {
2487                        correct_index = 1;
2488                        credits += (ext_depth(inode)) + 1;
2489                }
2490                credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2491
2492                err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2493                if (err)
2494                        goto out;
2495
2496                err = ext4_ext_get_access(handle, inode, path + depth);
2497                if (err)
2498                        goto out;
2499
2500                err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2501                                         a, b);
2502                if (err)
2503                        goto out;
2504
2505                if (num == 0)
2506                        /* this extent is removed; mark slot entirely unused */
2507                        ext4_ext_store_pblock(ex, 0);
2508
2509                ex->ee_len = cpu_to_le16(num);
2510                /*
2511                 * Do not mark uninitialized if all the blocks in the
2512                 * extent have been removed.
2513                 */
2514                if (uninitialized && num)
2515                        ext4_ext_mark_uninitialized(ex);
2516                /*
2517                 * If the extent was completely released,
2518                 * we need to remove it from the leaf
2519                 */
2520                if (num == 0) {
2521                        if (end != EXT_MAX_BLOCKS - 1) {
2522                                /*
2523                                 * For hole punching, we need to scoot all the
2524                                 * extents up when an extent is removed so that
2525                                 * we dont have blank extents in the middle
2526                                 */
2527                                memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2528                                        sizeof(struct ext4_extent));
2529
2530                                /* Now get rid of the one at the end */
2531                                memset(EXT_LAST_EXTENT(eh), 0,
2532                                        sizeof(struct ext4_extent));
2533                        }
2534                        le16_add_cpu(&eh->eh_entries, -1);
2535                } else
2536                        *partial_cluster = 0;
2537
2538                err = ext4_ext_dirty(handle, inode, path + depth);
2539                if (err)
2540                        goto out;
2541
2542                ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2543                                ext4_ext_pblock(ex));
2544                ex--;
2545                ex_ee_block = le32_to_cpu(ex->ee_block);
2546                ex_ee_len = ext4_ext_get_actual_len(ex);
2547        }
2548
2549        if (correct_index && eh->eh_entries)
2550                err = ext4_ext_correct_indexes(handle, inode, path);
2551
2552        /*
2553         * If there is still a entry in the leaf node, check to see if
2554         * it references the partial cluster.  This is the only place
2555         * where it could; if it doesn't, we can free the cluster.
2556         */
2557        if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2558            (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2559             *partial_cluster)) {
2560                int flags = EXT4_FREE_BLOCKS_FORGET;
2561
2562                if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2563                        flags |= EXT4_FREE_BLOCKS_METADATA;
2564
2565                ext4_free_blocks(handle, inode, NULL,
2566                                 EXT4_C2B(sbi, *partial_cluster),
2567                                 sbi->s_cluster_ratio, flags);
2568                *partial_cluster = 0;
2569        }
2570
2571        /* if this leaf is free, then we should
2572         * remove it from index block above */
2573        if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2574                err = ext4_ext_rm_idx(handle, inode, path, depth);
2575
2576out:
2577        return err;
2578}
2579
2580/*
2581 * ext4_ext_more_to_rm:
2582 * returns 1 if current index has to be freed (even partial)
2583 */
2584static int
2585ext4_ext_more_to_rm(struct ext4_ext_path *path)
2586{
2587        BUG_ON(path->p_idx == NULL);
2588
2589        if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2590                return 0;
2591
2592        /*
2593         * if truncate on deeper level happened, it wasn't partial,
2594         * so we have to consider current index for truncation
2595         */
2596        if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2597                return 0;
2598        return 1;
2599}
2600
2601static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2602                                 ext4_lblk_t end)
2603{
2604        struct super_block *sb = inode->i_sb;
2605        int depth = ext_depth(inode);
2606        struct ext4_ext_path *path = NULL;
2607        ext4_fsblk_t partial_cluster = 0;
2608        handle_t *handle;
2609        int i = 0, err = 0;
2610
2611        ext_debug("truncate since %u to %u\n", start, end);
2612
2613        /* probably first extent we're gonna free will be last in block */
2614        handle = ext4_journal_start(inode, depth + 1);
2615        if (IS_ERR(handle))
2616                return PTR_ERR(handle);
2617
2618again:
2619        ext4_ext_invalidate_cache(inode);
2620
2621        trace_ext4_ext_remove_space(inode, start, depth);
2622
2623        /*
2624         * Check if we are removing extents inside the extent tree. If that
2625         * is the case, we are going to punch a hole inside the extent tree
2626         * so we have to check whether we need to split the extent covering
2627         * the last block to remove so we can easily remove the part of it
2628         * in ext4_ext_rm_leaf().
2629         */
2630        if (end < EXT_MAX_BLOCKS - 1) {
2631                struct ext4_extent *ex;
2632                ext4_lblk_t ee_block;
2633
2634                /* find extent for this block */
2635                path = ext4_ext_find_extent(inode, end, NULL);
2636                if (IS_ERR(path)) {
2637                        ext4_journal_stop(handle);
2638                        return PTR_ERR(path);
2639                }
2640                depth = ext_depth(inode);
2641                /* Leaf not may not exist only if inode has no blocks at all */
2642                ex = path[depth].p_ext;
2643                if (!ex) {
2644                        if (depth) {
2645                                EXT4_ERROR_INODE(inode,
2646                                                 "path[%d].p_hdr == NULL",
2647                                                 depth);
2648                                err = -EIO;
2649                        }
2650                        goto out;
2651                }
2652
2653                ee_block = le32_to_cpu(ex->ee_block);
2654
2655                /*
2656                 * See if the last block is inside the extent, if so split
2657                 * the extent at 'end' block so we can easily remove the
2658                 * tail of the first part of the split extent in
2659                 * ext4_ext_rm_leaf().
2660                 */
2661                if (end >= ee_block &&
2662                    end < ee_block + ext4_ext_get_actual_len(ex) - 1) {
2663                        int split_flag = 0;
2664
2665                        if (ext4_ext_is_uninitialized(ex))
2666                                split_flag = EXT4_EXT_MARK_UNINIT1 |
2667                                             EXT4_EXT_MARK_UNINIT2;
2668
2669                        /*
2670                         * Split the extent in two so that 'end' is the last
2671                         * block in the first new extent
2672                         */
2673                        err = ext4_split_extent_at(handle, inode, path,
2674                                                end + 1, split_flag,
2675                                                EXT4_GET_BLOCKS_PRE_IO |
2676                                                EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
2677
2678                        if (err < 0)
2679                                goto out;
2680                }
2681        }
2682        /*
2683         * We start scanning from right side, freeing all the blocks
2684         * after i_size and walking into the tree depth-wise.
2685         */
2686        depth = ext_depth(inode);
2687        if (path) {
2688                int k = i = depth;
2689                while (--k > 0)
2690                        path[k].p_block =
2691                                le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2692        } else {
2693                path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1),
2694                               GFP_NOFS);
2695                if (path == NULL) {
2696                        ext4_journal_stop(handle);
2697                        return -ENOMEM;
2698                }
2699                path[0].p_depth = depth;
2700                path[0].p_hdr = ext_inode_hdr(inode);
2701                i = 0;
2702
2703                if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2704                        err = -EIO;
2705                        goto out;
2706                }
2707        }
2708        err = 0;
2709
2710        while (i >= 0 && err == 0) {
2711                if (i == depth) {
2712                        /* this is leaf block */
2713                        err = ext4_ext_rm_leaf(handle, inode, path,
2714                                               &partial_cluster, start,
2715                                               end);
2716                        /* root level has p_bh == NULL, brelse() eats this */
2717                        brelse(path[i].p_bh);
2718                        path[i].p_bh = NULL;
2719                        i--;
2720                        continue;
2721                }
2722
2723                /* this is index block */
2724                if (!path[i].p_hdr) {
2725                        ext_debug("initialize header\n");
2726                        path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2727                }
2728
2729                if (!path[i].p_idx) {
2730                        /* this level hasn't been touched yet */
2731                        path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2732                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2733                        ext_debug("init index ptr: hdr 0x%p, num %d\n",
2734                                  path[i].p_hdr,
2735                                  le16_to_cpu(path[i].p_hdr->eh_entries));
2736                } else {
2737                        /* we were already here, see at next index */
2738                        path[i].p_idx--;
2739                }
2740
2741                ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2742                                i, EXT_FIRST_INDEX(path[i].p_hdr),
2743                                path[i].p_idx);
2744                if (ext4_ext_more_to_rm(path + i)) {
2745                        struct buffer_head *bh;
2746                        /* go to the next level */
2747                        ext_debug("move to level %d (block %llu)\n",
2748                                  i + 1, ext4_idx_pblock(path[i].p_idx));
2749                        memset(path + i + 1, 0, sizeof(*path));
2750                        bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2751                        if (!bh) {
2752                                /* should we reset i_size? */
2753                                err = -EIO;
2754                                break;
2755                        }
2756                        if (WARN_ON(i + 1 > depth)) {
2757                                err = -EIO;
2758                                break;
2759                        }
2760                        if (ext4_ext_check_block(inode, ext_block_hdr(bh),
2761                                                        depth - i - 1, bh)) {
2762                                err = -EIO;
2763                                break;
2764                        }
2765                        path[i + 1].p_bh = bh;
2766
2767                        /* save actual number of indexes since this
2768                         * number is changed at the next iteration */
2769                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2770                        i++;
2771                } else {
2772                        /* we finished processing this index, go up */
2773                        if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2774                                /* index is empty, remove it;
2775                                 * handle must be already prepared by the
2776                                 * truncatei_leaf() */
2777                                err = ext4_ext_rm_idx(handle, inode, path, i);
2778                        }
2779                        /* root level has p_bh == NULL, brelse() eats this */
2780                        brelse(path[i].p_bh);
2781                        path[i].p_bh = NULL;
2782                        i--;
2783                        ext_debug("return to level %d\n", i);
2784                }
2785        }
2786
2787        trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2788                        path->p_hdr->eh_entries);
2789
2790        /* If we still have something in the partial cluster and we have removed
2791         * even the first extent, then we should free the blocks in the partial
2792         * cluster as well. */
2793        if (partial_cluster && path->p_hdr->eh_entries == 0) {
2794                int flags = EXT4_FREE_BLOCKS_FORGET;
2795
2796                if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2797                        flags |= EXT4_FREE_BLOCKS_METADATA;
2798
2799                ext4_free_blocks(handle, inode, NULL,
2800                                 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2801                                 EXT4_SB(sb)->s_cluster_ratio, flags);
2802                partial_cluster = 0;
2803        }
2804
2805        /* TODO: flexible tree reduction should be here */
2806        if (path->p_hdr->eh_entries == 0) {
2807                /*
2808                 * truncate to zero freed all the tree,
2809                 * so we need to correct eh_depth
2810                 */
2811                err = ext4_ext_get_access(handle, inode, path);
2812                if (err == 0) {
2813                        ext_inode_hdr(inode)->eh_depth = 0;
2814                        ext_inode_hdr(inode)->eh_max =
2815                                cpu_to_le16(ext4_ext_space_root(inode, 0));
2816                        err = ext4_ext_dirty(handle, inode, path);
2817                }
2818        }
2819out:
2820        ext4_ext_drop_refs(path);
2821        kfree(path);
2822        if (err == -EAGAIN) {
2823                path = NULL;
2824                goto again;
2825        }
2826        ext4_journal_stop(handle);
2827
2828        return err;
2829}
2830
2831/*
2832 * called at mount time
2833 */
2834void ext4_ext_init(struct super_block *sb)
2835{
2836        /*
2837         * possible initialization would be here
2838         */
2839
2840        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2841#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2842                printk(KERN_INFO "EXT4-fs: file extents enabled"
2843#ifdef AGGRESSIVE_TEST
2844                       ", aggressive tests"
2845#endif
2846#ifdef CHECK_BINSEARCH
2847                       ", check binsearch"
2848#endif
2849#ifdef EXTENTS_STATS
2850                       ", stats"
2851#endif
2852                       "\n");
2853#endif
2854#ifdef EXTENTS_STATS
2855                spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2856                EXT4_SB(sb)->s_ext_min = 1 << 30;
2857                EXT4_SB(sb)->s_ext_max = 0;
2858#endif
2859        }
2860}
2861
2862/*
2863 * called at umount time
2864 */
2865void ext4_ext_release(struct super_block *sb)
2866{
2867        if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2868                return;
2869
2870#ifdef EXTENTS_STATS
2871        if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2872                struct ext4_sb_info *sbi = EXT4_SB(sb);
2873                printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2874                        sbi->s_ext_blocks, sbi->s_ext_extents,
2875                        sbi->s_ext_blocks / sbi->s_ext_extents);
2876                printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2877                        sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2878        }
2879#endif
2880}
2881
2882/* FIXME!! we need to try to merge to left or right after zero-out  */
2883static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2884{
2885        ext4_fsblk_t ee_pblock;
2886        unsigned int ee_len;
2887        int ret;
2888
2889        ee_len    = ext4_ext_get_actual_len(ex);
2890        ee_pblock = ext4_ext_pblock(ex);
2891
2892        ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2893        if (ret > 0)
2894                ret = 0;
2895
2896        return ret;
2897}
2898
2899/*
2900 * ext4_split_extent_at() splits an extent at given block.
