linux/fs/ext4/extents.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   4 * Written by Alex Tomas <alex@clusterfs.com>
   5 *
   6 * Architecture independence:
   7 *   Copyright (c) 2005, Bull S.A.
   8 *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   9 */
  10
  11/*
  12 * Extents support for EXT4
  13 *
  14 * TODO:
  15 *   - ext4*_error() should be used in some situations
  16 *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  17 *   - smart tree reduction
  18 */
  19
  20#include <linux/fs.h>
  21#include <linux/time.h>
  22#include <linux/jbd2.h>
  23#include <linux/highuid.h>
  24#include <linux/pagemap.h>
  25#include <linux/quotaops.h>
  26#include <linux/string.h>
  27#include <linux/slab.h>
  28#include <linux/uaccess.h>
  29#include <linux/fiemap.h>
  30#include <linux/backing-dev.h>
  31#include <linux/iomap.h>
  32#include "ext4_jbd2.h"
  33#include "ext4_extents.h"
  34#include "xattr.h"
  35
  36#include <trace/events/ext4.h>
  37
  38/*
  39 * used by extent splitting.
  40 */
  41#define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
  42                                        due to ENOSPC */
  43#define EXT4_EXT_MARK_UNWRIT1   0x2  /* mark first half unwritten */
  44#define EXT4_EXT_MARK_UNWRIT2   0x4  /* mark second half unwritten */
  45
  46#define EXT4_EXT_DATA_VALID1    0x8  /* first half contains valid data */
  47#define EXT4_EXT_DATA_VALID2    0x10 /* second half contains valid data */
  48
  49static __le32 ext4_extent_block_csum(struct inode *inode,
  50                                     struct ext4_extent_header *eh)
  51{
  52        struct ext4_inode_info *ei = EXT4_I(inode);
  53        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  54        __u32 csum;
  55
  56        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
  57                           EXT4_EXTENT_TAIL_OFFSET(eh));
  58        return cpu_to_le32(csum);
  59}
  60
  61static int ext4_extent_block_csum_verify(struct inode *inode,
  62                                         struct ext4_extent_header *eh)
  63{
  64        struct ext4_extent_tail *et;
  65
  66        if (!ext4_has_metadata_csum(inode->i_sb))
  67                return 1;
  68
  69        et = find_ext4_extent_tail(eh);
  70        if (et->et_checksum != ext4_extent_block_csum(inode, eh))
  71                return 0;
  72        return 1;
  73}
  74
  75static void ext4_extent_block_csum_set(struct inode *inode,
  76                                       struct ext4_extent_header *eh)
  77{
  78        struct ext4_extent_tail *et;
  79
  80        if (!ext4_has_metadata_csum(inode->i_sb))
  81                return;
  82
  83        et = find_ext4_extent_tail(eh);
  84        et->et_checksum = ext4_extent_block_csum(inode, eh);
  85}
  86
  87static int ext4_split_extent_at(handle_t *handle,
  88                             struct inode *inode,
  89                             struct ext4_ext_path **ppath,
  90                             ext4_lblk_t split,
  91                             int split_flag,
  92                             int flags);
  93
  94static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
  95{
  96        /*
  97         * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
  98         * moment, get_block can be called only for blocks inside i_size since
  99         * page cache has been already dropped and writes are blocked by
 100         * i_mutex. So we can safely drop the i_data_sem here.
 101         */
 102        BUG_ON(EXT4_JOURNAL(inode) == NULL);
 103        ext4_discard_preallocations(inode, 0);
 104        up_write(&EXT4_I(inode)->i_data_sem);
 105        *dropped = 1;
 106        return 0;
 107}
 108
 109/*
 110 * Make sure 'handle' has at least 'check_cred' credits. If not, restart
 111 * transaction with 'restart_cred' credits. The function drops i_data_sem
 112 * when restarting transaction and gets it after transaction is restarted.
 113 *
 114 * The function returns 0 on success, 1 if transaction had to be restarted,
 115 * and < 0 in case of fatal error.
 116 */
 117int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
 118                                int check_cred, int restart_cred,
 119                                int revoke_cred)
 120{
 121        int ret;
 122        int dropped = 0;
 123
 124        ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
 125                revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
 126        if (dropped)
 127                down_write(&EXT4_I(inode)->i_data_sem);
 128        return ret;
 129}
 130
 131/*
 132 * could return:
 133 *  - EROFS
 134 *  - ENOMEM
 135 */
 136static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
 137                                struct ext4_ext_path *path)
 138{
 139        if (path->p_bh) {
 140                /* path points to block */
 141                BUFFER_TRACE(path->p_bh, "get_write_access");
 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 */
 155static int __ext4_ext_dirty(const char *where, unsigned int line,
 156                            handle_t *handle, struct inode *inode,
 157                            struct ext4_ext_path *path)
 158{
 159        int err;
 160
 161        WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
 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
 174#define ext4_ext_dirty(handle, inode, path) \
 175                __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
 176
 177static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 178                              struct ext4_ext_path *path,
 179                              ext4_lblk_t block)
 180{
 181        if (path) {
 182                int depth = path->p_depth;
 183                struct ext4_extent *ex;
 184
 185                /*
 186                 * Try to predict block placement assuming that we are
 187                 * filling in a file which will eventually be
 188                 * non-sparse --- i.e., in the case of libbfd writing
 189                 * an ELF object sections out-of-order but in a way
 190                 * the eventually results in a contiguous object or
 191                 * executable file, or some database extending a table
 192                 * space file.  However, this is actually somewhat
 193                 * non-ideal if we are writing a sparse file such as
 194                 * qemu or KVM writing a raw image file that is going
 195                 * to stay fairly sparse, since it will end up
 196                 * fragmenting the file system's free space.  Maybe we
 197                 * should have some hueristics or some way to allow
 198                 * userspace to pass a hint to file system,
 199                 * especially if the latter case turns out to be
 200                 * common.
 201                 */
 202                ex = path[depth].p_ext;
 203                if (ex) {
 204                        ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 205                        ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 206
 207                        if (block > ext_block)
 208                                return ext_pblk + (block - ext_block);
 209                        else
 210                                return ext_pblk - (ext_block - block);
 211                }
 212
 213                /* it looks like index is empty;
 214                 * try to find starting block from index itself */
 215                if (path[depth].p_bh)
 216                        return path[depth].p_bh->b_blocknr;
 217        }
 218
 219        /* OK. use inode's group */
 220        return ext4_inode_to_goal_block(inode);
 221}
 222
 223/*
 224 * Allocation for a meta data block
 225 */
 226static ext4_fsblk_t
 227ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 228                        struct ext4_ext_path *path,
 229                        struct ext4_extent *ex, int *err, unsigned int flags)
 230{
 231        ext4_fsblk_t goal, newblock;
 232
 233        goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 234        newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 235                                        NULL, err);
 236        return newblock;
 237}
 238
 239static inline int ext4_ext_space_block(struct inode *inode, int check)
 240{
 241        int size;
 242
 243        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 244                        / sizeof(struct ext4_extent);
 245#ifdef AGGRESSIVE_TEST
 246        if (!check && size > 6)
 247                size = 6;
 248#endif
 249        return size;
 250}
 251
 252static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 253{
 254        int size;
 255
 256        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 257                        / sizeof(struct ext4_extent_idx);
 258#ifdef AGGRESSIVE_TEST
 259        if (!check && size > 5)
 260                size = 5;
 261#endif
 262        return size;
 263}
 264
 265static inline int ext4_ext_space_root(struct inode *inode, int check)
 266{
 267        int size;
 268
 269        size = sizeof(EXT4_I(inode)->i_data);
 270        size -= sizeof(struct ext4_extent_header);
 271        size /= sizeof(struct ext4_extent);
 272#ifdef AGGRESSIVE_TEST
 273        if (!check && size > 3)
 274                size = 3;
 275#endif
 276        return size;
 277}
 278
 279static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 280{
 281        int size;
 282
 283        size = sizeof(EXT4_I(inode)->i_data);
 284        size -= sizeof(struct ext4_extent_header);
 285        size /= sizeof(struct ext4_extent_idx);
 286#ifdef AGGRESSIVE_TEST
 287        if (!check && size > 4)
 288                size = 4;
 289#endif
 290        return size;
 291}
 292
 293static inline int
 294ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
 295                           struct ext4_ext_path **ppath, ext4_lblk_t lblk,
 296                           int nofail)
 297{
 298        struct ext4_ext_path *path = *ppath;
 299        int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
 300        int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
 301
 302        if (nofail)
 303                flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
 304
 305        return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
 306                        EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
 307                        flags);
 308}
 309
 310static int
 311ext4_ext_max_entries(struct inode *inode, int depth)
 312{
 313        int max;
 314
 315        if (depth == ext_depth(inode)) {
 316                if (depth == 0)
 317                        max = ext4_ext_space_root(inode, 1);
 318                else
 319                        max = ext4_ext_space_root_idx(inode, 1);
 320        } else {
 321                if (depth == 0)
 322                        max = ext4_ext_space_block(inode, 1);
 323                else
 324                        max = ext4_ext_space_block_idx(inode, 1);
 325        }
 326
 327        return max;
 328}
 329
 330static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 331{
 332        ext4_fsblk_t block = ext4_ext_pblock(ext);
 333        int len = ext4_ext_get_actual_len(ext);
 334        ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
 335
 336        /*
 337         * We allow neither:
 338         *  - zero length
 339         *  - overflow/wrap-around
 340         */
 341        if (lblock + len <= lblock)
 342                return 0;
 343        return ext4_inode_block_valid(inode, block, len);
 344}
 345
 346static int ext4_valid_extent_idx(struct inode *inode,
 347                                struct ext4_extent_idx *ext_idx)
 348{
 349        ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 350
 351        return ext4_inode_block_valid(inode, block, 1);
 352}
 353
 354static int ext4_valid_extent_entries(struct inode *inode,
 355                                     struct ext4_extent_header *eh,
 356                                     ext4_fsblk_t *pblk, int depth)
 357{
 358        unsigned short entries;
 359        if (eh->eh_entries == 0)
 360                return 1;
 361
 362        entries = le16_to_cpu(eh->eh_entries);
 363
 364        if (depth == 0) {
 365                /* leaf entries */
 366                struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 367                ext4_lblk_t lblock = 0;
 368                ext4_lblk_t prev = 0;
 369                int len = 0;
 370                while (entries) {
 371                        if (!ext4_valid_extent(inode, ext))
 372                                return 0;
 373
 374                        /* Check for overlapping extents */
 375                        lblock = le32_to_cpu(ext->ee_block);
 376                        len = ext4_ext_get_actual_len(ext);
 377                        if ((lblock <= prev) && prev) {
 378                                *pblk = ext4_ext_pblock(ext);
 379                                return 0;
 380                        }
 381                        ext++;
 382                        entries--;
 383                        prev = lblock + len - 1;
 384                }
 385        } else {
 386                struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 387                while (entries) {
 388                        if (!ext4_valid_extent_idx(inode, ext_idx))
 389                                return 0;
 390                        ext_idx++;
 391                        entries--;
 392                }
 393        }
 394        return 1;
 395}
 396
 397static int __ext4_ext_check(const char *function, unsigned int line,
 398                            struct inode *inode, struct ext4_extent_header *eh,
 399                            int depth, ext4_fsblk_t pblk)
 400{
 401        const char *error_msg;
 402        int max = 0, err = -EFSCORRUPTED;
 403
 404        if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 405                error_msg = "invalid magic";
 406                goto corrupted;
 407        }
 408        if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 409                error_msg = "unexpected eh_depth";
 410                goto corrupted;
 411        }
 412        if (unlikely(eh->eh_max == 0)) {
 413                error_msg = "invalid eh_max";
 414                goto corrupted;
 415        }
 416        max = ext4_ext_max_entries(inode, depth);
 417        if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 418                error_msg = "too large eh_max";
 419                goto corrupted;
 420        }
 421        if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 422                error_msg = "invalid eh_entries";
 423                goto corrupted;
 424        }
 425        if (!ext4_valid_extent_entries(inode, eh, &pblk, depth)) {
 426                error_msg = "invalid extent entries";
 427                goto corrupted;
 428        }
 429        if (unlikely(depth > 32)) {
 430                error_msg = "too large eh_depth";
 431                goto corrupted;
 432        }
 433        /* Verify checksum on non-root extent tree nodes */
 434        if (ext_depth(inode) != depth &&
 435            !ext4_extent_block_csum_verify(inode, eh)) {
 436                error_msg = "extent tree corrupted";
 437                err = -EFSBADCRC;
 438                goto corrupted;
 439        }
 440        return 0;
 441
 442corrupted:
 443        ext4_error_inode_err(inode, function, line, 0, -err,
 444                             "pblk %llu bad header/extent: %s - magic %x, "
 445                             "entries %u, max %u(%u), depth %u(%u)",
 446                             (unsigned long long) pblk, error_msg,
 447                             le16_to_cpu(eh->eh_magic),
 448                             le16_to_cpu(eh->eh_entries),
 449                             le16_to_cpu(eh->eh_max),
 450                             max, le16_to_cpu(eh->eh_depth), depth);
 451        return err;
 452}
 453
 454#define ext4_ext_check(inode, eh, depth, pblk)                  \
 455        __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
 456
 457int ext4_ext_check_inode(struct inode *inode)
 458{
 459        return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
 460}
 461
 462static void ext4_cache_extents(struct inode *inode,
 463                               struct ext4_extent_header *eh)
 464{
 465        struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
 466        ext4_lblk_t prev = 0;
 467        int i;
 468
 469        for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
 470                unsigned int status = EXTENT_STATUS_WRITTEN;
 471                ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
 472                int len = ext4_ext_get_actual_len(ex);
 473
 474                if (prev && (prev != lblk))
 475                        ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
 476                                             EXTENT_STATUS_HOLE);
 477
 478                if (ext4_ext_is_unwritten(ex))
 479                        status = EXTENT_STATUS_UNWRITTEN;
 480                ext4_es_cache_extent(inode, lblk, len,
 481                                     ext4_ext_pblock(ex), status);
 482                prev = lblk + len;
 483        }
 484}
 485
 486static struct buffer_head *
 487__read_extent_tree_block(const char *function, unsigned int line,
 488                         struct inode *inode, ext4_fsblk_t pblk, int depth,
 489                         int flags)
 490{
 491        struct buffer_head              *bh;
 492        int                             err;
 493        gfp_t                           gfp_flags = __GFP_MOVABLE | GFP_NOFS;
 494
 495        if (flags & EXT4_EX_NOFAIL)
 496                gfp_flags |= __GFP_NOFAIL;
 497
 498        bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
 499        if (unlikely(!bh))
 500                return ERR_PTR(-ENOMEM);
 501
 502        if (!bh_uptodate_or_lock(bh)) {
 503                trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
 504                err = bh_submit_read(bh);
 505                if (err < 0)
 506                        goto errout;
 507        }
 508        if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
 509                return bh;
 510        err = __ext4_ext_check(function, line, inode,
 511                               ext_block_hdr(bh), depth, pblk);
 512        if (err)
 513                goto errout;
 514        set_buffer_verified(bh);
 515        /*
 516         * If this is a leaf block, cache all of its entries
 517         */
 518        if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
 519                struct ext4_extent_header *eh = ext_block_hdr(bh);
 520                ext4_cache_extents(inode, eh);
 521        }
 522        return bh;
 523errout:
 524        put_bh(bh);
 525        return ERR_PTR(err);
 526
 527}
 528
 529#define read_extent_tree_block(inode, pblk, depth, flags)               \
 530        __read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
 531                                 (depth), (flags))
 532
 533/*
 534 * This function is called to cache a file's extent information in the
 535 * extent status tree
 536 */
 537int ext4_ext_precache(struct inode *inode)
 538{
 539        struct ext4_inode_info *ei = EXT4_I(inode);
 540        struct ext4_ext_path *path = NULL;
 541        struct buffer_head *bh;
 542        int i = 0, depth, ret = 0;
 543
 544        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 545                return 0;       /* not an extent-mapped inode */
 546
 547        down_read(&ei->i_data_sem);
 548        depth = ext_depth(inode);
 549
 550        /* Don't cache anything if there are no external extent blocks */
 551        if (!depth) {
 552                up_read(&ei->i_data_sem);
 553                return ret;
 554        }
 555
 556        path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
 557                       GFP_NOFS);
 558        if (path == NULL) {
 559                up_read(&ei->i_data_sem);
 560                return -ENOMEM;
 561        }
 562
 563        path[0].p_hdr = ext_inode_hdr(inode);
 564        ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
 565        if (ret)
 566                goto out;
 567        path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
 568        while (i >= 0) {
 569                /*
 570                 * If this is a leaf block or we've reached the end of
 571                 * the index block, go up
 572                 */
 573                if ((i == depth) ||
 574                    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
 575                        brelse(path[i].p_bh);
 576                        path[i].p_bh = NULL;
 577                        i--;
 578                        continue;
 579                }
 580                bh = read_extent_tree_block(inode,
 581                                            ext4_idx_pblock(path[i].p_idx++),
 582                                            depth - i - 1,
 583                                            EXT4_EX_FORCE_CACHE);
 584                if (IS_ERR(bh)) {
 585                        ret = PTR_ERR(bh);
 586                        break;
 587                }
 588                i++;
 589                path[i].p_bh = bh;
 590                path[i].p_hdr = ext_block_hdr(bh);
 591                path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
 592        }
 593        ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
 594out:
 595        up_read(&ei->i_data_sem);
 596        ext4_ext_drop_refs(path);
 597        kfree(path);
 598        return ret;
 599}
 600
 601#ifdef EXT_DEBUG
 602static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 603{
 604        int k, l = path->p_depth;
 605
 606        ext_debug(inode, "path:");
 607        for (k = 0; k <= l; k++, path++) {
 608                if (path->p_idx) {
 609                        ext_debug(inode, "  %d->%llu",
 610                                  le32_to_cpu(path->p_idx->ei_block),
 611                                  ext4_idx_pblock(path->p_idx));
 612                } else if (path->p_ext) {
 613                        ext_debug(inode, "  %d:[%d]%d:%llu ",
 614                                  le32_to_cpu(path->p_ext->ee_block),
 615                                  ext4_ext_is_unwritten(path->p_ext),
 616                                  ext4_ext_get_actual_len(path->p_ext),
 617                                  ext4_ext_pblock(path->p_ext));
 618                } else
 619                        ext_debug(inode, "  []");
 620        }
 621        ext_debug(inode, "\n");
 622}
 623
 624static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 625{
 626        int depth = ext_depth(inode);
 627        struct ext4_extent_header *eh;
 628        struct ext4_extent *ex;
 629        int i;
 630
 631        if (!path)
 632                return;
 633
 634        eh = path[depth].p_hdr;
 635        ex = EXT_FIRST_EXTENT(eh);
 636
 637        ext_debug(inode, "Displaying leaf extents\n");
 638
 639        for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 640                ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 641                          ext4_ext_is_unwritten(ex),
 642                          ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 643        }
 644        ext_debug(inode, "\n");
 645}
 646
 647static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 648                        ext4_fsblk_t newblock, int level)
 649{
 650        int depth = ext_depth(inode);
 651        struct ext4_extent *ex;
 652
 653        if (depth != level) {
 654                struct ext4_extent_idx *idx;
 655                idx = path[level].p_idx;
 656                while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 657                        ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
 658                                  level, le32_to_cpu(idx->ei_block),
 659                                  ext4_idx_pblock(idx), newblock);
 660                        idx++;
 661                }
 662
 663                return;
 664        }
 665
 666        ex = path[depth].p_ext;
 667        while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 668                ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
 669                                le32_to_cpu(ex->ee_block),
 670                                ext4_ext_pblock(ex),
 671                                ext4_ext_is_unwritten(ex),
 672                                ext4_ext_get_actual_len(ex),
 673                                newblock);
 674                ex++;
 675        }
 676}
 677
 678#else
 679#define ext4_ext_show_path(inode, path)
 680#define ext4_ext_show_leaf(inode, path)
 681#define ext4_ext_show_move(inode, path, newblock, level)
 682#endif
 683
 684void ext4_ext_drop_refs(struct ext4_ext_path *path)
 685{
 686        int depth, i;
 687
 688        if (!path)
 689                return;
 690        depth = path->p_depth;
 691        for (i = 0; i <= depth; i++, path++) {
 692                brelse(path->p_bh);
 693                path->p_bh = NULL;
 694        }
 695}
 696
 697/*
 698 * ext4_ext_binsearch_idx:
 699 * binary search for the closest index of the given block
 700 * the header must be checked before calling this
 701 */
 702static void
 703ext4_ext_binsearch_idx(struct inode *inode,
 704                        struct ext4_ext_path *path, ext4_lblk_t block)
 705{
 706        struct ext4_extent_header *eh = path->p_hdr;
 707        struct ext4_extent_idx *r, *l, *m;
 708
 709
 710        ext_debug(inode, "binsearch for %u(idx):  ", block);
 711
 712        l = EXT_FIRST_INDEX(eh) + 1;
 713        r = EXT_LAST_INDEX(eh);
 714        while (l <= r) {
 715                m = l + (r - l) / 2;
 716                if (block < le32_to_cpu(m->ei_block))
 717                        r = m - 1;
 718                else
 719                        l = m + 1;
 720                ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 721                          le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
 722                          r, le32_to_cpu(r->ei_block));
 723        }
 724
 725        path->p_idx = l - 1;
 726        ext_debug(inode, "  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
 727                  ext4_idx_pblock(path->p_idx));
 728
 729#ifdef CHECK_BINSEARCH
 730        {
 731                struct ext4_extent_idx *chix, *ix;
 732                int k;
 733
 734                chix = ix = EXT_FIRST_INDEX(eh);
 735                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 736                        if (k != 0 && le32_to_cpu(ix->ei_block) <=
 737                            le32_to_cpu(ix[-1].ei_block)) {
 738                                printk(KERN_DEBUG "k=%d, ix=0x%p, "
 739                                       "first=0x%p\n", k,
 740                                       ix, EXT_FIRST_INDEX(eh));
 741                                printk(KERN_DEBUG "%u <= %u\n",
 742                                       le32_to_cpu(ix->ei_block),
 743                                       le32_to_cpu(ix[-1].ei_block));
 744                        }
 745                        BUG_ON(k && le32_to_cpu(ix->ei_block)
 746                                           <= le32_to_cpu(ix[-1].ei_block));
 747                        if (block < le32_to_cpu(ix->ei_block))
 748                                break;
 749                        chix = ix;
 750                }
 751                BUG_ON(chix != path->p_idx);
 752        }
 753#endif
 754
 755}
 756
 757/*
 758 * ext4_ext_binsearch:
 759 * binary search for closest extent of the given block
 760 * the header must be checked before calling this
 761 */
 762static void
 763ext4_ext_binsearch(struct inode *inode,
 764                struct ext4_ext_path *path, ext4_lblk_t block)
 765{
 766        struct ext4_extent_header *eh = path->p_hdr;
 767        struct ext4_extent *r, *l, *m;
 768
 769        if (eh->eh_entries == 0) {
 770                /*
 771                 * this leaf is empty:
 772                 * we get such a leaf in split/add case
 773                 */
 774                return;
 775        }
 776
 777        ext_debug(inode, "binsearch for %u:  ", block);
 778
 779        l = EXT_FIRST_EXTENT(eh) + 1;
 780        r = EXT_LAST_EXTENT(eh);
 781
 782        while (l <= r) {
 783                m = l + (r - l) / 2;
 784                if (block < le32_to_cpu(m->ee_block))
 785                        r = m - 1;
 786                else
 787                        l = m + 1;
 788                ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
 789                          le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
 790                          r, le32_to_cpu(r->ee_block));
 791        }
 792
 793        path->p_ext = l - 1;
 794        ext_debug(inode, "  -> %d:%llu:[%d]%d ",
 795                        le32_to_cpu(path->p_ext->ee_block),
 796                        ext4_ext_pblock(path->p_ext),
 797                        ext4_ext_is_unwritten(path->p_ext),
 798                        ext4_ext_get_actual_len(path->p_ext));
 799
 800#ifdef CHECK_BINSEARCH
 801        {
 802                struct ext4_extent *chex, *ex;
 803                int k;
 804
 805                chex = ex = EXT_FIRST_EXTENT(eh);
 806                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 807                        BUG_ON(k && le32_to_cpu(ex->ee_block)
 808                                          <= le32_to_cpu(ex[-1].ee_block));
 809                        if (block < le32_to_cpu(ex->ee_block))
 810                                break;
 811                        chex = ex;
 812                }
 813                BUG_ON(chex != path->p_ext);
 814        }
 815#endif
 816
 817}
 818
 819void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 820{
 821        struct ext4_extent_header *eh;
 822
 823        eh = ext_inode_hdr(inode);
 824        eh->eh_depth = 0;
 825        eh->eh_entries = 0;
 826        eh->eh_magic = EXT4_EXT_MAGIC;
 827        eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 828        ext4_mark_inode_dirty(handle, inode);
 829}
 830
 831struct ext4_ext_path *
 832ext4_find_extent(struct inode *inode, ext4_lblk_t block,
 833                 struct ext4_ext_path **orig_path, int flags)
 834{
 835        struct ext4_extent_header *eh;
 836        struct buffer_head *bh;
