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 = ext4_read_bh(bh, 0, NULL);
 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 not exists, return 0 and @phys is set to 0. We will return
1477 * 1 which means we found an allocated block and ret_ex is valid.
1478 * Or return a (< 0) 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        if (ret_ex)
1578                *ret_ex = *ex;
1579        if (bh)
1580                put_bh(bh);
1581        return 1;
1582}
1583
1584/*
1585 * ext4_ext_next_allocated_block:
1586 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1587 * NOTE: it considers block number from index entry as
1588 * allocated block. Thus, index entries have to be consistent
1589 * with leaves.
1590 */
1591ext4_lblk_t
1592ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1593{
1594        int depth;
1595
1596        BUG_ON(path == NULL);
1597        depth = path->p_depth;
1598
1599        if (depth == 0 && path->p_ext == NULL)
1600                return EXT_MAX_BLOCKS;
1601
1602        while (depth >= 0) {
1603                struct ext4_ext_path *p = &path[depth];
1604
1605                if (depth == path->p_depth) {
1606                        /* leaf */
1607                        if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1608                                return le32_to_cpu(p->p_ext[1].ee_block);
1609                } else {
1610                        /* index */
1611                        if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1612                                return le32_to_cpu(p->p_idx[1].ei_block);
1613                }
1614                depth--;
1615        }
1616
1617        return EXT_MAX_BLOCKS;
1618}
1619
1620/*
1621 * ext4_ext_next_leaf_block:
1622 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1623 */
1624static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1625{
1626        int depth;
1627
1628        BUG_ON(path == NULL);
1629        depth = path->p_depth;
1630
1631        /* zero-tree has no leaf blocks at all */
1632        if (depth == 0)
1633                return EXT_MAX_BLOCKS;
1634
1635        /* go to index block */
1636        depth--;
1637
1638        while (depth >= 0) {
1639                if (path[depth].p_idx !=
1640                                EXT_LAST_INDEX(path[depth].p_hdr))
1641                        return (ext4_lblk_t)
1642                                le32_to_cpu(path[depth].p_idx[1].ei_block);
1643                depth--;
1644        }
1645
1646        return EXT_MAX_BLOCKS;
1647}
1648
1649/*
1650 * ext4_ext_correct_indexes:
1651 * if leaf gets modified and modified extent is first in the leaf,
1652 * then we have to correct all indexes above.
1653 * TODO: do we need to correct tree in all cases?
1654 */
1655static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1656                                struct ext4_ext_path *path)
1657{
1658        struct ext4_extent_header *eh;
1659        int depth = ext_depth(inode);
1660        struct ext4_extent *ex;
1661        __le32 border;
1662        int k, err = 0;
1663
1664        eh = path[depth].p_hdr;
1665        ex = path[depth].p_ext;
1666
1667        if (unlikely(ex == NULL || eh == NULL)) {
1668                EXT4_ERROR_INODE(inode,
1669                                 "ex %p == NULL or eh %p == NULL", ex, eh);
1670                return -EFSCORRUPTED;
1671        }
1672
1673        if (depth == 0) {
1674                /* there is no tree at all */
1675                return 0;
1676        }
1677
1678        if (ex != EXT_FIRST_EXTENT(eh)) {
1679                /* we correct tree if first leaf got modified only */
1680                return 0;
1681        }
1682
1683        /*
1684         * TODO: we need correction if border is smaller than current one
1685         */
1686        k = depth - 1;
1687        border = path[depth].p_ext->ee_block;
1688        err = ext4_ext_get_access(handle, inode, path + k);
1689        if (err)
1690                return err;
1691        path[k].p_idx->ei_block = border;
1692        err = ext4_ext_dirty(handle, inode, path + k);
1693        if (err)
1694                return err;
1695
1696        while (k--) {
1697                /* change all left-side indexes */
1698                if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1699                        break;
1700                err = ext4_ext_get_access(handle, inode, path + k);
1701                if (err)
1702                        break;
1703                path[k].p_idx->ei_block = border;
1704                err = ext4_ext_dirty(handle, inode, path + k);
1705                if (err)
1706                        break;
1707        }
1708
1709        return err;
1710}
1711
1712static int ext4_can_extents_be_merged(struct inode *inode,
1713                                      struct ext4_extent *ex1,
1714                                      struct ext4_extent *ex2)
1715{
1716        unsigned short ext1_ee_len, ext2_ee_len;
1717
1718        if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1719                return 0;
1720
1721        ext1_ee_len = ext4_ext_get_actual_len(ex1);
1722        ext2_ee_len = ext4_ext_get_actual_len(ex2);
1723
1724        if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1725                        le32_to_cpu(ex2->ee_block))
1726                return 0;
1727
1728        if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1729                return 0;
1730
1731        if (ext4_ext_is_unwritten(ex1) &&
1732            ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1733                return 0;
1734#ifdef AGGRESSIVE_TEST
1735        if (ext1_ee_len >= 4)
1736                return 0;
1737#endif
1738
1739        if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1740                return 1;
1741        return 0;
1742}
1743
1744/*
1745 * This function tries to merge the "ex" extent to the next extent in the tree.
1746 * It always tries to merge towards right. If you want to merge towards
1747 * left, pass "ex - 1" as argument instead of "ex".
1748 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1749 * 1 if they got merged.
1750 */
1751static int ext4_ext_try_to_merge_right(struct inode *inode,
1752                                 struct ext4_ext_path *path,
1753                                 struct ext4_extent *ex)
1754{
1755        struct ext4_extent_header *eh;
1756        unsigned int depth, len;
1757        int merge_done = 0, unwritten;
1758
1759        depth = ext_depth(inode);
1760        BUG_ON(path[depth].p_hdr == NULL);
1761        eh = path[depth].p_hdr;
1762
1763        while (ex < EXT_LAST_EXTENT(eh)) {
1764                if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1765                        break;
1766                /* merge with next extent! */
1767                unwritten = ext4_ext_is_unwritten(ex);
1768                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1769                                + ext4_ext_get_actual_len(ex + 1));
1770                if (unwritten)
1771                        ext4_ext_mark_unwritten(ex);
1772
1773                if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1774                        len = (EXT_LAST_EXTENT(eh) - ex - 1)
1775                                * sizeof(struct ext4_extent);
1776                        memmove(ex + 1, ex + 2, len);
1777                }
1778                le16_add_cpu(&eh->eh_entries, -1);
1779                merge_done = 1;
1780                WARN_ON(eh->eh_entries == 0);
1781                if (!eh->eh_entries)
1782                        EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1783        }
1784
1785        return merge_done;
1786}
1787
1788/*
1789 * This function does a very simple check to see if we can collapse
1790 * an extent tree with a single extent tree leaf block into the inode.
1791 */
1792static void ext4_ext_try_to_merge_up(handle_t *handle,
1793                                     struct inode *inode,
1794                                     struct ext4_ext_path *path)
1795{
1796        size_t s;
1797        unsigned max_root = ext4_ext_space_root(inode, 0);
1798        ext4_fsblk_t blk;
1799
1800        if ((path[0].p_depth != 1) ||
1801            (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1802            (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1803                return;
1804
1805        /*
1806         * We need to modify the block allocation bitmap and the block
1807         * group descriptor to release the extent tree block.  If we
1808         * can't get the journal credits, give up.
1809         */
1810        if (ext4_journal_extend(handle, 2,
1811                        ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1812                return;
1813
1814        /*
1815         * Copy the extent data up to the inode
1816         */
1817        blk = ext4_idx_pblock(path[0].p_idx);
1818        s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1819                sizeof(struct ext4_extent_idx);
1820        s += sizeof(struct ext4_extent_header);
1821
1822        path[1].p_maxdepth = path[0].p_maxdepth;
1823        memcpy(path[0].p_hdr, path[1].p_hdr, s);
1824        path[0].p_depth = 0;
1825        path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1826                (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1827        path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1828
1829        brelse(path[1].p_bh);
1830        ext4_free_blocks(handle, inode, NULL, blk, 1,
1831                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1832}
1833
1834/*
1835 * This function tries to merge the @ex extent to neighbours in the tree, then
1836 * tries to collapse the extent tree into the inode.
1837 */
1838static void ext4_ext_try_to_merge(handle_t *handle,
1839                                  struct inode *inode,
1840                                  struct ext4_ext_path *path,
1841                                  struct ext4_extent *ex)
1842{
1843        struct ext4_extent_header *eh;
1844        unsigned int depth;
1845        int merge_done = 0;
1846
1847        depth = ext_depth(inode);
1848        BUG_ON(path[depth].p_hdr == NULL);
1849        eh = path[depth].p_hdr;
1850
1851        if (ex > EXT_FIRST_EXTENT(eh))
1852                merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1853
1854        if (!merge_done)
1855                (void) ext4_ext_try_to_merge_right(inode, path, ex);
1856
1857        ext4_ext_try_to_merge_up(handle, inode, path);
1858}
1859
1860/*
1861 * check if a portion of the "newext" extent overlaps with an
1862 * existing extent.
1863 *
1864 * If there is an overlap discovered, it updates the length of the newext
1865 * such that there will be no overlap, and then returns 1.
1866 * If there is no overlap found, it returns 0.
1867 */
1868static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1869                                           struct inode *inode,
1870                                           struct ext4_extent *newext,
1871                                           struct ext4_ext_path *path)
1872{
1873        ext4_lblk_t b1, b2;
1874        unsigned int depth, len1;
1875        unsigned int ret = 0;
1876
1877        b1 = le32_to_cpu(newext->ee_block);
1878        len1 = ext4_ext_get_actual_len(newext);
1879        depth = ext_depth(inode);
1880        if (!path[depth].p_ext)
1881                goto out;
1882        b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1883
1884        /*
1885         * get the next allocated block if the extent in the path
1886         * is before the requested block(s)
1887         */
1888        if (b2 < b1) {
1889                b2 = ext4_ext_next_allocated_block(path);
1890                if (b2 == EXT_MAX_BLOCKS)
1891                        goto out;
1892                b2 = EXT4_LBLK_CMASK(sbi, b2);
1893        }
1894
1895        /* check for wrap through zero on extent logical start block*/
1896        if (b1 + len1 < b1) {
1897                len1 = EXT_MAX_BLOCKS - b1;
1898                newext->ee_len = cpu_to_le16(len1);
1899                ret = 1;
1900        }
1901
1902        /* check for overlap */
1903        if (b1 + len1 > b2) {
1904                newext->ee_len = cpu_to_le16(b2 - b1);
1905                ret = 1;
1906        }
1907out:
1908        return ret;
1909}
1910
1911/*
1912 * ext4_ext_insert_extent:
1913 * tries to merge requested extent into the existing extent or
1914 * inserts requested extent as new one into the tree,
1915 * creating new leaf in the no-space case.