2901 *
2902 * @handle: the journal handle
2903 * @inode: the file inode
2904 * @path: the path to the extent
2905 * @split: the logical block where the extent is splitted.
2906 * @split_flags: indicates if the extent could be zeroout if split fails, and
2907 *               the states(init or uninit) of new extents.
2908 * @flags: flags used to insert new extent to extent tree.
2909 *
2910 *
2911 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2912 * of which are deterimined by split_flag.
2913 *
2914 * There are two cases:
2915 *  a> the extent are splitted into two extent.
2916 *  b> split is not needed, and just mark the extent.
2917 *
2918 * return 0 on success.
2919 */
2920static int ext4_split_extent_at(handle_t *handle,
2921                             struct inode *inode,
2922                             struct ext4_ext_path *path,
2923                             ext4_lblk_t split,
2924                             int split_flag,
2925                             int flags)
2926{
2927        ext4_fsblk_t newblock;
2928        ext4_lblk_t ee_block;
2929        struct ext4_extent *ex, newex, orig_ex;
2930        struct ext4_extent *ex2 = NULL;
2931        unsigned int ee_len, depth;
2932        int err = 0;
2933
2934        BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
2935               (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
2936
2937        ext_debug("ext4_split_extents_at: inode %lu, logical"
2938                "block %llu\n", inode->i_ino, (unsigned long long)split);
2939
2940        ext4_ext_show_leaf(inode, path);
2941
2942        depth = ext_depth(inode);
2943        ex = path[depth].p_ext;
2944        ee_block = le32_to_cpu(ex->ee_block);
2945        ee_len = ext4_ext_get_actual_len(ex);
2946        newblock = split - ee_block + ext4_ext_pblock(ex);
2947
2948        BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2949
2950        err = ext4_ext_get_access(handle, inode, path + depth);
2951        if (err)
2952                goto out;
2953
2954        if (split == ee_block) {
2955                /*
2956                 * case b: block @split is the block that the extent begins with
2957                 * then we just change the state of the extent, and splitting
2958                 * is not needed.
2959                 */
2960                if (split_flag & EXT4_EXT_MARK_UNINIT2)
2961                        ext4_ext_mark_uninitialized(ex);
2962                else
2963                        ext4_ext_mark_initialized(ex);
2964
2965                if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2966                        ext4_ext_try_to_merge(handle, inode, path, ex);
2967
2968                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2969                goto out;
2970        }
2971
2972        /* case a */
2973        memcpy(&orig_ex, ex, sizeof(orig_ex));
2974        ex->ee_len = cpu_to_le16(split - ee_block);
2975        if (split_flag & EXT4_EXT_MARK_UNINIT1)
2976                ext4_ext_mark_uninitialized(ex);
2977
2978        /*
2979         * path may lead to new leaf, not to original leaf any more
2980         * after ext4_ext_insert_extent() returns,
2981         */
2982        err = ext4_ext_dirty(handle, inode, path + depth);
2983        if (err)
2984                goto fix_extent_len;
2985
2986        ex2 = &newex;
2987        ex2->ee_block = cpu_to_le32(split);
2988        ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2989        ext4_ext_store_pblock(ex2, newblock);
2990        if (split_flag & EXT4_EXT_MARK_UNINIT2)
2991                ext4_ext_mark_uninitialized(ex2);
2992
2993        err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2994        if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2995                if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
2996                        if (split_flag & EXT4_EXT_DATA_VALID1)
2997                                err = ext4_ext_zeroout(inode, ex2);
2998                        else
2999                                err = ext4_ext_zeroout(inode, ex);
3000                } else
3001                        err = ext4_ext_zeroout(inode, &orig_ex);
3002
3003                if (err)
3004                        goto fix_extent_len;
3005                /* update the extent length and mark as initialized */
3006                ex->ee_len = cpu_to_le16(ee_len);
3007                ext4_ext_try_to_merge(handle, inode, path, ex);
3008                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3009                goto out;
3010        } else if (err)
3011                goto fix_extent_len;
3012
3013out:
3014        ext4_ext_show_leaf(inode, path);
3015        return err;
3016
3017fix_extent_len:
3018        ex->ee_len = orig_ex.ee_len;
3019        ext4_ext_dirty(handle, inode, path + depth);
3020        return err;
3021}
3022
3023/*
3024 * ext4_split_extents() splits an extent and mark extent which is covered
3025 * by @map as split_flags indicates
3026 *
3027 * It may result in splitting the extent into multiple extents (upto three)
3028 * There are three possibilities:
3029 *   a> There is no split required
3030 *   b> Splits in two extents: Split is happening at either end of the extent
3031 *   c> Splits in three extents: Somone is splitting in middle of the extent
3032 *
3033 */
3034static int ext4_split_extent(handle_t *handle,
3035                              struct inode *inode,
3036                              struct ext4_ext_path *path,
3037                              struct ext4_map_blocks *map,
3038                              int split_flag,
3039                              int flags)
3040{
3041        ext4_lblk_t ee_block;
3042        struct ext4_extent *ex;
3043        unsigned int ee_len, depth;
3044        int err = 0;
3045        int uninitialized;
3046        int split_flag1, flags1;
3047
3048        depth = ext_depth(inode);
3049        ex = path[depth].p_ext;
3050        ee_block = le32_to_cpu(ex->ee_block);
3051        ee_len = ext4_ext_get_actual_len(ex);
3052        uninitialized = ext4_ext_is_uninitialized(ex);
3053
3054        if (map->m_lblk + map->m_len < ee_block + ee_len) {
3055                split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3056                flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3057                if (uninitialized)
3058                        split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
3059                                       EXT4_EXT_MARK_UNINIT2;
3060                if (split_flag & EXT4_EXT_DATA_VALID2)
3061                        split_flag1 |= EXT4_EXT_DATA_VALID1;
3062                err = ext4_split_extent_at(handle, inode, path,
3063                                map->m_lblk + map->m_len, split_flag1, flags1);
3064                if (err)
3065                        goto out;
3066        }
3067
3068        ext4_ext_drop_refs(path);
3069        path = ext4_ext_find_extent(inode, map->m_lblk, path);
3070        if (IS_ERR(path))
3071                return PTR_ERR(path);
3072
3073        if (map->m_lblk >= ee_block) {
3074                split_flag1 = split_flag & (EXT4_EXT_MAY_ZEROOUT |
3075                                            EXT4_EXT_DATA_VALID2);
3076                if (uninitialized)
3077                        split_flag1 |= EXT4_EXT_MARK_UNINIT1;
3078                if (split_flag & EXT4_EXT_MARK_UNINIT2)
3079                        split_flag1 |= EXT4_EXT_MARK_UNINIT2;
3080                err = ext4_split_extent_at(handle, inode, path,
3081                                map->m_lblk, split_flag1, flags);
3082                if (err)
3083                        goto out;
3084        }
3085
3086        ext4_ext_show_leaf(inode, path);
3087out:
3088        return err ? err : map->m_len;
3089}
3090
3091/*
3092 * This function is called by ext4_ext_map_blocks() if someone tries to write
3093 * to an uninitialized extent. It may result in splitting the uninitialized
3094 * extent into multiple extents (up to three - one initialized and two
3095 * uninitialized).
3096 * There are three possibilities:
3097 *   a> There is no split required: Entire extent should be initialized
3098 *   b> Splits in two extents: Write is happening at either end of the extent
3099 *   c> Splits in three extents: Somone is writing in middle of the extent
3100 *
3101 * Pre-conditions:
3102 *  - The extent pointed to by 'path' is uninitialized.
3103 *  - The extent pointed to by 'path' contains a superset
3104 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3105 *
3106 * Post-conditions on success:
3107 *  - the returned value is the number of blocks beyond map->l_lblk
3108 *    that are allocated and initialized.
3109 *    It is guaranteed to be >= map->m_len.
3110 */
3111static int ext4_ext_convert_to_initialized(handle_t *handle,
3112                                           struct inode *inode,
3113                                           struct ext4_map_blocks *map,
3114                                           struct ext4_ext_path *path)
3115{
3116        struct ext4_sb_info *sbi;
3117        struct ext4_extent_header *eh;
3118        struct ext4_map_blocks split_map;
3119        struct ext4_extent zero_ex;
3120        struct ext4_extent *ex;
3121        ext4_lblk_t ee_block, eof_block;
3122        unsigned int ee_len, depth;
3123        int allocated, max_zeroout = 0;
3124        int err = 0;
3125        int split_flag = 0;
3126
3127        ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3128                "block %llu, max_blocks %u\n", inode->i_ino,
3129                (unsigned long long)map->m_lblk, map->m_len);
3130
3131        sbi = EXT4_SB(inode->i_sb);
3132        eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3133                inode->i_sb->s_blocksize_bits;
3134        if (eof_block < map->m_lblk + map->m_len)
3135                eof_block = map->m_lblk + map->m_len;
3136
3137        depth = ext_depth(inode);
3138        eh = path[depth].p_hdr;
3139        ex = path[depth].p_ext;
3140        ee_block = le32_to_cpu(ex->ee_block);
3141        ee_len = ext4_ext_get_actual_len(ex);
3142        allocated = ee_len - (map->m_lblk - ee_block);
3143
3144        trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3145
3146        /* Pre-conditions */
3147        BUG_ON(!ext4_ext_is_uninitialized(ex));
3148        BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3149
3150        /*
3151         * Attempt to transfer newly initialized blocks from the currently
3152         * uninitialized extent to its left neighbor. This is much cheaper
3153         * than an insertion followed by a merge as those involve costly
3154         * memmove() calls. This is the common case in steady state for
3155         * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3156         * writes.