 837        struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
 838        short int depth, i, ppos = 0;
 839        int ret;
 840        gfp_t gfp_flags = GFP_NOFS;
 841
 842        if (flags & EXT4_EX_NOFAIL)
 843                gfp_flags |= __GFP_NOFAIL;
 844
 845        eh = ext_inode_hdr(inode);
 846        depth = ext_depth(inode);
 847        if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
 848                EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
 849                                 depth);
 850                ret = -EFSCORRUPTED;
 851                goto err;
 852        }
 853
 854        if (path) {
 855                ext4_ext_drop_refs(path);
 856                if (depth > path[0].p_maxdepth) {
 857                        kfree(path);
 858                        *orig_path = path = NULL;
 859                }
 860        }
 861        if (!path) {
 862                /* account possible depth increase */
 863                path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
 864                                gfp_flags);
 865                if (unlikely(!path))
 866                        return ERR_PTR(-ENOMEM);
 867                path[0].p_maxdepth = depth + 1;
 868        }
 869        path[0].p_hdr = eh;
 870        path[0].p_bh = NULL;
 871
 872        i = depth;
 873        if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
 874                ext4_cache_extents(inode, eh);
 875        /* walk through the tree */
 876        while (i) {
 877                ext_debug(inode, "depth %d: num %d, max %d\n",
 878                          ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 879
 880                ext4_ext_binsearch_idx(inode, path + ppos, block);
 881                path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 882                path[ppos].p_depth = i;
 883                path[ppos].p_ext = NULL;
 884
 885                bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
 886                                            flags);
 887                if (IS_ERR(bh)) {
 888                        ret = PTR_ERR(bh);
 889                        goto err;
 890                }
 891
 892                eh = ext_block_hdr(bh);
 893                ppos++;
 894                path[ppos].p_bh = bh;
 895                path[ppos].p_hdr = eh;
 896        }
 897
 898        path[ppos].p_depth = i;
 899        path[ppos].p_ext = NULL;
 900        path[ppos].p_idx = NULL;
 901
 902        /* find extent */
 903        ext4_ext_binsearch(inode, path + ppos, block);
 904        /* if not an empty leaf */
 905        if (path[ppos].p_ext)
 906                path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 907
 908        ext4_ext_show_path(inode, path);
 909
 910        return path;
 911
 912err:
 913        ext4_ext_drop_refs(path);
 914        kfree(path);
 915        if (orig_path)
 916                *orig_path = NULL;
 917        return ERR_PTR(ret);
 918}
 919
 920/*
 921 * ext4_ext_insert_index:
 922 * insert new index [@logical;@ptr] into the block at @curp;
 923 * check where to insert: before @curp or after @curp
 924 */
 925static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 926                                 struct ext4_ext_path *curp,
 927                                 int logical, ext4_fsblk_t ptr)
 928{
 929        struct ext4_extent_idx *ix;
 930        int len, err;
 931
 932        err = ext4_ext_get_access(handle, inode, curp);
 933        if (err)
 934                return err;
 935
 936        if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 937                EXT4_ERROR_INODE(inode,
 938                                 "logical %d == ei_block %d!",
 939                                 logical, le32_to_cpu(curp->p_idx->ei_block));
 940                return -EFSCORRUPTED;
 941        }
 942
 943        if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 944                             >= le16_to_cpu(curp->p_hdr->eh_max))) {
 945                EXT4_ERROR_INODE(inode,
 946                                 "eh_entries %d >= eh_max %d!",
 947                                 le16_to_cpu(curp->p_hdr->eh_entries),
 948                                 le16_to_cpu(curp->p_hdr->eh_max));
 949                return -EFSCORRUPTED;
 950        }
 951
 952        if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 953                /* insert after */
 954                ext_debug(inode, "insert new index %d after: %llu\n",
 955                          logical, ptr);
 956                ix = curp->p_idx + 1;
 957        } else {
 958                /* insert before */
 959                ext_debug(inode, "insert new index %d before: %llu\n",
 960                          logical, ptr);
 961                ix = curp->p_idx;
 962        }
 963
 964        len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 965        BUG_ON(len < 0);
 966        if (len > 0) {
 967                ext_debug(inode, "insert new index %d: "
 968                                "move %d indices from 0x%p to 0x%p\n",
 969                                logical, len, ix, ix + 1);
 970                memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 971        }
 972
 973        if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 974                EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 975                return -EFSCORRUPTED;
 976        }
 977
 978        ix->ei_block = cpu_to_le32(logical);
 979        ext4_idx_store_pblock(ix, ptr);
 980        le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 981
 982        if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 983                EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 984                return -EFSCORRUPTED;
 985        }
 986
 987        err = ext4_ext_dirty(handle, inode, curp);
 988        ext4_std_error(inode->i_sb, err);
 989
 990        return err;
 991}
 992
 993/*
 994 * ext4_ext_split:
 995 * inserts new subtree into the path, using free index entry
 996 * at depth @at:
 997 * - allocates all needed blocks (new leaf and all intermediate index blocks)
 998 * - makes decision where to split
 999 * - moves remaining extents and index entries (right to the split point)
1000 *   into the newly allocated blocks
1001 * - initializes subtree
1002 */
1003static int ext4_ext_split(handle_t *handle, struct inode *inode,
1004                          unsigned int flags,
1005                          struct ext4_ext_path *path,
1006                          struct ext4_extent *newext, int at)
1007{
1008        struct buffer_head *bh = NULL;
1009        int depth = ext_depth(inode);
1010        struct ext4_extent_header *neh;
1011        struct ext4_extent_idx *fidx;
1012        int i = at, k, m, a;
1013        ext4_fsblk_t newblock, oldblock;
1014        __le32 border;
1015        ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1016        gfp_t gfp_flags = GFP_NOFS;
1017        int err = 0;
1018        size_t ext_size = 0;
1019
1020        if (flags & EXT4_EX_NOFAIL)
1021                gfp_flags |= __GFP_NOFAIL;
1022
1023        /* make decision: where to split? */
1024        /* FIXME: now decision is simplest: at current extent */
1025
1026        /* if current leaf will be split, then we should use
1027         * border from split point */
1028        if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1029                EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1030                return -EFSCORRUPTED;
1031        }
1032        if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1033                border = path[depth].p_ext[1].ee_block;
1034                ext_debug(inode, "leaf will be split."
1035                                " next leaf starts at %d\n",
1036                                  le32_to_cpu(border));
1037        } else {
1038                border = newext->ee_block;
1039                ext_debug(inode, "leaf will be added."
1040                                " next leaf starts at %d\n",
1041                                le32_to_cpu(border));
1042        }
1043
1044        /*
1045         * If error occurs, then we break processing
1046         * and mark filesystem read-only. index won't
1047         * be inserted and tree will be in consistent
1048         * state. Next mount will repair buffers too.
1049         */
1050
1051        /*
1052         * Get array to track all allocated blocks.
1053         * We need this to handle errors and free blocks
1054         * upon them.
1055         */
1056        ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1057        if (!ablocks)
1058                return -ENOMEM;
1059
1060        /* allocate all needed blocks */
1061        ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1062        for (a = 0; a < depth - at; a++) {
1063                newblock = ext4_ext_new_meta_block(handle, inode, path,
1064                                                   newext, &err, flags);
1065                if (newblock == 0)
1066                        goto cleanup;
1067                ablocks[a] = newblock;
1068        }
1069
1070        /* initialize new leaf */
1071        newblock = ablocks[--a];
1072        if (unlikely(newblock == 0)) {
1073                EXT4_ERROR_INODE(inode, "newblock == 0!");
1074                err = -EFSCORRUPTED;
1075                goto cleanup;
1076        }
1077        bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1078        if (unlikely(!bh)) {
1079                err = -ENOMEM;
1080                goto cleanup;
1081        }
1082        lock_buffer(bh);
1083
1084        err = ext4_journal_get_create_access(handle, bh);
1085        if (err)
1086                goto cleanup;
1087
1088        neh = ext_block_hdr(bh);
1089        neh->eh_entries = 0;
1090        neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1091        neh->eh_magic = EXT4_EXT_MAGIC;
1092        neh->eh_depth = 0;
1093
1094        /* move remainder of path[depth] to the new leaf */
1095        if (unlikely(path[depth].p_hdr->eh_entries !=
1096                     path[depth].p_hdr->eh_max)) {
1097                EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1098                                 path[depth].p_hdr->eh_entries,
1099                                 path[depth].p_hdr->eh_max);
1100                err = -EFSCORRUPTED;
1101                goto cleanup;
1102        }
1103        /* start copy from next extent */
1104        m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1105        ext4_ext_show_move(inode, path, newblock, depth);
1106        if (m) {
1107                struct ext4_extent *ex;
1108                ex = EXT_FIRST_EXTENT(neh);
1109                memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1110                le16_add_cpu(&neh->eh_entries, m);
1111        }
1112
1113        /* zero out unused area in the extent block */
1114        ext_size = sizeof(struct ext4_extent_header) +
1115                sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1116        memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1117        ext4_extent_block_csum_set(inode, neh);
1118        set_buffer_uptodate(bh);
1119        unlock_buffer(bh);
1120
1121        err = ext4_handle_dirty_metadata(handle, inode, bh);
1122        if (err)
1123                goto cleanup;
1124        brelse(bh);
1125        bh = NULL;
1126
1127        /* correct old leaf */
1128        if (m) {
1129                err = ext4_ext_get_access(handle, inode, path + depth);
1130                if (err)
1131                        goto cleanup;
1132                le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1133                err = ext4_ext_dirty(handle, inode, path + depth);
1134                if (err)
1135                        goto cleanup;
1136
1137        }
1138
1139        /* create intermediate indexes */
1140        k = depth - at - 1;
1141        if (unlikely(k < 0)) {
1142                EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1143                err = -EFSCORRUPTED;
1144                goto cleanup;
1145        }
1146        if (k)
1147                ext_debug(inode, "create %d intermediate indices\n", k);
1148        /* insert new index into current index block */
1149        /* current depth stored in i var */
1150        i = depth - 1;
1151        while (k--) {
1152                oldblock = newblock;
1153                newblock = ablocks[--a];
1154                bh = sb_getblk(inode->i_sb, newblock);
1155                if (unlikely(!bh)) {
1156                        err = -ENOMEM;
1157                        goto cleanup;
1158                }
1159                lock_buffer(bh);
1160
1161                err = ext4_journal_get_create_access(handle, bh);
1162                if (err)
1163                        goto cleanup;
1164
1165                neh = ext_block_hdr(bh);
1166                neh->eh_entries = cpu_to_le16(1);
1167                neh->eh_magic = EXT4_EXT_MAGIC;
1168                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1169                neh->eh_depth = cpu_to_le16(depth - i);
1170                fidx = EXT_FIRST_INDEX(neh);
1171                fidx->ei_block = border;
1172                ext4_idx_store_pblock(fidx, oldblock);
1173
1174                ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1175                                i, newblock, le32_to_cpu(border), oldblock);
1176
1177                /* move remainder of path[i] to the new index block */
1178                if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1179                                        EXT_LAST_INDEX(path[i].p_hdr))) {
1180                        EXT4_ERROR_INODE(inode,
1181                                         "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1182                                         le32_to_cpu(path[i].p_ext->ee_block));
1183                        err = -EFSCORRUPTED;
1184                        goto cleanup;
1185                }
1186                /* start copy indexes */
1187                m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1188                ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1189                                EXT_MAX_INDEX(path[i].p_hdr));
1190                ext4_ext_show_move(inode, path, newblock, i);
1191                if (m) {
1192                        memmove(++fidx, path[i].p_idx,
1193                                sizeof(struct ext4_extent_idx) * m);
1194                        le16_add_cpu(&neh->eh_entries, m);
1195                }
1196                /* zero out unused area in the extent block */
1197                ext_size = sizeof(struct ext4_extent_header) +
1198                   (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1199                memset(bh->b_data + ext_size, 0,
1200                        inode->i_sb->s_blocksize - ext_size);
1201                ext4_extent_block_csum_set(inode, neh);
1202                set_buffer_uptodate(bh);
1203                unlock_buffer(bh);
1204
1205                err = ext4_handle_dirty_metadata(handle, inode, bh);
1206                if (err)
1207                        goto cleanup;
1208                brelse(bh);
1209                bh = NULL;
1210
1211                /* correct old index */
1212                if (m) {
1213                        err = ext4_ext_get_access(handle, inode, path + i);
1214                        if (err)
1215                                goto cleanup;
1216                        le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1217                        err = ext4_ext_dirty(handle, inode, path + i);
1218                        if (err)
1219                                goto cleanup;
1220                }
1221
1222                i--;
1223        }
1224
1225        /* insert new index */
1226        err = ext4_ext_insert_index(handle, inode, path + at,
1227                                    le32_to_cpu(border), newblock);
1228
1229cleanup:
1230        if (bh) {
1231                if (buffer_locked(bh))
1232                        unlock_buffer(bh);
1233                brelse(bh);
1234        }
1235
1236        if (err) {
1237                /* free all allocated blocks in error case */
1238                for (i = 0; i < depth; i++) {
1239                        if (!ablocks[i])
1240                                continue;
1241                        ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1242                                         EXT4_FREE_BLOCKS_METADATA);
1243                }
1244        }
1245        kfree(ablocks);
1246
1247        return err;
1248}
1249
1250/*
1251 * ext4_ext_grow_indepth:
1252 * implements tree growing procedure:
1253 * - allocates new block
1254 * - moves top-level data (index block or leaf) into the new block
1255 * - initializes new top-level, creating index that points to the
1256 *   just created block
1257 */
1258static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1259                                 unsigned int flags)
1260{
1261        struct ext4_extent_header *neh;
1262        struct buffer_head *bh;
1263        ext4_fsblk_t newblock, goal = 0;
1264        struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1265        int err = 0;
1266        size_t ext_size = 0;
1267
1268        /* Try to prepend new index to old one */
1269        if (ext_depth(inode))
1270                goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1271        if (goal > le32_to_cpu(es->s_first_data_block)) {
1272                flags |= EXT4_MB_HINT_TRY_GOAL;
1273                goal--;
1274        } else
1275                goal = ext4_inode_to_goal_block(inode);
1276        newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1277                                        NULL, &err);
1278        if (newblock == 0)
1279                return err;
1280
1281        bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1282        if (unlikely(!bh))
1283                return -ENOMEM;
1284        lock_buffer(bh);
1285
1286        err = ext4_journal_get_create_access(handle, bh);
1287        if (err) {
1288                unlock_buffer(bh);
1289                goto out;
1290        }
1291
1292        ext_size = sizeof(EXT4_I(inode)->i_data);
1293        /* move top-level index/leaf into new block */
1294        memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1295        /* zero out unused area in the extent block */
1296        memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1297
1298        /* set size of new block */
1299        neh = ext_block_hdr(bh);
1300        /* old root could have indexes or leaves
1301         * so calculate e_max right way */
1302        if (ext_depth(inode))
1303                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1304        else
1305                neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1306        neh->eh_magic = EXT4_EXT_MAGIC;
1307        ext4_extent_block_csum_set(inode, neh);
1308        set_buffer_uptodate(bh);
1309        unlock_buffer(bh);
1310
1311        err = ext4_handle_dirty_metadata(handle, inode, bh);
1312        if (err)
1313                goto out;
1314
1315        /* Update top-level index: num,max,pointer */
1316        neh = ext_inode_hdr(inode);
1317        neh->eh_entries = cpu_to_le16(1);
1318        ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1319        if (neh->eh_depth == 0) {
1320                /* Root extent block becomes index block */
1321                neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1322                EXT_FIRST_INDEX(neh)->ei_block =
1323                        EXT_FIRST_EXTENT(neh)->ee_block;
1324        }
1325        ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1326                  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1327                  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1328                  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1329
1330        le16_add_cpu(&neh->eh_depth, 1);
1331        err = ext4_mark_inode_dirty(handle, inode);
1332out:
1333        brelse(bh);
1334
1335        return err;
1336}
1337
1338/*
1339 * ext4_ext_create_new_leaf:
1340 * finds empty index and adds new leaf.
1341 * if no free index is found, then it requests in-depth growing.
1342 */
1343static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1344                                    unsigned int mb_flags,
1345                                    unsigned int gb_flags,
1346                                    struct ext4_ext_path **ppath,
1347                                    struct ext4_extent *newext)
1348{
1349        struct ext4_ext_path *path = *ppath;
1350        struct ext4_ext_path *curp;
1351        int depth, i, err = 0;
1352
1353repeat:
1354        i = depth = ext_depth(inode);
1355
1356        /* walk up to the tree and look for free index entry */
1357        curp = path + depth;
1358        while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1359                i--;
1360                curp--;
1361        }
1362
1363        /* we use already allocated block for index block,
1364         * so subsequent data blocks should be contiguous */
1365        if (EXT_HAS_FREE_INDEX(curp)) {
1366                /* if we found index with free entry, then use that
1367                 * entry: create all needed subtree and add new leaf */
1368                err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1369                if (err)
1370                        goto out;
1371
1372                /* refill path */
1373                path = ext4_find_extent(inode,
1374                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1375                                    ppath, gb_flags);
1376                if (IS_ERR(path))
1377                        err = PTR_ERR(path);
1378        } else {
1379                /* tree is full, time to grow in depth */
1380                err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1381                if (err)
1382                        goto out;
1383
1384                /* refill path */
1385                path = ext4_find_extent(inode,
1386                                   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1387                                    ppath, gb_flags);
1388                if (IS_ERR(path)) {
1389                        err = PTR_ERR(path);
1390                        goto out;
1391                }
1392
1393                /*
1394                 * only first (depth 0 -> 1) produces free space;
1395                 * in all other cases we have to split the grown tree
1396                 */
1397                depth = ext_depth(inode);
1398                if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1399                        /* now we need to split */
1400                        goto repeat;
1401                }
1402        }
1403
1404out:
1405        return err;
1406}
1407
1408/*
1409 * search the closest allocated block to the left for *logical
1410 * and returns it at @logical + it's physical address at @phys
1411 * if *logical is the smallest allocated block, the function
1412 * returns 0 at @phys
1413 * return value contains 0 (success) or error code
1414 */
1415static int ext4_ext_search_left(struct inode *inode,
1416                                struct ext4_ext_path *path,
1417                                ext4_lblk_t *logical, ext4_fsblk_t *phys)
1418{
1419        struct ext4_extent_idx *ix;
1420        struct ext4_extent *ex;
1421        int depth, ee_len;
1422
1423        if (unlikely(path == NULL)) {
1424                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1425                return -EFSCORRUPTED;
1426        }
1427        depth = path->p_depth;
1428        *phys = 0;
1429
1430        if (depth == 0 && path->p_ext == NULL)
1431                return 0;
1432
1433        /* usually extent in the path covers blocks smaller
1434         * then *logical, but it can be that extent is the
1435         * first one in the file */
1436
1437        ex = path[depth].p_ext;
1438        ee_len = ext4_ext_get_actual_len(ex);
1439        if (*logical < le32_to_cpu(ex->ee_block)) {
1440                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1441                        EXT4_ERROR_INODE(inode,
1442                                         "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1443                                         *logical, le32_to_cpu(ex->ee_block));
1444                        return -EFSCORRUPTED;
1445                }
1446                while (--depth >= 0) {
1447                        ix = path[depth].p_idx;
1448                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1449                                EXT4_ERROR_INODE(inode,
1450                                  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1451                                  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1452                                  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1453                le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1454                                  depth);
1455                                return -EFSCORRUPTED;
1456                        }
1457                }
1458                return 0;
1459        }
1460
1461        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1462                EXT4_ERROR_INODE(inode,
1463                                 "logical %d < ee_block %d + ee_len %d!",
1464                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1465                return -EFSCORRUPTED;
1466        }
1467
1468        *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1469        *phys = ext4_ext_pblock(ex) + ee_len - 1;
1470        return 0;
1471}
1472
1473/*
1474 * search the closest allocated block to the right for *logical
1475 * and returns it at @logical + it's physical address at @phys
1476 * if *logical is the largest allocated block, the function
1477 * returns 0 at @phys
1478 * return value contains 0 (success) or error code
1479 */
1480static int ext4_ext_search_right(struct inode *inode,
1481                                 struct ext4_ext_path *path,
1482                                 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1483                                 struct ext4_extent **ret_ex)
1484{
1485        struct buffer_head *bh = NULL;
1486        struct ext4_extent_header *eh;
1487        struct ext4_extent_idx *ix;
1488        struct ext4_extent *ex;
1489        ext4_fsblk_t block;
1490        int depth;      /* Note, NOT eh_depth; depth from top of tree */
1491        int ee_len;
1492
1493        if (unlikely(path == NULL)) {
1494                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1495                return -EFSCORRUPTED;
1496        }
1497        depth = path->p_depth;
1498        *phys = 0;
1499
1500        if (depth == 0 && path->p_ext == NULL)
1501                return 0;
1502
1503        /* usually extent in the path covers blocks smaller
1504         * then *logical, but it can be that extent is the
1505         * first one in the file */
1506
1507        ex = path[depth].p_ext;
1508        ee_len = ext4_ext_get_actual_len(ex);
1509        if (*logical < le32_to_cpu(ex->ee_block)) {
1510                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1511                        EXT4_ERROR_INODE(inode,
1512                                         "first_extent(path[%d].p_hdr) != ex",
1513                                         depth);
1514                        return -EFSCORRUPTED;
1515                }
1516                while (--depth >= 0) {
1517                        ix = path[depth].p_idx;
1518                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1519                                EXT4_ERROR_INODE(inode,
1520                                                 "ix != EXT_FIRST_INDEX *logical %d!",
1521                                                 *logical);
1522                                return -EFSCORRUPTED;
1523                        }
1524                }
1525                goto found_extent;
1526        }
1527
1528        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1529                EXT4_ERROR_INODE(inode,
1530                                 "logical %d < ee_block %d + ee_len %d!",
1531                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1532                return -EFSCORRUPTED;
1533        }
1534
1535        if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1536                /* next allocated block in this leaf */
1537                ex++;
1538                goto found_extent;
1539        }
1540
1541        /* go up and search for index to the right */
1542        while (--depth >= 0) {
1543                ix = path[depth].p_idx;
1544                if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1545                        goto got_index;
1546        }
1547
1548        /* we've gone up to the root and found no index to the right */
1549        return 0;
1550
1551got_index:
1552        /* we've found index to the right, let's
1553         * follow it and find the closest allocated
1554         * block to the right */
1555        ix++;
1556        block = ext4_idx_pblock(ix);
1557        while (++depth < path->p_depth) {
1558                /* subtract from p_depth to get proper eh_depth */
1559                bh = read_extent_tree_block(inode, block,
1560                                            path->p_depth - depth, 0);
1561                if (IS_ERR(bh))
1562                        return PTR_ERR(bh);
1563                eh = ext_block_hdr(bh);
1564                ix = EXT_FIRST_INDEX(eh);
1565                block = ext4_idx_pblock(ix);
1566                put_bh(bh);
1567        }
1568
1569        bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1570        if (IS_ERR(bh))
1571                return PTR_ERR(bh);
1572        eh = ext_block_hdr(bh);
1573        ex = EXT_FIRST_EXTENT(eh);
1574found_extent:
1575        *logical = le32_to_cpu(ex->ee_block);
1576        *phys = ext4_ext_pblock(ex);
1577        *ret_ex = ex;
1578        if (bh)
1579                put_bh(bh);
1580        return 0;
1581}
1582
1583/*
1584 * ext4_ext_next_allocated_block:
1585 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1586 * NOTE: it considers block number from index entry as
1587 * allocated block. Thus, index entries have to be consistent
1588 * with leaves.