1916 */
1917int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1918                                struct ext4_ext_path **ppath,
1919                                struct ext4_extent *newext, int gb_flags)
1920{
1921        struct ext4_ext_path *path = *ppath;
1922        struct ext4_extent_header *eh;
1923        struct ext4_extent *ex, *fex;
1924        struct ext4_extent *nearex; /* nearest extent */
1925        struct ext4_ext_path *npath = NULL;
1926        int depth, len, err;
1927        ext4_lblk_t next;
1928        int mb_flags = 0, unwritten;
1929
1930        if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1931                mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1932        if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1933                EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1934                return -EFSCORRUPTED;
1935        }
1936        depth = ext_depth(inode);
1937        ex = path[depth].p_ext;
1938        eh = path[depth].p_hdr;
1939        if (unlikely(path[depth].p_hdr == NULL)) {
1940                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1941                return -EFSCORRUPTED;
1942        }
1943
1944        /* try to insert block into found extent and return */
1945        if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1946
1947                /*
1948                 * Try to see whether we should rather test the extent on
1949                 * right from ex, or from the left of ex. This is because
1950                 * ext4_find_extent() can return either extent on the
1951                 * left, or on the right from the searched position. This
1952                 * will make merging more effective.
1953                 */
1954                if (ex < EXT_LAST_EXTENT(eh) &&
1955                    (le32_to_cpu(ex->ee_block) +
1956                    ext4_ext_get_actual_len(ex) <
1957                    le32_to_cpu(newext->ee_block))) {
1958                        ex += 1;
1959                        goto prepend;
1960                } else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1961                           (le32_to_cpu(newext->ee_block) +
1962                           ext4_ext_get_actual_len(newext) <
1963                           le32_to_cpu(ex->ee_block)))
1964                        ex -= 1;
1965
1966                /* Try to append newex to the ex */
1967                if (ext4_can_extents_be_merged(inode, ex, newext)) {
1968                        ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
1969                                  "(from %llu)\n",
1970                                  ext4_ext_is_unwritten(newext),
1971                                  ext4_ext_get_actual_len(newext),
1972                                  le32_to_cpu(ex->ee_block),
1973                                  ext4_ext_is_unwritten(ex),
1974                                  ext4_ext_get_actual_len(ex),
1975                                  ext4_ext_pblock(ex));
1976                        err = ext4_ext_get_access(handle, inode,
1977                                                  path + depth);
1978                        if (err)
1979                                return err;
1980                        unwritten = ext4_ext_is_unwritten(ex);
1981                        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1982                                        + ext4_ext_get_actual_len(newext));
1983                        if (unwritten)
1984                                ext4_ext_mark_unwritten(ex);
1985                        eh = path[depth].p_hdr;
1986                        nearex = ex;
1987                        goto merge;
1988                }
1989
1990prepend:
1991                /* Try to prepend newex to the ex */
1992                if (ext4_can_extents_be_merged(inode, newext, ex)) {
1993                        ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
1994                                  "(from %llu)\n",
1995                                  le32_to_cpu(newext->ee_block),
1996                                  ext4_ext_is_unwritten(newext),
1997                                  ext4_ext_get_actual_len(newext),
1998                                  le32_to_cpu(ex->ee_block),
1999                                  ext4_ext_is_unwritten(ex),
2000                                  ext4_ext_get_actual_len(ex),
2001                                  ext4_ext_pblock(ex));
2002                        err = ext4_ext_get_access(handle, inode,
2003                                                  path + depth);
2004                        if (err)
2005                                return err;
2006
2007                        unwritten = ext4_ext_is_unwritten(ex);
2008                        ex->ee_block = newext->ee_block;
2009                        ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2010                        ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2011                                        + ext4_ext_get_actual_len(newext));
2012                        if (unwritten)
2013                                ext4_ext_mark_unwritten(ex);
2014                        eh = path[depth].p_hdr;
2015                        nearex = ex;
2016                        goto merge;
2017                }
2018        }
2019
2020        depth = ext_depth(inode);
2021        eh = path[depth].p_hdr;
2022        if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2023                goto has_space;
2024
2025        /* probably next leaf has space for us? */
2026        fex = EXT_LAST_EXTENT(eh);
2027        next = EXT_MAX_BLOCKS;
2028        if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2029                next = ext4_ext_next_leaf_block(path);
2030        if (next != EXT_MAX_BLOCKS) {
2031                ext_debug(inode, "next leaf block - %u\n", next);
2032                BUG_ON(npath != NULL);
2033                npath = ext4_find_extent(inode, next, NULL, gb_flags);
2034                if (IS_ERR(npath))
2035                        return PTR_ERR(npath);
2036                BUG_ON(npath->p_depth != path->p_depth);
2037                eh = npath[depth].p_hdr;
2038                if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2039                        ext_debug(inode, "next leaf isn't full(%d)\n",
2040                                  le16_to_cpu(eh->eh_entries));
2041                        path = npath;
2042                        goto has_space;
2043                }
2044                ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2045                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2046        }
2047
2048        /*
2049         * There is no free space in the found leaf.
2050         * We're gonna add a new leaf in the tree.
2051         */
2052        if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2053                mb_flags |= EXT4_MB_USE_RESERVED;
2054        err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2055                                       ppath, newext);
2056        if (err)
2057                goto cleanup;
2058        depth = ext_depth(inode);
2059        eh = path[depth].p_hdr;
2060
2061has_space:
2062        nearex = path[depth].p_ext;
2063
2064        err = ext4_ext_get_access(handle, inode, path + depth);
2065        if (err)
2066                goto cleanup;
2067
2068        if (!nearex) {
2069                /* there is no extent in this leaf, create first one */
2070                ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2071                                le32_to_cpu(newext->ee_block),
2072                                ext4_ext_pblock(newext),
2073                                ext4_ext_is_unwritten(newext),
2074                                ext4_ext_get_actual_len(newext));
2075                nearex = EXT_FIRST_EXTENT(eh);
2076        } else {
2077                if (le32_to_cpu(newext->ee_block)
2078                           > le32_to_cpu(nearex->ee_block)) {
2079                        /* Insert after */
2080                        ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2081                                        "nearest %p\n",
2082                                        le32_to_cpu(newext->ee_block),
2083                                        ext4_ext_pblock(newext),
2084                                        ext4_ext_is_unwritten(newext),
2085                                        ext4_ext_get_actual_len(newext),
2086                                        nearex);
2087                        nearex++;
2088                } else {
2089                        /* Insert before */
2090                        BUG_ON(newext->ee_block == nearex->ee_block);
2091                        ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2092                                        "nearest %p\n",
2093                                        le32_to_cpu(newext->ee_block),
2094                                        ext4_ext_pblock(newext),
2095                                        ext4_ext_is_unwritten(newext),
2096                                        ext4_ext_get_actual_len(newext),
2097                                        nearex);
2098                }
2099                len = EXT_LAST_EXTENT(eh) - nearex + 1;
2100                if (len > 0) {
2101                        ext_debug(inode, "insert %u:%llu:[%d]%d: "
2102                                        "move %d extents from 0x%p to 0x%p\n",
2103                                        le32_to_cpu(newext->ee_block),
2104                                        ext4_ext_pblock(newext),
2105                                        ext4_ext_is_unwritten(newext),
2106                                        ext4_ext_get_actual_len(newext),
2107                                        len, nearex, nearex + 1);
2108                        memmove(nearex + 1, nearex,
2109                                len * sizeof(struct ext4_extent));
2110                }
2111        }
2112
2113        le16_add_cpu(&eh->eh_entries, 1);
2114        path[depth].p_ext = nearex;
2115        nearex->ee_block = newext->ee_block;
2116        ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2117        nearex->ee_len = newext->ee_len;
2118
2119merge:
2120        /* try to merge extents */
2121        if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2122                ext4_ext_try_to_merge(handle, inode, path, nearex);
2123
2124
2125        /* time to correct all indexes above */
2126        err = ext4_ext_correct_indexes(handle, inode, path);
2127        if (err)
2128                goto cleanup;
2129
2130        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2131
2132cleanup:
2133        ext4_ext_drop_refs(npath);
2134        kfree(npath);
2135        return err;
2136}
2137
2138static int ext4_fill_es_cache_info(struct inode *inode,
2139                                   ext4_lblk_t block, ext4_lblk_t num,
2140                                   struct fiemap_extent_info *fieinfo)
2141{
2142        ext4_lblk_t next, end = block + num - 1;
2143        struct extent_status es;
2144        unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2145        unsigned int flags;
2146        int err;
2147
2148        while (block <= end) {
2149                next = 0;
2150                flags = 0;
2151                if (!ext4_es_lookup_extent(inode, block, &next, &es))
2152                        break;
2153                if (ext4_es_is_unwritten(&es))
2154                        flags |= FIEMAP_EXTENT_UNWRITTEN;
2155                if (ext4_es_is_delayed(&es))
2156                        flags |= (FIEMAP_EXTENT_DELALLOC |
2157                                  FIEMAP_EXTENT_UNKNOWN);
2158                if (ext4_es_is_hole(&es))
2159                        flags |= EXT4_FIEMAP_EXTENT_HOLE;
2160                if (next == 0)
2161                        flags |= FIEMAP_EXTENT_LAST;
2162                if (flags & (FIEMAP_EXTENT_DELALLOC|
2163                             EXT4_FIEMAP_EXTENT_HOLE))
2164                        es.es_pblk = 0;
2165                else
2166                        es.es_pblk = ext4_es_pblock(&es);
2167                err = fiemap_fill_next_extent(fieinfo,
2168                                (__u64)es.es_lblk << blksize_bits,
2169                                (__u64)es.es_pblk << blksize_bits,
2170                                (__u64)es.es_len << blksize_bits,
2171                                flags);
2172                if (next == 0)
2173                        break;
2174                block = next;
2175                if (err < 0)
2176                        return err;
2177                if (err == 1)
2178                        return 0;
2179        }
2180        return 0;
2181}
2182
2183
2184/*
2185 * ext4_ext_determine_hole - determine hole around given block
2186 * @inode:      inode we lookup in
2187 * @path:       path in extent tree to @lblk
2188 * @lblk:       pointer to logical block around which we want to determine hole
2189 *
2190 * Determine hole length (and start if easily possible) around given logical
2191 * block. We don't try too hard to find the beginning of the hole but @path
2192 * actually points to extent before @lblk, we provide it.
2193 *
2194 * The function returns the length of a hole starting at @lblk. We update @lblk
2195 * to the beginning of the hole if we managed to find it.
2196 */
2197static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2198                                           struct ext4_ext_path *path,
2199                                           ext4_lblk_t *lblk)
2200{
2201        int depth = ext_depth(inode);
2202        struct ext4_extent *ex;
2203        ext4_lblk_t len;
2204
2205        ex = path[depth].p_ext;
2206        if (ex == NULL) {
2207                /* there is no extent yet, so gap is [0;-] */
2208                *lblk = 0;
2209                len = EXT_MAX_BLOCKS;
2210        } else if (*lblk < le32_to_cpu(ex->ee_block)) {
2211                len = le32_to_cpu(ex->ee_block) - *lblk;
2212        } else if (*lblk >= le32_to_cpu(ex->ee_block)
2213                        + ext4_ext_get_actual_len(ex)) {
2214                ext4_lblk_t next;
2215
2216                *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2217                next = ext4_ext_next_allocated_block(path);
2218                BUG_ON(next == *lblk);
2219                len = next - *lblk;
2220        } else {
2221                BUG();
2222        }
2223        return len;
2224}
2225
2226/*
2227 * ext4_ext_put_gap_in_cache:
2228 * calculate boundaries of the gap that the requested block fits into
2229 * and cache this gap
2230 */
2231static void
2232ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2233                          ext4_lblk_t hole_len)
2234{
2235        struct extent_status es;
2236
2237        ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2238                                  hole_start + hole_len - 1, &es);
2239        if (es.es_len) {
2240                /* There's delayed extent containing lblock? */
2241                if (es.es_lblk <= hole_start)
2242                        return;
2243                hole_len = min(es.es_lblk - hole_start, hole_len);
2244        }
2245        ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2246        ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2247                              EXTENT_STATUS_HOLE);
2248}
2249
2250/*
2251 * ext4_ext_rm_idx:
2252 * removes index from the index block.