3157         *
3158         * Limitations of the current logic:
3159         *  - L1: we only deal with writes at the start of the extent.
3160         *    The approach could be extended to writes at the end
3161         *    of the extent but this scenario was deemed less common.
3162         *  - L2: we do not deal with writes covering the whole extent.
3163         *    This would require removing the extent if the transfer
3164         *    is possible.
3165         *  - L3: we only attempt to merge with an extent stored in the
3166         *    same extent tree node.
3167         */
3168        if ((map->m_lblk == ee_block) &&        /*L1*/
3169                (map->m_len < ee_len) &&        /*L2*/
3170                (ex > EXT_FIRST_EXTENT(eh))) {  /*L3*/
3171                struct ext4_extent *prev_ex;
3172                ext4_lblk_t prev_lblk;
3173                ext4_fsblk_t prev_pblk, ee_pblk;
3174                unsigned int prev_len, write_len;
3175
3176                prev_ex = ex - 1;
3177                prev_lblk = le32_to_cpu(prev_ex->ee_block);
3178                prev_len = ext4_ext_get_actual_len(prev_ex);
3179                prev_pblk = ext4_ext_pblock(prev_ex);
3180                ee_pblk = ext4_ext_pblock(ex);
3181                write_len = map->m_len;
3182
3183                /*
3184                 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3185                 * upon those conditions:
3186                 * - C1: prev_ex is initialized,
3187                 * - C2: prev_ex is logically abutting ex,
3188                 * - C3: prev_ex is physically abutting ex,
3189                 * - C4: prev_ex can receive the additional blocks without
3190                 *   overflowing the (initialized) length limit.
3191                 */
3192                if ((!ext4_ext_is_uninitialized(prev_ex)) &&            /*C1*/
3193                        ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3194                        ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3195                        (prev_len < (EXT_INIT_MAX_LEN - write_len))) {  /*C4*/
3196                        err = ext4_ext_get_access(handle, inode, path + depth);
3197                        if (err)
3198                                goto out;
3199
3200                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3201                                map, ex, prev_ex);
3202
3203                        /* Shift the start of ex by 'write_len' blocks */
3204                        ex->ee_block = cpu_to_le32(ee_block + write_len);
3205                        ext4_ext_store_pblock(ex, ee_pblk + write_len);
3206                        ex->ee_len = cpu_to_le16(ee_len - write_len);
3207                        ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3208
3209                        /* Extend prev_ex by 'write_len' blocks */
3210                        prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3211
3212                        /* Mark the block containing both extents as dirty */
3213                        ext4_ext_dirty(handle, inode, path + depth);
3214
3215                        /* Update path to point to the right extent */
3216                        path[depth].p_ext = prev_ex;
3217
3218                        /* Result: number of initialized blocks past m_lblk */
3219                        allocated = write_len;
3220                        goto out;
3221                }
3222        }
3223
3224        WARN_ON(map->m_lblk < ee_block);
3225        /*
3226         * It is safe to convert extent to initialized via explicit
3227         * zeroout only if extent is fully insde i_size or new_size.
3228         */
3229        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3230
3231        if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3232                max_zeroout = sbi->s_extent_max_zeroout_kb >>
3233                        inode->i_sb->s_blocksize_bits;
3234
3235        /* If extent is less than s_max_zeroout_kb, zeroout directly */
3236        if (max_zeroout && (ee_len <= max_zeroout)) {
3237                err = ext4_ext_zeroout(inode, ex);
3238                if (err)
3239                        goto out;
3240
3241                err = ext4_ext_get_access(handle, inode, path + depth);
3242                if (err)
3243                        goto out;
3244                ext4_ext_mark_initialized(ex);
3245                ext4_ext_try_to_merge(handle, inode, path, ex);
3246                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3247                goto out;
3248        }
3249
3250        /*
3251         * four cases:
3252         * 1. split the extent into three extents.
3253         * 2. split the extent into two extents, zeroout the first half.
3254         * 3. split the extent into two extents, zeroout the second half.
3255         * 4. split the extent into two extents with out zeroout.
3256         */
3257        split_map.m_lblk = map->m_lblk;
3258        split_map.m_len = map->m_len;
3259
3260        if (max_zeroout && (allocated > map->m_len)) {
3261                if (allocated <= max_zeroout) {
3262                        /* case 3 */
3263                        zero_ex.ee_block =
3264                                         cpu_to_le32(map->m_lblk);
3265                        zero_ex.ee_len = cpu_to_le16(allocated);
3266                        ext4_ext_store_pblock(&zero_ex,
3267                                ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3268                        err = ext4_ext_zeroout(inode, &zero_ex);
3269                        if (err)
3270                                goto out;
3271                        split_map.m_lblk = map->m_lblk;
3272                        split_map.m_len = allocated;
3273                } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) {
3274                        /* case 2 */
3275                        if (map->m_lblk != ee_block) {
3276                                zero_ex.ee_block = ex->ee_block;
3277                                zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3278                                                        ee_block);
3279                                ext4_ext_store_pblock(&zero_ex,
3280                                                      ext4_ext_pblock(ex));
3281                                err = ext4_ext_zeroout(inode, &zero_ex);
3282                                if (err)
3283                                        goto out;
3284                        }
3285
3286                        split_map.m_lblk = ee_block;
3287                        split_map.m_len = map->m_lblk - ee_block + map->m_len;
3288                        allocated = map->m_len;
3289                }
3290        }
3291
3292        allocated = ext4_split_extent(handle, inode, path,
3293                                      &split_map, split_flag, 0);
3294        if (allocated < 0)
3295                err = allocated;
3296
3297out:
3298        return err ? err : allocated;
3299}
3300
3301/*
3302 * This function is called by ext4_ext_map_blocks() from
3303 * ext4_get_blocks_dio_write() when DIO to write
3304 * to an uninitialized extent.
3305 *
3306 * Writing to an uninitialized extent may result in splitting the uninitialized
3307 * extent into multiple initialized/uninitialized extents (up to three)
3308 * There are three possibilities:
3309 *   a> There is no split required: Entire extent should be uninitialized
3310 *   b> Splits in two extents: Write is happening at either end of the extent
3311 *   c> Splits in three extents: Somone is writing in middle of the extent
3312 *
3313 * One of more index blocks maybe needed if the extent tree grow after
3314 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3315 * complete, we need to split the uninitialized extent before DIO submit
3316 * the IO. The uninitialized extent called at this time will be split
3317 * into three uninitialized extent(at most). After IO complete, the part
3318 * being filled will be convert to initialized by the end_io callback function
3319 * via ext4_convert_unwritten_extents().
3320 *
3321 * Returns the size of uninitialized extent to be written on success.
3322 */
3323static int ext4_split_unwritten_extents(handle_t *handle,
3324                                        struct inode *inode,
3325                                        struct ext4_map_blocks *map,
3326                                        struct ext4_ext_path *path,
3327                                        int flags)
3328{
3329        ext4_lblk_t eof_block;
3330        ext4_lblk_t ee_block;
3331        struct ext4_extent *ex;
3332        unsigned int ee_len;
3333        int split_flag = 0, depth;
3334
3335        ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3336                "block %llu, max_blocks %u\n", inode->i_ino,
3337                (unsigned long long)map->m_lblk, map->m_len);
3338
3339        eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3340                inode->i_sb->s_blocksize_bits;
3341        if (eof_block < map->m_lblk + map->m_len)
3342                eof_block = map->m_lblk + map->m_len;
3343        /*
3344         * It is safe to convert extent to initialized via explicit
3345         * zeroout only if extent is fully insde i_size or new_size.
3346         */
3347        depth = ext_depth(inode);
3348        ex = path[depth].p_ext;
3349        ee_block = le32_to_cpu(ex->ee_block);
3350        ee_len = ext4_ext_get_actual_len(ex);
3351
3352        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3353        split_flag |= EXT4_EXT_MARK_UNINIT2;
3354        if (flags & EXT4_GET_BLOCKS_CONVERT)
3355                split_flag |= EXT4_EXT_DATA_VALID2;
3356        flags |= EXT4_GET_BLOCKS_PRE_IO;
3357        return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3358}
3359
3360static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3361                                                struct inode *inode,
3362                                                struct ext4_map_blocks *map,
3363                                                struct ext4_ext_path *path)
3364{
3365        struct ext4_extent *ex;
3366        ext4_lblk_t ee_block;
3367        unsigned int ee_len;
3368        int depth;
3369        int err = 0;
3370
3371        depth = ext_depth(inode);
3372        ex = path[depth].p_ext;
3373        ee_block = le32_to_cpu(ex->ee_block);
3374        ee_len = ext4_ext_get_actual_len(ex);
3375
3376        ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3377                "block %llu, max_blocks %u\n", inode->i_ino,
3378                  (unsigned long long)ee_block, ee_len);
3379
3380        /* If extent is larger than requested then split is required */
3381        if (ee_block != map->m_lblk || ee_len > map->m_len) {
3382                err = ext4_split_unwritten_extents(handle, inode, map, path,
3383                                                   EXT4_GET_BLOCKS_CONVERT);
3384                if (err < 0)
3385                        goto out;
3386                ext4_ext_drop_refs(path);
3387                path = ext4_ext_find_extent(inode, map->m_lblk, path);
3388                if (IS_ERR(path)) {
3389                        err = PTR_ERR(path);
3390                        goto out;
3391                }
3392                depth = ext_depth(inode);
3393                ex = path[depth].p_ext;
3394        }
3395
3396        err = ext4_ext_get_access(handle, inode, path + depth);
3397        if (err)
3398                goto out;
3399        /* first mark the extent as initialized */
3400        ext4_ext_mark_initialized(ex);
3401
3402        /* note: ext4_ext_correct_indexes() isn't needed here because
3403         * borders are not changed
3404         */
3405        ext4_ext_try_to_merge(handle, inode, path, ex);
3406
3407        /* Mark modified extent as dirty */
3408        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3409out:
3410        ext4_ext_show_leaf(inode, path);
3411        return err;
3412}
3413
3414static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3415                        sector_t block, int count)
3416{
3417        int i;
3418        for (i = 0; i < count; i++)
3419                unmap_underlying_metadata(bdev, block + i);
3420}
3421
3422/*
3423 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3424 */
3425static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3426                              ext4_lblk_t lblk,
3427                              struct ext4_ext_path *path,
3428                              unsigned int len)
3429{
3430        int i, depth;
3431        struct ext4_extent_header *eh;
3432        struct ext4_extent *last_ex;
3433
3434        if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3435                return 0;
3436
3437        depth = ext_depth(inode);
3438        eh = path[depth].p_hdr;
3439
3440        /*
3441         * We're going to remove EOFBLOCKS_FL entirely in future so we
3442         * do not care for this case anymore. Simply remove the flag
3443         * if there are no extents.