1589 */
1590ext4_lblk_t
1591ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1592{
1593        int depth;
1594
1595        BUG_ON(path == NULL);
1596        depth = path->p_depth;
1597
1598        if (depth == 0 && path->p_ext == NULL)
1599                return EXT_MAX_BLOCKS;
1600
1601        while (depth >= 0) {
1602                struct ext4_ext_path *p = &path[depth];
1603
1604                if (depth == path->p_depth) {
1605                        /* leaf */
1606                        if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1607                                return le32_to_cpu(p->p_ext[1].ee_block);
1608                } else {
1609                        /* index */
1610                        if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1611                                return le32_to_cpu(p->p_idx[1].ei_block);
1612                }
1613                depth--;
1614        }
1615
1616        return EXT_MAX_BLOCKS;
1617}
1618
1619/*
1620 * ext4_ext_next_leaf_block:
1621 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1622 */
1623static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1624{
1625        int depth;
1626
1627        BUG_ON(path == NULL);
1628        depth = path->p_depth;
1629
1630        /* zero-tree has no leaf blocks at all */
1631        if (depth == 0)
1632                return EXT_MAX_BLOCKS;
1633
1634        /* go to index block */
1635        depth--;
1636
1637        while (depth >= 0) {
1638                if (path[depth].p_idx !=
1639                                EXT_LAST_INDEX(path[depth].p_hdr))
1640                        return (ext4_lblk_t)
1641                                le32_to_cpu(path[depth].p_idx[1].ei_block);
1642                depth--;
1643        }
1644
1645        return EXT_MAX_BLOCKS;
1646}
1647
1648/*
1649 * ext4_ext_correct_indexes:
1650 * if leaf gets modified and modified extent is first in the leaf,
1651 * then we have to correct all indexes above.
1652 * TODO: do we need to correct tree in all cases?
1653 */
1654static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1655                                struct ext4_ext_path *path)
1656{
1657        struct ext4_extent_header *eh;
1658        int depth = ext_depth(inode);
1659        struct ext4_extent *ex;
1660        __le32 border;
1661        int k, err = 0;
1662
1663        eh = path[depth].p_hdr;
1664        ex = path[depth].p_ext;
1665
1666        if (unlikely(ex == NULL || eh == NULL)) {
1667                EXT4_ERROR_INODE(inode,
1668                                 "ex %p == NULL or eh %p == NULL", ex, eh);
1669                return -EFSCORRUPTED;
1670        }
1671
1672        if (depth == 0) {
1673                /* there is no tree at all */
1674                return 0;
1675        }
1676
1677        if (ex != EXT_FIRST_EXTENT(eh)) {
1678                /* we correct tree if first leaf got modified only */
1679                return 0;
1680        }
1681
1682        /*
1683         * TODO: we need correction if border is smaller than current one
1684         */
1685        k = depth - 1;
1686        border = path[depth].p_ext->ee_block;
1687        err = ext4_ext_get_access(handle, inode, path + k);
1688        if (err)
1689                return err;
1690        path[k].p_idx->ei_block = border;
1691        err = ext4_ext_dirty(handle, inode, path + k);
1692        if (err)
1693                return err;
1694
1695        while (k--) {
1696                /* change all left-side indexes */
1697                if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1698                        break;
1699                err = ext4_ext_get_access(handle, inode, path + k);
1700                if (err)
1701                        break;
1702                path[k].p_idx->ei_block = border;
1703                err = ext4_ext_dirty(handle, inode, path + k);
1704                if (err)
1705                        break;
1706        }
1707
1708        return err;
1709}
1710
1711static int ext4_can_extents_be_merged(struct inode *inode,
1712                                      struct ext4_extent *ex1,
1713                                      struct ext4_extent *ex2)
1714{
1715        unsigned short ext1_ee_len, ext2_ee_len;
1716
1717        if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1718                return 0;
1719
1720        ext1_ee_len = ext4_ext_get_actual_len(ex1);
1721        ext2_ee_len = ext4_ext_get_actual_len(ex2);
1722
1723        if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1724                        le32_to_cpu(ex2->ee_block))
1725                return 0;
1726
1727        if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1728                return 0;
1729
1730        if (ext4_ext_is_unwritten(ex1) &&
1731            ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1732                return 0;
1733#ifdef AGGRESSIVE_TEST
1734        if (ext1_ee_len >= 4)
1735                return 0;
1736#endif
1737
1738        if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1739                return 1;
1740        return 0;
1741}
1742
1743/*
1744 * This function tries to merge the "ex" extent to the next extent in the tree.
1745 * It always tries to merge towards right. If you want to merge towards
1746 * left, pass "ex - 1" as argument instead of "ex".
1747 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1748 * 1 if they got merged.
1749 */
1750static int ext4_ext_try_to_merge_right(struct inode *inode,
1751                                 struct ext4_ext_path *path,
1752                                 struct ext4_extent *ex)
1753{
1754        struct ext4_extent_header *eh;
1755        unsigned int depth, len;
1756        int merge_done = 0, unwritten;
1757
1758        depth = ext_depth(inode);
1759        BUG_ON(path[depth].p_hdr == NULL);
1760        eh = path[depth].p_hdr;
1761
1762        while (ex < EXT_LAST_EXTENT(eh)) {
1763                if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1764                        break;
1765                /* merge with next extent! */
1766                unwritten = ext4_ext_is_unwritten(ex);
1767                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1768                                + ext4_ext_get_actual_len(ex + 1));
1769                if (unwritten)
1770                        ext4_ext_mark_unwritten(ex);
1771
1772                if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1773                        len = (EXT_LAST_EXTENT(eh) - ex - 1)
1774                                * sizeof(struct ext4_extent);
1775                        memmove(ex + 1, ex + 2, len);
1776                }
1777                le16_add_cpu(&eh->eh_entries, -1);
1778                merge_done = 1;
1779                WARN_ON(eh->eh_entries == 0);
1780                if (!eh->eh_entries)
1781                        EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1782        }
1783
1784        return merge_done;
1785}
1786
1787/*
1788 * This function does a very simple check to see if we can collapse
1789 * an extent tree with a single extent tree leaf block into the inode.
1790 */
1791static void ext4_ext_try_to_merge_up(handle_t *handle,
1792                                     struct inode *inode,
1793                                     struct ext4_ext_path *path)
1794{
1795        size_t s;
1796        unsigned max_root = ext4_ext_space_root(inode, 0);
1797        ext4_fsblk_t blk;
1798
1799        if ((path[0].p_depth != 1) ||
1800            (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1801            (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1802                return;
1803
1804        /*
1805         * We need to modify the block allocation bitmap and the block
1806         * group descriptor to release the extent tree block.  If we
1807         * can't get the journal credits, give up.
1808         */
1809        if (ext4_journal_extend(handle, 2,
1810                        ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1811                return;
1812
1813        /*
1814         * Copy the extent data up to the inode
1815         */
1816        blk = ext4_idx_pblock(path[0].p_idx);
1817        s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1818                sizeof(struct ext4_extent_idx);
1819        s += sizeof(struct ext4_extent_header);
1820
1821        path[1].p_maxdepth = path[0].p_maxdepth;
1822        memcpy(path[0].p_hdr, path[1].p_hdr, s);
1823        path[0].p_depth = 0;
1824        path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1825                (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1826        path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1827
1828        brelse(path[1].p_bh);
1829        ext4_free_blocks(handle, inode, NULL, blk, 1,
1830                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1831}
1832
1833/*
1834 * This function tries to merge the @ex extent to neighbours in the tree, then
1835 * tries to collapse the extent tree into the inode.
1836 */
1837static void ext4_ext_try_to_merge(handle_t *handle,
1838                                  struct inode *inode,
1839                                  struct ext4_ext_path *path,
1840                                  struct ext4_extent *ex)
1841{
1842        struct ext4_extent_header *eh;
1843        unsigned int depth;
1844        int merge_done = 0;
1845
1846        depth = ext_depth(inode);
1847        BUG_ON(path[depth].p_hdr == NULL);
1848        eh = path[depth].p_hdr;
1849
1850        if (ex > EXT_FIRST_EXTENT(eh))
1851                merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1852
1853        if (!merge_done)
1854                (void) ext4_ext_try_to_merge_right(inode, path, ex);
1855
1856        ext4_ext_try_to_merge_up(handle, inode, path);
1857}
1858
1859/*
1860 * check if a portion of the "newext" extent overlaps with an
1861 * existing extent.
1862 *
1863 * If there is an overlap discovered, it updates the length of the newext
1864 * such that there will be no overlap, and then returns 1.
1865 * If there is no overlap found, it returns 0.
1866 */
1867static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1868                                           struct inode *inode,
1869                                           struct ext4_extent *newext,
1870                                           struct ext4_ext_path *path)
1871{
1872        ext4_lblk_t b1, b2;
1873        unsigned int depth, len1;
1874        unsigned int ret = 0;
1875
1876        b1 = le32_to_cpu(newext->ee_block);
1877        len1 = ext4_ext_get_actual_len(newext);
1878        depth = ext_depth(inode);
1879        if (!path[depth].p_ext)
1880                goto out;
1881        b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1882
1883        /*
1884         * get the next allocated block if the extent in the path
1885         * is before the requested block(s)
1886         */
1887        if (b2 < b1) {
1888                b2 = ext4_ext_next_allocated_block(path);
1889                if (b2 == EXT_MAX_BLOCKS)
1890                        goto out;
1891                b2 = EXT4_LBLK_CMASK(sbi, b2);
1892        }
1893
1894        /* check for wrap through zero on extent logical start block*/
1895        if (b1 + len1 < b1) {
1896                len1 = EXT_MAX_BLOCKS - b1;
1897                newext->ee_len = cpu_to_le16(len1);
1898                ret = 1;
1899        }
1900
1901        /* check for overlap */
1902        if (b1 + len1 > b2) {
1903                newext->ee_len = cpu_to_le16(b2 - b1);
1904                ret = 1;
1905        }
1906out:
1907        return ret;
1908}
1909
1910/*
1911 * ext4_ext_insert_extent:
1912 * tries to merge requested extent into the existing extent or
1913 * inserts requested extent as new one into the tree,
1914 * creating new leaf in the no-space case.
1915 */
1916int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1917                                struct ext4_ext_path **ppath,
1918                                struct ext4_extent *newext, int gb_flags)
1919{
1920        struct ext4_ext_path *path = *ppath;
1921        struct ext4_extent_header *eh;
1922        struct ext4_extent *ex, *fex;
1923        struct ext4_extent *nearex; /* nearest extent */
1924        struct ext4_ext_path *npath = NULL;
1925        int depth, len, err;
1926        ext4_lblk_t next;
1927        int mb_flags = 0, unwritten;
1928
1929        if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1930                mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1931        if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1932                EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1933                return -EFSCORRUPTED;
1934        }
1935        depth = ext_depth(inode);
1936        ex = path[depth].p_ext;
1937        eh = path[depth].p_hdr;
1938        if (unlikely(path[depth].p_hdr == NULL)) {
1939                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1940                return -EFSCORRUPTED;
1941        }
1942
1943        /* try to insert block into found extent and return */
1944        if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1945
1946                /*
1947                 * Try to see whether we should rather test the extent on
1948                 * right from ex, or from the left of ex. This is because
1949                 * ext4_find_extent() can return either extent on the
1950                 * left, or on the right from the searched position. This
1951                 * will make merging more effective.
1952                 */
1953                if (ex < EXT_LAST_EXTENT(eh) &&
1954                    (le32_to_cpu(ex->ee_block) +
1955                    ext4_ext_get_actual_len(ex) <
1956                    le32_to_cpu(newext->ee_block))) {
1957                        ex += 1;
1958                        goto prepend;
1959                } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1960                           (le32_to_cpu(newext->ee_block) +
1961                           ext4_ext_get_actual_len(newext) <
1962                           le32_to_cpu(ex->ee_block)))
1963                        ex -= 1;
1964
1965                /* Try to append newex to the ex */
1966                if (ext4_can_extents_be_merged(inode, ex, newext)) {
1967                        ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
1968                                  "(from %llu)\n",
1969                                  ext4_ext_is_unwritten(newext),
1970                                  ext4_ext_get_actual_len(newext),
1971                                  le32_to_cpu(ex->ee_block),
1972                                  ext4_ext_is_unwritten(ex),
1973                                  ext4_ext_get_actual_len(ex),
1974                                  ext4_ext_pblock(ex));
1975                        err = ext4_ext_get_access(handle, inode,
1976                                                  path + depth);
1977                        if (err)
1978                                return err;
1979                        unwritten = ext4_ext_is_unwritten(ex);
1980                        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1981                                        + ext4_ext_get_actual_len(newext));
1982                        if (unwritten)
1983                                ext4_ext_mark_unwritten(ex);
1984                        eh = path[depth].p_hdr;
1985                        nearex = ex;
1986                        goto merge;
1987                }
1988
1989prepend:
1990                /* Try to prepend newex to the ex */
1991                if (ext4_can_extents_be_merged(inode, newext, ex)) {
1992                        ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
1993                                  "(from %llu)\n",
1994                                  le32_to_cpu(newext->ee_block),
1995                                  ext4_ext_is_unwritten(newext),
1996                                  ext4_ext_get_actual_len(newext),
1997                                  le32_to_cpu(ex->ee_block),
1998                                  ext4_ext_is_unwritten(ex),
1999                                  ext4_ext_get_actual_len(ex),
2000                                  ext4_ext_pblock(ex));
2001                        err = ext4_ext_get_access(handle, inode,
2002                                                  path + depth);
2003                        if (err)
2004                                return err;
2005
2006                        unwritten = ext4_ext_is_unwritten(ex);
2007                        ex->ee_block = newext->ee_block;
2008                        ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2009                        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2010                                        + ext4_ext_get_actual_len(newext));
2011                        if (unwritten)
2012                                ext4_ext_mark_unwritten(ex);
2013                        eh = path[depth].p_hdr;
2014                        nearex = ex;
2015                        goto merge;
2016                }
2017        }
2018
2019        depth = ext_depth(inode);
2020        eh = path[depth].p_hdr;
2021        if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2022                goto has_space;
2023
2024        /* probably next leaf has space for us? */
2025        fex = EXT_LAST_EXTENT(eh);
2026        next = EXT_MAX_BLOCKS;
2027        if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2028                next = ext4_ext_next_leaf_block(path);
2029        if (next != EXT_MAX_BLOCKS) {
2030                ext_debug(inode, "next leaf block - %u\n", next);
2031                BUG_ON(npath != NULL);
2032                npath = ext4_find_extent(inode, next, NULL, gb_flags);
2033                if (IS_ERR(npath))
2034                        return PTR_ERR(npath);
2035                BUG_ON(npath->p_depth != path->p_depth);
2036                eh = npath[depth].p_hdr;
2037                if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2038                        ext_debug(inode, "next leaf isn't full(%d)\n",
2039                                  le16_to_cpu(eh->eh_entries));
2040                        path = npath;
2041                        goto has_space;
2042                }
2043                ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2044                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2045        }
2046
2047        /*
2048         * There is no free space in the found leaf.
2049         * We're gonna add a new leaf in the tree.
2050         */
2051        if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2052                mb_flags |= EXT4_MB_USE_RESERVED;
2053        err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2054                                       ppath, newext);
2055        if (err)
2056                goto cleanup;
2057        depth = ext_depth(inode);
2058        eh = path[depth].p_hdr;
2059
2060has_space:
2061        nearex = path[depth].p_ext;
2062
2063        err = ext4_ext_get_access(handle, inode, path + depth);
2064        if (err)
2065                goto cleanup;
2066
2067        if (!nearex) {
2068                /* there is no extent in this leaf, create first one */
2069                ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2070                                le32_to_cpu(newext->ee_block),
2071                                ext4_ext_pblock(newext),
2072                                ext4_ext_is_unwritten(newext),
2073                                ext4_ext_get_actual_len(newext));
2074                nearex = EXT_FIRST_EXTENT(eh);
2075        } else {
2076                if (le32_to_cpu(newext->ee_block)
2077                           > le32_to_cpu(nearex->ee_block)) {
2078                        /* Insert after */
2079                        ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2080                                        "nearest %p\n",
2081                                        le32_to_cpu(newext->ee_block),
2082                                        ext4_ext_pblock(newext),
2083                                        ext4_ext_is_unwritten(newext),
2084                                        ext4_ext_get_actual_len(newext),
2085                                        nearex);
2086                        nearex++;
2087                } else {
2088                        /* Insert before */
2089                        BUG_ON(newext->ee_block == nearex->ee_block);
2090                        ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2091                                        "nearest %p\n",
2092                                        le32_to_cpu(newext->ee_block),
2093                                        ext4_ext_pblock(newext),
2094                                        ext4_ext_is_unwritten(newext),
2095                                        ext4_ext_get_actual_len(newext),
2096                                        nearex);
2097                }
2098                len = EXT_LAST_EXTENT(eh) - nearex + 1;
2099                if (len > 0) {
2100                        ext_debug(inode, "insert %u:%llu:[%d]%d: "
2101                                        "move %d extents from 0x%p to 0x%p\n",
2102                                        le32_to_cpu(newext->ee_block),
2103                                        ext4_ext_pblock(newext),
2104                                        ext4_ext_is_unwritten(newext),
2105                                        ext4_ext_get_actual_len(newext),
2106                                        len, nearex, nearex + 1);
2107                        memmove(nearex + 1, nearex,
2108                                len * sizeof(struct ext4_extent));
2109                }
2110        }
2111
2112        le16_add_cpu(&eh->eh_entries, 1);
2113        path[depth].p_ext = nearex;
2114        nearex->ee_block = newext->ee_block;
2115        ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2116        nearex->ee_len = newext->ee_len;
2117
2118merge:
2119        /* try to merge extents */
2120        if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2121                ext4_ext_try_to_merge(handle, inode, path, nearex);
2122
2123
2124        /* time to correct all indexes above */
2125        err = ext4_ext_correct_indexes(handle, inode, path);
2126        if (err)
2127                goto cleanup;
2128
2129        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2130
2131cleanup:
2132        ext4_ext_drop_refs(npath);
2133        kfree(npath);
2134        return err;
2135}
2136
2137static int ext4_fill_es_cache_info(struct inode *inode,
2138                                   ext4_lblk_t block, ext4_lblk_t num,
2139                                   struct fiemap_extent_info *fieinfo)
2140{
2141        ext4_lblk_t next, end = block + num - 1;
2142        struct extent_status es;
2143        unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2144        unsigned int flags;
2145        int err;
2146
2147        while (block <= end) {
2148                next = 0;
2149                flags = 0;
2150                if (!ext4_es_lookup_extent(inode, block, &next, &es))
2151                        break;
2152                if (ext4_es_is_unwritten(&es))
2153                        flags |= FIEMAP_EXTENT_UNWRITTEN;
2154                if (ext4_es_is_delayed(&es))
2155                        flags |= (FIEMAP_EXTENT_DELALLOC |
2156                                  FIEMAP_EXTENT_UNKNOWN);
2157                if (ext4_es_is_hole(&es))
2158                        flags |= EXT4_FIEMAP_EXTENT_HOLE;
2159                if (next == 0)
2160                        flags |= FIEMAP_EXTENT_LAST;
2161                if (flags & (FIEMAP_EXTENT_DELALLOC|
2162                             EXT4_FIEMAP_EXTENT_HOLE))
2163                        es.es_pblk = 0;
2164                else
2165                        es.es_pblk = ext4_es_pblock(&es);
2166                err = fiemap_fill_next_extent(fieinfo,
2167                                (__u64)es.es_lblk << blksize_bits,
2168                                (__u64)es.es_pblk << blksize_bits,
2169                                (__u64)es.es_len << blksize_bits,
2170                                flags);
2171                if (next == 0)
2172                        break;
2173                block = next;
2174                if (err < 0)
2175                        return err;
2176                if (err == 1)
2177                        return 0;
2178        }
2179        return 0;
2180}
2181
2182
2183/*
2184 * ext4_ext_determine_hole - determine hole around given block
2185 * @inode:      inode we lookup in
2186 * @path:       path in extent tree to @lblk
2187 * @lblk:       pointer to logical block around which we want to determine hole
2188 *
2189 * Determine hole length (and start if easily possible) around given logical
2190 * block. We don't try too hard to find the beginning of the hole but @path
2191 * actually points to extent before @lblk, we provide it.
2192 *
2193 * The function returns the length of a hole starting at @lblk. We update @lblk
2194 * to the beginning of the hole if we managed to find it.
2195 */
2196static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2197                                           struct ext4_ext_path *path,
2198                                           ext4_lblk_t *lblk)
2199{
2200        int depth = ext_depth(inode);
2201        struct ext4_extent *ex;
2202        ext4_lblk_t len;
2203
2204        ex = path[depth].p_ext;
2205        if (ex == NULL) {
2206                /* there is no extent yet, so gap is [0;-] */
2207                *lblk = 0;
2208                len = EXT_MAX_BLOCKS;
2209        } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2210                len = le32_to_cpu(ex->ee_block) - *lblk;
2211        } else if (*lblk >= le32_to_cpu(ex->ee_block)
2212                        + ext4_ext_get_actual_len(ex)) {
2213                ext4_lblk_t next;
2214
2215                *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2216                next = ext4_ext_next_allocated_block(path);
2217                BUG_ON(next == *lblk);
2218                len = next - *lblk;
2219        } else {
2220                BUG();
2221        }
2222        return len;
2223}
2224
2225/*
2226 * ext4_ext_put_gap_in_cache:
2227 * calculate boundaries of the gap that the requested block fits into
2228 * and cache this gap
2229 */
2230static void
2231ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2232                          ext4_lblk_t hole_len)
2233{
2234        struct extent_status es;
2235
2236        ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2237                                  hole_start + hole_len - 1, &es);
2238        if (es.es_len) {
2239                /* There's delayed extent containing lblock? */
2240                if (es.es_lblk <= hole_start)
2241                        return;
2242                hole_len = min(es.es_lblk - hole_start, hole_len);
2243        }
2244        ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2245        ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2246                              EXTENT_STATUS_HOLE);
2247}
2248
2249/*
2250 * ext4_ext_rm_idx:
2251 * removes index from the index block.