2253 */
2254static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2255                        struct ext4_ext_path *path, int depth)
2256{
2257        int err;
2258        ext4_fsblk_t leaf;
2259
2260        /* free index block */
2261        depth--;
2262        path = path + depth;
2263        leaf = ext4_idx_pblock(path->p_idx);
2264        if (unlikely(path->p_hdr->eh_entries == 0)) {
2265                EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2266                return -EFSCORRUPTED;
2267        }
2268        err = ext4_ext_get_access(handle, inode, path);
2269        if (err)
2270                return err;
2271
2272        if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2273                int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2274                len *= sizeof(struct ext4_extent_idx);
2275                memmove(path->p_idx, path->p_idx + 1, len);
2276        }
2277
2278        le16_add_cpu(&path->p_hdr->eh_entries, -1);
2279        err = ext4_ext_dirty(handle, inode, path);
2280        if (err)
2281                return err;
2282        ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2283        trace_ext4_ext_rm_idx(inode, leaf);
2284
2285        ext4_free_blocks(handle, inode, NULL, leaf, 1,
2286                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2287
2288        while (--depth >= 0) {
2289                if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2290                        break;
2291                path--;
2292                err = ext4_ext_get_access(handle, inode, path);
2293                if (err)
2294                        break;
2295                path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2296                err = ext4_ext_dirty(handle, inode, path);
2297                if (err)
2298                        break;
2299        }
2300        return err;
2301}
2302
2303/*
2304 * ext4_ext_calc_credits_for_single_extent:
2305 * This routine returns max. credits that needed to insert an extent
2306 * to the extent tree.
2307 * When pass the actual path, the caller should calculate credits
2308 * under i_data_sem.
2309 */
2310int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2311                                                struct ext4_ext_path *path)
2312{
2313        if (path) {
2314                int depth = ext_depth(inode);
2315                int ret = 0;
2316
2317                /* probably there is space in leaf? */
2318                if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2319                                < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2320
2321                        /*
2322                         *  There are some space in the leaf tree, no
2323                         *  need to account for leaf block credit
2324                         *
2325                         *  bitmaps and block group descriptor blocks
2326                         *  and other metadata blocks still need to be
2327                         *  accounted.
2328                         */
2329                        /* 1 bitmap, 1 block group descriptor */
2330                        ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2331                        return ret;
2332                }
2333        }
2334
2335        return ext4_chunk_trans_blocks(inode, nrblocks);
2336}
2337
2338/*
2339 * How many index/leaf blocks need to change/allocate to add @extents extents?
2340 *
2341 * If we add a single extent, then in the worse case, each tree level
2342 * index/leaf need to be changed in case of the tree split.
2343 *
2344 * If more extents are inserted, they could cause the whole tree split more
2345 * than once, but this is really rare.
2346 */
2347int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2348{
2349        int index;
2350        int depth;
2351
2352        /* If we are converting the inline data, only one is needed here. */
2353        if (ext4_has_inline_data(inode))
2354                return 1;
2355
2356        depth = ext_depth(inode);
2357
2358        if (extents <= 1)
2359                index = depth * 2;
2360        else
2361                index = depth * 3;
2362
2363        return index;
2364}
2365
2366static inline int get_default_free_blocks_flags(struct inode *inode)
2367{
2368        if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2369            ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2370                return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2371        else if (ext4_should_journal_data(inode))
2372                return EXT4_FREE_BLOCKS_FORGET;
2373        return 0;
2374}
2375
2376/*
2377 * ext4_rereserve_cluster - increment the reserved cluster count when
2378 *                          freeing a cluster with a pending reservation
2379 *
2380 * @inode - file containing the cluster
2381 * @lblk - logical block in cluster to be reserved
2382 *
2383 * Increments the reserved cluster count and adjusts quota in a bigalloc
2384 * file system when freeing a partial cluster containing at least one
2385 * delayed and unwritten block.  A partial cluster meeting that
2386 * requirement will have a pending reservation.  If so, the
2387 * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2388 * defer reserved and allocated space accounting to a subsequent call
2389 * to this function.
2390 */
2391static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2392{
2393        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2394        struct ext4_inode_info *ei = EXT4_I(inode);
2395
2396        dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2397
2398        spin_lock(&ei->i_block_reservation_lock);
2399        ei->i_reserved_data_blocks++;
2400        percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2401        spin_unlock(&ei->i_block_reservation_lock);
2402
2403        percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2404        ext4_remove_pending(inode, lblk);
2405}
2406
2407static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2408                              struct ext4_extent *ex,
2409                              struct partial_cluster *partial,
2410                              ext4_lblk_t from, ext4_lblk_t to)
2411{
2412        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2413        unsigned short ee_len = ext4_ext_get_actual_len(ex);
2414        ext4_fsblk_t last_pblk, pblk;
2415        ext4_lblk_t num;
2416        int flags;
2417
2418        /* only extent tail removal is allowed */
2419        if (from < le32_to_cpu(ex->ee_block) ||
2420            to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2421                ext4_error(sbi->s_sb,
2422                           "strange request: removal(2) %u-%u from %u:%u",
2423                           from, to, le32_to_cpu(ex->ee_block), ee_len);
2424                return 0;
2425        }
2426
2427#ifdef EXTENTS_STATS
2428        spin_lock(&sbi->s_ext_stats_lock);
2429        sbi->s_ext_blocks += ee_len;
2430        sbi->s_ext_extents++;
2431        if (ee_len < sbi->s_ext_min)
2432                sbi->s_ext_min = ee_len;
2433        if (ee_len > sbi->s_ext_max)
2434                sbi->s_ext_max = ee_len;
2435        if (ext_depth(inode) > sbi->s_depth_max)
2436                sbi->s_depth_max = ext_depth(inode);
2437        spin_unlock(&sbi->s_ext_stats_lock);
2438#endif
2439
2440        trace_ext4_remove_blocks(inode, ex, from, to, partial);
2441
2442        /*
2443         * if we have a partial cluster, and it's different from the
2444         * cluster of the last block in the extent, we free it
2445         */
2446        last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2447
2448        if (partial->state != initial &&
2449            partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2450                if (partial->state == tofree) {
2451                        flags = get_default_free_blocks_flags(inode);
2452                        if (ext4_is_pending(inode, partial->lblk))
2453                                flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2454                        ext4_free_blocks(handle, inode, NULL,
2455                                         EXT4_C2B(sbi, partial->pclu),
2456                                         sbi->s_cluster_ratio, flags);
2457                        if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2458                                ext4_rereserve_cluster(inode, partial->lblk);
2459                }
2460                partial->state = initial;
2461        }
2462
2463        num = le32_to_cpu(ex->ee_block) + ee_len - from;
2464        pblk = ext4_ext_pblock(ex) + ee_len - num;
2465
2466        /*
2467         * We free the partial cluster at the end of the extent (if any),
2468         * unless the cluster is used by another extent (partial_cluster
2469         * state is nofree).  If a partial cluster exists here, it must be
2470         * shared with the last block in the extent.
2471         */
2472        flags = get_default_free_blocks_flags(inode);
2473
2474        /* partial, left end cluster aligned, right end unaligned */
2475        if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2476            (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2477            (partial->state != nofree)) {
2478                if (ext4_is_pending(inode, to))
2479                        flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2480                ext4_free_blocks(handle, inode, NULL,
2481                                 EXT4_PBLK_CMASK(sbi, last_pblk),
2482                                 sbi->s_cluster_ratio, flags);
2483                if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2484                        ext4_rereserve_cluster(inode, to);
2485                partial->state = initial;
2486                flags = get_default_free_blocks_flags(inode);
2487        }
2488
2489        flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2490
2491        /*
2492         * For bigalloc file systems, we never free a partial cluster
2493         * at the beginning of the extent.  Instead, we check to see if we
2494         * need to free it on a subsequent call to ext4_remove_blocks,
2495         * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2496         */
2497        flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2498        ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2499
2500        /* reset the partial cluster if we've freed past it */
2501        if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2502                partial->state = initial;
2503
2504        /*
2505         * If we've freed the entire extent but the beginning is not left
2506         * cluster aligned and is not marked as ineligible for freeing we
2507         * record the partial cluster at the beginning of the extent.  It
2508         * wasn't freed by the preceding ext4_free_blocks() call, and we
2509         * need to look farther to the left to determine if it's to be freed
2510         * (not shared with another extent). Else, reset the partial
2511         * cluster - we're either  done freeing or the beginning of the
2512         * extent is left cluster aligned.
2513         */
2514        if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2515                if (partial->state == initial) {
2516                        partial->pclu = EXT4_B2C(sbi, pblk);
2517                        partial->lblk = from;
2518                        partial->state = tofree;
2519                }
2520        } else {
2521                partial->state = initial;
2522        }
2523
2524        return 0;
2525}
2526
2527/*
2528 * ext4_ext_rm_leaf() Removes the extents associated with the
2529 * blocks appearing between "start" and "end".  Both "start"
2530 * and "end" must appear in the same extent or EIO is returned.
2531 *
2532 * @handle: The journal handle
2533 * @inode:  The files inode
2534 * @path:   The path to the leaf
2535 * @partial_cluster: The cluster which we'll have to free if all extents
2536 *                   has been released from it.  However, if this value is
2537 *                   negative, it's a cluster just to the right of the
2538 *                   punched region and it must not be freed.
2539 * @start:  The first block to remove
2540 * @end:   The last block to remove
2541 */
2542static int
2543ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2544                 struct ext4_ext_path *path,
2545                 struct partial_cluster *partial,
2546                 ext4_lblk_t start, ext4_lblk_t end)
2547{
2548        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2549        int err = 0, correct_index = 0;
2550        int depth = ext_depth(inode), credits, revoke_credits;
2551        struct ext4_extent_header *eh;
2552        ext4_lblk_t a, b;
2553        unsigned num;
2554        ext4_lblk_t ex_ee_block;
2555        unsigned short ex_ee_len;
2556        unsigned unwritten = 0;
2557        struct ext4_extent *ex;
2558        ext4_fsblk_t pblk;
2559
2560        /* the header must be checked already in ext4_ext_remove_space() */
2561        ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2562        if (!path[depth].p_hdr)
2563                path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2564        eh = path[depth].p_hdr;
2565        if (unlikely(path[depth].p_hdr == NULL)) {
2566                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2567                return -EFSCORRUPTED;
2568        }
2569        /* find where to start removing */
2570        ex = path[depth].p_ext;
2571        if (!ex)
2572                ex = EXT_LAST_EXTENT(eh);
2573
2574        ex_ee_block = le32_to_cpu(ex->ee_block);
2575        ex_ee_len = ext4_ext_get_actual_len(ex);
2576
2577        trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2578
2579        while (ex >= EXT_FIRST_EXTENT(eh) &&
2580                        ex_ee_block + ex_ee_len > start) {
2581
2582                if (ext4_ext_is_unwritten(ex))
2583                        unwritten = 1;
2584                else
2585                        unwritten = 0;
2586
2587                ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2588                          unwritten, ex_ee_len);
2589                path[depth].p_ext = ex;
2590
2591                a = ex_ee_block > start ? ex_ee_block : start;
2592                b = ex_ee_block+ex_ee_len - 1 < end ?