3444         */
3445        if (unlikely(!eh->eh_entries))
3446                goto out;
3447        last_ex = EXT_LAST_EXTENT(eh);
3448        /*
3449         * We should clear the EOFBLOCKS_FL flag if we are writing the
3450         * last block in the last extent in the file.  We test this by
3451         * first checking to see if the caller to
3452         * ext4_ext_get_blocks() was interested in the last block (or
3453         * a block beyond the last block) in the current extent.  If
3454         * this turns out to be false, we can bail out from this
3455         * function immediately.
3456         */
3457        if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3458            ext4_ext_get_actual_len(last_ex))
3459                return 0;
3460        /*
3461         * If the caller does appear to be planning to write at or
3462         * beyond the end of the current extent, we then test to see
3463         * if the current extent is the last extent in the file, by
3464         * checking to make sure it was reached via the rightmost node
3465         * at each level of the tree.
3466         */
3467        for (i = depth-1; i >= 0; i--)
3468                if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3469                        return 0;
3470out:
3471        ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3472        return ext4_mark_inode_dirty(handle, inode);
3473}
3474
3475/**
3476 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3477 *
3478 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3479 * whether there are any buffers marked for delayed allocation. It returns '1'
3480 * on the first delalloc'ed buffer head found. If no buffer head in the given
3481 * range is marked for delalloc, it returns 0.
3482 * lblk_start should always be <= lblk_end.
3483 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3484 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3485 * block sooner). This is useful when blocks are truncated sequentially from
3486 * lblk_start towards lblk_end.
3487 */
3488static int ext4_find_delalloc_range(struct inode *inode,
3489                                    ext4_lblk_t lblk_start,
3490                                    ext4_lblk_t lblk_end,
3491                                    int search_hint_reverse)
3492{
3493        struct address_space *mapping = inode->i_mapping;
3494        struct buffer_head *head, *bh = NULL;
3495        struct page *page;
3496        ext4_lblk_t i, pg_lblk;
3497        pgoff_t index;
3498
3499        if (!test_opt(inode->i_sb, DELALLOC))
3500                return 0;
3501
3502        /* reverse search wont work if fs block size is less than page size */
3503        if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3504                search_hint_reverse = 0;
3505
3506        if (search_hint_reverse)
3507                i = lblk_end;
3508        else
3509                i = lblk_start;
3510
3511        index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3512
3513        while ((i >= lblk_start) && (i <= lblk_end)) {
3514                page = find_get_page(mapping, index);
3515                if (!page)
3516                        goto nextpage;
3517
3518                if (!page_has_buffers(page))
3519                        goto nextpage;
3520
3521                head = page_buffers(page);
3522                if (!head)
3523                        goto nextpage;
3524
3525                bh = head;
3526                pg_lblk = index << (PAGE_CACHE_SHIFT -
3527                                                inode->i_blkbits);
3528                do {
3529                        if (unlikely(pg_lblk < lblk_start)) {
3530                                /*
3531                                 * This is possible when fs block size is less
3532                                 * than page size and our cluster starts/ends in
3533                                 * middle of the page. So we need to skip the
3534                                 * initial few blocks till we reach the 'lblk'
3535                                 */
3536                                pg_lblk++;
3537                                continue;
3538                        }
3539
3540                        /* Check if the buffer is delayed allocated and that it
3541                         * is not yet mapped. (when da-buffers are mapped during
3542                         * their writeout, their da_mapped bit is set.)
3543                         */
3544                        if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3545                                page_cache_release(page);
3546                                trace_ext4_find_delalloc_range(inode,
3547                                                lblk_start, lblk_end,
3548                                                search_hint_reverse,
3549                                                1, i);
3550                                return 1;
3551                        }
3552                        if (search_hint_reverse)
3553                                i--;
3554                        else
3555                                i++;
3556                } while ((i >= lblk_start) && (i <= lblk_end) &&
3557                                ((bh = bh->b_this_page) != head));
3558nextpage:
3559                if (page)
3560                        page_cache_release(page);
3561                /*
3562                 * Move to next page. 'i' will be the first lblk in the next
3563                 * page.
3564                 */
3565                if (search_hint_reverse)
3566                        index--;
3567                else
3568                        index++;
3569                i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3570        }
3571
3572        trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3573                                        search_hint_reverse, 0, 0);
3574        return 0;
3575}
3576
3577int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3578                               int search_hint_reverse)
3579{
3580        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3581        ext4_lblk_t lblk_start, lblk_end;
3582        lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3583        lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3584
3585        return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3586                                        search_hint_reverse);
3587}
3588
3589/**
3590 * Determines how many complete clusters (out of those specified by the 'map')
3591 * are under delalloc and were reserved quota for.
3592 * This function is called when we are writing out the blocks that were
3593 * originally written with their allocation delayed, but then the space was
3594 * allocated using fallocate() before the delayed allocation could be resolved.
3595 * The cases to look for are:
3596 * ('=' indicated delayed allocated blocks
3597 *  '-' indicates non-delayed allocated blocks)
3598 * (a) partial clusters towards beginning and/or end outside of allocated range
3599 *     are not delalloc'ed.
3600 *      Ex:
3601 *      |----c---=|====c====|====c====|===-c----|
3602 *               |++++++ allocated ++++++|
3603 *      ==> 4 complete clusters in above example
3604 *
3605 * (b) partial cluster (outside of allocated range) towards either end is
3606 *     marked for delayed allocation. In this case, we will exclude that
3607 *     cluster.
3608 *      Ex:
3609 *      |----====c========|========c========|
3610 *           |++++++ allocated ++++++|
3611 *      ==> 1 complete clusters in above example
3612 *
3613 *      Ex:
3614 *      |================c================|
3615 *            |++++++ allocated ++++++|
3616 *      ==> 0 complete clusters in above example
3617 *
3618 * The ext4_da_update_reserve_space will be called only if we
3619 * determine here that there were some "entire" clusters that span
3620 * this 'allocated' range.
3621 * In the non-bigalloc case, this function will just end up returning num_blks
3622 * without ever calling ext4_find_delalloc_range.
3623 */
3624static unsigned int
3625get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3626                           unsigned int num_blks)
3627{
3628        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3629        ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3630        ext4_lblk_t lblk_from, lblk_to, c_offset;
3631        unsigned int allocated_clusters = 0;
3632
3633        alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3634        alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3635
3636        /* max possible clusters for this allocation */
3637        allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3638
3639        trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3640
3641        /* Check towards left side */
3642        c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3643        if (c_offset) {
3644                lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3645                lblk_to = lblk_from + c_offset - 1;
3646
3647                if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3648                        allocated_clusters--;
3649        }
3650
3651        /* Now check towards right. */
3652        c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3653        if (allocated_clusters && c_offset) {
3654                lblk_from = lblk_start + num_blks;
3655                lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3656
3657                if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3658                        allocated_clusters--;
3659        }
3660
3661        return allocated_clusters;
3662}
3663
3664static int
3665ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3666                        struct ext4_map_blocks *map,
3667                        struct ext4_ext_path *path, int flags,
3668                        unsigned int allocated, ext4_fsblk_t newblock)
3669{
3670        int ret = 0;
3671        int err = 0;
3672        ext4_io_end_t *io = ext4_inode_aio(inode);
3673
3674        ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3675                  "block %llu, max_blocks %u, flags %x, allocated %u\n",
3676                  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3677                  flags, allocated);
3678        ext4_ext_show_leaf(inode, path);
3679
3680        trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3681                                                    newblock);
3682
3683        /* get_block() before submit the IO, split the extent */
3684        if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3685                ret = ext4_split_unwritten_extents(handle, inode, map,
3686                                                   path, flags);
3687                if (ret <= 0)
3688                        goto out;
3689                /*
3690                 * Flag the inode(non aio case) or end_io struct (aio case)
3691                 * that this IO needs to conversion to written when IO is
3692                 * completed
3693                 */
3694                if (io)
3695                        ext4_set_io_unwritten_flag(inode, io);
3696                else
3697                        ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3698                if (ext4_should_dioread_nolock(inode))
3699                        map->m_flags |= EXT4_MAP_UNINIT;
3700                goto out;
3701        }
3702        /* IO end_io complete, convert the filled extent to written */
3703        if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3704                ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
3705                                                        path);
3706                if (ret >= 0) {
3707                        ext4_update_inode_fsync_trans(handle, inode, 1);
3708                        err = check_eofblocks_fl(handle, inode, map->m_lblk,
3709                                                 path, map->m_len);
3710                } else
3711                        err = ret;
3712                goto out2;
3713        }
3714        /* buffered IO case */
3715        /*
3716         * repeat fallocate creation request
3717         * we already have an unwritten extent
3718         */
3719        if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3720                goto map_out;
3721
3722        /* buffered READ or buffered write_begin() lookup */
3723        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3724                /*
3725                 * We have blocks reserved already.  We
3726                 * return allocated blocks so that delalloc
3727                 * won't do block reservation for us.  But
3728                 * the buffer head will be unmapped so that
3729                 * a read from the block returns 0s.
3730                 */
3731                map->m_flags |= EXT4_MAP_UNWRITTEN;
3732                goto out1;
3733        }
3734
3735        /* buffered write, writepage time, convert*/
3736        ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3737        if (ret >= 0)
3738                ext4_update_inode_fsync_trans(handle, inode, 1);
3739out:
3740        if (ret <= 0) {
3741                err = ret;
3742                goto out2;
3743        } else
3744                allocated = ret;
3745        map->m_flags |= EXT4_MAP_NEW;
3746        /*
3747         * if we allocated more blocks than requested
3748         * we need to make sure we unmap the extra block
3749         * allocated. The actual needed block will get
3750         * unmapped later when we find the buffer_head marked
3751         * new.
3752         */
3753        if (allocated > map->m_len) {
3754                unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3755                                        newblock + map->m_len,
3756                                        allocated - map->m_len);
3757                allocated = map->m_len;
3758        }
3759
3760        /*
3761         * If we have done fallocate with the offset that is already
3762         * delayed allocated, we would have block reservation
3763         * and quota reservation done in the delayed write path.