2252 */
2253static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2254                        struct ext4_ext_path *path, int depth)
2255{
2256        int err;
2257        ext4_fsblk_t leaf;
2258
2259        /* free index block */
2260        depth--;
2261        path = path + depth;
2262        leaf = ext4_idx_pblock(path->p_idx);
2263        if (unlikely(path->p_hdr->eh_entries == 0)) {
2264                EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2265                return -EFSCORRUPTED;
2266        }
2267        err = ext4_ext_get_access(handle, inode, path);
2268        if (err)
2269                return err;
2270
2271        if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2272                int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2273                len *= sizeof(struct ext4_extent_idx);
2274                memmove(path->p_idx, path->p_idx + 1, len);
2275        }
2276
2277        le16_add_cpu(&path->p_hdr->eh_entries, -1);
2278        err = ext4_ext_dirty(handle, inode, path);
2279        if (err)
2280                return err;
2281        ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2282        trace_ext4_ext_rm_idx(inode, leaf);
2283
2284        ext4_free_blocks(handle, inode, NULL, leaf, 1,
2285                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2286
2287        while (--depth >= 0) {
2288                if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2289                        break;
2290                path--;
2291                err = ext4_ext_get_access(handle, inode, path);
2292                if (err)
2293                        break;
2294                path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2295                err = ext4_ext_dirty(handle, inode, path);
2296                if (err)
2297                        break;
2298        }
2299        return err;
2300}
2301
2302/*
2303 * ext4_ext_calc_credits_for_single_extent:
2304 * This routine returns max. credits that needed to insert an extent
2305 * to the extent tree.
2306 * When pass the actual path, the caller should calculate credits
2307 * under i_data_sem.
2308 */
2309int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2310                                                struct ext4_ext_path *path)
2311{
2312        if (path) {
2313                int depth = ext_depth(inode);
2314                int ret = 0;
2315
2316                /* probably there is space in leaf? */
2317                if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2318                                < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2319
2320                        /*
2321                         *  There are some space in the leaf tree, no
2322                         *  need to account for leaf block credit
2323                         *
2324                         *  bitmaps and block group descriptor blocks
2325                         *  and other metadata blocks still need to be
2326                         *  accounted.
2327                         */
2328                        /* 1 bitmap, 1 block group descriptor */
2329                        ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2330                        return ret;
2331                }
2332        }
2333
2334        return ext4_chunk_trans_blocks(inode, nrblocks);
2335}
2336
2337/*
2338 * How many index/leaf blocks need to change/allocate to add @extents extents?
2339 *
2340 * If we add a single extent, then in the worse case, each tree level
2341 * index/leaf need to be changed in case of the tree split.
2342 *
2343 * If more extents are inserted, they could cause the whole tree split more
2344 * than once, but this is really rare.
2345 */
2346int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2347{
2348        int index;
2349        int depth;
2350
2351        /* If we are converting the inline data, only one is needed here. */
2352        if (ext4_has_inline_data(inode))
2353                return 1;
2354
2355        depth = ext_depth(inode);
2356
2357        if (extents <= 1)
2358                index = depth * 2;
2359        else
2360                index = depth * 3;
2361
2362        return index;
2363}
2364
2365static inline int get_default_free_blocks_flags(struct inode *inode)
2366{
2367        if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2368            ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2369                return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2370        else if (ext4_should_journal_data(inode))
2371                return EXT4_FREE_BLOCKS_FORGET;
2372        return 0;
2373}
2374
2375/*
2376 * ext4_rereserve_cluster - increment the reserved cluster count when
2377 *                          freeing a cluster with a pending reservation
2378 *
2379 * @inode - file containing the cluster
2380 * @lblk - logical block in cluster to be reserved
2381 *
2382 * Increments the reserved cluster count and adjusts quota in a bigalloc
2383 * file system when freeing a partial cluster containing at least one
2384 * delayed and unwritten block.  A partial cluster meeting that
2385 * requirement will have a pending reservation.  If so, the
2386 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2387 * defer reserved and allocated space accounting to a subsequent call
2388 * to this function.
2389 */
2390static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2391{
2392        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2393        struct ext4_inode_info *ei = EXT4_I(inode);
2394
2395        dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2396
2397        spin_lock(&ei->i_block_reservation_lock);
2398        ei->i_reserved_data_blocks++;
2399        percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2400        spin_unlock(&ei->i_block_reservation_lock);
2401
2402        percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2403        ext4_remove_pending(inode, lblk);
2404}
2405
2406static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2407                              struct ext4_extent *ex,
2408                              struct partial_cluster *partial,
2409                              ext4_lblk_t from, ext4_lblk_t to)
2410{
2411        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2412        unsigned short ee_len = ext4_ext_get_actual_len(ex);
2413        ext4_fsblk_t last_pblk, pblk;
2414        ext4_lblk_t num;
2415        int flags;
2416
2417        /* only extent tail removal is allowed */
2418        if (from < le32_to_cpu(ex->ee_block) ||
2419            to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2420                ext4_error(sbi->s_sb,
2421                           "strange request: removal(2) %u-%u from %u:%u",
2422                           from, to, le32_to_cpu(ex->ee_block), ee_len);
2423                return 0;
2424        }
2425
2426#ifdef EXTENTS_STATS
2427        spin_lock(&sbi->s_ext_stats_lock);
2428        sbi->s_ext_blocks += ee_len;
2429        sbi->s_ext_extents++;
2430        if (ee_len < sbi->s_ext_min)
2431                sbi->s_ext_min = ee_len;
2432        if (ee_len > sbi->s_ext_max)
2433                sbi->s_ext_max = ee_len;
2434        if (ext_depth(inode) > sbi->s_depth_max)
2435                sbi->s_depth_max = ext_depth(inode);
2436        spin_unlock(&sbi->s_ext_stats_lock);
2437#endif
2438
2439        trace_ext4_remove_blocks(inode, ex, from, to, partial);
2440
2441        /*
2442         * if we have a partial cluster, and it's different from the
2443         * cluster of the last block in the extent, we free it
2444         */
2445        last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2446
2447        if (partial->state != initial &&
2448            partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2449                if (partial->state == tofree) {
2450                        flags = get_default_free_blocks_flags(inode);
2451                        if (ext4_is_pending(inode, partial->lblk))
2452                                flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2453                        ext4_free_blocks(handle, inode, NULL,
2454                                         EXT4_C2B(sbi, partial->pclu),
2455                                         sbi->s_cluster_ratio, flags);
2456                        if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2457                                ext4_rereserve_cluster(inode, partial->lblk);
2458                }
2459                partial->state = initial;
2460        }
2461
2462        num = le32_to_cpu(ex->ee_block) + ee_len - from;
2463        pblk = ext4_ext_pblock(ex) + ee_len - num;
2464
2465        /*
2466         * We free the partial cluster at the end of the extent (if any),
2467         * unless the cluster is used by another extent (partial_cluster
2468         * state is nofree).  If a partial cluster exists here, it must be
2469         * shared with the last block in the extent.
2470         */
2471        flags = get_default_free_blocks_flags(inode);
2472
2473        /* partial, left end cluster aligned, right end unaligned */
2474        if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2475            (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2476            (partial->state != nofree)) {
2477                if (ext4_is_pending(inode, to))
2478                        flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2479                ext4_free_blocks(handle, inode, NULL,
2480                                 EXT4_PBLK_CMASK(sbi, last_pblk),
2481                                 sbi->s_cluster_ratio, flags);
2482                if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2483                        ext4_rereserve_cluster(inode, to);
2484                partial->state = initial;
2485                flags = get_default_free_blocks_flags(inode);
2486        }
2487
2488        flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2489
2490        /*
2491         * For bigalloc file systems, we never free a partial cluster
2492         * at the beginning of the extent.  Instead, we check to see if we
2493         * need to free it on a subsequent call to ext4_remove_blocks,
2494         * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2495         */
2496        flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2497        ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2498
2499        /* reset the partial cluster if we've freed past it */
2500        if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2501                partial->state = initial;
2502
2503        /*
2504         * If we've freed the entire extent but the beginning is not left
2505         * cluster aligned and is not marked as ineligible for freeing we
2506         * record the partial cluster at the beginning of the extent.  It
2507         * wasn't freed by the preceding ext4_free_blocks() call, and we
2508         * need to look farther to the left to determine if it's to be freed
2509         * (not shared with another extent). Else, reset the partial
2510         * cluster - we're either  done freeing or the beginning of the
2511         * extent is left cluster aligned.
2512         */
2513        if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2514                if (partial->state == initial) {
2515                        partial->pclu = EXT4_B2C(sbi, pblk);
2516                        partial->lblk = from;
2517                        partial->state = tofree;
2518                }
2519        } else {
2520                partial->state = initial;
2521        }
2522
2523        return 0;
2524}
2525
2526/*
2527 * ext4_ext_rm_leaf() Removes the extents associated with the
2528 * blocks appearing between "start" and "end".  Both "start"
2529 * and "end" must appear in the same extent or EIO is returned.
2530 *
2531 * @handle: The journal handle
2532 * @inode:  The files inode
2533 * @path:   The path to the leaf
2534 * @partial_cluster: The cluster which we'll have to free if all extents
2535 *                   has been released from it.  However, if this value is
2536 *                   negative, it's a cluster just to the right of the
2537 *                   punched region and it must not be freed.
2538 * @start:  The first block to remove
2539 * @end:   The last block to remove
2540 */
2541static int
2542ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2543                 struct ext4_ext_path *path,
2544                 struct partial_cluster *partial,
2545                 ext4_lblk_t start, ext4_lblk_t end)
2546{
2547        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2548        int err = 0, correct_index = 0;
2549        int depth = ext_depth(inode), credits, revoke_credits;
2550        struct ext4_extent_header *eh;
2551        ext4_lblk_t a, b;
2552        unsigned num;
2553        ext4_lblk_t ex_ee_block;
2554        unsigned short ex_ee_len;
2555        unsigned unwritten = 0;
2556        struct ext4_extent *ex;
2557        ext4_fsblk_t pblk;
2558
2559        /* the header must be checked already in ext4_ext_remove_space() */
2560        ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2561        if (!path[depth].p_hdr)
2562                path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2563        eh = path[depth].p_hdr;
2564        if (unlikely(path[depth].p_hdr == NULL)) {
2565                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2566                return -EFSCORRUPTED;
2567        }
2568        /* find where to start removing */
2569        ex = path[depth].p_ext;
2570        if (!ex)
2571                ex = EXT_LAST_EXTENT(eh);
2572
2573        ex_ee_block = le32_to_cpu(ex->ee_block);
2574        ex_ee_len = ext4_ext_get_actual_len(ex);
2575
2576        trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2577
2578        while (ex >= EXT_FIRST_EXTENT(eh) &&
2579                        ex_ee_block + ex_ee_len > start) {
2580
2581                if (ext4_ext_is_unwritten(ex))
2582                        unwritten = 1;
2583                else
2584                        unwritten = 0;
2585
2586                ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2587                          unwritten, ex_ee_len);
2588                path[depth].p_ext = ex;
2589
2590                a = ex_ee_block > start ? ex_ee_block : start;
2591                b = ex_ee_block+ex_ee_len - 1 < end ?
2592                        ex_ee_block+ex_ee_len - 1 : end;
2593
2594                ext_debug(inode, "  border %u:%u\n", a, b);
2595
2596                /* If this extent is beyond the end of the hole, skip it */
2597                if (end < ex_ee_block) {
2598                        /*
2599                         * We're going to skip this extent and move to another,
2600                         * so note that its first cluster is in use to avoid
2601                         * freeing it when removing blocks.  Eventually, the
2602                         * right edge of the truncated/punched region will
2603                         * be just to the left.
2604                         */
2605                        if (sbi->s_cluster_ratio > 1) {
2606                                pblk = ext4_ext_pblock(ex);
2607                                partial->pclu = EXT4_B2C(sbi, pblk);
2608                                partial->state = nofree;
2609                        }
2610                        ex--;
2611                        ex_ee_block = le32_to_cpu(ex->ee_block);
2612                        ex_ee_len = ext4_ext_get_actual_len(ex);
2613                        continue;
2614                } else if (b != ex_ee_block + ex_ee_len - 1) {
2615                        EXT4_ERROR_INODE(inode,
2616                                         "can not handle truncate %u:%u "
2617                                         "on extent %u:%u",
2618                                         start, end, ex_ee_block,
2619                                         ex_ee_block + ex_ee_len - 1);
2620                        err = -EFSCORRUPTED;
2621                        goto out;
2622                } else if (a != ex_ee_block) {
2623                        /* remove tail of the extent */
2624                        num = a - ex_ee_block;
2625                } else {
2626                        /* remove whole extent: excellent! */
2627                        num = 0;
2628                }
2629                /*
2630                 * 3 for leaf, sb, and inode plus 2 (bmap and group
2631                 * descriptor) for each block group; assume two block
2632                 * groups plus ex_ee_len/blocks_per_block_group for
2633                 * the worst case
2634                 */
2635                credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2636                if (ex == EXT_FIRST_EXTENT(eh)) {
2637                        correct_index = 1;
2638                        credits += (ext_depth(inode)) + 1;
2639                }
2640                credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2641                /*
2642                 * We may end up freeing some index blocks and data from the
2643                 * punched range. Note that partial clusters are accounted for
2644                 * by ext4_free_data_revoke_credits().
2645                 */
2646                revoke_credits =
2647                        ext4_free_metadata_revoke_credits(inode->i_sb,
2648                                                          ext_depth(inode)) +
2649                        ext4_free_data_revoke_credits(inode, b - a + 1);
2650
2651                err = ext4_datasem_ensure_credits(handle, inode, credits,
2652                                                  credits, revoke_credits);
2653                if (err) {
2654                        if (err > 0)
2655                                err = -EAGAIN;
2656                        goto out;
2657                }
2658
2659                err = ext4_ext_get_access(handle, inode, path + depth);
2660                if (err)
2661                        goto out;
2662
2663                err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2664                if (err)
2665                        goto out;
2666
2667                if (num == 0)
2668                        /* this extent is removed; mark slot entirely unused */
2669                        ext4_ext_store_pblock(ex, 0);
2670
2671                ex->ee_len = cpu_to_le16(num);
2672                /*
2673                 * Do not mark unwritten if all the blocks in the
2674                 * extent have been removed.
2675                 */
2676                if (unwritten && num)
2677                        ext4_ext_mark_unwritten(ex);
2678                /*
2679                 * If the extent was completely released,
2680                 * we need to remove it from the leaf
2681                 */
2682                if (num == 0) {
2683                        if (end != EXT_MAX_BLOCKS - 1) {
2684                                /*
2685                                 * For hole punching, we need to scoot all the
2686                                 * extents up when an extent is removed so that
2687                                 * we dont have blank extents in the middle
2688                                 */
2689                                memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2690                                        sizeof(struct ext4_extent));
2691
2692                                /* Now get rid of the one at the end */
2693                                memset(EXT_LAST_EXTENT(eh), 0,
2694                                        sizeof(struct ext4_extent));
2695                        }
2696                        le16_add_cpu(&eh->eh_entries, -1);
2697                }
2698
2699                err = ext4_ext_dirty(handle, inode, path + depth);
2700                if (err)
2701                        goto out;
2702
2703                ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2704                                ext4_ext_pblock(ex));
2705                ex--;
2706                ex_ee_block = le32_to_cpu(ex->ee_block);
2707                ex_ee_len = ext4_ext_get_actual_len(ex);
2708        }
2709
2710        if (correct_index && eh->eh_entries)
2711                err = ext4_ext_correct_indexes(handle, inode, path);
2712
2713        /*
2714         * If there's a partial cluster and at least one extent remains in
2715         * the leaf, free the partial cluster if it isn't shared with the
2716         * current extent.  If it is shared with the current extent
2717         * we reset the partial cluster because we've reached the start of the
2718         * truncated/punched region and we're done removing blocks.
2719         */
2720        if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2721                pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2722                if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2723                        int flags = get_default_free_blocks_flags(inode);
2724
2725                        if (ext4_is_pending(inode, partial->lblk))
2726                                flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2727                        ext4_free_blocks(handle, inode, NULL,
2728                                         EXT4_C2B(sbi, partial->pclu),
2729                                         sbi->s_cluster_ratio, flags);
2730                        if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2731                                ext4_rereserve_cluster(inode, partial->lblk);
2732                }
2733                partial->state = initial;
2734        }
2735
2736        /* if this leaf is free, then we should
2737         * remove it from index block above */
2738        if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2739                err = ext4_ext_rm_idx(handle, inode, path, depth);
2740
2741out:
2742        return err;
2743}
2744
2745/*
2746 * ext4_ext_more_to_rm:
2747 * returns 1 if current index has to be freed (even partial)
2748 */
2749static int
2750ext4_ext_more_to_rm(struct ext4_ext_path *path)
2751{
2752        BUG_ON(path->p_idx == NULL);
2753
2754        if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2755                return 0;
2756
2757        /*
2758         * if truncate on deeper level happened, it wasn't partial,
2759         * so we have to consider current index for truncation
2760         */
2761        if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2762                return 0;
2763        return 1;
2764}
2765
2766int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2767                          ext4_lblk_t end)
2768{
2769        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2770        int depth = ext_depth(inode);
2771        struct ext4_ext_path *path = NULL;
2772        struct partial_cluster partial;
2773        handle_t *handle;
2774        int i = 0, err = 0;
2775
2776        partial.pclu = 0;
2777        partial.lblk = 0;
2778        partial.state = initial;
2779
2780        ext_debug(inode, "truncate since %u to %u\n", start, end);
2781
2782        /* probably first extent we're gonna free will be last in block */
2783        handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2784                        depth + 1,
2785                        ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2786        if (IS_ERR(handle))
2787                return PTR_ERR(handle);
2788
2789again:
2790        trace_ext4_ext_remove_space(inode, start, end, depth);
2791
2792        /*
2793         * Check if we are removing extents inside the extent tree. If that
2794         * is the case, we are going to punch a hole inside the extent tree
2795         * so we have to check whether we need to split the extent covering
2796         * the last block to remove so we can easily remove the part of it
2797         * in ext4_ext_rm_leaf().
2798         */
2799        if (end < EXT_MAX_BLOCKS - 1) {
2800                struct ext4_extent *ex;
2801                ext4_lblk_t ee_block, ex_end, lblk;
2802                ext4_fsblk_t pblk;
2803
2804                /* find extent for or closest extent to this block */
2805                path = ext4_find_extent(inode, end, NULL,
2806                                        EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2807                if (IS_ERR(path)) {
2808                        ext4_journal_stop(handle);
2809                        return PTR_ERR(path);
2810                }
2811                depth = ext_depth(inode);
2812                /* Leaf not may not exist only if inode has no blocks at all */
2813                ex = path[depth].p_ext;
2814                if (!ex) {
2815                        if (depth) {
2816                                EXT4_ERROR_INODE(inode,
2817                                                 "path[%d].p_hdr == NULL",
2818                                                 depth);
2819                                err = -EFSCORRUPTED;
2820                        }
2821                        goto out;
2822                }
2823
2824                ee_block = le32_to_cpu(ex->ee_block);
2825                ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2826
2827                /*
2828                 * See if the last block is inside the extent, if so split
2829                 * the extent at 'end' block so we can easily remove the
2830                 * tail of the first part of the split extent in
2831                 * ext4_ext_rm_leaf().
2832                 */
2833                if (end >= ee_block && end < ex_end) {
2834
2835                        /*
2836                         * If we're going to split the extent, note that
2837                         * the cluster containing the block after 'end' is
2838                         * in use to avoid freeing it when removing blocks.
2839                         */
2840                        if (sbi->s_cluster_ratio > 1) {
2841                                pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2842                                partial.pclu = EXT4_B2C(sbi, pblk);
2843                                partial.state = nofree;
2844                        }
2845
2846                        /*
2847                         * Split the extent in two so that 'end' is the last
2848                         * block in the first new extent. Also we should not
2849                         * fail removing space due to ENOSPC so try to use
2850                         * reserved block if that happens.
2851                         */
2852                        err = ext4_force_split_extent_at(handle, inode, &path,
2853                                                         end + 1, 1);
2854                        if (err < 0)
2855                                goto out;
2856
2857                } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2858                           partial.state == initial) {
2859                        /*
2860                         * If we're punching, there's an extent to the right.
2861                         * If the partial cluster hasn't been set, set it to
2862                         * that extent's first cluster and its state to nofree
2863                         * so it won't be freed should it contain blocks to be
2864                         * removed. If it's already set (tofree/nofree), we're
2865                         * retrying and keep the original partial cluster info
2866                         * so a cluster marked tofree as a result of earlier
2867                         * extent removal is not lost.
2868                         */
2869                        lblk = ex_end + 1;
2870                        err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2871                                                    &ex);
2872                        if (err)
2873                                goto out;
2874                        if (pblk) {
2875                                partial.pclu = EXT4_B2C(sbi, pblk);
2876                                partial.state = nofree;
2877                        }
2878                }
2879        }
2880        /*
2881         * We start scanning from right side, freeing all the blocks
2882         * after i_size and walking into the tree depth-wise.
2883         */
2884        depth = ext_depth(inode);
2885        if (path) {
2886                int k = i = depth;
2887                while (--k > 0)
2888                        path[k].p_block =
2889                                le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2890        } else {
2891                path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2892                               GFP_NOFS | __GFP_NOFAIL);
2893                if (path == NULL) {
2894                        ext4_journal_stop(handle);
2895                        return -ENOMEM;
2896                }
2897                path[0].p_maxdepth = path[0].p_depth = depth;
2898                path[0].p_hdr = ext_inode_hdr(inode);
2899                i = 0;
2900
2901                if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2902                        err = -EFSCORRUPTED;
2903                        goto out;
2904                }
2905        }
2906        err = 0;
2907
2908        while (i >= 0 && err == 0) {
2909                if (i == depth) {
2910                        /* this is leaf block */
2911                        err = ext4_ext_rm_leaf(handle, inode, path,
2912                                               &partial, start, end);
2913                        /* root level has p_bh == NULL, brelse() eats this */
2914                        brelse(path[i].p_bh);
2915                        path[i].p_bh = NULL;
2916                        i--;
2917                        continue;
2918                }
2919
2920                /* this is index block */
2921                if (!path[i].p_hdr) {
2922                        ext_debug(inode, "initialize header\n");
2923                        path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2924                }
2925
2926                if (!path[i].p_idx) {
2927                        /* this level hasn't been touched yet */
2928                        path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2929                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2930                        ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2931                                  path[i].p_hdr,
2932                                  le16_to_cpu(path[i].p_hdr->eh_entries));
2933                } else {
2934                        /* we were already here, see at next index */
2935                        path[i].p_idx--;
2936                }
2937
2938                ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2939                                i, EXT_FIRST_INDEX(path[i].p_hdr),
2940                                path[i].p_idx);
2941                if (ext4_ext_more_to_rm(path + i)) {
2942                        struct buffer_head *bh;
2943                        /* go to the next level */
2944                        ext_debug(inode, "move to level %d (block %llu)\n",
2945                                  i + 1, ext4_idx_pblock(path[i].p_idx));
2946                        memset(path + i + 1, 0, sizeof(*path));
2947                        bh = read_extent_tree_block(inode,
2948                                ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2949                                EXT4_EX_NOCACHE);
2950                        if (IS_ERR(bh)) {
2951                                /* should we reset i_size? */
2952                                err = PTR_ERR(bh);
2953                                break;
2954                        }
2955                        /* Yield here to deal with large extent trees.