2593                        ex_ee_block+ex_ee_len - 1 : end;
2594
2595                ext_debug(inode, "  border %u:%u\n", a, b);
2596
2597                /* If this extent is beyond the end of the hole, skip it */
2598                if (end < ex_ee_block) {
2599                        /*
2600                         * We're going to skip this extent and move to another,
2601                         * so note that its first cluster is in use to avoid
2602                         * freeing it when removing blocks.  Eventually, the
2603                         * right edge of the truncated/punched region will
2604                         * be just to the left.
2605                         */
2606                        if (sbi->s_cluster_ratio > 1) {
2607                                pblk = ext4_ext_pblock(ex);
2608                                partial->pclu = EXT4_B2C(sbi, pblk);
2609                                partial->state = nofree;
2610                        }
2611                        ex--;
2612                        ex_ee_block = le32_to_cpu(ex->ee_block);
2613                        ex_ee_len = ext4_ext_get_actual_len(ex);
2614                        continue;
2615                } else if (b != ex_ee_block + ex_ee_len - 1) {
2616                        EXT4_ERROR_INODE(inode,
2617                                         "can not handle truncate %u:%u "
2618                                         "on extent %u:%u",
2619                                         start, end, ex_ee_block,
2620                                         ex_ee_block + ex_ee_len - 1);
2621                        err = -EFSCORRUPTED;
2622                        goto out;
2623                } else if (a != ex_ee_block) {
2624                        /* remove tail of the extent */
2625                        num = a - ex_ee_block;
2626                } else {
2627                        /* remove whole extent: excellent! */
2628                        num = 0;
2629                }
2630                /*
2631                 * 3 for leaf, sb, and inode plus 2 (bmap and group
2632                 * descriptor) for each block group; assume two block
2633                 * groups plus ex_ee_len/blocks_per_block_group for
2634                 * the worst case
2635                 */
2636                credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2637                if (ex == EXT_FIRST_EXTENT(eh)) {
2638                        correct_index = 1;
2639                        credits += (ext_depth(inode)) + 1;
2640                }
2641                credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2642                /*
2643                 * We may end up freeing some index blocks and data from the
2644                 * punched range. Note that partial clusters are accounted for
2645                 * by ext4_free_data_revoke_credits().
2646                 */
2647                revoke_credits =
2648                        ext4_free_metadata_revoke_credits(inode->i_sb,
2649                                                          ext_depth(inode)) +
2650                        ext4_free_data_revoke_credits(inode, b - a + 1);
2651
2652                err = ext4_datasem_ensure_credits(handle, inode, credits,
2653                                                  credits, revoke_credits);
2654                if (err) {
2655                        if (err > 0)
2656                                err = -EAGAIN;
2657                        goto out;
2658                }
2659
2660                err = ext4_ext_get_access(handle, inode, path + depth);
2661                if (err)
2662                        goto out;
2663
2664                err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2665                if (err)
2666                        goto out;
2667
2668                if (num == 0)
2669                        /* this extent is removed; mark slot entirely unused */
2670                        ext4_ext_store_pblock(ex, 0);
2671
2672                ex->ee_len = cpu_to_le16(num);
2673                /*
2674                 * Do not mark unwritten if all the blocks in the
2675                 * extent have been removed.
2676                 */
2677                if (unwritten && num)
2678                        ext4_ext_mark_unwritten(ex);
2679                /*
2680                 * If the extent was completely released,
2681                 * we need to remove it from the leaf
2682                 */
2683                if (num == 0) {
2684                        if (end != EXT_MAX_BLOCKS - 1) {
2685                                /*
2686                                 * For hole punching, we need to scoot all the
2687                                 * extents up when an extent is removed so that
2688                                 * we dont have blank extents in the middle
2689                                 */
2690                                memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2691                                        sizeof(struct ext4_extent));
2692
2693                                /* Now get rid of the one at the end */
2694                                memset(EXT_LAST_EXTENT(eh), 0,
2695                                        sizeof(struct ext4_extent));
2696                        }
2697                        le16_add_cpu(&eh->eh_entries, -1);
2698                }
2699
2700                err = ext4_ext_dirty(handle, inode, path + depth);
2701                if (err)
2702                        goto out;
2703
2704                ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2705                                ext4_ext_pblock(ex));
2706                ex--;
2707                ex_ee_block = le32_to_cpu(ex->ee_block);
2708                ex_ee_len = ext4_ext_get_actual_len(ex);
2709        }
2710
2711        if (correct_index && eh->eh_entries)
2712                err = ext4_ext_correct_indexes(handle, inode, path);
2713
2714        /*
2715         * If there's a partial cluster and at least one extent remains in
2716         * the leaf, free the partial cluster if it isn't shared with the
2717         * current extent.  If it is shared with the current extent
2718         * we reset the partial cluster because we've reached the start of the
2719         * truncated/punched region and we're done removing blocks.
2720         */
2721        if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2722                pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2723                if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2724                        int flags = get_default_free_blocks_flags(inode);
2725
2726                        if (ext4_is_pending(inode, partial->lblk))
2727                                flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2728                        ext4_free_blocks(handle, inode, NULL,
2729                                         EXT4_C2B(sbi, partial->pclu),
2730                                         sbi->s_cluster_ratio, flags);
2731                        if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2732                                ext4_rereserve_cluster(inode, partial->lblk);
2733                }
2734                partial->state = initial;
2735        }
2736
2737        /* if this leaf is free, then we should
2738         * remove it from index block above */
2739        if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2740                err = ext4_ext_rm_idx(handle, inode, path, depth);
2741
2742out:
2743        return err;
2744}
2745
2746/*
2747 * ext4_ext_more_to_rm:
2748 * returns 1 if current index has to be freed (even partial)
2749 */
2750static int
2751ext4_ext_more_to_rm(struct ext4_ext_path *path)
2752{
2753        BUG_ON(path->p_idx == NULL);
2754
2755        if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2756                return 0;
2757
2758        /*
2759         * if truncate on deeper level happened, it wasn't partial,
2760         * so we have to consider current index for truncation
2761         */
2762        if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2763                return 0;
2764        return 1;
2765}
2766
2767int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2768                          ext4_lblk_t end)
2769{
2770        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2771        int depth = ext_depth(inode);
2772        struct ext4_ext_path *path = NULL;
2773        struct partial_cluster partial;
2774        handle_t *handle;
2775        int i = 0, err = 0;
2776
2777        partial.pclu = 0;
2778        partial.lblk = 0;
2779        partial.state = initial;
2780
2781        ext_debug(inode, "truncate since %u to %u\n", start, end);
2782
2783        /* probably first extent we're gonna free will be last in block */
2784        handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2785                        depth + 1,
2786                        ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2787        if (IS_ERR(handle))
2788                return PTR_ERR(handle);
2789
2790again:
2791        trace_ext4_ext_remove_space(inode, start, end, depth);
2792
2793        /*
2794         * Check if we are removing extents inside the extent tree. If that
2795         * is the case, we are going to punch a hole inside the extent tree
2796         * so we have to check whether we need to split the extent covering
2797         * the last block to remove so we can easily remove the part of it
2798         * in ext4_ext_rm_leaf().
2799         */
2800        if (end < EXT_MAX_BLOCKS - 1) {
2801                struct ext4_extent *ex;
2802                ext4_lblk_t ee_block, ex_end, lblk;
2803                ext4_fsblk_t pblk;
2804
2805                /* find extent for or closest extent to this block */
2806                path = ext4_find_extent(inode, end, NULL,
2807                                        EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2808                if (IS_ERR(path)) {
2809                        ext4_journal_stop(handle);
2810                        return PTR_ERR(path);
2811                }
2812                depth = ext_depth(inode);
2813                /* Leaf not may not exist only if inode has no blocks at all */
2814                ex = path[depth].p_ext;
2815                if (!ex) {
2816                        if (depth) {
2817                                EXT4_ERROR_INODE(inode,
2818                                                 "path[%d].p_hdr == NULL",
2819                                                 depth);
2820                                err = -EFSCORRUPTED;
2821                        }
2822                        goto out;
2823                }
2824
2825                ee_block = le32_to_cpu(ex->ee_block);
2826                ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2827
2828                /*
2829                 * See if the last block is inside the extent, if so split
2830                 * the extent at 'end' block so we can easily remove the
2831                 * tail of the first part of the split extent in
2832                 * ext4_ext_rm_leaf().
2833                 */
2834                if (end >= ee_block && end < ex_end) {
2835
2836                        /*
2837                         * If we're going to split the extent, note that
2838                         * the cluster containing the block after 'end' is
2839                         * in use to avoid freeing it when removing blocks.
2840                         */
2841                        if (sbi->s_cluster_ratio > 1) {
2842                                pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2843                                partial.pclu = EXT4_B2C(sbi, pblk);
2844                                partial.state = nofree;
2845                        }
2846
2847                        /*
2848                         * Split the extent in two so that 'end' is the last
2849                         * block in the first new extent. Also we should not
2850                         * fail removing space due to ENOSPC so try to use
2851                         * reserved block if that happens.
2852                         */
2853                        err = ext4_force_split_extent_at(handle, inode, &path,
2854                                                         end + 1, 1);
2855                        if (err < 0)
2856                                goto out;
2857
2858                } else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2859                           partial.state == initial) {
2860                        /*
2861                         * If we're punching, there's an extent to the right.
2862                         * If the partial cluster hasn't been set, set it to
2863                         * that extent's first cluster and its state to nofree
2864                         * so it won't be freed should it contain blocks to be
2865                         * removed. If it's already set (tofree/nofree), we're
2866                         * retrying and keep the original partial cluster info
2867                         * so a cluster marked tofree as a result of earlier
2868                         * extent removal is not lost.
2869                         */
2870                        lblk = ex_end + 1;
2871                        err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2872                                                    NULL);
2873                        if (err < 0)
2874                                goto out;
2875                        if (pblk) {
2876                                partial.pclu = EXT4_B2C(sbi, pblk);
2877                                partial.state = nofree;
2878                        }
2879                }
2880        }
2881        /*
2882         * We start scanning from right side, freeing all the blocks
2883         * after i_size and walking into the tree depth-wise.