3764         * But fallocate would have already updated quota and block
3765         * count for this offset. So cancel these reservation
3766         */
3767        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3768                unsigned int reserved_clusters;
3769                reserved_clusters = get_reserved_cluster_alloc(inode,
3770                                map->m_lblk, map->m_len);
3771                if (reserved_clusters)
3772                        ext4_da_update_reserve_space(inode,
3773                                                     reserved_clusters,
3774                                                     0);
3775        }
3776
3777map_out:
3778        map->m_flags |= EXT4_MAP_MAPPED;
3779        if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3780                err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3781                                         map->m_len);
3782                if (err < 0)
3783                        goto out2;
3784        }
3785out1:
3786        if (allocated > map->m_len)
3787                allocated = map->m_len;
3788        ext4_ext_show_leaf(inode, path);
3789        map->m_pblk = newblock;
3790        map->m_len = allocated;
3791out2:
3792        if (path) {
3793                ext4_ext_drop_refs(path);
3794                kfree(path);
3795        }
3796        return err ? err : allocated;
3797}
3798
3799/*
3800 * get_implied_cluster_alloc - check to see if the requested
3801 * allocation (in the map structure) overlaps with a cluster already
3802 * allocated in an extent.
3803 *      @sb     The filesystem superblock structure
3804 *      @map    The requested lblk->pblk mapping
3805 *      @ex     The extent structure which might contain an implied
3806 *                      cluster allocation
3807 *
3808 * This function is called by ext4_ext_map_blocks() after we failed to
3809 * find blocks that were already in the inode's extent tree.  Hence,
3810 * we know that the beginning of the requested region cannot overlap
3811 * the extent from the inode's extent tree.  There are three cases we
3812 * want to catch.  The first is this case:
3813 *
3814 *               |--- cluster # N--|
3815 *    |--- extent ---|  |---- requested region ---|
3816 *                      |==========|
3817 *
3818 * The second case that we need to test for is this one:
3819 *
3820 *   |--------- cluster # N ----------------|
3821 *         |--- requested region --|   |------- extent ----|
3822 *         |=======================|
3823 *
3824 * The third case is when the requested region lies between two extents
3825 * within the same cluster:
3826 *          |------------- cluster # N-------------|
3827 * |----- ex -----|                  |---- ex_right ----|
3828 *                  |------ requested region ------|
3829 *                  |================|
3830 *
3831 * In each of the above cases, we need to set the map->m_pblk and
3832 * map->m_len so it corresponds to the return the extent labelled as
3833 * "|====|" from cluster #N, since it is already in use for data in
3834 * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3835 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3836 * as a new "allocated" block region.  Otherwise, we will return 0 and
3837 * ext4_ext_map_blocks() will then allocate one or more new clusters
3838 * by calling ext4_mb_new_blocks().
3839 */
3840static int get_implied_cluster_alloc(struct super_block *sb,
3841                                     struct ext4_map_blocks *map,
3842                                     struct ext4_extent *ex,
3843                                     struct ext4_ext_path *path)
3844{
3845        struct ext4_sb_info *sbi = EXT4_SB(sb);
3846        ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3847        ext4_lblk_t ex_cluster_start, ex_cluster_end;
3848        ext4_lblk_t rr_cluster_start;
3849        ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3850        ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3851        unsigned short ee_len = ext4_ext_get_actual_len(ex);
3852
3853        /* The extent passed in that we are trying to match */
3854        ex_cluster_start = EXT4_B2C(sbi, ee_block);
3855        ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3856
3857        /* The requested region passed into ext4_map_blocks() */
3858        rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3859
3860        if ((rr_cluster_start == ex_cluster_end) ||
3861            (rr_cluster_start == ex_cluster_start)) {
3862                if (rr_cluster_start == ex_cluster_end)
3863                        ee_start += ee_len - 1;
3864                map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3865                        c_offset;
3866                map->m_len = min(map->m_len,
3867                                 (unsigned) sbi->s_cluster_ratio - c_offset);
3868                /*
3869                 * Check for and handle this case:
3870                 *
3871                 *   |--------- cluster # N-------------|
3872                 *                     |------- extent ----|
3873                 *         |--- requested region ---|
3874                 *         |===========|
3875                 */
3876
3877                if (map->m_lblk < ee_block)
3878                        map->m_len = min(map->m_len, ee_block - map->m_lblk);
3879
3880                /*
3881                 * Check for the case where there is already another allocated
3882                 * block to the right of 'ex' but before the end of the cluster.
3883                 *
3884                 *          |------------- cluster # N-------------|
3885                 * |----- ex -----|                  |---- ex_right ----|
3886                 *                  |------ requested region ------|
3887                 *                  |================|
3888                 */
3889                if (map->m_lblk > ee_block) {
3890                        ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3891                        map->m_len = min(map->m_len, next - map->m_lblk);
3892                }
3893
3894                trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3895                return 1;
3896        }
3897
3898        trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3899        return 0;
3900}
3901
3902
3903/*
3904 * Block allocation/map/preallocation routine for extents based files
3905 *
3906 *
3907 * Need to be called with
3908 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3909 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3910 *
3911 * return > 0, number of of blocks already mapped/allocated
3912 *          if create == 0 and these are pre-allocated blocks
3913 *              buffer head is unmapped
3914 *          otherwise blocks are mapped
3915 *
3916 * return = 0, if plain look up failed (blocks have not been allocated)
3917 *          buffer head is unmapped
3918 *
3919 * return < 0, error case.
3920 */
3921int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3922                        struct ext4_map_blocks *map, int flags)
3923{
3924        struct ext4_ext_path *path = NULL;
3925        struct ext4_extent newex, *ex, *ex2;
3926        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3927        ext4_fsblk_t newblock = 0;
3928        int free_on_err = 0, err = 0, depth, ret;
3929        unsigned int allocated = 0, offset = 0;
3930        unsigned int allocated_clusters = 0;
3931        struct ext4_allocation_request ar;
3932        ext4_io_end_t *io = ext4_inode_aio(inode);
3933        ext4_lblk_t cluster_offset;
3934        int set_unwritten = 0;
3935
3936        ext_debug("blocks %u/%u requested for inode %lu\n",
3937                  map->m_lblk, map->m_len, inode->i_ino);
3938        trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3939
3940        /* check in cache */
3941        if (ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3942                if (!newex.ee_start_lo && !newex.ee_start_hi) {
3943                        if ((sbi->s_cluster_ratio > 1) &&
3944                            ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3945                                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3946
3947                        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3948                                /*
3949                                 * block isn't allocated yet and
3950                                 * user doesn't want to allocate it
3951                                 */
3952                                goto out2;
3953                        }
3954                        /* we should allocate requested block */
3955                } else {
3956                        /* block is already allocated */
3957                        if (sbi->s_cluster_ratio > 1)
3958                                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3959                        newblock = map->m_lblk
3960                                   - le32_to_cpu(newex.ee_block)
3961                                   + ext4_ext_pblock(&newex);
3962                        /* number of remaining blocks in the extent */
3963                        allocated = ext4_ext_get_actual_len(&newex) -
3964                                (map->m_lblk - le32_to_cpu(newex.ee_block));
3965                        goto out;
3966                }
3967        }
3968
3969        /* find extent for this block */
3970        path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3971        if (IS_ERR(path)) {
3972                err = PTR_ERR(path);
3973                path = NULL;
3974                goto out2;
3975        }
3976
3977        depth = ext_depth(inode);
3978
3979        /*
3980         * consistent leaf must not be empty;
3981         * this situation is possible, though, _during_ tree modification;
3982         * this is why assert can't be put in ext4_ext_find_extent()
3983         */
3984        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3985                EXT4_ERROR_INODE(inode, "bad extent address "
3986                                 "lblock: %lu, depth: %d pblock %lld",
3987                                 (unsigned long) map->m_lblk, depth,
3988                                 path[depth].p_block);
3989                err = -EIO;
3990                goto out2;
3991        }
3992
3993        ex = path[depth].p_ext;
3994        if (ex) {
3995                ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3996                ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3997                unsigned short ee_len;
3998
3999                /*
4000                 * Uninitialized extents are treated as holes, except that
4001                 * we split out initialized portions during a write.
4002                 */
4003                ee_len = ext4_ext_get_actual_len(ex);
4004
4005                trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4006
4007                /* if found extent covers block, simply return it */
4008                if (in_range(map->m_lblk, ee_block, ee_len)) {
4009                        newblock = map->m_lblk - ee_block + ee_start;
4010                        /* number of remaining blocks in the extent */
4011                        allocated = ee_len - (map->m_lblk - ee_block);
4012                        ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4013                                  ee_block, ee_len, newblock);
4014
4015                        /*
4016                         * Do not put uninitialized extent
4017                         * in the cache
4018                         */
4019                        if (!ext4_ext_is_uninitialized(ex)) {
4020                                ext4_ext_put_in_cache(inode, ee_block,
4021                                        ee_len, ee_start);
4022                                goto out;
4023                        }
4024                        ret = ext4_ext_handle_uninitialized_extents(
4025                                handle, inode, map, path, flags,
4026                                allocated, newblock);
4027                        return ret;
4028                }
4029        }
4030
4031        if ((sbi->s_cluster_ratio > 1) &&
4032            ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
4033                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4034
4035        /*
4036         * requested block isn't allocated yet;
4037         * we couldn't try to create block if create flag is zero
4038         */
4039        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4040                /*
4041                 * put just found gap into cache to speed up
4042                 * subsequent requests
4043                 */
4044                ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
4045                goto out2;
4046        }
4047
4048        /*
4049         * Okay, we need to do block allocation.
4050         */
4051        map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
4052        newex.ee_block = cpu_to_le32(map->m_lblk);
4053        cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
4054
4055        /*
4056         * If we are doing bigalloc, check to see if the extent returned
4057         * by ext4_ext_find_extent() implies a cluster we can use.
4058         */
4059        if (cluster_offset && ex &&
4060            get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4061                ar.len = allocated = map->m_len;
4062                newblock = map->m_pblk;
4063                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4064                goto got_allocated_blocks;
4065        }
4066
4067        /* find neighbour allocated blocks */
4068        ar.lleft = map->m_lblk;
4069        err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4070        if (err)
4071                goto out2;
4072        ar.lright = map->m_lblk;
4073        ex2 = NULL;
4074        err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4075        if (err)
4076                goto out2;
4077
4078        /* Check if the extent after searching to the right implies a
4079         * cluster we can use. */
4080        if ((sbi->s_cluster_ratio > 1) && ex2 &&
4081            get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4082                ar.len = allocated = map->m_len;
4083                newblock = map->m_pblk;
4084                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
4085                goto got_allocated_blocks;
4086        }
4087
4088        /*
4089         * See if request is beyond maximum number of blocks we can have in
4090         * a single extent. For an initialized extent this limit is
4091         * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4092         * EXT_UNINIT_MAX_LEN.