2956                         * Should be a no-op if we did IO above. */
2957                        cond_resched();
2958                        if (WARN_ON(i + 1 > depth)) {
2959                                err = -EFSCORRUPTED;
2960                                break;
2961                        }
2962                        path[i + 1].p_bh = bh;
2963
2964                        /* save actual number of indexes since this
2965                         * number is changed at the next iteration */
2966                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2967                        i++;
2968                } else {
2969                        /* we finished processing this index, go up */
2970                        if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2971                                /* index is empty, remove it;
2972                                 * handle must be already prepared by the
2973                                 * truncatei_leaf() */
2974                                err = ext4_ext_rm_idx(handle, inode, path, i);
2975                        }
2976                        /* root level has p_bh == NULL, brelse() eats this */
2977                        brelse(path[i].p_bh);
2978                        path[i].p_bh = NULL;
2979                        i--;
2980                        ext_debug(inode, "return to level %d\n", i);
2981                }
2982        }
2983
2984        trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
2985                                         path->p_hdr->eh_entries);
2986
2987        /*
2988         * if there's a partial cluster and we have removed the first extent
2989         * in the file, then we also free the partial cluster, if any
2990         */
2991        if (partial.state == tofree && err == 0) {
2992                int flags = get_default_free_blocks_flags(inode);
2993
2994                if (ext4_is_pending(inode, partial.lblk))
2995                        flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2996                ext4_free_blocks(handle, inode, NULL,
2997                                 EXT4_C2B(sbi, partial.pclu),
2998                                 sbi->s_cluster_ratio, flags);
2999                if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3000                        ext4_rereserve_cluster(inode, partial.lblk);
3001                partial.state = initial;
3002        }
3003
3004        /* TODO: flexible tree reduction should be here */
3005        if (path->p_hdr->eh_entries == 0) {
3006                /*
3007                 * truncate to zero freed all the tree,
3008                 * so we need to correct eh_depth
3009                 */
3010                err = ext4_ext_get_access(handle, inode, path);
3011                if (err == 0) {
3012                        ext_inode_hdr(inode)->eh_depth = 0;
3013                        ext_inode_hdr(inode)->eh_max =
3014                                cpu_to_le16(ext4_ext_space_root(inode, 0));
3015                        err = ext4_ext_dirty(handle, inode, path);
3016                }
3017        }
3018out:
3019        ext4_ext_drop_refs(path);
3020        kfree(path);
3021        path = NULL;
3022        if (err == -EAGAIN)
3023                goto again;
3024        ext4_journal_stop(handle);
3025
3026        return err;
3027}
3028
3029/*
3030 * called at mount time
3031 */
3032void ext4_ext_init(struct super_block *sb)
3033{
3034        /*
3035         * possible initialization would be here
3036         */
3037
3038        if (ext4_has_feature_extents(sb)) {
3039#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3040                printk(KERN_INFO "EXT4-fs: file extents enabled"
3041#ifdef AGGRESSIVE_TEST
3042                       ", aggressive tests"
3043#endif
3044#ifdef CHECK_BINSEARCH
3045                       ", check binsearch"
3046#endif
3047#ifdef EXTENTS_STATS
3048                       ", stats"
3049#endif
3050                       "\n");
3051#endif
3052#ifdef EXTENTS_STATS
3053                spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3054                EXT4_SB(sb)->s_ext_min = 1 << 30;
3055                EXT4_SB(sb)->s_ext_max = 0;
3056#endif
3057        }
3058}
3059
3060/*
3061 * called at umount time
3062 */
3063void ext4_ext_release(struct super_block *sb)
3064{
3065        if (!ext4_has_feature_extents(sb))
3066                return;
3067
3068#ifdef EXTENTS_STATS
3069        if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3070                struct ext4_sb_info *sbi = EXT4_SB(sb);
3071                printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3072                        sbi->s_ext_blocks, sbi->s_ext_extents,
3073                        sbi->s_ext_blocks / sbi->s_ext_extents);
3074                printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3075                        sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3076        }
3077#endif
3078}
3079
3080static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3081{
3082        ext4_lblk_t  ee_block;
3083        ext4_fsblk_t ee_pblock;
3084        unsigned int ee_len;
3085
3086        ee_block  = le32_to_cpu(ex->ee_block);
3087        ee_len    = ext4_ext_get_actual_len(ex);
3088        ee_pblock = ext4_ext_pblock(ex);
3089
3090        if (ee_len == 0)
3091                return 0;
3092
3093        return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3094                                     EXTENT_STATUS_WRITTEN);
3095}
3096
3097/* FIXME!! we need to try to merge to left or right after zero-out  */
3098static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3099{
3100        ext4_fsblk_t ee_pblock;
3101        unsigned int ee_len;
3102
3103        ee_len    = ext4_ext_get_actual_len(ex);
3104        ee_pblock = ext4_ext_pblock(ex);
3105        return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3106                                  ee_len);
3107}
3108
3109/*
3110 * ext4_split_extent_at() splits an extent at given block.
3111 *
3112 * @handle: the journal handle
3113 * @inode: the file inode
3114 * @path: the path to the extent
3115 * @split: the logical block where the extent is splitted.
3116 * @split_flags: indicates if the extent could be zeroout if split fails, and
3117 *               the states(init or unwritten) of new extents.
3118 * @flags: flags used to insert new extent to extent tree.
3119 *
3120 *
3121 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3122 * of which are determined by split_flag.
3123 *
3124 * There are two cases:
3125 *  a> the extent are splitted into two extent.
3126 *  b> split is not needed, and just mark the extent.
3127 *
3128 * return 0 on success.
3129 */
3130static int ext4_split_extent_at(handle_t *handle,
3131                             struct inode *inode,
3132                             struct ext4_ext_path **ppath,
3133                             ext4_lblk_t split,
3134                             int split_flag,
3135                             int flags)
3136{
3137        struct ext4_ext_path *path = *ppath;
3138        ext4_fsblk_t newblock;
3139        ext4_lblk_t ee_block;
3140        struct ext4_extent *ex, newex, orig_ex, zero_ex;
3141        struct ext4_extent *ex2 = NULL;
3142        unsigned int ee_len, depth;
3143        int err = 0;
3144
3145        BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3146               (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3147
3148        ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3149
3150        ext4_ext_show_leaf(inode, path);
3151
3152        depth = ext_depth(inode);
3153        ex = path[depth].p_ext;
3154        ee_block = le32_to_cpu(ex->ee_block);
3155        ee_len = ext4_ext_get_actual_len(ex);
3156        newblock = split - ee_block + ext4_ext_pblock(ex);
3157
3158        BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3159        BUG_ON(!ext4_ext_is_unwritten(ex) &&
3160               split_flag & (EXT4_EXT_MAY_ZEROOUT |
3161                             EXT4_EXT_MARK_UNWRIT1 |
3162                             EXT4_EXT_MARK_UNWRIT2));
3163
3164        err = ext4_ext_get_access(handle, inode, path + depth);
3165        if (err)
3166                goto out;
3167
3168        if (split == ee_block) {
3169                /*
3170                 * case b: block @split is the block that the extent begins with
3171                 * then we just change the state of the extent, and splitting
3172                 * is not needed.
3173                 */
3174                if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3175                        ext4_ext_mark_unwritten(ex);
3176                else
3177                        ext4_ext_mark_initialized(ex);
3178
3179                if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3180                        ext4_ext_try_to_merge(handle, inode, path, ex);
3181
3182                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3183                goto out;
3184        }
3185
3186        /* case a */
3187        memcpy(&orig_ex, ex, sizeof(orig_ex));
3188        ex->ee_len = cpu_to_le16(split - ee_block);
3189        if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3190                ext4_ext_mark_unwritten(ex);
3191
3192        /*
3193         * path may lead to new leaf, not to original leaf any more
3194         * after ext4_ext_insert_extent() returns,
3195         */
3196        err = ext4_ext_dirty(handle, inode, path + depth);
3197        if (err)
3198                goto fix_extent_len;
3199
3200        ex2 = &newex;
3201        ex2->ee_block = cpu_to_le32(split);
3202        ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3203        ext4_ext_store_pblock(ex2, newblock);
3204        if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3205                ext4_ext_mark_unwritten(ex2);
3206
3207        err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3208        if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3209                if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3210                        if (split_flag & EXT4_EXT_DATA_VALID1) {
3211                                err = ext4_ext_zeroout(inode, ex2);
3212                                zero_ex.ee_block = ex2->ee_block;
3213                                zero_ex.ee_len = cpu_to_le16(
3214                                                ext4_ext_get_actual_len(ex2));
3215                                ext4_ext_store_pblock(&zero_ex,
3216                                                      ext4_ext_pblock(ex2));
3217                        } else {
3218                                err = ext4_ext_zeroout(inode, ex);
3219                                zero_ex.ee_block = ex->ee_block;
3220                                zero_ex.ee_len = cpu_to_le16(
3221                                                ext4_ext_get_actual_len(ex));
3222                                ext4_ext_store_pblock(&zero_ex,
3223                                                      ext4_ext_pblock(ex));
3224                        }
3225                } else {
3226                        err = ext4_ext_zeroout(inode, &orig_ex);
3227                        zero_ex.ee_block = orig_ex.ee_block;
3228                        zero_ex.ee_len = cpu_to_le16(
3229                                                ext4_ext_get_actual_len(&orig_ex));
3230                        ext4_ext_store_pblock(&zero_ex,
3231                                              ext4_ext_pblock(&orig_ex));
3232                }
3233
3234                if (err)
3235                        goto fix_extent_len;
3236                /* update the extent length and mark as initialized */
3237                ex->ee_len = cpu_to_le16(ee_len);
3238                ext4_ext_try_to_merge(handle, inode, path, ex);
3239                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3240                if (err)
3241                        goto fix_extent_len;
3242
3243                /* update extent status tree */
3244                err = ext4_zeroout_es(inode, &zero_ex);
3245
3246                goto out;
3247        } else if (err)
3248                goto fix_extent_len;
3249
3250out:
3251        ext4_ext_show_leaf(inode, path);
3252        return err;
3253
3254fix_extent_len:
3255        ex->ee_len = orig_ex.ee_len;
3256        /*
3257         * Ignore ext4_ext_dirty return value since we are already in error path
3258         * and err is a non-zero error code.
3259         */
3260        ext4_ext_dirty(handle, inode, path + path->p_depth);
3261        return err;
3262}
3263
3264/*
3265 * ext4_split_extents() splits an extent and mark extent which is covered
3266 * by @map as split_flags indicates
3267 *
3268 * It may result in splitting the extent into multiple extents (up to three)
3269 * There are three possibilities:
3270 *   a> There is no split required
3271 *   b> Splits in two extents: Split is happening at either end of the extent
3272 *   c> Splits in three extents: Somone is splitting in middle of the extent
3273 *
3274 */
3275static int ext4_split_extent(handle_t *handle,
3276                              struct inode *inode,
3277                              struct ext4_ext_path **ppath,
3278                              struct ext4_map_blocks *map,
3279                              int split_flag,
3280                              int flags)
3281{
3282        struct ext4_ext_path *path = *ppath;
3283        ext4_lblk_t ee_block;
3284        struct ext4_extent *ex;
3285        unsigned int ee_len, depth;
3286        int err = 0;
3287        int unwritten;
3288        int split_flag1, flags1;
3289        int allocated = map->m_len;
3290
3291        depth = ext_depth(inode);
3292        ex = path[depth].p_ext;
3293        ee_block = le32_to_cpu(ex->ee_block);
3294        ee_len = ext4_ext_get_actual_len(ex);
3295        unwritten = ext4_ext_is_unwritten(ex);
3296
3297        if (map->m_lblk + map->m_len < ee_block + ee_len) {
3298                split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3299                flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3300                if (unwritten)
3301                        split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3302                                       EXT4_EXT_MARK_UNWRIT2;
3303                if (split_flag & EXT4_EXT_DATA_VALID2)
3304                        split_flag1 |= EXT4_EXT_DATA_VALID1;
3305                err = ext4_split_extent_at(handle, inode, ppath,
3306                                map->m_lblk + map->m_len, split_flag1, flags1);
3307                if (err)
3308                        goto out;
3309        } else {
3310                allocated = ee_len - (map->m_lblk - ee_block);
3311        }
3312        /*
3313         * Update path is required because previous ext4_split_extent_at() may
3314         * result in split of original leaf or extent zeroout.
3315         */
3316        path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3317        if (IS_ERR(path))
3318                return PTR_ERR(path);
3319        depth = ext_depth(inode);
3320        ex = path[depth].p_ext;
3321        if (!ex) {
3322                EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3323                                 (unsigned long) map->m_lblk);
3324                return -EFSCORRUPTED;
3325        }
3326        unwritten = ext4_ext_is_unwritten(ex);
3327        split_flag1 = 0;
3328
3329        if (map->m_lblk >= ee_block) {
3330                split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3331                if (unwritten) {
3332                        split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3333                        split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3334                                                     EXT4_EXT_MARK_UNWRIT2);
3335                }
3336                err = ext4_split_extent_at(handle, inode, ppath,
3337                                map->m_lblk, split_flag1, flags);
3338                if (err)
3339                        goto out;
3340        }
3341
3342        ext4_ext_show_leaf(inode, path);
3343out:
3344        return err ? err : allocated;
3345}
3346
3347/*
3348 * This function is called by ext4_ext_map_blocks() if someone tries to write
3349 * to an unwritten extent. It may result in splitting the unwritten
3350 * extent into multiple extents (up to three - one initialized and two
3351 * unwritten).
3352 * There are three possibilities:
3353 *   a> There is no split required: Entire extent should be initialized
3354 *   b> Splits in two extents: Write is happening at either end of the extent
3355 *   c> Splits in three extents: Somone is writing in middle of the extent
3356 *
3357 * Pre-conditions:
3358 *  - The extent pointed to by 'path' is unwritten.
3359 *  - The extent pointed to by 'path' contains a superset
3360 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3361 *
3362 * Post-conditions on success:
3363 *  - the returned value is the number of blocks beyond map->l_lblk
3364 *    that are allocated and initialized.
3365 *    It is guaranteed to be >= map->m_len.
3366 */
3367static int ext4_ext_convert_to_initialized(handle_t *handle,
3368                                           struct inode *inode,
3369                                           struct ext4_map_blocks *map,
3370                                           struct ext4_ext_path **ppath,
3371                                           int flags)
3372{
3373        struct ext4_ext_path *path = *ppath;
3374        struct ext4_sb_info *sbi;
3375        struct ext4_extent_header *eh;
3376        struct ext4_map_blocks split_map;
3377        struct ext4_extent zero_ex1, zero_ex2;
3378        struct ext4_extent *ex, *abut_ex;
3379        ext4_lblk_t ee_block, eof_block;
3380        unsigned int ee_len, depth, map_len = map->m_len;
3381        int allocated = 0, max_zeroout = 0;
3382        int err = 0;
3383        int split_flag = EXT4_EXT_DATA_VALID2;
3384
3385        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3386                  (unsigned long long)map->m_lblk, map_len);
3387
3388        sbi = EXT4_SB(inode->i_sb);
3389        eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3390                        >> inode->i_sb->s_blocksize_bits;
3391        if (eof_block < map->m_lblk + map_len)
3392                eof_block = map->m_lblk + map_len;
3393
3394        depth = ext_depth(inode);
3395        eh = path[depth].p_hdr;
3396        ex = path[depth].p_ext;
3397        ee_block = le32_to_cpu(ex->ee_block);
3398        ee_len = ext4_ext_get_actual_len(ex);
3399        zero_ex1.ee_len = 0;
3400        zero_ex2.ee_len = 0;
3401
3402        trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3403
3404        /* Pre-conditions */
3405        BUG_ON(!ext4_ext_is_unwritten(ex));
3406        BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3407
3408        /*
3409         * Attempt to transfer newly initialized blocks from the currently
3410         * unwritten extent to its neighbor. This is much cheaper
3411         * than an insertion followed by a merge as those involve costly
3412         * memmove() calls. Transferring to the left is the common case in
3413         * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3414         * followed by append writes.
3415         *
3416         * Limitations of the current logic:
3417         *  - L1: we do not deal with writes covering the whole extent.
3418         *    This would require removing the extent if the transfer
3419         *    is possible.
3420         *  - L2: we only attempt to merge with an extent stored in the
3421         *    same extent tree node.
3422         */
3423        if ((map->m_lblk == ee_block) &&
3424                /* See if we can merge left */
3425                (map_len < ee_len) &&           /*L1*/
3426                (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3427                ext4_lblk_t prev_lblk;
3428                ext4_fsblk_t prev_pblk, ee_pblk;
3429                unsigned int prev_len;
3430
3431                abut_ex = ex - 1;
3432                prev_lblk = le32_to_cpu(abut_ex->ee_block);
3433                prev_len = ext4_ext_get_actual_len(abut_ex);
3434                prev_pblk = ext4_ext_pblock(abut_ex);
3435                ee_pblk = ext4_ext_pblock(ex);
3436
3437                /*
3438                 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3439                 * upon those conditions:
3440                 * - C1: abut_ex is initialized,
3441                 * - C2: abut_ex is logically abutting ex,
3442                 * - C3: abut_ex is physically abutting ex,
3443                 * - C4: abut_ex can receive the additional blocks without
3444                 *   overflowing the (initialized) length limit.
3445                 */
3446                if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3447                        ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3448                        ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3449                        (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3450                        err = ext4_ext_get_access(handle, inode, path + depth);
3451                        if (err)
3452                                goto out;
3453
3454                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3455                                map, ex, abut_ex);
3456
3457                        /* Shift the start of ex by 'map_len' blocks */
3458                        ex->ee_block = cpu_to_le32(ee_block + map_len);
3459                        ext4_ext_store_pblock(ex, ee_pblk + map_len);
3460                        ex->ee_len = cpu_to_le16(ee_len - map_len);
3461                        ext4_ext_mark_unwritten(ex); /* Restore the flag */
3462
3463                        /* Extend abut_ex by 'map_len' blocks */
3464                        abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3465
3466                        /* Result: number of initialized blocks past m_lblk */
3467                        allocated = map_len;
3468                }
3469        } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3470                   (map_len < ee_len) &&        /*L1*/
3471                   ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3472                /* See if we can merge right */
3473                ext4_lblk_t next_lblk;
3474                ext4_fsblk_t next_pblk, ee_pblk;
3475                unsigned int next_len;
3476
3477                abut_ex = ex + 1;
3478                next_lblk = le32_to_cpu(abut_ex->ee_block);
3479                next_len = ext4_ext_get_actual_len(abut_ex);
3480                next_pblk = ext4_ext_pblock(abut_ex);
3481                ee_pblk = ext4_ext_pblock(ex);
3482
3483                /*
3484                 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3485                 * upon those conditions:
3486                 * - C1: abut_ex is initialized,
3487                 * - C2: abut_ex is logically abutting ex,
3488                 * - C3: abut_ex is physically abutting ex,
3489                 * - C4: abut_ex can receive the additional blocks without
3490                 *   overflowing the (initialized) length limit.
3491                 */
3492                if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3493                    ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3494                    ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3495                    (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3496                        err = ext4_ext_get_access(handle, inode, path + depth);
3497                        if (err)
3498                                goto out;
3499
3500                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3501                                map, ex, abut_ex);
3502
3503                        /* Shift the start of abut_ex by 'map_len' blocks */
3504                        abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3505                        ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3506                        ex->ee_len = cpu_to_le16(ee_len - map_len);
3507                        ext4_ext_mark_unwritten(ex); /* Restore the flag */
3508
3509                        /* Extend abut_ex by 'map_len' blocks */
3510                        abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3511
3512                        /* Result: number of initialized blocks past m_lblk */
3513                        allocated = map_len;
3514                }
3515        }
3516        if (allocated) {
3517                /* Mark the block containing both extents as dirty */
3518                err = ext4_ext_dirty(handle, inode, path + depth);
3519
3520                /* Update path to point to the right extent */
3521                path[depth].p_ext = abut_ex;
3522                goto out;
3523        } else
3524                allocated = ee_len - (map->m_lblk - ee_block);
3525
3526        WARN_ON(map->m_lblk < ee_block);
3527        /*
3528         * It is safe to convert extent to initialized via explicit
3529         * zeroout only if extent is fully inside i_size or new_size.
3530         */
3531        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3532
3533        if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3534                max_zeroout = sbi->s_extent_max_zeroout_kb >>
3535                        (inode->i_sb->s_blocksize_bits - 10);
3536
3537        /*
3538         * five cases:
3539         * 1. split the extent into three extents.
3540         * 2. split the extent into two extents, zeroout the head of the first
3541         *    extent.
3542         * 3. split the extent into two extents, zeroout the tail of the second
3543         *    extent.
3544         * 4. split the extent into two extents with out zeroout.
3545         * 5. no splitting needed, just possibly zeroout the head and / or the
3546         *    tail of the extent.
3547         */
3548        split_map.m_lblk = map->m_lblk;
3549        split_map.m_len = map->m_len;
3550
3551        if (max_zeroout && (allocated > split_map.m_len)) {
3552                if (allocated <= max_zeroout) {
3553                        /* case 3 or 5 */
3554                        zero_ex1.ee_block =
3555                                 cpu_to_le32(split_map.m_lblk +
3556                                             split_map.m_len);
3557                        zero_ex1.ee_len =
3558                                cpu_to_le16(allocated - split_map.m_len);
3559                        ext4_ext_store_pblock(&zero_ex1,
3560                                ext4_ext_pblock(ex) + split_map.m_lblk +
3561                                split_map.m_len - ee_block);
3562                        err = ext4_ext_zeroout(inode, &zero_ex1);
3563                        if (err)
3564                                goto out;
3565                        split_map.m_len = allocated;
3566                }
3567                if (split_map.m_lblk - ee_block + split_map.m_len <
3568                                                                max_zeroout) {
3569                        /* case 2 or 5 */
3570                        if (split_map.m_lblk != ee_block) {
3571                                zero_ex2.ee_block = ex->ee_block;
3572                                zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3573                                                        ee_block);
3574                                ext4_ext_store_pblock(&zero_ex2,
3575                                                      ext4_ext_pblock(ex));
3576                                err = ext4_ext_zeroout(inode, &zero_ex2);
3577                                if (err)
3578                                        goto out;
3579                        }
3580
3581                        split_map.m_len += split_map.m_lblk - ee_block;
3582                        split_map.m_lblk = ee_block;
3583                        allocated = map->m_len;
3584                }
3585        }
3586
3587        err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3588                                flags);
3589        if (err > 0)
3590                err = 0;
3591out:
3592        /* If we have gotten a failure, don't zero out status tree */
3593        if (!err) {
3594                err = ext4_zeroout_es(inode, &zero_ex1);
3595                if (!err)
3596                        err = ext4_zeroout_es(inode, &zero_ex2);
3597        }
3598        return err ? err : allocated;
3599}
3600
3601/*
3602 * This function is called by ext4_ext_map_blocks() from
3603 * ext4_get_blocks_dio_write() when DIO to write
3604 * to an unwritten extent.
3605 *
3606 * Writing to an unwritten extent may result in splitting the unwritten
3607 * extent into multiple initialized/unwritten extents (up to three)
3608 * There are three possibilities:
3609 *   a> There is no split required: Entire extent should be unwritten
3610 *   b> Splits in two extents: Write is happening at either end of the extent
3611 *   c> Splits in three extents: Somone is writing in middle of the extent
3612 *
3613 * This works the same way in the case of initialized -> unwritten conversion.
3614 *
3615 * One of more index blocks maybe needed if the extent tree grow after
3616 * the unwritten extent split. To prevent ENOSPC occur at the IO
3617 * complete, we need to split the unwritten extent before DIO submit
3618 * the IO. The unwritten extent called at this time will be split
3619 * into three unwritten extent(at most). After IO complete, the part
3620 * being filled will be convert to initialized by the end_io callback function
3621 * via ext4_convert_unwritten_extents().