2884         */
2885        depth = ext_depth(inode);
2886        if (path) {
2887                int k = i = depth;
2888                while (--k > 0)
2889                        path[k].p_block =
2890                                le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2891        } else {
2892                path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2893                               GFP_NOFS | __GFP_NOFAIL);
2894                if (path == NULL) {
2895                        ext4_journal_stop(handle);
2896                        return -ENOMEM;
2897                }
2898                path[0].p_maxdepth = path[0].p_depth = depth;
2899                path[0].p_hdr = ext_inode_hdr(inode);
2900                i = 0;
2901
2902                if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2903                        err = -EFSCORRUPTED;
2904                        goto out;
2905                }
2906        }
2907        err = 0;
2908
2909        while (i >= 0 && err == 0) {
2910                if (i == depth) {
2911                        /* this is leaf block */
2912                        err = ext4_ext_rm_leaf(handle, inode, path,
2913                                               &partial, start, end);
2914                        /* root level has p_bh == NULL, brelse() eats this */
2915                        brelse(path[i].p_bh);
2916                        path[i].p_bh = NULL;
2917                        i--;
2918                        continue;
2919                }
2920
2921                /* this is index block */
2922                if (!path[i].p_hdr) {
2923                        ext_debug(inode, "initialize header\n");
2924                        path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2925                }
2926
2927                if (!path[i].p_idx) {
2928                        /* this level hasn't been touched yet */
2929                        path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2930                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2931                        ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2932                                  path[i].p_hdr,
2933                                  le16_to_cpu(path[i].p_hdr->eh_entries));
2934                } else {
2935                        /* we were already here, see at next index */
2936                        path[i].p_idx--;
2937                }
2938
2939                ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2940                                i, EXT_FIRST_INDEX(path[i].p_hdr),
2941                                path[i].p_idx);
2942                if (ext4_ext_more_to_rm(path + i)) {
2943                        struct buffer_head *bh;
2944                        /* go to the next level */
2945                        ext_debug(inode, "move to level %d (block %llu)\n",
2946                                  i + 1, ext4_idx_pblock(path[i].p_idx));
2947                        memset(path + i + 1, 0, sizeof(*path));
2948                        bh = read_extent_tree_block(inode,
2949                                ext4_idx_pblock(path[i].p_idx), depth - i - 1,
2950                                EXT4_EX_NOCACHE);
2951                        if (IS_ERR(bh)) {
2952                                /* should we reset i_size? */
2953                                err = PTR_ERR(bh);
2954                                break;
2955                        }
2956                        /* Yield here to deal with large extent trees.
2957                         * Should be a no-op if we did IO above. */
2958                        cond_resched();
2959                        if (WARN_ON(i + 1 > depth)) {
2960                                err = -EFSCORRUPTED;
2961                                break;
2962                        }
2963                        path[i + 1].p_bh = bh;
2964
2965                        /* save actual number of indexes since this
2966                         * number is changed at the next iteration */
2967                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2968                        i++;
2969                } else {
2970                        /* we finished processing this index, go up */
2971                        if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2972                                /* index is empty, remove it;
2973                                 * handle must be already prepared by the
2974                                 * truncatei_leaf() */
2975                                err = ext4_ext_rm_idx(handle, inode, path, i);
2976                        }
2977                        /* root level has p_bh == NULL, brelse() eats this */
2978                        brelse(path[i].p_bh);
2979                        path[i].p_bh = NULL;
2980                        i--;
2981                        ext_debug(inode, "return to level %d\n", i);
2982                }
2983        }
2984
2985        trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
2986                                         path->p_hdr->eh_entries);
2987
2988        /*
2989         * if there's a partial cluster and we have removed the first extent
2990         * in the file, then we also free the partial cluster, if any
2991         */
2992        if (partial.state == tofree && err == 0) {
2993                int flags = get_default_free_blocks_flags(inode);
2994
2995                if (ext4_is_pending(inode, partial.lblk))
2996                        flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2997                ext4_free_blocks(handle, inode, NULL,
2998                                 EXT4_C2B(sbi, partial.pclu),
2999                                 sbi->s_cluster_ratio, flags);
3000                if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3001                        ext4_rereserve_cluster(inode, partial.lblk);
3002                partial.state = initial;
3003        }
3004
3005        /* TODO: flexible tree reduction should be here */
3006        if (path->p_hdr->eh_entries == 0) {
3007                /*
3008                 * truncate to zero freed all the tree,
3009                 * so we need to correct eh_depth
3010                 */
3011                err = ext4_ext_get_access(handle, inode, path);
3012                if (err == 0) {
3013                        ext_inode_hdr(inode)->eh_depth = 0;
3014                        ext_inode_hdr(inode)->eh_max =
3015                                cpu_to_le16(ext4_ext_space_root(inode, 0));
3016                        err = ext4_ext_dirty(handle, inode, path);
3017                }
3018        }
3019out:
3020        ext4_ext_drop_refs(path);
3021        kfree(path);
3022        path = NULL;
3023        if (err == -EAGAIN)
3024                goto again;
3025        ext4_journal_stop(handle);
3026
3027        return err;
3028}
3029
3030/*
3031 * called at mount time
3032 */
3033void ext4_ext_init(struct super_block *sb)
3034{
3035        /*
3036         * possible initialization would be here
3037         */
3038
3039        if (ext4_has_feature_extents(sb)) {
3040#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3041                printk(KERN_INFO "EXT4-fs: file extents enabled"
3042#ifdef AGGRESSIVE_TEST
3043                       ", aggressive tests"
3044#endif
3045#ifdef CHECK_BINSEARCH
3046                       ", check binsearch"
3047#endif
3048#ifdef EXTENTS_STATS
3049                       ", stats"
3050#endif
3051                       "\n");
3052#endif
3053#ifdef EXTENTS_STATS
3054                spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3055                EXT4_SB(sb)->s_ext_min = 1 << 30;
3056                EXT4_SB(sb)->s_ext_max = 0;
3057#endif
3058        }
3059}
3060
3061/*
3062 * called at umount time
3063 */
3064void ext4_ext_release(struct super_block *sb)
3065{
3066        if (!ext4_has_feature_extents(sb))
3067                return;
3068
3069#ifdef EXTENTS_STATS
3070        if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3071                struct ext4_sb_info *sbi = EXT4_SB(sb);
3072                printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3073                        sbi->s_ext_blocks, sbi->s_ext_extents,
3074                        sbi->s_ext_blocks / sbi->s_ext_extents);
3075                printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3076                        sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3077        }
3078#endif
3079}
3080
3081static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3082{
3083        ext4_lblk_t  ee_block;
3084        ext4_fsblk_t ee_pblock;
3085        unsigned int ee_len;
3086
3087        ee_block  = le32_to_cpu(ex->ee_block);
3088        ee_len    = ext4_ext_get_actual_len(ex);
3089        ee_pblock = ext4_ext_pblock(ex);
3090
3091        if (ee_len == 0)
3092                return 0;
3093
3094        return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3095                                     EXTENT_STATUS_WRITTEN);
3096}
3097
3098/* FIXME!! we need to try to merge to left or right after zero-out  */
3099static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3100{
3101        ext4_fsblk_t ee_pblock;
3102        unsigned int ee_len;
3103
3104        ee_len    = ext4_ext_get_actual_len(ex);
3105        ee_pblock = ext4_ext_pblock(ex);
3106        return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3107                                  ee_len);
3108}
3109
3110/*
3111 * ext4_split_extent_at() splits an extent at given block.
3112 *
3113 * @handle: the journal handle
3114 * @inode: the file inode
3115 * @path: the path to the extent
3116 * @split: the logical block where the extent is splitted.
3117 * @split_flags: indicates if the extent could be zeroout if split fails, and
3118 *               the states(init or unwritten) of new extents.
3119 * @flags: flags used to insert new extent to extent tree.
3120 *
3121 *
3122 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3123 * of which are determined by split_flag.
3124 *
3125 * There are two cases:
3126 *  a> the extent are splitted into two extent.
3127 *  b> split is not needed, and just mark the extent.
3128 *
3129 * return 0 on success.
3130 */
3131static int ext4_split_extent_at(handle_t *handle,
3132                             struct inode *inode,
3133                             struct ext4_ext_path **ppath,
3134                             ext4_lblk_t split,
3135                             int split_flag,
3136                             int flags)
3137{
3138        struct ext4_ext_path *path = *ppath;
3139        ext4_fsblk_t newblock;
3140        ext4_lblk_t ee_block;
3141        struct ext4_extent *ex, newex, orig_ex, zero_ex;
3142        struct ext4_extent *ex2 = NULL;
3143        unsigned int ee_len, depth;
3144        int err = 0;
3145
3146        BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3147               (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3148
3149        ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3150
3151        ext4_ext_show_leaf(inode, path);
3152
3153        depth = ext_depth(inode);
3154        ex = path[depth].p_ext;
3155        ee_block = le32_to_cpu(ex->ee_block);
3156        ee_len = ext4_ext_get_actual_len(ex);
3157        newblock = split - ee_block + ext4_ext_pblock(ex);
3158
3159        BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3160        BUG_ON(!ext4_ext_is_unwritten(ex) &&
3161               split_flag & (EXT4_EXT_MAY_ZEROOUT |
3162                             EXT4_EXT_MARK_UNWRIT1 |
3163                             EXT4_EXT_MARK_UNWRIT2));
3164
3165        err = ext4_ext_get_access(handle, inode, path + depth);
3166        if (err)
3167                goto out;
3168
3169        if (split == ee_block) {
3170                /*
3171                 * case b: block @split is the block that the extent begins with
3172                 * then we just change the state of the extent, and splitting
3173                 * is not needed.
3174                 */
3175                if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3176                        ext4_ext_mark_unwritten(ex);
3177                else
3178                        ext4_ext_mark_initialized(ex);
3179
3180                if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3181                        ext4_ext_try_to_merge(handle, inode, path, ex);
3182
3183                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3184                goto out;
3185        }
3186
3187        /* case a */
3188        memcpy(&orig_ex, ex, sizeof(orig_ex));
3189        ex->ee_len = cpu_to_le16(split - ee_block);
3190        if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3191                ext4_ext_mark_unwritten(ex);
3192
3193        /*
3194         * path may lead to new leaf, not to original leaf any more
3195         * after ext4_ext_insert_extent() returns,
3196         */
3197        err = ext4_ext_dirty(handle, inode, path + depth);
3198        if (err)
3199                goto fix_extent_len;
3200
3201        ex2 = &newex;
3202        ex2->ee_block = cpu_to_le32(split);
3203        ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3204        ext4_ext_store_pblock(ex2, newblock);
3205        if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3206                ext4_ext_mark_unwritten(ex2);
3207
3208        err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3209        if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3210                if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3211                        if (split_flag & EXT4_EXT_DATA_VALID1) {
3212                                err = ext4_ext_zeroout(inode, ex2);
3213                                zero_ex.ee_block = ex2->ee_block;
3214                                zero_ex.ee_len = cpu_to_le16(
3215                                                ext4_ext_get_actual_len(ex2));
3216                                ext4_ext_store_pblock(&zero_ex,
3217                                                      ext4_ext_pblock(ex2));
3218                        } else {
3219                                err = ext4_ext_zeroout(inode, ex);
3220                                zero_ex.ee_block = ex->ee_block;
3221                                zero_ex.ee_len = cpu_to_le16(
3222                                                ext4_ext_get_actual_len(ex));
3223                                ext4_ext_store_pblock(&zero_ex,
3224                                                      ext4_ext_pblock(ex));
3225                        }
3226                } else {
3227                        err = ext4_ext_zeroout(inode, &orig_ex);
3228                        zero_ex.ee_block = orig_ex.ee_block;
3229                        zero_ex.ee_len = cpu_to_le16(
3230                                                ext4_ext_get_actual_len(&orig_ex));
3231                        ext4_ext_store_pblock(&zero_ex,
3232                                              ext4_ext_pblock(&orig_ex));
3233                }
3234
3235                if (err)
3236                        goto fix_extent_len;
3237                /* update the extent length and mark as initialized */
3238                ex->ee_len = cpu_to_le16(ee_len);
3239                ext4_ext_try_to_merge(handle, inode, path, ex);
3240                err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3241                if (err)
3242                        goto fix_extent_len;
3243
3244                /* update extent status tree */
3245                err = ext4_zeroout_es(inode, &zero_ex);
3246
3247                goto out;
3248        } else if (err)
3249                goto fix_extent_len;
3250
3251out:
3252        ext4_ext_show_leaf(inode, path);
3253        return err;
3254
3255fix_extent_len:
3256        ex->ee_len = orig_ex.ee_len;
3257        /*
3258         * Ignore ext4_ext_dirty return value since we are already in error path
3259         * and err is a non-zero error code.