4093         */
4094        if (map->m_len > EXT_INIT_MAX_LEN &&
4095            !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4096                map->m_len = EXT_INIT_MAX_LEN;
4097        else if (map->m_len > EXT_UNINIT_MAX_LEN &&
4098                 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
4099                map->m_len = EXT_UNINIT_MAX_LEN;
4100
4101        /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4102        newex.ee_len = cpu_to_le16(map->m_len);
4103        err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4104        if (err)
4105                allocated = ext4_ext_get_actual_len(&newex);
4106        else
4107                allocated = map->m_len;
4108
4109        /* allocate new block */
4110        ar.inode = inode;
4111        ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4112        ar.logical = map->m_lblk;
4113        /*
4114         * We calculate the offset from the beginning of the cluster
4115         * for the logical block number, since when we allocate a
4116         * physical cluster, the physical block should start at the
4117         * same offset from the beginning of the cluster.  This is
4118         * needed so that future calls to get_implied_cluster_alloc()
4119         * work correctly.
4120         */
4121        offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
4122        ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4123        ar.goal -= offset;
4124        ar.logical -= offset;
4125        if (S_ISREG(inode->i_mode))
4126                ar.flags = EXT4_MB_HINT_DATA;
4127        else
4128                /* disable in-core preallocation for non-regular files */
4129                ar.flags = 0;
4130        if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4131                ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4132        newblock = ext4_mb_new_blocks(handle, &ar, &err);
4133        if (!newblock)
4134                goto out2;
4135        ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4136                  ar.goal, newblock, allocated);
4137        free_on_err = 1;
4138        allocated_clusters = ar.len;
4139        ar.len = EXT4_C2B(sbi, ar.len) - offset;
4140        if (ar.len > allocated)
4141                ar.len = allocated;
4142
4143got_allocated_blocks:
4144        /* try to insert new extent into found leaf and return */
4145        ext4_ext_store_pblock(&newex, newblock + offset);
4146        newex.ee_len = cpu_to_le16(ar.len);
4147        /* Mark uninitialized */
4148        if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4149                ext4_ext_mark_uninitialized(&newex);
4150                /*
4151                 * io_end structure was created for every IO write to an
4152                 * uninitialized extent. To avoid unnecessary conversion,
4153                 * here we flag the IO that really needs the conversion.
4154                 * For non asycn direct IO case, flag the inode state
4155                 * that we need to perform conversion when IO is done.
4156                 */
4157                if ((flags & EXT4_GET_BLOCKS_PRE_IO))
4158                        set_unwritten = 1;
4159                if (ext4_should_dioread_nolock(inode))
4160                        map->m_flags |= EXT4_MAP_UNINIT;
4161        }
4162
4163        err = 0;
4164        if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4165                err = check_eofblocks_fl(handle, inode, map->m_lblk,
4166                                         path, ar.len);
4167        if (!err)
4168                err = ext4_ext_insert_extent(handle, inode, path,
4169                                             &newex, flags);
4170
4171        if (!err && set_unwritten) {
4172                if (io)
4173                        ext4_set_io_unwritten_flag(inode, io);
4174                else
4175                        ext4_set_inode_state(inode,
4176                                             EXT4_STATE_DIO_UNWRITTEN);
4177        }
4178
4179        if (err && free_on_err) {
4180                int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4181                        EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4182                /* free data blocks we just allocated */
4183                /* not a good idea to call discard here directly,
4184                 * but otherwise we'd need to call it every free() */
4185                ext4_discard_preallocations(inode);
4186                ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4187                                 ext4_ext_get_actual_len(&newex), fb_flags);
4188                goto out2;
4189        }
4190
4191        /* previous routine could use block we allocated */
4192        newblock = ext4_ext_pblock(&newex);
4193        allocated = ext4_ext_get_actual_len(&newex);
4194        if (allocated > map->m_len)
4195                allocated = map->m_len;
4196        map->m_flags |= EXT4_MAP_NEW;
4197
4198        /*
4199         * Update reserved blocks/metadata blocks after successful
4200         * block allocation which had been deferred till now.
4201         */
4202        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4203                unsigned int reserved_clusters;
4204                /*
4205                 * Check how many clusters we had reserved this allocated range
4206                 */
4207                reserved_clusters = get_reserved_cluster_alloc(inode,
4208                                                map->m_lblk, allocated);
4209                if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4210                        if (reserved_clusters) {
4211                                /*
4212                                 * We have clusters reserved for this range.
4213                                 * But since we are not doing actual allocation
4214                                 * and are simply using blocks from previously
4215                                 * allocated cluster, we should release the
4216                                 * reservation and not claim quota.
4217                                 */
4218                                ext4_da_update_reserve_space(inode,
4219                                                reserved_clusters, 0);
4220                        }
4221                } else {
4222                        BUG_ON(allocated_clusters < reserved_clusters);
4223                        /* We will claim quota for all newly allocated blocks.*/
4224                        ext4_da_update_reserve_space(inode, allocated_clusters,
4225                                                        1);
4226                        if (reserved_clusters < allocated_clusters) {
4227                                struct ext4_inode_info *ei = EXT4_I(inode);
4228                                int reservation = allocated_clusters -
4229                                                  reserved_clusters;
4230                                /*
4231                                 * It seems we claimed few clusters outside of
4232                                 * the range of this allocation. We should give
4233                                 * it back to the reservation pool. This can
4234                                 * happen in the following case:
4235                                 *
4236                                 * * Suppose s_cluster_ratio is 4 (i.e., each
4237                                 *   cluster has 4 blocks. Thus, the clusters
4238                                 *   are [0-3],[4-7],[8-11]...
4239                                 * * First comes delayed allocation write for
4240                                 *   logical blocks 10 & 11. Since there were no
4241                                 *   previous delayed allocated blocks in the
4242                                 *   range [8-11], we would reserve 1 cluster
4243                                 *   for this write.
4244                                 * * Next comes write for logical blocks 3 to 8.
4245                                 *   In this case, we will reserve 2 clusters
4246                                 *   (for [0-3] and [4-7]; and not for [8-11] as
4247                                 *   that range has a delayed allocated blocks.
4248                                 *   Thus total reserved clusters now becomes 3.
4249                                 * * Now, during the delayed allocation writeout
4250                                 *   time, we will first write blocks [3-8] and
4251                                 *   allocate 3 clusters for writing these
4252                                 *   blocks. Also, we would claim all these
4253                                 *   three clusters above.
4254                                 * * Now when we come here to writeout the
4255                                 *   blocks [10-11], we would expect to claim
4256                                 *   the reservation of 1 cluster we had made
4257                                 *   (and we would claim it since there are no
4258                                 *   more delayed allocated blocks in the range
4259                                 *   [8-11]. But our reserved cluster count had
4260                                 *   already gone to 0.
4261                                 *
4262                                 *   Thus, at the step 4 above when we determine
4263                                 *   that there are still some unwritten delayed
4264                                 *   allocated blocks outside of our current
4265                                 *   block range, we should increment the
4266                                 *   reserved clusters count so that when the
4267                                 *   remaining blocks finally gets written, we
4268                                 *   could claim them.
4269                                 */
4270                                dquot_reserve_block(inode,
4271                                                EXT4_C2B(sbi, reservation));
4272                                spin_lock(&ei->i_block_reservation_lock);
4273                                ei->i_reserved_data_blocks += reservation;
4274                                spin_unlock(&ei->i_block_reservation_lock);
4275                        }
4276                }
4277        }
4278
4279        /*
4280         * Cache the extent and update transaction to commit on fdatasync only
4281         * when it is _not_ an uninitialized extent.
4282         */
4283        if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4284                ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4285                ext4_update_inode_fsync_trans(handle, inode, 1);
4286        } else
4287                ext4_update_inode_fsync_trans(handle, inode, 0);
4288out:
4289        if (allocated > map->m_len)
4290                allocated = map->m_len;
4291        ext4_ext_show_leaf(inode, path);
4292        map->m_flags |= EXT4_MAP_MAPPED;
4293        map->m_pblk = newblock;
4294        map->m_len = allocated;
4295out2:
4296        if (path) {
4297                ext4_ext_drop_refs(path);
4298                kfree(path);
4299        }
4300
4301        trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4302                newblock, map->m_len, err ? err : allocated);
4303
4304        return err ? err : allocated;
4305}
4306
4307void ext4_ext_truncate(struct inode *inode)
4308{
4309        struct address_space *mapping = inode->i_mapping;
4310        struct super_block *sb = inode->i_sb;
4311        ext4_lblk_t last_block;
4312        handle_t *handle;
4313        loff_t page_len;
4314        int err = 0;
4315
4316        /*
4317         * finish any pending end_io work so we won't run the risk of
4318         * converting any truncated blocks to initialized later
4319         */
4320        ext4_flush_unwritten_io(inode);
4321
4322        /*
4323         * probably first extent we're gonna free will be last in block
4324         */
4325        err = ext4_writepage_trans_blocks(inode);
4326        handle = ext4_journal_start(inode, err);
4327        if (IS_ERR(handle))
4328                return;
4329
4330        if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4331                page_len = PAGE_CACHE_SIZE -
4332                        (inode->i_size & (PAGE_CACHE_SIZE - 1));
4333
4334                err = ext4_discard_partial_page_buffers(handle,
4335                        mapping, inode->i_size, page_len, 0);
4336
4337                if (err)
4338                        goto out_stop;
4339        }
4340
4341        if (ext4_orphan_add(handle, inode))
4342                goto out_stop;
4343
4344        down_write(&EXT4_I(inode)->i_data_sem);
4345        ext4_ext_invalidate_cache(inode);
4346
4347        ext4_discard_preallocations(inode);
4348
4349        /*
4350         * TODO: optimization is possible here.
4351         * Probably we need not scan at all,
4352         * because page truncation is enough.
4353         */
4354
4355        /* we have to know where to truncate from in crash case */
4356        EXT4_I(inode)->i_disksize = inode->i_size;
4357        ext4_mark_inode_dirty(handle, inode);
4358
4359        last_block = (inode->i_size + sb->s_blocksize - 1)
4360                        >> EXT4_BLOCK_SIZE_BITS(sb);
4361        err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4362
4363        /* In a multi-transaction truncate, we only make the final
4364         * transaction synchronous.
4365         */
4366        if (IS_SYNC(inode))
4367                ext4_handle_sync(handle);
4368
4369        up_write(&EXT4_I(inode)->i_data_sem);
4370
4371out_stop:
4372        /*
4373         * If this was a simple ftruncate() and the file will remain alive,
4374         * then we need to clear up the orphan record which we created above.
4375         * However, if this was a real unlink then we were called by
4376         * ext4_delete_inode(), and we allow that function to clean up the
4377         * orphan info for us.
4378         */
4379        if (inode->i_nlink)
4380                ext4_orphan_del(handle, inode);
4381
4382        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4383        ext4_mark_inode_dirty(handle, inode);
4384        ext4_journal_stop(handle);
4385}
4386
4387static void ext4_falloc_update_inode(struct inode *inode,
4388                                int mode, loff_t new_size, int update_ctime)
4389{
4390        struct timespec now;
4391
4392        if (update_ctime) {
4393                now = current_fs_time(inode->i_sb);
4394                if (!timespec_equal(&inode->i_ctime, &now))
4395                        inode->i_ctime = now;
4396        }
4397        /*
4398         * Update only when preallocation was requested beyond
4399         * the file size.