3622 *
3623 * Returns the size of unwritten extent to be written on success.
3624 */
3625static int ext4_split_convert_extents(handle_t *handle,
3626                                        struct inode *inode,
3627                                        struct ext4_map_blocks *map,
3628                                        struct ext4_ext_path **ppath,
3629                                        int flags)
3630{
3631        struct ext4_ext_path *path = *ppath;
3632        ext4_lblk_t eof_block;
3633        ext4_lblk_t ee_block;
3634        struct ext4_extent *ex;
3635        unsigned int ee_len;
3636        int split_flag = 0, depth;
3637
3638        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3639                  (unsigned long long)map->m_lblk, map->m_len);
3640
3641        eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3642                        >> inode->i_sb->s_blocksize_bits;
3643        if (eof_block < map->m_lblk + map->m_len)
3644                eof_block = map->m_lblk + map->m_len;
3645        /*
3646         * It is safe to convert extent to initialized via explicit
3647         * zeroout only if extent is fully inside i_size or new_size.
3648         */
3649        depth = ext_depth(inode);
3650        ex = path[depth].p_ext;
3651        ee_block = le32_to_cpu(ex->ee_block);
3652        ee_len = ext4_ext_get_actual_len(ex);
3653
3654        /* Convert to unwritten */
3655        if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3656                split_flag |= EXT4_EXT_DATA_VALID1;
3657        /* Convert to initialized */
3658        } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3659                split_flag |= ee_block + ee_len <= eof_block ?
3660                              EXT4_EXT_MAY_ZEROOUT : 0;
3661                split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3662        }
3663        flags |= EXT4_GET_BLOCKS_PRE_IO;
3664        return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3665}
3666
3667static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3668                                                struct inode *inode,
3669                                                struct ext4_map_blocks *map,
3670                                                struct ext4_ext_path **ppath)
3671{
3672        struct ext4_ext_path *path = *ppath;
3673        struct ext4_extent *ex;
3674        ext4_lblk_t ee_block;
3675        unsigned int ee_len;
3676        int depth;
3677        int err = 0;
3678
3679        depth = ext_depth(inode);
3680        ex = path[depth].p_ext;
3681        ee_block = le32_to_cpu(ex->ee_block);
3682        ee_len = ext4_ext_get_actual_len(ex);
3683
3684        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3685                  (unsigned long long)ee_block, ee_len);
3686
3687        /* If extent is larger than requested it is a clear sign that we still
3688         * have some extent state machine issues left. So extent_split is still
3689         * required.
3690         * TODO: Once all related issues will be fixed this situation should be
3691         * illegal.
3692         */
3693        if (ee_block != map->m_lblk || ee_len > map->m_len) {
3694#ifdef CONFIG_EXT4_DEBUG
3695                ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3696                             " len %u; IO logical block %llu, len %u",
3697                             inode->i_ino, (unsigned long long)ee_block, ee_len,
3698                             (unsigned long long)map->m_lblk, map->m_len);
3699#endif
3700                err = ext4_split_convert_extents(handle, inode, map, ppath,
3701                                                 EXT4_GET_BLOCKS_CONVERT);
3702                if (err < 0)
3703                        return err;
3704                path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3705                if (IS_ERR(path))
3706                        return PTR_ERR(path);
3707                depth = ext_depth(inode);
3708                ex = path[depth].p_ext;
3709        }
3710
3711        err = ext4_ext_get_access(handle, inode, path + depth);
3712        if (err)
3713                goto out;
3714        /* first mark the extent as initialized */
3715        ext4_ext_mark_initialized(ex);
3716
3717        /* note: ext4_ext_correct_indexes() isn't needed here because
3718         * borders are not changed
3719         */
3720        ext4_ext_try_to_merge(handle, inode, path, ex);
3721
3722        /* Mark modified extent as dirty */
3723        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3724out:
3725        ext4_ext_show_leaf(inode, path);
3726        return err;
3727}
3728
3729static int
3730convert_initialized_extent(handle_t *handle, struct inode *inode,
3731                           struct ext4_map_blocks *map,
3732                           struct ext4_ext_path **ppath,
3733                           unsigned int *allocated)
3734{
3735        struct ext4_ext_path *path = *ppath;
3736        struct ext4_extent *ex;
3737        ext4_lblk_t ee_block;
3738        unsigned int ee_len;
3739        int depth;
3740        int err = 0;
3741
3742        /*
3743         * Make sure that the extent is no bigger than we support with
3744         * unwritten extent
3745         */
3746        if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3747                map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3748
3749        depth = ext_depth(inode);
3750        ex = path[depth].p_ext;
3751        ee_block = le32_to_cpu(ex->ee_block);
3752        ee_len = ext4_ext_get_actual_len(ex);
3753
3754        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3755                  (unsigned long long)ee_block, ee_len);
3756
3757        if (ee_block != map->m_lblk || ee_len > map->m_len) {
3758                err = ext4_split_convert_extents(handle, inode, map, ppath,
3759                                EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3760                if (err < 0)
3761                        return err;
3762                path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3763                if (IS_ERR(path))
3764                        return PTR_ERR(path);
3765                depth = ext_depth(inode);
3766                ex = path[depth].p_ext;
3767                if (!ex) {
3768                        EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3769                                         (unsigned long) map->m_lblk);
3770                        return -EFSCORRUPTED;
3771                }
3772        }
3773
3774        err = ext4_ext_get_access(handle, inode, path + depth);
3775        if (err)
3776                return err;
3777        /* first mark the extent as unwritten */
3778        ext4_ext_mark_unwritten(ex);
3779
3780        /* note: ext4_ext_correct_indexes() isn't needed here because
3781         * borders are not changed
3782         */
3783        ext4_ext_try_to_merge(handle, inode, path, ex);
3784
3785        /* Mark modified extent as dirty */
3786        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3787        if (err)
3788                return err;
3789        ext4_ext_show_leaf(inode, path);
3790
3791        ext4_update_inode_fsync_trans(handle, inode, 1);
3792
3793        map->m_flags |= EXT4_MAP_UNWRITTEN;
3794        if (*allocated > map->m_len)
3795                *allocated = map->m_len;
3796        map->m_len = *allocated;
3797        return 0;
3798}
3799
3800static int
3801ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3802                        struct ext4_map_blocks *map,
3803                        struct ext4_ext_path **ppath, int flags,
3804                        unsigned int allocated, ext4_fsblk_t newblock)
3805{
3806        struct ext4_ext_path __maybe_unused *path = *ppath;
3807        int ret = 0;
3808        int err = 0;
3809
3810        ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3811                  (unsigned long long)map->m_lblk, map->m_len, flags,
3812                  allocated);
3813        ext4_ext_show_leaf(inode, path);
3814
3815        /*
3816         * When writing into unwritten space, we should not fail to
3817         * allocate metadata blocks for the new extent block if needed.
3818         */
3819        flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3820
3821        trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3822                                                    allocated, newblock);
3823
3824        /* get_block() before submitting IO, split the extent */
3825        if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3826                ret = ext4_split_convert_extents(handle, inode, map, ppath,
3827                                         flags | EXT4_GET_BLOCKS_CONVERT);
3828                if (ret < 0) {
3829                        err = ret;
3830                        goto out2;
3831                }
3832                /*
3833                 * shouldn't get a 0 return when splitting an extent unless
3834                 * m_len is 0 (bug) or extent has been corrupted
3835                 */
3836                if (unlikely(ret == 0)) {
3837                        EXT4_ERROR_INODE(inode,
3838                                         "unexpected ret == 0, m_len = %u",
3839                                         map->m_len);
3840                        err = -EFSCORRUPTED;
3841                        goto out2;
3842                }
3843                map->m_flags |= EXT4_MAP_UNWRITTEN;
3844                goto out;
3845        }
3846        /* IO end_io complete, convert the filled extent to written */
3847        if (flags & EXT4_GET_BLOCKS_CONVERT) {
3848                err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3849                                                           ppath);
3850                if (err < 0)
3851                        goto out2;
3852                ext4_update_inode_fsync_trans(handle, inode, 1);
3853                goto map_out;
3854        }
3855        /* buffered IO cases */
3856        /*
3857         * repeat fallocate creation request
3858         * we already have an unwritten extent
3859         */
3860        if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3861                map->m_flags |= EXT4_MAP_UNWRITTEN;
3862                goto map_out;
3863        }
3864
3865        /* buffered READ or buffered write_begin() lookup */
3866        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3867                /*
3868                 * We have blocks reserved already.  We
3869                 * return allocated blocks so that delalloc
3870                 * won't do block reservation for us.  But
3871                 * the buffer head will be unmapped so that
3872                 * a read from the block returns 0s.
3873                 */
3874                map->m_flags |= EXT4_MAP_UNWRITTEN;
3875                goto out1;
3876        }
3877
3878        /*
3879         * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3880         * For buffered writes, at writepage time, etc.  Convert a
3881         * discovered unwritten extent to written.
3882         */
3883        ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3884        if (ret < 0) {
3885                err = ret;
3886                goto out2;
3887        }
3888        ext4_update_inode_fsync_trans(handle, inode, 1);
3889        /*
3890         * shouldn't get a 0 return when converting an unwritten extent
3891         * unless m_len is 0 (bug) or extent has been corrupted
3892         */
3893        if (unlikely(ret == 0)) {
3894                EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3895                                 map->m_len);
3896                err = -EFSCORRUPTED;
3897                goto out2;
3898        }
3899
3900out:
3901        allocated = ret;
3902        map->m_flags |= EXT4_MAP_NEW;
3903map_out:
3904        map->m_flags |= EXT4_MAP_MAPPED;
3905out1:
3906        map->m_pblk = newblock;
3907        if (allocated > map->m_len)
3908                allocated = map->m_len;
3909        map->m_len = allocated;
3910        ext4_ext_show_leaf(inode, path);
3911out2:
3912        return err ? err : allocated;
3913}
3914
3915/*
3916 * get_implied_cluster_alloc - check to see if the requested
3917 * allocation (in the map structure) overlaps with a cluster already
3918 * allocated in an extent.
3919 *      @sb     The filesystem superblock structure
3920 *      @map    The requested lblk->pblk mapping
3921 *      @ex     The extent structure which might contain an implied
3922 *                      cluster allocation
3923 *
3924 * This function is called by ext4_ext_map_blocks() after we failed to
3925 * find blocks that were already in the inode's extent tree.  Hence,
3926 * we know that the beginning of the requested region cannot overlap
3927 * the extent from the inode's extent tree.  There are three cases we
3928 * want to catch.  The first is this case:
3929 *
3930 *               |--- cluster # N--|
3931 *    |--- extent ---|  |---- requested region ---|
3932 *                      |==========|
3933 *
3934 * The second case that we need to test for is this one:
3935 *
3936 *   |--------- cluster # N ----------------|
3937 *         |--- requested region --|   |------- extent ----|
3938 *         |=======================|
3939 *
3940 * The third case is when the requested region lies between two extents
3941 * within the same cluster:
3942 *          |------------- cluster # N-------------|
3943 * |----- ex -----|                  |---- ex_right ----|
3944 *                  |------ requested region ------|
3945 *                  |================|
3946 *
3947 * In each of the above cases, we need to set the map->m_pblk and
3948 * map->m_len so it corresponds to the return the extent labelled as
3949 * "|====|" from cluster #N, since it is already in use for data in
3950 * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3951 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3952 * as a new "allocated" block region.  Otherwise, we will return 0 and
3953 * ext4_ext_map_blocks() will then allocate one or more new clusters
3954 * by calling ext4_mb_new_blocks().
3955 */
3956static int get_implied_cluster_alloc(struct super_block *sb,
3957                                     struct ext4_map_blocks *map,
3958                                     struct ext4_extent *ex,
3959                                     struct ext4_ext_path *path)
3960{
3961        struct ext4_sb_info *sbi = EXT4_SB(sb);
3962        ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3963        ext4_lblk_t ex_cluster_start, ex_cluster_end;
3964        ext4_lblk_t rr_cluster_start;
3965        ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3966        ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3967        unsigned short ee_len = ext4_ext_get_actual_len(ex);
3968
3969        /* The extent passed in that we are trying to match */
3970        ex_cluster_start = EXT4_B2C(sbi, ee_block);
3971        ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3972
3973        /* The requested region passed into ext4_map_blocks() */
3974        rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3975
3976        if ((rr_cluster_start == ex_cluster_end) ||
3977            (rr_cluster_start == ex_cluster_start)) {
3978                if (rr_cluster_start == ex_cluster_end)
3979                        ee_start += ee_len - 1;
3980                map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3981                map->m_len = min(map->m_len,
3982                                 (unsigned) sbi->s_cluster_ratio - c_offset);
3983                /*
3984                 * Check for and handle this case:
3985                 *
3986                 *   |--------- cluster # N-------------|
3987                 *                     |------- extent ----|
3988                 *         |--- requested region ---|
3989                 *         |===========|
3990                 */
3991
3992                if (map->m_lblk < ee_block)
3993                        map->m_len = min(map->m_len, ee_block - map->m_lblk);
3994
3995                /*
3996                 * Check for the case where there is already another allocated
3997                 * block to the right of 'ex' but before the end of the cluster.
3998                 *
3999                 *          |------------- cluster # N-------------|
4000                 * |----- ex -----|                  |---- ex_right ----|
4001                 *                  |------ requested region ------|
4002                 *                  |================|
4003                 */
4004                if (map->m_lblk > ee_block) {
4005                        ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4006                        map->m_len = min(map->m_len, next - map->m_lblk);
4007                }
4008
4009                trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4010                return 1;
4011        }
4012
4013        trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4014        return 0;
4015}
4016
4017
4018/*
4019 * Block allocation/map/preallocation routine for extents based files
4020 *
4021 *
4022 * Need to be called with
4023 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4024 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4025 *
4026 * return > 0, number of of blocks already mapped/allocated
4027 *          if create == 0 and these are pre-allocated blocks
4028 *              buffer head is unmapped
4029 *          otherwise blocks are mapped
4030 *
4031 * return = 0, if plain look up failed (blocks have not been allocated)
4032 *          buffer head is unmapped
4033 *
4034 * return < 0, error case.
4035 */
4036int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4037                        struct ext4_map_blocks *map, int flags)
4038{
4039        struct ext4_ext_path *path = NULL;
4040        struct ext4_extent newex, *ex, *ex2;
4041        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4042        ext4_fsblk_t newblock = 0, pblk;
4043        int err = 0, depth, ret;
4044        unsigned int allocated = 0, offset = 0;
4045        unsigned int allocated_clusters = 0;
4046        struct ext4_allocation_request ar;
4047        ext4_lblk_t cluster_offset;
4048
4049        ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4050        trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4051
4052        /* find extent for this block */
4053        path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4054        if (IS_ERR(path)) {
4055                err = PTR_ERR(path);
4056                path = NULL;
4057                goto out;
4058        }
4059
4060        depth = ext_depth(inode);
4061
4062        /*
4063         * consistent leaf must not be empty;
4064         * this situation is possible, though, _during_ tree modification;
4065         * this is why assert can't be put in ext4_find_extent()
4066         */
4067        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4068                EXT4_ERROR_INODE(inode, "bad extent address "
4069                                 "lblock: %lu, depth: %d pblock %lld",
4070                                 (unsigned long) map->m_lblk, depth,
4071                                 path[depth].p_block);
4072                err = -EFSCORRUPTED;
4073                goto out;
4074        }
4075
4076        ex = path[depth].p_ext;
4077        if (ex) {
4078                ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4079                ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4080                unsigned short ee_len;
4081
4082
4083                /*
4084                 * unwritten extents are treated as holes, except that
4085                 * we split out initialized portions during a write.
4086                 */
4087                ee_len = ext4_ext_get_actual_len(ex);
4088
4089                trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4090
4091                /* if found extent covers block, simply return it */
4092                if (in_range(map->m_lblk, ee_block, ee_len)) {
4093                        newblock = map->m_lblk - ee_block + ee_start;
4094                        /* number of remaining blocks in the extent */
4095                        allocated = ee_len - (map->m_lblk - ee_block);
4096                        ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4097                                  map->m_lblk, ee_block, ee_len, newblock);
4098
4099                        /*
4100                         * If the extent is initialized check whether the
4101                         * caller wants to convert it to unwritten.
4102                         */
4103                        if ((!ext4_ext_is_unwritten(ex)) &&
4104                            (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4105                                err = convert_initialized_extent(handle,
4106                                        inode, map, &path, &allocated);
4107                                goto out;
4108                        } else if (!ext4_ext_is_unwritten(ex)) {
4109                                map->m_flags |= EXT4_MAP_MAPPED;
4110                                map->m_pblk = newblock;
4111                                if (allocated > map->m_len)
4112                                        allocated = map->m_len;
4113                                map->m_len = allocated;
4114                                ext4_ext_show_leaf(inode, path);
4115                                goto out;
4116                        }
4117
4118                        ret = ext4_ext_handle_unwritten_extents(
4119                                handle, inode, map, &path, flags,
4120                                allocated, newblock);
4121                        if (ret < 0)
4122                                err = ret;
4123                        else
4124                                allocated = ret;
4125                        goto out;
4126                }
4127        }
4128
4129        /*
4130         * requested block isn't allocated yet;
4131         * we couldn't try to create block if create flag is zero
4132         */
4133        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4134                ext4_lblk_t hole_start, hole_len;
4135
4136                hole_start = map->m_lblk;
4137                hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4138                /*
4139                 * put just found gap into cache to speed up
4140                 * subsequent requests
4141                 */
4142                ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4143
4144                /* Update hole_len to reflect hole size after map->m_lblk */
4145                if (hole_start != map->m_lblk)
4146                        hole_len -= map->m_lblk - hole_start;
4147                map->m_pblk = 0;
4148                map->m_len = min_t(unsigned int, map->m_len, hole_len);
4149
4150                goto out;
4151        }
4152
4153        /*
4154         * Okay, we need to do block allocation.
4155         */
4156        newex.ee_block = cpu_to_le32(map->m_lblk);
4157        cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4158
4159        /*
4160         * If we are doing bigalloc, check to see if the extent returned
4161         * by ext4_find_extent() implies a cluster we can use.
4162         */
4163        if (cluster_offset && ex &&
4164            get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4165                ar.len = allocated = map->m_len;
4166                newblock = map->m_pblk;
4167                goto got_allocated_blocks;
4168        }
4169
4170        /* find neighbour allocated blocks */
4171        ar.lleft = map->m_lblk;
4172        err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4173        if (err)
4174                goto out;
4175        ar.lright = map->m_lblk;
4176        ex2 = NULL;
4177        err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4178        if (err)
4179                goto out;
4180
4181        /* Check if the extent after searching to the right implies a
4182         * cluster we can use. */
4183        if ((sbi->s_cluster_ratio > 1) && ex2 &&
4184            get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4185                ar.len = allocated = map->m_len;
4186                newblock = map->m_pblk;
4187                goto got_allocated_blocks;
4188        }
4189
4190        /*
4191         * See if request is beyond maximum number of blocks we can have in
4192         * a single extent. For an initialized extent this limit is
4193         * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4194         * EXT_UNWRITTEN_MAX_LEN.
4195         */
4196        if (map->m_len > EXT_INIT_MAX_LEN &&
4197            !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4198                map->m_len = EXT_INIT_MAX_LEN;
4199        else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4200                 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4201                map->m_len = EXT_UNWRITTEN_MAX_LEN;
4202
4203        /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4204        newex.ee_len = cpu_to_le16(map->m_len);
4205        err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4206        if (err)
4207                allocated = ext4_ext_get_actual_len(&newex);
4208        else
4209                allocated = map->m_len;
4210
4211        /* allocate new block */
4212        ar.inode = inode;
4213        ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4214        ar.logical = map->m_lblk;
4215        /*
4216         * We calculate the offset from the beginning of the cluster
4217         * for the logical block number, since when we allocate a
4218         * physical cluster, the physical block should start at the
4219         * same offset from the beginning of the cluster.  This is
4220         * needed so that future calls to get_implied_cluster_alloc()
4221         * work correctly.
4222         */
4223        offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4224        ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4225        ar.goal -= offset;
4226        ar.logical -= offset;
4227        if (S_ISREG(inode->i_mode))
4228                ar.flags = EXT4_MB_HINT_DATA;
4229        else
4230                /* disable in-core preallocation for non-regular files */
4231                ar.flags = 0;
4232        if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4233                ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4234        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4235                ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4236        if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4237                ar.flags |= EXT4_MB_USE_RESERVED;
4238        newblock = ext4_mb_new_blocks(handle, &ar, &err);
4239        if (!newblock)
4240                goto out;
4241        allocated_clusters = ar.len;
4242        ar.len = EXT4_C2B(sbi, ar.len) - offset;
4243        ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4244                  ar.goal, newblock, ar.len, allocated);
4245        if (ar.len > allocated)
4246                ar.len = allocated;
4247
4248got_allocated_blocks:
4249        /* try to insert new extent into found leaf and return */
4250        pblk = newblock + offset;
4251        ext4_ext_store_pblock(&newex, pblk);
4252        newex.ee_len = cpu_to_le16(ar.len);
4253        /* Mark unwritten */
4254        if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4255                ext4_ext_mark_unwritten(&newex);
4256                map->m_flags |= EXT4_MAP_UNWRITTEN;
4257        }
4258
4259        err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4260        if (err) {
4261                if (allocated_clusters) {
4262                        int fb_flags = 0;
4263
4264                        /*
4265                         * free data blocks we just allocated.
4266                         * not a good idea to call discard here directly,
4267                         * but otherwise we'd need to call it every free().
4268                         */
4269                        ext4_discard_preallocations(inode, 0);
4270                        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4271                                fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4272                        ext4_free_blocks(handle, inode, NULL, newblock,
4273                                         EXT4_C2B(sbi, allocated_clusters),
4274                                         fb_flags);
4275                }
4276                goto out;
4277        }
4278
4279        /*
4280         * Reduce the reserved cluster count to reflect successful deferred
4281         * allocation of delayed allocated clusters or direct allocation of
4282         * clusters discovered to be delayed allocated.  Once allocated, a
4283         * cluster is not included in the reserved count.
4284         */
4285        if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4286                if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4287                        /*
4288                         * When allocating delayed allocated clusters, simply
4289                         * reduce the reserved cluster count and claim quota
4290                         */
4291                        ext4_da_update_reserve_space(inode, allocated_clusters,
4292                                                        1);
4293                } else {
4294                        ext4_lblk_t lblk, len;
4295                        unsigned int n;
4296
4297                        /*
4298                         * When allocating non-delayed allocated clusters
4299                         * (from fallocate, filemap, DIO, or clusters
4300                         * allocated when delalloc has been disabled by
4301                         * ext4_nonda_switch), reduce the reserved cluster
4302                         * count by the number of allocated clusters that
4303                         * have previously been delayed allocated.  Quota
4304                         * has been claimed by ext4_mb_new_blocks() above,
4305                         * so release the quota reservations made for any
4306                         * previously delayed allocated clusters.
4307                         */
4308                        lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4309                        len = allocated_clusters << sbi->s_cluster_bits;
4310                        n = ext4_es_delayed_clu(inode, lblk, len);
4311                        if (n > 0)
4312                                ext4_da_update_reserve_space(inode, (int) n, 0);
4313                }
4314        }
4315
4316        /*
4317         * Cache the extent and update transaction to commit on fdatasync only
4318         * when it is _not_ an unwritten extent.