3260         */
3261        ext4_ext_dirty(handle, inode, path + path->p_depth);
3262        return err;
3263}
3264
3265/*
3266 * ext4_split_extents() splits an extent and mark extent which is covered
3267 * by @map as split_flags indicates
3268 *
3269 * It may result in splitting the extent into multiple extents (up to three)
3270 * There are three possibilities:
3271 *   a> There is no split required
3272 *   b> Splits in two extents: Split is happening at either end of the extent
3273 *   c> Splits in three extents: Somone is splitting in middle of the extent
3274 *
3275 */
3276static int ext4_split_extent(handle_t *handle,
3277                              struct inode *inode,
3278                              struct ext4_ext_path **ppath,
3279                              struct ext4_map_blocks *map,
3280                              int split_flag,
3281                              int flags)
3282{
3283        struct ext4_ext_path *path = *ppath;
3284        ext4_lblk_t ee_block;
3285        struct ext4_extent *ex;
3286        unsigned int ee_len, depth;
3287        int err = 0;
3288        int unwritten;
3289        int split_flag1, flags1;
3290        int allocated = map->m_len;
3291
3292        depth = ext_depth(inode);
3293        ex = path[depth].p_ext;
3294        ee_block = le32_to_cpu(ex->ee_block);
3295        ee_len = ext4_ext_get_actual_len(ex);
3296        unwritten = ext4_ext_is_unwritten(ex);
3297
3298        if (map->m_lblk + map->m_len < ee_block + ee_len) {
3299                split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3300                flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3301                if (unwritten)
3302                        split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3303                                       EXT4_EXT_MARK_UNWRIT2;
3304                if (split_flag & EXT4_EXT_DATA_VALID2)
3305                        split_flag1 |= EXT4_EXT_DATA_VALID1;
3306                err = ext4_split_extent_at(handle, inode, ppath,
3307                                map->m_lblk + map->m_len, split_flag1, flags1);
3308                if (err)
3309                        goto out;
3310        } else {
3311                allocated = ee_len - (map->m_lblk - ee_block);
3312        }
3313        /*
3314         * Update path is required because previous ext4_split_extent_at() may
3315         * result in split of original leaf or extent zeroout.
3316         */
3317        path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3318        if (IS_ERR(path))
3319                return PTR_ERR(path);
3320        depth = ext_depth(inode);
3321        ex = path[depth].p_ext;
3322        if (!ex) {
3323                EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3324                                 (unsigned long) map->m_lblk);
3325                return -EFSCORRUPTED;
3326        }
3327        unwritten = ext4_ext_is_unwritten(ex);
3328        split_flag1 = 0;
3329
3330        if (map->m_lblk >= ee_block) {
3331                split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3332                if (unwritten) {
3333                        split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3334                        split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3335                                                     EXT4_EXT_MARK_UNWRIT2);
3336                }
3337                err = ext4_split_extent_at(handle, inode, ppath,
3338                                map->m_lblk, split_flag1, flags);
3339                if (err)
3340                        goto out;
3341        }
3342
3343        ext4_ext_show_leaf(inode, path);
3344out:
3345        return err ? err : allocated;
3346}
3347
3348/*
3349 * This function is called by ext4_ext_map_blocks() if someone tries to write
3350 * to an unwritten extent. It may result in splitting the unwritten
3351 * extent into multiple extents (up to three - one initialized and two
3352 * unwritten).
3353 * There are three possibilities:
3354 *   a> There is no split required: Entire extent should be initialized
3355 *   b> Splits in two extents: Write is happening at either end of the extent
3356 *   c> Splits in three extents: Somone is writing in middle of the extent
3357 *
3358 * Pre-conditions:
3359 *  - The extent pointed to by 'path' is unwritten.
3360 *  - The extent pointed to by 'path' contains a superset
3361 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3362 *
3363 * Post-conditions on success:
3364 *  - the returned value is the number of blocks beyond map->l_lblk
3365 *    that are allocated and initialized.
3366 *    It is guaranteed to be >= map->m_len.
3367 */
3368static int ext4_ext_convert_to_initialized(handle_t *handle,
3369                                           struct inode *inode,
3370                                           struct ext4_map_blocks *map,
3371                                           struct ext4_ext_path **ppath,
3372                                           int flags)
3373{
3374        struct ext4_ext_path *path = *ppath;
3375        struct ext4_sb_info *sbi;
3376        struct ext4_extent_header *eh;
3377        struct ext4_map_blocks split_map;
3378        struct ext4_extent zero_ex1, zero_ex2;
3379        struct ext4_extent *ex, *abut_ex;
3380        ext4_lblk_t ee_block, eof_block;
3381        unsigned int ee_len, depth, map_len = map->m_len;
3382        int allocated = 0, max_zeroout = 0;
3383        int err = 0;
3384        int split_flag = EXT4_EXT_DATA_VALID2;
3385
3386        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3387                  (unsigned long long)map->m_lblk, map_len);
3388
3389        sbi = EXT4_SB(inode->i_sb);
3390        eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3391                        >> inode->i_sb->s_blocksize_bits;
3392        if (eof_block < map->m_lblk + map_len)
3393                eof_block = map->m_lblk + map_len;
3394
3395        depth = ext_depth(inode);
3396        eh = path[depth].p_hdr;
3397        ex = path[depth].p_ext;
3398        ee_block = le32_to_cpu(ex->ee_block);
3399        ee_len = ext4_ext_get_actual_len(ex);
3400        zero_ex1.ee_len = 0;
3401        zero_ex2.ee_len = 0;
3402
3403        trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3404
3405        /* Pre-conditions */
3406        BUG_ON(!ext4_ext_is_unwritten(ex));
3407        BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3408
3409        /*
3410         * Attempt to transfer newly initialized blocks from the currently
3411         * unwritten extent to its neighbor. This is much cheaper
3412         * than an insertion followed by a merge as those involve costly
3413         * memmove() calls. Transferring to the left is the common case in
3414         * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3415         * followed by append writes.
3416         *
3417         * Limitations of the current logic:
3418         *  - L1: we do not deal with writes covering the whole extent.
3419         *    This would require removing the extent if the transfer
3420         *    is possible.
3421         *  - L2: we only attempt to merge with an extent stored in the
3422         *    same extent tree node.
3423         */
3424        if ((map->m_lblk == ee_block) &&
3425                /* See if we can merge left */
3426                (map_len < ee_len) &&           /*L1*/
3427                (ex > EXT_FIRST_EXTENT(eh))) {  /*L2*/
3428                ext4_lblk_t prev_lblk;
3429                ext4_fsblk_t prev_pblk, ee_pblk;
3430                unsigned int prev_len;
3431
3432                abut_ex = ex - 1;
3433                prev_lblk = le32_to_cpu(abut_ex->ee_block);
3434                prev_len = ext4_ext_get_actual_len(abut_ex);
3435                prev_pblk = ext4_ext_pblock(abut_ex);
3436                ee_pblk = ext4_ext_pblock(ex);
3437
3438                /*
3439                 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3440                 * upon those conditions:
3441                 * - C1: abut_ex is initialized,
3442                 * - C2: abut_ex is logically abutting ex,
3443                 * - C3: abut_ex is physically abutting ex,
3444                 * - C4: abut_ex can receive the additional blocks without
3445                 *   overflowing the (initialized) length limit.
3446                 */
3447                if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3448                        ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3449                        ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3450                        (prev_len < (EXT_INIT_MAX_LEN - map_len))) {    /*C4*/
3451                        err = ext4_ext_get_access(handle, inode, path + depth);
3452                        if (err)
3453                                goto out;
3454
3455                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3456                                map, ex, abut_ex);
3457
3458                        /* Shift the start of ex by 'map_len' blocks */
3459                        ex->ee_block = cpu_to_le32(ee_block + map_len);
3460                        ext4_ext_store_pblock(ex, ee_pblk + map_len);
3461                        ex->ee_len = cpu_to_le16(ee_len - map_len);
3462                        ext4_ext_mark_unwritten(ex); /* Restore the flag */
3463
3464                        /* Extend abut_ex by 'map_len' blocks */
3465                        abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3466
3467                        /* Result: number of initialized blocks past m_lblk */
3468                        allocated = map_len;
3469                }
3470        } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3471                   (map_len < ee_len) &&        /*L1*/
3472                   ex < EXT_LAST_EXTENT(eh)) {  /*L2*/
3473                /* See if we can merge right */
3474                ext4_lblk_t next_lblk;
3475                ext4_fsblk_t next_pblk, ee_pblk;
3476                unsigned int next_len;
3477
3478                abut_ex = ex + 1;
3479                next_lblk = le32_to_cpu(abut_ex->ee_block);
3480                next_len = ext4_ext_get_actual_len(abut_ex);
3481                next_pblk = ext4_ext_pblock(abut_ex);
3482                ee_pblk = ext4_ext_pblock(ex);
3483
3484                /*
3485                 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3486                 * upon those conditions:
3487                 * - C1: abut_ex is initialized,
3488                 * - C2: abut_ex is logically abutting ex,
3489                 * - C3: abut_ex is physically abutting ex,
3490                 * - C4: abut_ex can receive the additional blocks without
3491                 *   overflowing the (initialized) length limit.
3492                 */
3493                if ((!ext4_ext_is_unwritten(abut_ex)) &&                /*C1*/
3494                    ((map->m_lblk + map_len) == next_lblk) &&           /*C2*/
3495                    ((ee_pblk + ee_len) == next_pblk) &&                /*C3*/
3496                    (next_len < (EXT_INIT_MAX_LEN - map_len))) {        /*C4*/
3497                        err = ext4_ext_get_access(handle, inode, path + depth);
3498                        if (err)
3499                                goto out;
3500
3501                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3502                                map, ex, abut_ex);
3503
3504                        /* Shift the start of abut_ex by 'map_len' blocks */
3505                        abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3506                        ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3507                        ex->ee_len = cpu_to_le16(ee_len - map_len);
3508                        ext4_ext_mark_unwritten(ex); /* Restore the flag */
3509
3510                        /* Extend abut_ex by 'map_len' blocks */
3511                        abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3512
3513                        /* Result: number of initialized blocks past m_lblk */
3514                        allocated = map_len;
3515                }
3516        }
3517        if (allocated) {
3518                /* Mark the block containing both extents as dirty */
3519                err = ext4_ext_dirty(handle, inode, path + depth);
3520
3521                /* Update path to point to the right extent */
3522                path[depth].p_ext = abut_ex;
3523                goto out;
3524        } else
3525                allocated = ee_len - (map->m_lblk - ee_block);
3526
3527        WARN_ON(map->m_lblk < ee_block);
3528        /*
3529         * It is safe to convert extent to initialized via explicit
3530         * zeroout only if extent is fully inside i_size or new_size.
3531         */
3532        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3533
3534        if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3535                max_zeroout = sbi->s_extent_max_zeroout_kb >>
3536                        (inode->i_sb->s_blocksize_bits - 10);
3537
3538        /*
3539         * five cases:
3540         * 1. split the extent into three extents.