4400         */
4401        if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4402                if (new_size > i_size_read(inode))
4403                        i_size_write(inode, new_size);
4404                if (new_size > EXT4_I(inode)->i_disksize)
4405                        ext4_update_i_disksize(inode, new_size);
4406        } else {
4407                /*
4408                 * Mark that we allocate beyond EOF so the subsequent truncate
4409                 * can proceed even if the new size is the same as i_size.
4410                 */
4411                if (new_size > i_size_read(inode))
4412                        ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4413        }
4414
4415}
4416
4417/*
4418 * preallocate space for a file. This implements ext4's fallocate file
4419 * operation, which gets called from sys_fallocate system call.
4420 * For block-mapped files, posix_fallocate should fall back to the method
4421 * of writing zeroes to the required new blocks (the same behavior which is
4422 * expected for file systems which do not support fallocate() system call).
4423 */
4424long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4425{
4426        struct inode *inode = file->f_path.dentry->d_inode;
4427        handle_t *handle;
4428        loff_t new_size;
4429        unsigned int max_blocks;
4430        int ret = 0;
4431        int ret2 = 0;
4432        int retries = 0;
4433        int flags;
4434        struct ext4_map_blocks map;
4435        unsigned int credits, blkbits = inode->i_blkbits;
4436
4437        /*
4438         * currently supporting (pre)allocate mode for extent-based
4439         * files _only_
4440         */
4441        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4442                return -EOPNOTSUPP;
4443
4444        /* Return error if mode is not supported */
4445        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4446                return -EOPNOTSUPP;
4447
4448        if (mode & FALLOC_FL_PUNCH_HOLE)
4449                return ext4_punch_hole(file, offset, len);
4450
4451        trace_ext4_fallocate_enter(inode, offset, len, mode);
4452        map.m_lblk = offset >> blkbits;
4453        /*
4454         * We can't just convert len to max_blocks because
4455         * If blocksize = 4096 offset = 3072 and len = 2048
4456         */
4457        max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4458                - map.m_lblk;
4459        /*
4460         * credits to insert 1 extent into extent tree
4461         */
4462        credits = ext4_chunk_trans_blocks(inode, max_blocks);
4463        mutex_lock(&inode->i_mutex);
4464        ret = inode_newsize_ok(inode, (len + offset));
4465        if (ret) {
4466                mutex_unlock(&inode->i_mutex);
4467                trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4468                return ret;
4469        }
4470        flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4471        if (mode & FALLOC_FL_KEEP_SIZE)
4472                flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4473        /*
4474         * Don't normalize the request if it can fit in one extent so
4475         * that it doesn't get unnecessarily split into multiple
4476         * extents.
4477         */
4478        if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4479                flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4480
4481        /* Prevent race condition between unwritten */
4482        ext4_flush_unwritten_io(inode);
4483retry:
4484        while (ret >= 0 && ret < max_blocks) {
4485                map.m_lblk = map.m_lblk + ret;
4486                map.m_len = max_blocks = max_blocks - ret;
4487                handle = ext4_journal_start(inode, credits);
4488                if (IS_ERR(handle)) {
4489                        ret = PTR_ERR(handle);
4490                        break;
4491                }
4492                ret = ext4_map_blocks(handle, inode, &map, flags);
4493                if (ret <= 0) {
4494#ifdef EXT4FS_DEBUG
4495                        WARN_ON(ret <= 0);
4496                        printk(KERN_ERR "%s: ext4_ext_map_blocks "
4497                                    "returned error inode#%lu, block=%u, "
4498                                    "max_blocks=%u", __func__,
4499                                    inode->i_ino, map.m_lblk, max_blocks);
4500#endif
4501                        ext4_mark_inode_dirty(handle, inode);
4502                        ret2 = ext4_journal_stop(handle);
4503                        break;
4504                }
4505                if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4506                                                blkbits) >> blkbits))
4507                        new_size = offset + len;
4508                else
4509                        new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4510
4511                ext4_falloc_update_inode(inode, mode, new_size,
4512                                         (map.m_flags & EXT4_MAP_NEW));
4513                ext4_mark_inode_dirty(handle, inode);
4514                if ((file->f_flags & O_SYNC) && ret >= max_blocks)
4515                        ext4_handle_sync(handle);
4516                ret2 = ext4_journal_stop(handle);
4517                if (ret2)
4518                        break;
4519        }
4520        if (ret == -ENOSPC &&
4521                        ext4_should_retry_alloc(inode->i_sb, &retries)) {
4522                ret = 0;
4523                goto retry;
4524        }
4525        mutex_unlock(&inode->i_mutex);
4526        trace_ext4_fallocate_exit(inode, offset, max_blocks,
4527                                ret > 0 ? ret2 : ret);
4528        return ret > 0 ? ret2 : ret;
4529}
4530
4531/*
4532 * This function convert a range of blocks to written extents
4533 * The caller of this function will pass the start offset and the size.
4534 * all unwritten extents within this range will be converted to
4535 * written extents.
4536 *
4537 * This function is called from the direct IO end io call back
4538 * function, to convert the fallocated extents after IO is completed.
4539 * Returns 0 on success.
4540 */
4541int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4542                                    ssize_t len)
4543{
4544        handle_t *handle;
4545        unsigned int max_blocks;
4546        int ret = 0;
4547        int ret2 = 0;
4548        struct ext4_map_blocks map;
4549        unsigned int credits, blkbits = inode->i_blkbits;
4550
4551        map.m_lblk = offset >> blkbits;
4552        /*
4553         * We can't just convert len to max_blocks because
4554         * If blocksize = 4096 offset = 3072 and len = 2048
4555         */
4556        max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4557                      map.m_lblk);
4558        /*
4559         * credits to insert 1 extent into extent tree
4560         */
4561        credits = ext4_chunk_trans_blocks(inode, max_blocks);
4562        while (ret >= 0 && ret < max_blocks) {
4563                map.m_lblk += ret;
4564                map.m_len = (max_blocks -= ret);
4565                handle = ext4_journal_start(inode, credits);
4566                if (IS_ERR(handle)) {
4567                        ret = PTR_ERR(handle);
4568                        break;
4569                }
4570                ret = ext4_map_blocks(handle, inode, &map,
4571                                      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4572                if (ret <= 0) {
4573                        WARN_ON(ret <= 0);
4574                        ext4_msg(inode->i_sb, KERN_ERR,
4575                                 "%s:%d: inode #%lu: block %u: len %u: "
4576                                 "ext4_ext_map_blocks returned %d",
4577                                 __func__, __LINE__, inode->i_ino, map.m_lblk,
4578                                 map.m_len, ret);
4579                }
4580                ext4_mark_inode_dirty(handle, inode);
4581                ret2 = ext4_journal_stop(handle);
4582                if (ret <= 0 || ret2 )
4583                        break;
4584        }
4585        return ret > 0 ? ret2 : ret;
4586}
4587
4588/*
4589 * Callback function called for each extent to gather FIEMAP information.
4590 */
4591static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4592                       struct ext4_ext_cache *newex, struct ext4_extent *ex,
4593                       void *data)
4594{
4595        __u64   logical;
4596        __u64   physical;
4597        __u64   length;
4598        __u32   flags = 0;
4599        int             ret = 0;
4600        struct fiemap_extent_info *fieinfo = data;
4601        unsigned char blksize_bits;
4602
4603        blksize_bits = inode->i_sb->s_blocksize_bits;
4604        logical = (__u64)newex->ec_block << blksize_bits;
4605
4606        if (newex->ec_start == 0) {
4607                /*
4608                 * No extent in extent-tree contains block @newex->ec_start,
4609                 * then the block may stay in 1)a hole or 2)delayed-extent.
4610                 *
4611                 * Holes or delayed-extents are processed as follows.
4612                 * 1. lookup dirty pages with specified range in pagecache.
4613                 *    If no page is got, then there is no delayed-extent and
4614                 *    return with EXT_CONTINUE.
4615                 * 2. find the 1st mapped buffer,
4616                 * 3. check if the mapped buffer is both in the request range
4617                 *    and a delayed buffer. If not, there is no delayed-extent,
4618                 *    then return.
4619                 * 4. a delayed-extent is found, the extent will be collected.
4620                 */
4621                ext4_lblk_t     end = 0;
4622                pgoff_t         last_offset;
4623                pgoff_t         offset;
4624                pgoff_t         index;
4625                pgoff_t         start_index = 0;
4626                struct page     **pages = NULL;
4627                struct buffer_head *bh = NULL;
4628                struct buffer_head *head = NULL;
4629                unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4630
4631                pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4632                if (pages == NULL)
4633                        return -ENOMEM;
4634
4635                offset = logical >> PAGE_SHIFT;
4636repeat:
4637                last_offset = offset;
4638                head = NULL;
4639                ret = find_get_pages_tag(inode->i_mapping, &offset,
4640                                        PAGECACHE_TAG_DIRTY, nr_pages, pages);
4641
4642                if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4643                        /* First time, try to find a mapped buffer. */
4644                        if (ret == 0) {
4645out:
4646                                for (index = 0; index < ret; index++)
4647                                        page_cache_release(pages[index]);
4648                                /* just a hole. */
4649                                kfree(pages);
4650                                return EXT_CONTINUE;
4651                        }
4652                        index = 0;
4653
4654next_page:
4655                        /* Try to find the 1st mapped buffer. */
4656                        end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4657                                  blksize_bits;
4658                        if (!page_has_buffers(pages[index]))
4659                                goto out;
4660                        head = page_buffers(pages[index]);
4661                        if (!head)
4662                                goto out;
4663
4664                        index++;
4665                        bh = head;
4666                        do {
4667                                if (end >= newex->ec_block +
4668                                        newex->ec_len)
4669                                        /* The buffer is out of
4670                                         * the request range.
4671                                         */
4672                                        goto out;
4673
4674                                if (buffer_mapped(bh) &&
4675                                    end >= newex->ec_block) {
4676                                        start_index = index - 1;
4677                                        /* get the 1st mapped buffer. */
4678                                        goto found_mapped_buffer;
4679                                }
4680
4681                                bh = bh->b_this_page;
4682                                end++;
4683                        } while (bh != head);
4684
4685                        /* No mapped buffer in the range found in this page,
4686                         * We need to look up next page.
4687                         */
4688                        if (index >= ret) {
4689                                /* There is no page left, but we need to limit
4690                                 * newex->ec_len.
4691                                 */
4692                                newex->ec_len = end - newex->ec_block;
4693                                goto out;
4694                        }
4695                        goto next_page;
4696                } else {
4697                        /*Find contiguous delayed buffers. */
4698                        if (ret > 0 && pages[0]->index == last_offset)
4699                                head = page_buffers(pages[0]);
4700                        bh = head;
4701                        index = 1;
4702                        start_index = 0;
4703                }
4704
4705found_mapped_buffer:
4706                if (bh != NULL && buffer_delay(bh)) {
4707                        /* 1st or contiguous delayed buffer found. */
4708                        if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4709                                /*
4710                                 * 1st delayed buffer found, record
4711                                 * the start of extent.