4319         */
4320        if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4321                ext4_update_inode_fsync_trans(handle, inode, 1);
4322        else
4323                ext4_update_inode_fsync_trans(handle, inode, 0);
4324
4325        map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4326        map->m_pblk = pblk;
4327        map->m_len = ar.len;
4328        allocated = map->m_len;
4329        ext4_ext_show_leaf(inode, path);
4330
4331out:
4332        ext4_ext_drop_refs(path);
4333        kfree(path);
4334
4335        trace_ext4_ext_map_blocks_exit(inode, flags, map,
4336                                       err ? err : allocated);
4337        return err ? err : allocated;
4338}
4339
4340int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4341{
4342        struct super_block *sb = inode->i_sb;
4343        ext4_lblk_t last_block;
4344        int err = 0;
4345
4346        /*
4347         * TODO: optimization is possible here.
4348         * Probably we need not scan at all,
4349         * because page truncation is enough.
4350         */
4351
4352        /* we have to know where to truncate from in crash case */
4353        EXT4_I(inode)->i_disksize = inode->i_size;
4354        err = ext4_mark_inode_dirty(handle, inode);
4355        if (err)
4356                return err;
4357
4358        last_block = (inode->i_size + sb->s_blocksize - 1)
4359                        >> EXT4_BLOCK_SIZE_BITS(sb);
4360retry:
4361        err = ext4_es_remove_extent(inode, last_block,
4362                                    EXT_MAX_BLOCKS - last_block);
4363        if (err == -ENOMEM) {
4364                cond_resched();
4365                congestion_wait(BLK_RW_ASYNC, HZ/50);
4366                goto retry;
4367        }
4368        if (err)
4369                return err;
4370retry_remove_space:
4371        err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4372        if (err == -ENOMEM) {
4373                cond_resched();
4374                congestion_wait(BLK_RW_ASYNC, HZ/50);
4375                goto retry_remove_space;
4376        }
4377        return err;
4378}
4379
4380static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4381                                  ext4_lblk_t len, loff_t new_size,
4382                                  int flags)
4383{
4384        struct inode *inode = file_inode(file);
4385        handle_t *handle;
4386        int ret = 0;
4387        int ret2 = 0, ret3 = 0;
4388        int retries = 0;
4389        int depth = 0;
4390        struct ext4_map_blocks map;
4391        unsigned int credits;
4392        loff_t epos;
4393
4394        BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4395        map.m_lblk = offset;
4396        map.m_len = len;
4397        /*
4398         * Don't normalize the request if it can fit in one extent so
4399         * that it doesn't get unnecessarily split into multiple
4400         * extents.
4401         */
4402        if (len <= EXT_UNWRITTEN_MAX_LEN)
4403                flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4404
4405        /*
4406         * credits to insert 1 extent into extent tree
4407         */
4408        credits = ext4_chunk_trans_blocks(inode, len);
4409        depth = ext_depth(inode);
4410
4411retry:
4412        while (ret >= 0 && len) {
4413                /*
4414                 * Recalculate credits when extent tree depth changes.
4415                 */
4416                if (depth != ext_depth(inode)) {
4417                        credits = ext4_chunk_trans_blocks(inode, len);
4418                        depth = ext_depth(inode);
4419                }
4420
4421                handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4422                                            credits);
4423                if (IS_ERR(handle)) {
4424                        ret = PTR_ERR(handle);
4425                        break;
4426                }
4427                ret = ext4_map_blocks(handle, inode, &map, flags);
4428                if (ret <= 0) {
4429                        ext4_debug("inode #%lu: block %u: len %u: "
4430                                   "ext4_ext_map_blocks returned %d",
4431                                   inode->i_ino, map.m_lblk,
4432                                   map.m_len, ret);
4433                        ext4_mark_inode_dirty(handle, inode);
4434                        ret2 = ext4_journal_stop(handle);
4435                        break;
4436                }
4437                map.m_lblk += ret;
4438                map.m_len = len = len - ret;
4439                epos = (loff_t)map.m_lblk << inode->i_blkbits;
4440                inode->i_ctime = current_time(inode);
4441                if (new_size) {
4442                        if (epos > new_size)
4443                                epos = new_size;
4444                        if (ext4_update_inode_size(inode, epos) & 0x1)
4445                                inode->i_mtime = inode->i_ctime;
4446                }
4447                ret2 = ext4_mark_inode_dirty(handle, inode);
4448                ext4_update_inode_fsync_trans(handle, inode, 1);
4449                ret3 = ext4_journal_stop(handle);
4450                ret2 = ret3 ? ret3 : ret2;
4451                if (unlikely(ret2))
4452                        break;
4453        }
4454        if (ret == -ENOSPC &&
4455                        ext4_should_retry_alloc(inode->i_sb, &retries)) {
4456                ret = 0;
4457                goto retry;
4458        }
4459
4460        return ret > 0 ? ret2 : ret;
4461}
4462
4463static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4464
4465static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4466
4467static long ext4_zero_range(struct file *file, loff_t offset,
4468                            loff_t len, int mode)
4469{
4470        struct inode *inode = file_inode(file);
4471        handle_t *handle = NULL;
4472        unsigned int max_blocks;
4473        loff_t new_size = 0;
4474        int ret = 0;
4475        int flags;
4476        int credits;
4477        int partial_begin, partial_end;
4478        loff_t start, end;
4479        ext4_lblk_t lblk;
4480        unsigned int blkbits = inode->i_blkbits;
4481
4482        trace_ext4_zero_range(inode, offset, len, mode);
4483
4484        /* Call ext4_force_commit to flush all data in case of data=journal. */
4485        if (ext4_should_journal_data(inode)) {
4486                ret = ext4_force_commit(inode->i_sb);
4487                if (ret)
4488                        return ret;
4489        }
4490
4491        /*
4492         * Round up offset. This is not fallocate, we need to zero out
4493         * blocks, so convert interior block aligned part of the range to
4494         * unwritten and possibly manually zero out unaligned parts of the
4495         * range.
4496         */
4497        start = round_up(offset, 1 << blkbits);
4498        end = round_down((offset + len), 1 << blkbits);
4499
4500        if (start < offset || end > offset + len)
4501                return -EINVAL;
4502        partial_begin = offset & ((1 << blkbits) - 1);
4503        partial_end = (offset + len) & ((1 << blkbits) - 1);
4504
4505        lblk = start >> blkbits;
4506        max_blocks = (end >> blkbits);
4507        if (max_blocks < lblk)
4508                max_blocks = 0;
4509        else
4510                max_blocks -= lblk;
4511
4512        inode_lock(inode);
4513
4514        /*
4515         * Indirect files do not support unwritten extents
4516         */
4517        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4518                ret = -EOPNOTSUPP;
4519                goto out_mutex;
4520        }
4521
4522        if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4523            (offset + len > inode->i_size ||
4524             offset + len > EXT4_I(inode)->i_disksize)) {
4525                new_size = offset + len;
4526                ret = inode_newsize_ok(inode, new_size);
4527                if (ret)
4528                        goto out_mutex;
4529        }
4530
4531        flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4532
4533        /* Wait all existing dio workers, newcomers will block on i_mutex */
4534        inode_dio_wait(inode);
4535
4536        /* Preallocate the range including the unaligned edges */
4537        if (partial_begin || partial_end) {
4538                ret = ext4_alloc_file_blocks(file,
4539                                round_down(offset, 1 << blkbits) >> blkbits,
4540                                (round_up((offset + len), 1 << blkbits) -
4541                                 round_down(offset, 1 << blkbits)) >> blkbits,
4542                                new_size, flags);
4543                if (ret)
4544                        goto out_mutex;
4545
4546        }
4547
4548        /* Zero range excluding the unaligned edges */
4549        if (max_blocks > 0) {
4550                flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4551                          EXT4_EX_NOCACHE);
4552
4553                /*
4554                 * Prevent page faults from reinstantiating pages we have
4555                 * released from page cache.
4556                 */
4557                down_write(&EXT4_I(inode)->i_mmap_sem);
4558
4559                ret = ext4_break_layouts(inode);
4560                if (ret) {
4561                        up_write(&EXT4_I(inode)->i_mmap_sem);
4562                        goto out_mutex;
4563                }
4564
4565                ret = ext4_update_disksize_before_punch(inode, offset, len);
4566                if (ret) {
4567                        up_write(&EXT4_I(inode)->i_mmap_sem);
4568                        goto out_mutex;
4569                }
4570                /* Now release the pages and zero block aligned part of pages */
4571                truncate_pagecache_range(inode, start, end - 1);
4572                inode->i_mtime = inode->i_ctime = current_time(inode);
4573
4574                ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4575                                             flags);
4576                up_write(&EXT4_I(inode)->i_mmap_sem);
4577                if (ret)
4578                        goto out_mutex;
4579        }
4580        if (!partial_begin && !partial_end)
4581                goto out_mutex;
4582
4583        /*
4584         * In worst case we have to writeout two nonadjacent unwritten
4585         * blocks and update the inode
4586         */
4587        credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4588        if (ext4_should_journal_data(inode))
4589                credits += 2;
4590        handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4591        if (IS_ERR(handle)) {
4592                ret = PTR_ERR(handle);
4593                ext4_std_error(inode->i_sb, ret);
4594                goto out_mutex;
4595        }
4596
4597        inode->i_mtime = inode->i_ctime = current_time(inode);
4598        if (new_size)
4599                ext4_update_inode_size(inode, new_size);
4600        ret = ext4_mark_inode_dirty(handle, inode);
4601        if (unlikely(ret))
4602                goto out_handle;
4603
4604        /* Zero out partial block at the edges of the range */
4605        ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4606        if (ret >= 0)
4607                ext4_update_inode_fsync_trans(handle, inode, 1);
4608
4609        if (file->f_flags & O_SYNC)
4610                ext4_handle_sync(handle);
4611
4612out_handle:
4613        ext4_journal_stop(handle);
4614out_mutex:
4615        inode_unlock(inode);
4616        return ret;
4617}
4618
4619/*
4620 * preallocate space for a file. This implements ext4's fallocate file
4621 * operation, which gets called from sys_fallocate system call.
4622 * For block-mapped files, posix_fallocate should fall back to the method
4623 * of writing zeroes to the required new blocks (the same behavior which is
4624 * expected for file systems which do not support fallocate() system call).
4625 */
4626long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4627{
4628        struct inode *inode = file_inode(file);
4629        loff_t new_size = 0;
4630        unsigned int max_blocks;
4631        int ret = 0;
4632        int flags;
4633        ext4_lblk_t lblk;
4634        unsigned int blkbits = inode->i_blkbits;
4635
4636        /*
4637         * Encrypted inodes can't handle collapse range or insert
4638         * range since we would need to re-encrypt blocks with a
4639         * different IV or XTS tweak (which are based on the logical
4640         * block number).
4641         */
4642        if (IS_ENCRYPTED(inode) &&
4643            (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4644                return -EOPNOTSUPP;
4645
4646        /* Return error if mode is not supported */
4647        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4648                     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4649                     FALLOC_FL_INSERT_RANGE))
4650                return -EOPNOTSUPP;
4651
4652        if (mode & FALLOC_FL_PUNCH_HOLE)
4653                return ext4_punch_hole(inode, offset, len);
4654
4655        ret = ext4_convert_inline_data(inode);
4656        if (ret)
4657                return ret;
4658
4659        if (mode & FALLOC_FL_COLLAPSE_RANGE)
4660                return ext4_collapse_range(inode, offset, len);
4661
4662        if (mode & FALLOC_FL_INSERT_RANGE)
4663                return ext4_insert_range(inode, offset, len);
4664
4665        if (mode & FALLOC_FL_ZERO_RANGE)
4666                return ext4_zero_range(file, offset, len, mode);
4667
4668        trace_ext4_fallocate_enter(inode, offset, len, mode);
4669        lblk = offset >> blkbits;
4670
4671        max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4672        flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4673
4674        inode_lock(inode);
4675
4676        /*
4677         * We only support preallocation for extent-based files only
4678         */
4679        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4680                ret = -EOPNOTSUPP;
4681                goto out;
4682        }
4683
4684        if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4685            (offset + len > inode->i_size ||
4686             offset + len > EXT4_I(inode)->i_disksize)) {
4687                new_size = offset + len;
4688                ret = inode_newsize_ok(inode, new_size);
4689                if (ret)
4690                        goto out;
4691        }
4692
4693        /* Wait all existing dio workers, newcomers will block on i_mutex */
4694        inode_dio_wait(inode);
4695
4696        ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4697        if (ret)
4698                goto out;
4699
4700        if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4701                ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4702                                                EXT4_I(inode)->i_sync_tid);
4703        }
4704out:
4705        inode_unlock(inode);
4706        trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4707        return ret;
4708}
4709
4710/*
4711 * This function convert a range of blocks to written extents
4712 * The caller of this function will pass the start offset and the size.
4713 * all unwritten extents within this range will be converted to
4714 * written extents.
4715 *
4716 * This function is called from the direct IO end io call back
4717 * function, to convert the fallocated extents after IO is completed.
4718 * Returns 0 on success.
4719 */
4720int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4721                                   loff_t offset, ssize_t len)
4722{
4723        unsigned int max_blocks;
4724        int ret = 0, ret2 = 0, ret3 = 0;
4725        struct ext4_map_blocks map;
4726        unsigned int blkbits = inode->i_blkbits;
4727        unsigned int credits = 0;
4728
4729        map.m_lblk = offset >> blkbits;
4730        max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4731
4732        if (!handle) {
4733                /*
4734                 * credits to insert 1 extent into extent tree
4735                 */
4736                credits = ext4_chunk_trans_blocks(inode, max_blocks);
4737        }
4738        while (ret >= 0 && ret < max_blocks) {
4739                map.m_lblk += ret;
4740                map.m_len = (max_blocks -= ret);
4741                if (credits) {
4742                        handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4743                                                    credits);
4744                        if (IS_ERR(handle)) {
4745                                ret = PTR_ERR(handle);
4746                                break;
4747                        }
4748                }
4749                ret = ext4_map_blocks(handle, inode, &map,
4750                                      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4751                if (ret <= 0)
4752                        ext4_warning(inode->i_sb,
4753                                     "inode #%lu: block %u: len %u: "
4754                                     "ext4_ext_map_blocks returned %d",
4755                                     inode->i_ino, map.m_lblk,
4756                                     map.m_len, ret);
4757                ret2 = ext4_mark_inode_dirty(handle, inode);
4758                if (credits) {
4759                        ret3 = ext4_journal_stop(handle);
4760                        if (unlikely(ret3))
4761                                ret2 = ret3;
4762                }
4763
4764                if (ret <= 0 || ret2)
4765                        break;
4766        }
4767        return ret > 0 ? ret2 : ret;
4768}
4769
4770int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4771{
4772        int ret, err = 0;
4773        struct ext4_io_end_vec *io_end_vec;
4774
4775        /*
4776         * This is somewhat ugly but the idea is clear: When transaction is
4777         * reserved, everything goes into it. Otherwise we rather start several
4778         * smaller transactions for conversion of each extent separately.
4779         */
4780        if (handle) {
4781                handle = ext4_journal_start_reserved(handle,
4782                                                     EXT4_HT_EXT_CONVERT);
4783                if (IS_ERR(handle))
4784                        return PTR_ERR(handle);
4785        }
4786
4787        list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4788                ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4789                                                     io_end_vec->offset,
4790                                                     io_end_vec->size);
4791                if (ret)
4792                        break;
4793        }
4794
4795        if (handle)
4796                err = ext4_journal_stop(handle);
4797
4798        return ret < 0 ? ret : err;
4799}
4800
4801static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4802{
4803        __u64 physical = 0;
4804        __u64 length = 0;
4805        int blockbits = inode->i_sb->s_blocksize_bits;
4806        int error = 0;
4807        u16 iomap_type;
4808
4809        /* in-inode? */
4810        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4811                struct ext4_iloc iloc;
4812                int offset;     /* offset of xattr in inode */
4813
4814                error = ext4_get_inode_loc(inode, &iloc);
4815                if (error)
4816                        return error;
4817                physical = (__u64)iloc.bh->b_blocknr << blockbits;
4818                offset = EXT4_GOOD_OLD_INODE_SIZE +
4819                                EXT4_I(inode)->i_extra_isize;
4820                physical += offset;
4821                length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4822                brelse(iloc.bh);
4823                iomap_type = IOMAP_INLINE;
4824        } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4825                physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4826                length = inode->i_sb->s_blocksize;
4827                iomap_type = IOMAP_MAPPED;
4828        } else {
4829                /* no in-inode or external block for xattr, so return -ENOENT */
4830                error = -ENOENT;
4831                goto out;
4832        }
4833
4834        iomap->addr = physical;
4835        iomap->offset = 0;
4836        iomap->length = length;
4837        iomap->type = iomap_type;
4838        iomap->flags = 0;
4839out:
4840        return error;
4841}
4842
4843static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4844                                  loff_t length, unsigned flags,
4845                                  struct iomap *iomap, struct iomap *srcmap)
4846{
4847        int error;
4848
4849        error = ext4_iomap_xattr_fiemap(inode, iomap);
4850        if (error == 0 && (offset >= iomap->length))
4851                error = -ENOENT;
4852        return error;
4853}
4854
4855static const struct iomap_ops ext4_iomap_xattr_ops = {
4856        .iomap_begin            = ext4_iomap_xattr_begin,
4857};
4858
4859static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4860{
4861        u64 maxbytes;
4862
4863        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4864                maxbytes = inode->i_sb->s_maxbytes;
4865        else
4866                maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4867
4868        if (*len == 0)
4869                return -EINVAL;
4870        if (start > maxbytes)
4871                return -EFBIG;
4872
4873        /*
4874         * Shrink request scope to what the fs can actually handle.
4875         */
4876        if (*len > maxbytes || (maxbytes - *len) < start)
4877                *len = maxbytes - start;
4878        return 0;
4879}
4880
4881int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4882                u64 start, u64 len)
4883{
4884        int error = 0;
4885
4886        if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4887                error = ext4_ext_precache(inode);
4888                if (error)
4889                        return error;
4890                fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4891        }
4892
4893        /*
4894         * For bitmap files the maximum size limit could be smaller than
4895         * s_maxbytes, so check len here manually instead of just relying on the
4896         * generic check.
4897         */
4898        error = ext4_fiemap_check_ranges(inode, start, &len);
4899        if (error)
4900                return error;
4901
4902        if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4903                fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4904                return iomap_fiemap(inode, fieinfo, start, len,
4905                                    &ext4_iomap_xattr_ops);
4906        }
4907
4908        return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4909}
4910
4911int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4912                      __u64 start, __u64 len)
4913{
4914        ext4_lblk_t start_blk, len_blks;
4915        __u64 last_blk;
4916        int error = 0;
4917
4918        if (ext4_has_inline_data(inode)) {
4919                int has_inline;
4920
4921                down_read(&EXT4_I(inode)->xattr_sem);
4922                has_inline = ext4_has_inline_data(inode);
4923                up_read(&EXT4_I(inode)->xattr_sem);
4924                if (has_inline)
4925                        return 0;
4926        }
4927
4928        if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4929                error = ext4_ext_precache(inode);
4930                if (error)
4931                        return error;
4932                fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4933        }
4934
4935        error = fiemap_prep(inode, fieinfo, start, &len, 0);
4936        if (error)
4937                return error;
4938
4939        error = ext4_fiemap_check_ranges(inode, start, &len);
4940        if (error)
4941                return error;
4942
4943        start_blk = start >> inode->i_sb->s_blocksize_bits;
4944        last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4945        if (last_blk >= EXT_MAX_BLOCKS)
4946                last_blk = EXT_MAX_BLOCKS-1;
4947        len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4948
4949        /*
4950         * Walk the extent tree gathering extent information
4951         * and pushing extents back to the user.
4952         */
4953        return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
4954}
4955
4956/*
4957 * ext4_access_path:
4958 * Function to access the path buffer for marking it dirty.
4959 * It also checks if there are sufficient credits left in the journal handle
4960 * to update path.
4961 */
4962static int
4963ext4_access_path(handle_t *handle, struct inode *inode,
4964                struct ext4_ext_path *path)
4965{
4966        int credits, err;
4967
4968        if (!ext4_handle_valid(handle))
4969                return 0;
4970
4971        /*
4972         * Check if need to extend journal credits
4973         * 3 for leaf, sb, and inode plus 2 (bmap and group
4974         * descriptor) for each block group; assume two block
4975         * groups
4976         */
4977        credits = ext4_writepage_trans_blocks(inode);
4978        err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
4979        if (err < 0)
4980                return err;
4981
4982        err = ext4_ext_get_access(handle, inode, path);
4983        return err;
4984}
4985
4986/*
4987 * ext4_ext_shift_path_extents:
4988 * Shift the extents of a path structure lying between path[depth].p_ext
4989 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
4990 * if it is right shift or left shift operation.
4991 */
4992static int
4993ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
4994                            struct inode *inode, handle_t *handle,
4995                            enum SHIFT_DIRECTION SHIFT)
4996{
4997        int depth, err = 0;
4998        struct ext4_extent *ex_start, *ex_last;
4999        bool update = false;
5000        depth = path->p_depth;
5001
5002        while (depth >= 0) {
5003                if (depth == path->p_depth) {
5004                        ex_start = path[depth].p_ext;
5005                        if (!ex_start)
5006                                return -EFSCORRUPTED;
5007
5008                        ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5009
5010                        err = ext4_access_path(handle, inode, path + depth);
5011                        if (err)
5012                                goto out;
5013
5014                        if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5015                                update = true;
5016
5017                        while (ex_start <= ex_last) {
5018                                if (SHIFT == SHIFT_LEFT) {
5019                                        le32_add_cpu(&ex_start->ee_block,
5020                                                -shift);
5021                                        /* Try to merge to the left. */
5022                                        if ((ex_start >
5023                                            EXT_FIRST_EXTENT(path[depth].p_hdr))
5024                                            &&
5025                                            ext4_ext_try_to_merge_right(inode,
5026                                            path, ex_start - 1))
5027                                                ex_last--;
5028                                        else
5029                                                ex_start++;
5030                                } else {
5031                                        le32_add_cpu(&ex_last->ee_block, shift);
5032                                        ext4_ext_try_to_merge_right(inode, path,
5033                                                ex_last);
5034                                        ex_last--;
5035                                }
5036                        }
5037                        err = ext4_ext_dirty(handle, inode, path + depth);
5038                        if (err)
5039                                goto out;
5040
5041                        if (--depth < 0 || !update)
5042                                break;
5043                }
5044
5045                /* Update index too */
5046                err = ext4_access_path(handle, inode, path + depth);
5047                if (err)
5048                        goto out;
5049
5050                if (SHIFT == SHIFT_LEFT)
5051                        le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5052                else
5053                        le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5054                err = ext4_ext_dirty(handle, inode, path + depth);
5055                if (err)
5056                        goto out;
5057
5058                /* we are done if current index is not a starting index */
5059                if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5060                        break;
5061
5062                depth--;
5063        }
5064
5065out:
5066        return err;
5067}
5068
5069/*
5070 * ext4_ext_shift_extents:
5071 * All the extents which lies in the range from @start to the last allocated
5072 * block for the @inode are shifted either towards left or right (depending
5073 * upon @SHIFT) by @shift blocks.
5074 * On success, 0 is returned, error otherwise.
5075 */
5076static int
5077ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5078                       ext4_lblk_t start, ext4_lblk_t shift,
5079                       enum SHIFT_DIRECTION SHIFT)
5080{
5081        struct ext4_ext_path *path;
5082        int ret = 0, depth;
5083        struct ext4_extent *extent;
5084        ext4_lblk_t stop, *iterator, ex_start, ex_end;
5085
5086        /* Let path point to the last extent */
5087        path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5088                                EXT4_EX_NOCACHE);
5089        if (IS_ERR(path))
5090                return PTR_ERR(path);
5091
5092        depth = path->p_depth;
5093        extent = path[depth].p_ext;
5094        if (!extent)
5095                goto out;
5096
5097        stop = le32_to_cpu(extent->ee_block);
5098
5099       /*
5100        * For left shifts, make sure the hole on the left is big enough to
5101        * accommodate the shift.  For right shifts, make sure the last extent
5102        * won't be shifted beyond EXT_MAX_BLOCKS.