3541         * 2. split the extent into two extents, zeroout the head of the first
3542         *    extent.
3543         * 3. split the extent into two extents, zeroout the tail of the second
3544         *    extent.
3545         * 4. split the extent into two extents with out zeroout.
3546         * 5. no splitting needed, just possibly zeroout the head and / or the
3547         *    tail of the extent.
3548         */
3549        split_map.m_lblk = map->m_lblk;
3550        split_map.m_len = map->m_len;
3551
3552        if (max_zeroout && (allocated > split_map.m_len)) {
3553                if (allocated <= max_zeroout) {
3554                        /* case 3 or 5 */
3555                        zero_ex1.ee_block =
3556                                 cpu_to_le32(split_map.m_lblk +
3557                                             split_map.m_len);
3558                        zero_ex1.ee_len =
3559                                cpu_to_le16(allocated - split_map.m_len);
3560                        ext4_ext_store_pblock(&zero_ex1,
3561                                ext4_ext_pblock(ex) + split_map.m_lblk +
3562                                split_map.m_len - ee_block);
3563                        err = ext4_ext_zeroout(inode, &zero_ex1);
3564                        if (err)
3565                                goto out;
3566                        split_map.m_len = allocated;
3567                }
3568                if (split_map.m_lblk - ee_block + split_map.m_len <
3569                                                                max_zeroout) {
3570                        /* case 2 or 5 */
3571                        if (split_map.m_lblk != ee_block) {
3572                                zero_ex2.ee_block = ex->ee_block;
3573                                zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3574                                                        ee_block);
3575                                ext4_ext_store_pblock(&zero_ex2,
3576                                                      ext4_ext_pblock(ex));
3577                                err = ext4_ext_zeroout(inode, &zero_ex2);
3578                                if (err)
3579                                        goto out;
3580                        }
3581
3582                        split_map.m_len += split_map.m_lblk - ee_block;
3583                        split_map.m_lblk = ee_block;
3584                        allocated = map->m_len;
3585                }
3586        }
3587
3588        err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3589                                flags);
3590        if (err > 0)
3591                err = 0;
3592out:
3593        /* If we have gotten a failure, don't zero out status tree */
3594        if (!err) {
3595                err = ext4_zeroout_es(inode, &zero_ex1);
3596                if (!err)
3597                        err = ext4_zeroout_es(inode, &zero_ex2);
3598        }
3599        return err ? err : allocated;
3600}
3601
3602/*
3603 * This function is called by ext4_ext_map_blocks() from
3604 * ext4_get_blocks_dio_write() when DIO to write
3605 * to an unwritten extent.
3606 *
3607 * Writing to an unwritten extent may result in splitting the unwritten
3608 * extent into multiple initialized/unwritten extents (up to three)
3609 * There are three possibilities:
3610 *   a> There is no split required: Entire extent should be unwritten
3611 *   b> Splits in two extents: Write is happening at either end of the extent
3612 *   c> Splits in three extents: Somone is writing in middle of the extent
3613 *
3614 * This works the same way in the case of initialized -> unwritten conversion.
3615 *
3616 * One of more index blocks maybe needed if the extent tree grow after
3617 * the unwritten extent split. To prevent ENOSPC occur at the IO
3618 * complete, we need to split the unwritten extent before DIO submit
3619 * the IO. The unwritten extent called at this time will be split
3620 * into three unwritten extent(at most). After IO complete, the part
3621 * being filled will be convert to initialized by the end_io callback function
3622 * via ext4_convert_unwritten_extents().
3623 *
3624 * Returns the size of unwritten extent to be written on success.
3625 */
3626static int ext4_split_convert_extents(handle_t *handle,
3627                                        struct inode *inode,
3628                                        struct ext4_map_blocks *map,
3629                                        struct ext4_ext_path **ppath,
3630                                        int flags)
3631{
3632        struct ext4_ext_path *path = *ppath;
3633        ext4_lblk_t eof_block;
3634        ext4_lblk_t ee_block;
3635        struct ext4_extent *ex;
3636        unsigned int ee_len;
3637        int split_flag = 0, depth;
3638
3639        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3640                  (unsigned long long)map->m_lblk, map->m_len);
3641
3642        eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3643                        >> inode->i_sb->s_blocksize_bits;
3644        if (eof_block < map->m_lblk + map->m_len)
3645                eof_block = map->m_lblk + map->m_len;
3646        /*
3647         * It is safe to convert extent to initialized via explicit
3648         * zeroout only if extent is fully inside i_size or new_size.
3649         */
3650        depth = ext_depth(inode);
3651        ex = path[depth].p_ext;
3652        ee_block = le32_to_cpu(ex->ee_block);
3653        ee_len = ext4_ext_get_actual_len(ex);
3654
3655        /* Convert to unwritten */
3656        if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3657                split_flag |= EXT4_EXT_DATA_VALID1;
3658        /* Convert to initialized */
3659        } else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3660                split_flag |= ee_block + ee_len <= eof_block ?
3661                              EXT4_EXT_MAY_ZEROOUT : 0;
3662                split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3663        }
3664        flags |= EXT4_GET_BLOCKS_PRE_IO;
3665        return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3666}
3667
3668static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3669                                                struct inode *inode,
3670                                                struct ext4_map_blocks *map,
3671                                                struct ext4_ext_path **ppath)
3672{
3673        struct ext4_ext_path *path = *ppath;
3674        struct ext4_extent *ex;
3675        ext4_lblk_t ee_block;
3676        unsigned int ee_len;
3677        int depth;
3678        int err = 0;
3679
3680        depth = ext_depth(inode);
3681        ex = path[depth].p_ext;
3682        ee_block = le32_to_cpu(ex->ee_block);
3683        ee_len = ext4_ext_get_actual_len(ex);
3684
3685        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3686                  (unsigned long long)ee_block, ee_len);
3687
3688        /* If extent is larger than requested it is a clear sign that we still
3689         * have some extent state machine issues left. So extent_split is still
3690         * required.
3691         * TODO: Once all related issues will be fixed this situation should be
3692         * illegal.
3693         */
3694        if (ee_block != map->m_lblk || ee_len > map->m_len) {
3695#ifdef CONFIG_EXT4_DEBUG
3696                ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3697                             " len %u; IO logical block %llu, len %u",
3698                             inode->i_ino, (unsigned long long)ee_block, ee_len,
3699                             (unsigned long long)map->m_lblk, map->m_len);
3700#endif
3701                err = ext4_split_convert_extents(handle, inode, map, ppath,
3702                                                 EXT4_GET_BLOCKS_CONVERT);
3703                if (err < 0)
3704                        return err;
3705                path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3706                if (IS_ERR(path))
3707                        return PTR_ERR(path);
3708                depth = ext_depth(inode);
3709                ex = path[depth].p_ext;
3710        }
3711
3712        err = ext4_ext_get_access(handle, inode, path + depth);
3713        if (err)
3714                goto out;
3715        /* first mark the extent as initialized */
3716        ext4_ext_mark_initialized(ex);
3717
3718        /* note: ext4_ext_correct_indexes() isn't needed here because
3719         * borders are not changed
3720         */
3721        ext4_ext_try_to_merge(handle, inode, path, ex);
3722
3723        /* Mark modified extent as dirty */
3724        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3725out:
3726        ext4_ext_show_leaf(inode, path);
3727        return err;
3728}
3729
3730static int
3731convert_initialized_extent(handle_t *handle, struct inode *inode,
3732                           struct ext4_map_blocks *map,
3733                           struct ext4_ext_path **ppath,
3734                           unsigned int *allocated)
3735{
3736        struct ext4_ext_path *path = *ppath;
3737        struct ext4_extent *ex;
3738        ext4_lblk_t ee_block;
3739        unsigned int ee_len;
3740        int depth;
3741        int err = 0;
3742
3743        /*
3744         * Make sure that the extent is no bigger than we support with
3745         * unwritten extent
3746         */
3747        if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3748                map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3749
3750        depth = ext_depth(inode);
3751        ex = path[depth].p_ext;
3752        ee_block = le32_to_cpu(ex->ee_block);
3753        ee_len = ext4_ext_get_actual_len(ex);
3754
3755        ext_debug(inode, "logical block %llu, max_blocks %u\n",
3756                  (unsigned long long)ee_block, ee_len);
3757
3758        if (ee_block != map->m_lblk || ee_len > map->m_len) {
3759                err = ext4_split_convert_extents(handle, inode, map, ppath,
3760                                EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3761                if (err < 0)
3762                        return err;
3763                path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3764                if (IS_ERR(path))
3765                        return PTR_ERR(path);
3766                depth = ext_depth(inode);
3767                ex = path[depth].p_ext;
3768                if (!ex) {
3769                        EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3770                                         (unsigned long) map->m_lblk);
3771                        return -EFSCORRUPTED;
3772                }
3773        }
3774
3775        err = ext4_ext_get_access(handle, inode, path + depth);
3776        if (err)
3777                return err;
3778        /* first mark the extent as unwritten */
3779        ext4_ext_mark_unwritten(ex);
3780
3781        /* note: ext4_ext_correct_indexes() isn't needed here because
3782         * borders are not changed
3783         */
3784        ext4_ext_try_to_merge(handle, inode, path, ex);
3785
3786        /* Mark modified extent as dirty */
3787        err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3788        if (err)
3789                return err;
3790        ext4_ext_show_leaf(inode, path);
3791
3792        ext4_update_inode_fsync_trans(handle, inode, 1);
3793
3794        map->m_flags |= EXT4_MAP_UNWRITTEN;
3795        if (*allocated > map->m_len)
3796                *allocated = map->m_len;
3797        map->m_len = *allocated;
3798        return 0;
3799}
3800
3801static int
3802ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3803                        struct ext4_map_blocks *map,
3804                        struct ext4_ext_path **ppath, int flags,
3805                        unsigned int allocated, ext4_fsblk_t newblock)
3806{
3807        struct ext4_ext_path __maybe_unused *path = *ppath;
3808        int ret = 0;
3809        int err = 0;
3810
3811        ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3812                  (unsigned long long)map->m_lblk, map->m_len, flags,
3813                  allocated);
3814        ext4_ext_show_leaf(inode, path);
3815
3816        /*
3817         * When writing into unwritten space, we should not fail to
3818         * allocate metadata blocks for the new extent block if needed.
3819         */
3820        flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3821
3822        trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3823                                                    allocated, newblock);
3824
3825        /* get_block() before submitting IO, split the extent */
3826        if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3827                ret = ext4_split_convert_extents(handle, inode, map, ppath,
3828                                         flags | EXT4_GET_BLOCKS_CONVERT);
3829                if (ret < 0) {
3830                        err = ret;
3831                        goto out2;
3832                }
3833                /*
3834                 * shouldn't get a 0 return when splitting an extent unless
3835                 * m_len is 0 (bug) or extent has been corrupted
3836                 */
3837                if (unlikely(ret == 0)) {
3838                        EXT4_ERROR_INODE(inode,
3839                                         "unexpected ret == 0, m_len = %u",
3840                                         map->m_len);
3841                        err = -EFSCORRUPTED;
3842                        goto out2;
3843                }
3844                map->m_flags |= EXT4_MAP_UNWRITTEN;
3845                goto out;
3846        }
3847        /* IO end_io complete, convert the filled extent to written */
3848        if (flags & EXT4_GET_BLOCKS_CONVERT) {
3849                err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3850                                                           ppath);
3851                if (err < 0)
3852                        goto out2;
3853                ext4_update_inode_fsync_trans(handle, inode, 1);
3854                goto map_out;
3855        }
3856        /* buffered IO cases */
3857        /*
3858         * repeat fallocate creation request
3859         * we already have an unwritten extent
3860         */
3861        if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3862                map->m_flags |= EXT4_MAP_UNWRITTEN;
3863                goto map_out;
3864        }
3865
3866        /* buffered READ or buffered write_begin() lookup */
3867        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3868                /*
3869                 * We have blocks reserved already.  We
3870                 * return allocated blocks so that delalloc
3871                 * won't do block reservation for us.  But
3872                 * the buffer head will be unmapped so that
3873                 * a read from the block returns 0s.