4712                                 */
4713                                flags |= FIEMAP_EXTENT_DELALLOC;
4714                                newex->ec_block = end;
4715                                logical = (__u64)end << blksize_bits;
4716                        }
4717                        /* Find contiguous delayed buffers. */
4718                        do {
4719                                if (!buffer_delay(bh))
4720                                        goto found_delayed_extent;
4721                                bh = bh->b_this_page;
4722                                end++;
4723                        } while (bh != head);
4724
4725                        for (; index < ret; index++) {
4726                                if (!page_has_buffers(pages[index])) {
4727                                        bh = NULL;
4728                                        break;
4729                                }
4730                                head = page_buffers(pages[index]);
4731                                if (!head) {
4732                                        bh = NULL;
4733                                        break;
4734                                }
4735
4736                                if (pages[index]->index !=
4737                                    pages[start_index]->index + index
4738                                    - start_index) {
4739                                        /* Blocks are not contiguous. */
4740                                        bh = NULL;
4741                                        break;
4742                                }
4743                                bh = head;
4744                                do {
4745                                        if (!buffer_delay(bh))
4746                                                /* Delayed-extent ends. */
4747                                                goto found_delayed_extent;
4748                                        bh = bh->b_this_page;
4749                                        end++;
4750                                } while (bh != head);
4751                        }
4752                } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4753                        /* a hole found. */
4754                        goto out;
4755
4756found_delayed_extent:
4757                newex->ec_len = min(end - newex->ec_block,
4758                                                (ext4_lblk_t)EXT_INIT_MAX_LEN);
4759                if (ret == nr_pages && bh != NULL &&
4760                        newex->ec_len < EXT_INIT_MAX_LEN &&
4761                        buffer_delay(bh)) {
4762                        /* Have not collected an extent and continue. */
4763                        for (index = 0; index < ret; index++)
4764                                page_cache_release(pages[index]);
4765                        goto repeat;
4766                }
4767
4768                for (index = 0; index < ret; index++)
4769                        page_cache_release(pages[index]);
4770                kfree(pages);
4771        }
4772
4773        physical = (__u64)newex->ec_start << blksize_bits;
4774        length =   (__u64)newex->ec_len << blksize_bits;
4775
4776        if (ex && ext4_ext_is_uninitialized(ex))
4777                flags |= FIEMAP_EXTENT_UNWRITTEN;
4778
4779        if (next == EXT_MAX_BLOCKS)
4780                flags |= FIEMAP_EXTENT_LAST;
4781
4782        ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4783                                        length, flags);
4784        if (ret < 0)
4785                return ret;
4786        if (ret == 1)
4787                return EXT_BREAK;
4788        return EXT_CONTINUE;
4789}
4790/* fiemap flags we can handle specified here */
4791#define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4792
4793static int ext4_xattr_fiemap(struct inode *inode,
4794                                struct fiemap_extent_info *fieinfo)
4795{
4796        __u64 physical = 0;
4797        __u64 length;
4798        __u32 flags = FIEMAP_EXTENT_LAST;
4799        int blockbits = inode->i_sb->s_blocksize_bits;
4800        int error = 0;
4801
4802        /* in-inode? */
4803        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4804                struct ext4_iloc iloc;
4805                int offset;     /* offset of xattr in inode */
4806
4807                error = ext4_get_inode_loc(inode, &iloc);
4808                if (error)
4809                        return error;
4810                physical = iloc.bh->b_blocknr << blockbits;
4811                offset = EXT4_GOOD_OLD_INODE_SIZE +
4812                                EXT4_I(inode)->i_extra_isize;
4813                physical += offset;
4814                length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4815                flags |= FIEMAP_EXTENT_DATA_INLINE;
4816                brelse(iloc.bh);
4817        } else { /* external block */
4818                physical = EXT4_I(inode)->i_file_acl << blockbits;
4819                length = inode->i_sb->s_blocksize;
4820        }
4821
4822        if (physical)
4823                error = fiemap_fill_next_extent(fieinfo, 0, physical,
4824                                                length, flags);
4825        return (error < 0 ? error : 0);
4826}
4827
4828/*
4829 * ext4_ext_punch_hole
4830 *
4831 * Punches a hole of "length" bytes in a file starting
4832 * at byte "offset"
4833 *
4834 * @inode:  The inode of the file to punch a hole in
4835 * @offset: The starting byte offset of the hole
4836 * @length: The length of the hole
4837 *
4838 * Returns the number of blocks removed or negative on err
4839 */
4840int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4841{
4842        struct inode *inode = file->f_path.dentry->d_inode;
4843        struct super_block *sb = inode->i_sb;
4844        ext4_lblk_t first_block, stop_block;
4845        struct address_space *mapping = inode->i_mapping;
4846        handle_t *handle;
4847        loff_t first_page, last_page, page_len;
4848        loff_t first_page_offset, last_page_offset;
4849        int credits, err = 0;
4850
4851        /*
4852         * Write out all dirty pages to avoid race conditions
4853         * Then release them.
4854         */
4855        if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4856                err = filemap_write_and_wait_range(mapping,
4857                        offset, offset + length - 1);
4858
4859                if (err)
4860                        return err;
4861        }
4862
4863        mutex_lock(&inode->i_mutex);
4864        /* It's not possible punch hole on append only file */
4865        if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4866                err = -EPERM;
4867                goto out_mutex;
4868        }
4869        if (IS_SWAPFILE(inode)) {
4870                err = -ETXTBSY;
4871                goto out_mutex;
4872        }
4873
4874        /* No need to punch hole beyond i_size */
4875        if (offset >= inode->i_size)
4876                goto out_mutex;
4877
4878        /*
4879         * If the hole extends beyond i_size, set the hole
4880         * to end after the page that contains i_size
4881         */
4882        if (offset + length > inode->i_size) {
4883                length = inode->i_size +
4884                   PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4885                   offset;
4886        }
4887
4888        first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4889        last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4890
4891        first_page_offset = first_page << PAGE_CACHE_SHIFT;
4892        last_page_offset = last_page << PAGE_CACHE_SHIFT;
4893
4894        /* Now release the pages */
4895        if (last_page_offset > first_page_offset) {
4896                truncate_pagecache_range(inode, first_page_offset,
4897                                         last_page_offset - 1);
4898        }
4899
4900        /* Wait all existing dio workers, newcomers will block on i_mutex */
4901        ext4_inode_block_unlocked_dio(inode);
4902        err = ext4_flush_unwritten_io(inode);
4903        if (err)
4904                goto out_dio;
4905        inode_dio_wait(inode);
4906
4907        credits = ext4_writepage_trans_blocks(inode);
4908        handle = ext4_journal_start(inode, credits);
4909        if (IS_ERR(handle)) {
4910                err = PTR_ERR(handle);
4911                goto out_dio;
4912        }
4913
4914
4915        /*
4916         * Now we need to zero out the non-page-aligned data in the
4917         * pages at the start and tail of the hole, and unmap the buffer
4918         * heads for the block aligned regions of the page that were
4919         * completely zeroed.
4920         */
4921        if (first_page > last_page) {
4922                /*
4923                 * If the file space being truncated is contained within a page
4924                 * just zero out and unmap the middle of that page
4925                 */
4926                err = ext4_discard_partial_page_buffers(handle,
4927                        mapping, offset, length, 0);
4928
4929                if (err)
4930                        goto out;
4931        } else {
4932                /*
4933                 * zero out and unmap the partial page that contains
4934                 * the start of the hole
4935                 */
4936                page_len  = first_page_offset - offset;
4937                if (page_len > 0) {
4938                        err = ext4_discard_partial_page_buffers(handle, mapping,
4939                                                   offset, page_len, 0);
4940                        if (err)
4941                                goto out;
4942                }
4943
4944                /*
4945                 * zero out and unmap the partial page that contains
4946                 * the end of the hole
4947                 */
4948                page_len = offset + length - last_page_offset;
4949                if (page_len > 0) {
4950                        err = ext4_discard_partial_page_buffers(handle, mapping,
4951                                        last_page_offset, page_len, 0);
4952                        if (err)
4953                                goto out;
4954                }
4955        }
4956
4957        /*
4958         * If i_size is contained in the last page, we need to
4959         * unmap and zero the partial page after i_size
4960         */
4961        if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4962           inode->i_size % PAGE_CACHE_SIZE != 0) {
4963
4964                page_len = PAGE_CACHE_SIZE -
4965                        (inode->i_size & (PAGE_CACHE_SIZE - 1));
4966
4967                if (page_len > 0) {
4968                        err = ext4_discard_partial_page_buffers(handle,
4969                          mapping, inode->i_size, page_len, 0);
4970
4971                        if (err)
4972                                goto out;
4973                }
4974        }
4975
4976        first_block = (offset + sb->s_blocksize - 1) >>
4977                EXT4_BLOCK_SIZE_BITS(sb);
4978        stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4979
4980        /* If there are no blocks to remove, return now */
4981        if (first_block >= stop_block)
4982                goto out;
4983
4984        down_write(&EXT4_I(inode)->i_data_sem);
4985        ext4_ext_invalidate_cache(inode);
4986        ext4_discard_preallocations(inode);
4987
4988        err = ext4_ext_remove_space(inode, first_block, stop_block - 1);
4989
4990        ext4_ext_invalidate_cache(inode);
4991        ext4_discard_preallocations(inode);
4992
4993        if (IS_SYNC(inode))
4994                ext4_handle_sync(handle);
4995
4996        up_write(&EXT4_I(inode)->i_data_sem);
4997
4998out:
4999        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
5000        ext4_mark_inode_dirty(handle, inode);
5001        ext4_journal_stop(handle);
5002out_dio:
5003        ext4_inode_resume_unlocked_dio(inode);
5004out_mutex:
5005        mutex_unlock(&inode->i_mutex);
5006        return err;
5007}
5008int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5009                __u64 start, __u64 len)
5010{
5011        ext4_lblk_t start_blk;
5012        int error = 0;
5013
5014        /* fallback to generic here if not in extents fmt */
5015        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5016                return generic_block_fiemap(inode, fieinfo, start, len,
5017                        ext4_get_block);
5018
5019        if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
5020                return -EBADR;
5021
5022        if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5023                error = ext4_xattr_fiemap(inode, fieinfo);
5024        } else {
5025                ext4_lblk_t len_blks;
5026                __u64 last_blk;
5027
5028                start_blk = start >> inode->i_sb->s_blocksize_bits;
5029                last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5030                if (last_blk >= EXT_MAX_BLOCKS)
5031                        last_blk = EXT_MAX_BLOCKS-1;
5032                len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5033
5034                /*
5035                 * Walk the extent tree gathering extent information.
5036                 * ext4_ext_fiemap_cb will push extents back to user.
5037                 */
5038                error = ext4_ext_walk_space(inode, start_blk, len_blks,
5039                                          ext4_ext_fiemap_cb, fieinfo);
5040        }
5041
5042        return error;
5043}
5044
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.