5103        */
5104        if (SHIFT == SHIFT_LEFT) {
5105                path = ext4_find_extent(inode, start - 1, &path,
5106                                        EXT4_EX_NOCACHE);
5107                if (IS_ERR(path))
5108                        return PTR_ERR(path);
5109                depth = path->p_depth;
5110                extent =  path[depth].p_ext;
5111                if (extent) {
5112                        ex_start = le32_to_cpu(extent->ee_block);
5113                        ex_end = le32_to_cpu(extent->ee_block) +
5114                                ext4_ext_get_actual_len(extent);
5115                } else {
5116                        ex_start = 0;
5117                        ex_end = 0;
5118                }
5119
5120                if ((start == ex_start && shift > ex_start) ||
5121                    (shift > start - ex_end)) {
5122                        ret = -EINVAL;
5123                        goto out;
5124                }
5125        } else {
5126                if (shift > EXT_MAX_BLOCKS -
5127                    (stop + ext4_ext_get_actual_len(extent))) {
5128                        ret = -EINVAL;
5129                        goto out;
5130                }
5131        }
5132
5133        /*
5134         * In case of left shift, iterator points to start and it is increased
5135         * till we reach stop. In case of right shift, iterator points to stop
5136         * and it is decreased till we reach start.
5137         */
5138        if (SHIFT == SHIFT_LEFT)
5139                iterator = &start;
5140        else
5141                iterator = &stop;
5142
5143        /*
5144         * Its safe to start updating extents.  Start and stop are unsigned, so
5145         * in case of right shift if extent with 0 block is reached, iterator
5146         * becomes NULL to indicate the end of the loop.
5147         */
5148        while (iterator && start <= stop) {
5149                path = ext4_find_extent(inode, *iterator, &path,
5150                                        EXT4_EX_NOCACHE);
5151                if (IS_ERR(path))
5152                        return PTR_ERR(path);
5153                depth = path->p_depth;
5154                extent = path[depth].p_ext;
5155                if (!extent) {
5156                        EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5157                                         (unsigned long) *iterator);
5158                        return -EFSCORRUPTED;
5159                }
5160                if (SHIFT == SHIFT_LEFT && *iterator >
5161                    le32_to_cpu(extent->ee_block)) {
5162                        /* Hole, move to the next extent */
5163                        if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5164                                path[depth].p_ext++;
5165                        } else {
5166                                *iterator = ext4_ext_next_allocated_block(path);
5167                                continue;
5168                        }
5169                }
5170
5171                if (SHIFT == SHIFT_LEFT) {
5172                        extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5173                        *iterator = le32_to_cpu(extent->ee_block) +
5174                                        ext4_ext_get_actual_len(extent);
5175                } else {
5176                        extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5177                        if (le32_to_cpu(extent->ee_block) > 0)
5178                                *iterator = le32_to_cpu(extent->ee_block) - 1;
5179                        else
5180                                /* Beginning is reached, end of the loop */
5181                                iterator = NULL;
5182                        /* Update path extent in case we need to stop */
5183                        while (le32_to_cpu(extent->ee_block) < start)
5184                                extent++;
5185                        path[depth].p_ext = extent;
5186                }
5187                ret = ext4_ext_shift_path_extents(path, shift, inode,
5188                                handle, SHIFT);
5189                if (ret)
5190                        break;
5191        }
5192out:
5193        ext4_ext_drop_refs(path);
5194        kfree(path);
5195        return ret;
5196}
5197
5198/*
5199 * ext4_collapse_range:
5200 * This implements the fallocate's collapse range functionality for ext4
5201 * Returns: 0 and non-zero on error.
5202 */
5203static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5204{
5205        struct super_block *sb = inode->i_sb;
5206        ext4_lblk_t punch_start, punch_stop;
5207        handle_t *handle;
5208        unsigned int credits;
5209        loff_t new_size, ioffset;
5210        int ret;
5211
5212        /*
5213         * We need to test this early because xfstests assumes that a
5214         * collapse range of (0, 1) will return EOPNOTSUPP if the file
5215         * system does not support collapse range.
5216         */
5217        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5218                return -EOPNOTSUPP;
5219
5220        /* Collapse range works only on fs cluster size aligned regions. */
5221        if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5222                return -EINVAL;
5223
5224        trace_ext4_collapse_range(inode, offset, len);
5225
5226        punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5227        punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5228
5229        /* Call ext4_force_commit to flush all data in case of data=journal. */
5230        if (ext4_should_journal_data(inode)) {
5231                ret = ext4_force_commit(inode->i_sb);
5232                if (ret)
5233                        return ret;
5234        }
5235
5236        inode_lock(inode);
5237        /*
5238         * There is no need to overlap collapse range with EOF, in which case
5239         * it is effectively a truncate operation
5240         */
5241        if (offset + len >= inode->i_size) {
5242                ret = -EINVAL;
5243                goto out_mutex;
5244        }
5245
5246        /* Currently just for extent based files */
5247        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5248                ret = -EOPNOTSUPP;
5249                goto out_mutex;
5250        }
5251
5252        /* Wait for existing dio to complete */
5253        inode_dio_wait(inode);
5254
5255        /*
5256         * Prevent page faults from reinstantiating pages we have released from
5257         * page cache.
5258         */
5259        down_write(&EXT4_I(inode)->i_mmap_sem);
5260
5261        ret = ext4_break_layouts(inode);
5262        if (ret)
5263                goto out_mmap;
5264
5265        /*
5266         * Need to round down offset to be aligned with page size boundary
5267         * for page size > block size.
5268         */
5269        ioffset = round_down(offset, PAGE_SIZE);
5270        /*
5271         * Write tail of the last page before removed range since it will get
5272         * removed from the page cache below.
5273         */
5274        ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5275        if (ret)
5276                goto out_mmap;
5277        /*
5278         * Write data that will be shifted to preserve them when discarding
5279         * page cache below. We are also protected from pages becoming dirty
5280         * by i_mmap_sem.
5281         */
5282        ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5283                                           LLONG_MAX);
5284        if (ret)
5285                goto out_mmap;
5286        truncate_pagecache(inode, ioffset);
5287
5288        credits = ext4_writepage_trans_blocks(inode);
5289        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5290        if (IS_ERR(handle)) {
5291                ret = PTR_ERR(handle);
5292                goto out_mmap;
5293        }
5294
5295        down_write(&EXT4_I(inode)->i_data_sem);
5296        ext4_discard_preallocations(inode, 0);
5297
5298        ret = ext4_es_remove_extent(inode, punch_start,
5299                                    EXT_MAX_BLOCKS - punch_start);
5300        if (ret) {
5301                up_write(&EXT4_I(inode)->i_data_sem);
5302                goto out_stop;
5303        }
5304
5305        ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5306        if (ret) {
5307                up_write(&EXT4_I(inode)->i_data_sem);
5308                goto out_stop;
5309        }
5310        ext4_discard_preallocations(inode, 0);
5311
5312        ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5313                                     punch_stop - punch_start, SHIFT_LEFT);
5314        if (ret) {
5315                up_write(&EXT4_I(inode)->i_data_sem);
5316                goto out_stop;
5317        }
5318
5319        new_size = inode->i_size - len;
5320        i_size_write(inode, new_size);
5321        EXT4_I(inode)->i_disksize = new_size;
5322
5323        up_write(&EXT4_I(inode)->i_data_sem);
5324        if (IS_SYNC(inode))
5325                ext4_handle_sync(handle);
5326        inode->i_mtime = inode->i_ctime = current_time(inode);
5327        ret = ext4_mark_inode_dirty(handle, inode);
5328        ext4_update_inode_fsync_trans(handle, inode, 1);
5329
5330out_stop:
5331        ext4_journal_stop(handle);
5332out_mmap:
5333        up_write(&EXT4_I(inode)->i_mmap_sem);
5334out_mutex:
5335        inode_unlock(inode);
5336        return ret;
5337}
5338
5339/*
5340 * ext4_insert_range:
5341 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5342 * The data blocks starting from @offset to the EOF are shifted by @len
5343 * towards right to create a hole in the @inode. Inode size is increased
5344 * by len bytes.
5345 * Returns 0 on success, error otherwise.
5346 */
5347static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5348{
5349        struct super_block *sb = inode->i_sb;
5350        handle_t *handle;
5351        struct ext4_ext_path *path;
5352        struct ext4_extent *extent;
5353        ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5354        unsigned int credits, ee_len;
5355        int ret = 0, depth, split_flag = 0;
5356        loff_t ioffset;
5357
5358        /*
5359         * We need to test this early because xfstests assumes that an
5360         * insert range of (0, 1) will return EOPNOTSUPP if the file
5361         * system does not support insert range.
5362         */
5363        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5364                return -EOPNOTSUPP;
5365
5366        /* Insert range works only on fs cluster size aligned regions. */
5367        if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5368                return -EINVAL;
5369
5370        trace_ext4_insert_range(inode, offset, len);
5371
5372        offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5373        len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5374
5375        /* Call ext4_force_commit to flush all data in case of data=journal */
5376        if (ext4_should_journal_data(inode)) {
5377                ret = ext4_force_commit(inode->i_sb);
5378                if (ret)
5379                        return ret;
5380        }
5381
5382        inode_lock(inode);
5383        /* Currently just for extent based files */
5384        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5385                ret = -EOPNOTSUPP;
5386                goto out_mutex;
5387        }
5388
5389        /* Check whether the maximum file size would be exceeded */
5390        if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5391                ret = -EFBIG;
5392                goto out_mutex;
5393        }
5394
5395        /* Offset must be less than i_size */
5396        if (offset >= inode->i_size) {
5397                ret = -EINVAL;
5398                goto out_mutex;
5399        }
5400
5401        /* Wait for existing dio to complete */
5402        inode_dio_wait(inode);
5403
5404        /*
5405         * Prevent page faults from reinstantiating pages we have released from
5406         * page cache.
5407         */
5408        down_write(&EXT4_I(inode)->i_mmap_sem);
5409
5410        ret = ext4_break_layouts(inode);
5411        if (ret)
5412                goto out_mmap;
5413
5414        /*
5415         * Need to round down to align start offset to page size boundary
5416         * for page size > block size.
5417         */
5418        ioffset = round_down(offset, PAGE_SIZE);
5419        /* Write out all dirty pages */
5420        ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5421                        LLONG_MAX);
5422        if (ret)
5423                goto out_mmap;
5424        truncate_pagecache(inode, ioffset);
5425
5426        credits = ext4_writepage_trans_blocks(inode);
5427        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5428        if (IS_ERR(handle)) {
5429                ret = PTR_ERR(handle);
5430                goto out_mmap;
5431        }
5432
5433        /* Expand file to avoid data loss if there is error while shifting */
5434        inode->i_size += len;
5435        EXT4_I(inode)->i_disksize += len;
5436        inode->i_mtime = inode->i_ctime = current_time(inode);
5437        ret = ext4_mark_inode_dirty(handle, inode);
5438        if (ret)
5439                goto out_stop;
5440
5441        down_write(&EXT4_I(inode)->i_data_sem);
5442        ext4_discard_preallocations(inode, 0);
5443
5444        path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5445        if (IS_ERR(path)) {
5446                up_write(&EXT4_I(inode)->i_data_sem);
5447                goto out_stop;
5448        }
5449
5450        depth = ext_depth(inode);
5451        extent = path[depth].p_ext;
5452        if (extent) {
5453                ee_start_lblk = le32_to_cpu(extent->ee_block);
5454                ee_len = ext4_ext_get_actual_len(extent);
5455
5456                /*
5457                 * If offset_lblk is not the starting block of extent, split
5458                 * the extent @offset_lblk
5459                 */
5460                if ((offset_lblk > ee_start_lblk) &&
5461                                (offset_lblk < (ee_start_lblk + ee_len))) {
5462                        if (ext4_ext_is_unwritten(extent))
5463                                split_flag = EXT4_EXT_MARK_UNWRIT1 |
5464                                        EXT4_EXT_MARK_UNWRIT2;
5465                        ret = ext4_split_extent_at(handle, inode, &path,
5466                                        offset_lblk, split_flag,
5467                                        EXT4_EX_NOCACHE |
5468                                        EXT4_GET_BLOCKS_PRE_IO |
5469                                        EXT4_GET_BLOCKS_METADATA_NOFAIL);
5470                }
5471
5472                ext4_ext_drop_refs(path);
5473                kfree(path);
5474                if (ret < 0) {
5475                        up_write(&EXT4_I(inode)->i_data_sem);
5476                        goto out_stop;
5477                }
5478        } else {
5479                ext4_ext_drop_refs(path);
5480                kfree(path);
5481        }
5482
5483        ret = ext4_es_remove_extent(inode, offset_lblk,
5484                        EXT_MAX_BLOCKS - offset_lblk);
5485        if (ret) {
5486                up_write(&EXT4_I(inode)->i_data_sem);
5487                goto out_stop;
5488        }
5489
5490        /*
5491         * if offset_lblk lies in a hole which is at start of file, use
5492         * ee_start_lblk to shift extents
5493         */
5494        ret = ext4_ext_shift_extents(inode, handle,
5495                ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5496                len_lblk, SHIFT_RIGHT);
5497
5498        up_write(&EXT4_I(inode)->i_data_sem);
5499        if (IS_SYNC(inode))
5500                ext4_handle_sync(handle);
5501        if (ret >= 0)
5502                ext4_update_inode_fsync_trans(handle, inode, 1);
5503
5504out_stop:
5505        ext4_journal_stop(handle);
5506out_mmap:
5507        up_write(&EXT4_I(inode)->i_mmap_sem);
5508out_mutex:
5509        inode_unlock(inode);
5510        return ret;
5511}
5512
5513/**
5514 * ext4_swap_extents() - Swap extents between two inodes
5515 * @handle: handle for this transaction
5516 * @inode1:     First inode
5517 * @inode2:     Second inode
5518 * @lblk1:      Start block for first inode
5519 * @lblk2:      Start block for second inode
5520 * @count:      Number of blocks to swap
5521 * @unwritten: Mark second inode's extents as unwritten after swap
5522 * @erp:        Pointer to save error value
5523 *
5524 * This helper routine does exactly what is promise "swap extents". All other
5525 * stuff such as page-cache locking consistency, bh mapping consistency or
5526 * extent's data copying must be performed by caller.
5527 * Locking:
5528 *              i_mutex is held for both inodes
5529 *              i_data_sem is locked for write for both inodes
5530 * Assumptions:
5531 *              All pages from requested range are locked for both inodes
5532 */
5533int
5534ext4_swap_extents(handle_t *handle, struct inode *inode1,
5535                  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5536                  ext4_lblk_t count, int unwritten, int *erp)
5537{
5538        struct ext4_ext_path *path1 = NULL;
5539        struct ext4_ext_path *path2 = NULL;
5540        int replaced_count = 0;
5541
5542        BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5543        BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5544        BUG_ON(!inode_is_locked(inode1));
5545        BUG_ON(!inode_is_locked(inode2));
5546
5547        *erp = ext4_es_remove_extent(inode1, lblk1, count);
5548        if (unlikely(*erp))
5549                return 0;
5550        *erp = ext4_es_remove_extent(inode2, lblk2, count);
5551        if (unlikely(*erp))
5552                return 0;
5553
5554        while (count) {
5555                struct ext4_extent *ex1, *ex2, tmp_ex;
5556                ext4_lblk_t e1_blk, e2_blk;
5557                int e1_len, e2_len, len;
5558                int split = 0;
5559
5560                path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5561                if (IS_ERR(path1)) {
5562                        *erp = PTR_ERR(path1);
5563                        path1 = NULL;
5564                finish:
5565                        count = 0;
5566                        goto repeat;
5567                }
5568                path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5569                if (IS_ERR(path2)) {
5570                        *erp = PTR_ERR(path2);
5571                        path2 = NULL;
5572                        goto finish;
5573                }
5574                ex1 = path1[path1->p_depth].p_ext;
5575                ex2 = path2[path2->p_depth].p_ext;
5576                /* Do we have something to swap ? */
5577                if (unlikely(!ex2 || !ex1))
5578                        goto finish;
5579
5580                e1_blk = le32_to_cpu(ex1->ee_block);
5581                e2_blk = le32_to_cpu(ex2->ee_block);
5582                e1_len = ext4_ext_get_actual_len(ex1);
5583                e2_len = ext4_ext_get_actual_len(ex2);
5584
5585                /* Hole handling */
5586                if (!in_range(lblk1, e1_blk, e1_len) ||
5587                    !in_range(lblk2, e2_blk, e2_len)) {
5588                        ext4_lblk_t next1, next2;
5589
5590                        /* if hole after extent, then go to next extent */
5591                        next1 = ext4_ext_next_allocated_block(path1);
5592                        next2 = ext4_ext_next_allocated_block(path2);
5593                        /* If hole before extent, then shift to that extent */
5594                        if (e1_blk > lblk1)
5595                                next1 = e1_blk;
5596                        if (e2_blk > lblk2)
5597                                next2 = e2_blk;
5598                        /* Do we have something to swap */
5599                        if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5600                                goto finish;
5601                        /* Move to the rightest boundary */
5602                        len = next1 - lblk1;
5603                        if (len < next2 - lblk2)
5604                                len = next2 - lblk2;
5605                        if (len > count)
5606                                len = count;
5607                        lblk1 += len;
5608                        lblk2 += len;
5609                        count -= len;
5610                        goto repeat;
5611                }
5612
5613                /* Prepare left boundary */
5614                if (e1_blk < lblk1) {
5615                        split = 1;
5616                        *erp = ext4_force_split_extent_at(handle, inode1,
5617                                                &path1, lblk1, 0);
5618                        if (unlikely(*erp))
5619                                goto finish;
5620                }
5621                if (e2_blk < lblk2) {
5622                        split = 1;
5623                        *erp = ext4_force_split_extent_at(handle, inode2,
5624                                                &path2,  lblk2, 0);
5625                        if (unlikely(*erp))
5626                                goto finish;
5627                }
5628                /* ext4_split_extent_at() may result in leaf extent split,
5629                 * path must to be revalidated. */
5630                if (split)
5631                        goto repeat;
5632
5633                /* Prepare right boundary */
5634                len = count;
5635                if (len > e1_blk + e1_len - lblk1)
5636                        len = e1_blk + e1_len - lblk1;
5637                if (len > e2_blk + e2_len - lblk2)
5638                        len = e2_blk + e2_len - lblk2;
5639
5640                if (len != e1_len) {
5641                        split = 1;
5642                        *erp = ext4_force_split_extent_at(handle, inode1,
5643                                                &path1, lblk1 + len, 0);
5644                        if (unlikely(*erp))
5645                                goto finish;
5646                }
5647                if (len != e2_len) {
5648                        split = 1;
5649                        *erp = ext4_force_split_extent_at(handle, inode2,
5650                                                &path2, lblk2 + len, 0);
5651                        if (*erp)
5652                                goto finish;
5653                }
5654                /* ext4_split_extent_at() may result in leaf extent split,
5655                 * path must to be revalidated. */
5656                if (split)
5657                        goto repeat;
5658
5659                BUG_ON(e2_len != e1_len);
5660                *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5661                if (unlikely(*erp))
5662                        goto finish;
5663                *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5664                if (unlikely(*erp))
5665                        goto finish;
5666
5667                /* Both extents are fully inside boundaries. Swap it now */
5668                tmp_ex = *ex1;
5669                ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5670                ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5671                ex1->ee_len = cpu_to_le16(e2_len);
5672                ex2->ee_len = cpu_to_le16(e1_len);
5673                if (unwritten)
5674                        ext4_ext_mark_unwritten(ex2);
5675                if (ext4_ext_is_unwritten(&tmp_ex))
5676                        ext4_ext_mark_unwritten(ex1);
5677
5678                ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5679                ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5680                *erp = ext4_ext_dirty(handle, inode2, path2 +
5681                                      path2->p_depth);
5682                if (unlikely(*erp))
5683                        goto finish;
5684                *erp = ext4_ext_dirty(handle, inode1, path1 +
5685                                      path1->p_depth);
5686                /*
5687                 * Looks scarry ah..? second inode already points to new blocks,
5688                 * and it was successfully dirtied. But luckily error may happen
5689                 * only due to journal error, so full transaction will be
5690                 * aborted anyway.
5691                 */
5692                if (unlikely(*erp))
5693                        goto finish;
5694                lblk1 += len;
5695                lblk2 += len;
5696                replaced_count += len;
5697                count -= len;
5698
5699        repeat:
5700                ext4_ext_drop_refs(path1);
5701                kfree(path1);
5702                ext4_ext_drop_refs(path2);
5703                kfree(path2);
5704                path1 = path2 = NULL;
5705        }
5706        return replaced_count;
5707}
5708
5709/*
5710 * ext4_clu_mapped - determine whether any block in a logical cluster has
5711 *                   been mapped to a physical cluster
5712 *
5713 * @inode - file containing the logical cluster
5714 * @lclu - logical cluster of interest
5715 *
5716 * Returns 1 if any block in the logical cluster is mapped, signifying
5717 * that a physical cluster has been allocated for it.  Otherwise,
5718 * returns 0.  Can also return negative error codes.  Derived from
5719 * ext4_ext_map_blocks().
5720 */
5721int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5722{
5723        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5724        struct ext4_ext_path *path;
5725        int depth, mapped = 0, err = 0;
5726        struct ext4_extent *extent;
5727        ext4_lblk_t first_lblk, first_lclu, last_lclu;
5728
5729        /* search for the extent closest to the first block in the cluster */
5730        path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5731        if (IS_ERR(path)) {
5732                err = PTR_ERR(path);
5733                path = NULL;
5734                goto out;
5735        }
5736
5737        depth = ext_depth(inode);
5738
5739        /*
5740         * A consistent leaf must not be empty.  This situation is possible,
5741         * though, _during_ tree modification, and it's why an assert can't
5742         * be put in ext4_find_extent().
5743         */
5744        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5745                EXT4_ERROR_INODE(inode,
5746                    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5747                                 (unsigned long) EXT4_C2B(sbi, lclu),
5748                                 depth, path[depth].p_block);
5749                err = -EFSCORRUPTED;
5750                goto out;
5751        }
5752
5753        extent = path[depth].p_ext;
5754
5755        /* can't be mapped if the extent tree is empty */
5756        if (extent == NULL)
5757                goto out;
5758
5759        first_lblk = le32_to_cpu(extent->ee_block);
5760        first_lclu = EXT4_B2C(sbi, first_lblk);
5761
5762        /*
5763         * Three possible outcomes at this point - found extent spanning
5764         * the target cluster, to the left of the target cluster, or to the
5765         * right of the target cluster.  The first two cases are handled here.
5766         * The last case indicates the target cluster is not mapped.
5767         */
5768        if (lclu >= first_lclu) {
5769                last_lclu = EXT4_B2C(sbi, first_lblk +
5770                                     ext4_ext_get_actual_len(extent) - 1);
5771                if (lclu <= last_lclu) {
5772                        mapped = 1;
5773                } else {
5774                        first_lblk = ext4_ext_next_allocated_block(path);
5775                        first_lclu = EXT4_B2C(sbi, first_lblk);
5776                        if (lclu == first_lclu)
5777                                mapped = 1;
5778                }
5779        }
5780
5781out:
5782        ext4_ext_drop_refs(path);
5783        kfree(path);
5784
5785        return err ? err : mapped;
5786}
5787