3874                 */
3875                map->m_flags |= EXT4_MAP_UNWRITTEN;
3876                goto out1;
3877        }
3878
3879        /*
3880         * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3881         * For buffered writes, at writepage time, etc.  Convert a
3882         * discovered unwritten extent to written.
3883         */
3884        ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3885        if (ret < 0) {
3886                err = ret;
3887                goto out2;
3888        }
3889        ext4_update_inode_fsync_trans(handle, inode, 1);
3890        /*
3891         * shouldn't get a 0 return when converting an unwritten extent
3892         * unless m_len is 0 (bug) or extent has been corrupted
3893         */
3894        if (unlikely(ret == 0)) {
3895                EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3896                                 map->m_len);
3897                err = -EFSCORRUPTED;
3898                goto out2;
3899        }
3900
3901out:
3902        allocated = ret;
3903        map->m_flags |= EXT4_MAP_NEW;
3904map_out:
3905        map->m_flags |= EXT4_MAP_MAPPED;
3906out1:
3907        map->m_pblk = newblock;
3908        if (allocated > map->m_len)
3909                allocated = map->m_len;
3910        map->m_len = allocated;
3911        ext4_ext_show_leaf(inode, path);
3912out2:
3913        return err ? err : allocated;
3914}
3915
3916/*
3917 * get_implied_cluster_alloc - check to see if the requested
3918 * allocation (in the map structure) overlaps with a cluster already
3919 * allocated in an extent.
3920 *      @sb     The filesystem superblock structure
3921 *      @map    The requested lblk->pblk mapping
3922 *      @ex     The extent structure which might contain an implied
3923 *                      cluster allocation
3924 *
3925 * This function is called by ext4_ext_map_blocks() after we failed to
3926 * find blocks that were already in the inode's extent tree.  Hence,
3927 * we know that the beginning of the requested region cannot overlap
3928 * the extent from the inode's extent tree.  There are three cases we
3929 * want to catch.  The first is this case:
3930 *
3931 *               |--- cluster # N--|
3932 *    |--- extent ---|  |---- requested region ---|
3933 *                      |==========|
3934 *
3935 * The second case that we need to test for is this one:
3936 *
3937 *   |--------- cluster # N ----------------|
3938 *         |--- requested region --|   |------- extent ----|
3939 *         |=======================|
3940 *
3941 * The third case is when the requested region lies between two extents
3942 * within the same cluster:
3943 *          |------------- cluster # N-------------|
3944 * |----- ex -----|                  |---- ex_right ----|
3945 *                  |------ requested region ------|
3946 *                  |================|
3947 *
3948 * In each of the above cases, we need to set the map->m_pblk and
3949 * map->m_len so it corresponds to the return the extent labelled as
3950 * "|====|" from cluster #N, since it is already in use for data in
3951 * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3952 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3953 * as a new "allocated" block region.  Otherwise, we will return 0 and
3954 * ext4_ext_map_blocks() will then allocate one or more new clusters
3955 * by calling ext4_mb_new_blocks().
3956 */
3957static int get_implied_cluster_alloc(struct super_block *sb,
3958                                     struct ext4_map_blocks *map,
3959                                     struct ext4_extent *ex,
3960                                     struct ext4_ext_path *path)
3961{
3962        struct ext4_sb_info *sbi = EXT4_SB(sb);
3963        ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
3964        ext4_lblk_t ex_cluster_start, ex_cluster_end;
3965        ext4_lblk_t rr_cluster_start;
3966        ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3967        ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3968        unsigned short ee_len = ext4_ext_get_actual_len(ex);
3969
3970        /* The extent passed in that we are trying to match */
3971        ex_cluster_start = EXT4_B2C(sbi, ee_block);
3972        ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3973
3974        /* The requested region passed into ext4_map_blocks() */
3975        rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3976
3977        if ((rr_cluster_start == ex_cluster_end) ||
3978            (rr_cluster_start == ex_cluster_start)) {
3979                if (rr_cluster_start == ex_cluster_end)
3980                        ee_start += ee_len - 1;
3981                map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
3982                map->m_len = min(map->m_len,
3983                                 (unsigned) sbi->s_cluster_ratio - c_offset);
3984                /*
3985                 * Check for and handle this case:
3986                 *
3987                 *   |--------- cluster # N-------------|
3988                 *                     |------- extent ----|
3989                 *         |--- requested region ---|
3990                 *         |===========|
3991                 */
3992
3993                if (map->m_lblk < ee_block)
3994                        map->m_len = min(map->m_len, ee_block - map->m_lblk);
3995
3996                /*
3997                 * Check for the case where there is already another allocated
3998                 * block to the right of 'ex' but before the end of the cluster.
3999                 *
4000                 *          |------------- cluster # N-------------|
4001                 * |----- ex -----|                  |---- ex_right ----|
4002                 *                  |------ requested region ------|
4003                 *                  |================|
4004                 */
4005                if (map->m_lblk > ee_block) {
4006                        ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4007                        map->m_len = min(map->m_len, next - map->m_lblk);
4008                }
4009
4010                trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4011                return 1;
4012        }
4013
4014        trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4015        return 0;
4016}
4017
4018
4019/*
4020 * Block allocation/map/preallocation routine for extents based files
4021 *
4022 *
4023 * Need to be called with
4024 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4025 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4026 *
4027 * return > 0, number of blocks already mapped/allocated
4028 *          if create == 0 and these are pre-allocated blocks
4029 *              buffer head is unmapped
4030 *          otherwise blocks are mapped
4031 *
4032 * return = 0, if plain look up failed (blocks have not been allocated)
4033 *          buffer head is unmapped
4034 *
4035 * return < 0, error case.
4036 */
4037int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4038                        struct ext4_map_blocks *map, int flags)
4039{
4040        struct ext4_ext_path *path = NULL;
4041        struct ext4_extent newex, *ex, ex2;
4042        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4043        ext4_fsblk_t newblock = 0, pblk;
4044        int err = 0, depth, ret;
4045        unsigned int allocated = 0, offset = 0;
4046        unsigned int allocated_clusters = 0;
4047        struct ext4_allocation_request ar;
4048        ext4_lblk_t cluster_offset;
4049
4050        ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4051        trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4052
4053        /* find extent for this block */
4054        path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4055        if (IS_ERR(path)) {
4056                err = PTR_ERR(path);
4057                path = NULL;
4058                goto out;
4059        }
4060
4061        depth = ext_depth(inode);
4062
4063        /*
4064         * consistent leaf must not be empty;
4065         * this situation is possible, though, _during_ tree modification;
4066         * this is why assert can't be put in ext4_find_extent()
4067         */
4068        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4069                EXT4_ERROR_INODE(inode, "bad extent address "
4070                                 "lblock: %lu, depth: %d pblock %lld",
4071                                 (unsigned long) map->m_lblk, depth,
4072                                 path[depth].p_block);
4073                err = -EFSCORRUPTED;
4074                goto out;
4075        }
4076
4077        ex = path[depth].p_ext;
4078        if (ex) {
4079                ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4080                ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4081                unsigned short ee_len;
4082
4083
4084                /*
4085                 * unwritten extents are treated as holes, except that
4086                 * we split out initialized portions during a write.
4087                 */
4088                ee_len = ext4_ext_get_actual_len(ex);
4089
4090                trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4091
4092                /* if found extent covers block, simply return it */
4093                if (in_range(map->m_lblk, ee_block, ee_len)) {
4094                        newblock = map->m_lblk - ee_block + ee_start;
4095                        /* number of remaining blocks in the extent */
4096                        allocated = ee_len - (map->m_lblk - ee_block);
4097                        ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4098                                  map->m_lblk, ee_block, ee_len, newblock);
4099
4100                        /*
4101                         * If the extent is initialized check whether the
4102                         * caller wants to convert it to unwritten.
4103                         */
4104                        if ((!ext4_ext_is_unwritten(ex)) &&
4105                            (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4106                                err = convert_initialized_extent(handle,
4107                                        inode, map, &path, &allocated);
4108                                goto out;
4109                        } else if (!ext4_ext_is_unwritten(ex)) {
4110                                map->m_flags |= EXT4_MAP_MAPPED;
4111                                map->m_pblk = newblock;
4112                                if (allocated > map->m_len)
4113                                        allocated = map->m_len;
4114                                map->m_len = allocated;
4115                                ext4_ext_show_leaf(inode, path);
4116                                goto out;
4117                        }
4118
4119                        ret = ext4_ext_handle_unwritten_extents(
4120                                handle, inode, map, &path, flags,
4121                                allocated, newblock);
4122                        if (ret < 0)
4123                                err = ret;
4124                        else
4125                                allocated = ret;
4126                        goto out;
4127                }
4128        }
4129
4130        /*
4131         * requested block isn't allocated yet;
4132         * we couldn't try to create block if create flag is zero
4133         */
4134        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4135                ext4_lblk_t hole_start, hole_len;
4136
4137                hole_start = map->m_lblk;
4138                hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4139                /*
4140                 * put just found gap into cache to speed up
4141                 * subsequent requests
4142                 */
4143                ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4144
4145                /* Update hole_len to reflect hole size after map->m_lblk */
4146                if (hole_start != map->m_lblk)
4147                        hole_len -= map->m_lblk - hole_start;
4148                map->m_pblk = 0;
4149                map->m_len = min_t(unsigned int, map->m_len, hole_len);
4150
4151                goto out;
4152        }
4153
4154        /*
4155         * Okay, we need to do block allocation.
4156         */
4157        newex.ee_block = cpu_to_le32(map->m_lblk);
4158        cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4159
4160        /*
4161         * If we are doing bigalloc, check to see if the extent returned
4162         * by ext4_find_extent() implies a cluster we can use.
4163         */
4164        if (cluster_offset && ex &&
4165            get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4166                ar.len = allocated = map->m_len;
4167                newblock = map->m_pblk;
4168                goto got_allocated_blocks;
4169        }
4170
4171        /* find neighbour allocated blocks */
4172        ar.lleft = map->m_lblk;
4173        err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4174        if (err)
4175                goto out;
4176        ar.lright = map->m_lblk;
4177        err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4178        if (err < 0)
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) && err &&
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);
4330out:
4331        ext4_ext_drop_refs(path);
4332        kfree(path);
4333
4334        trace_ext4_ext_map_blocks_exit(inode, flags, map,
4335                                       err ? err : allocated);
4336        return err ? err : allocated;
4337}
4338
4339int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4340{
4341        struct super_block *sb = inode->i_sb;
4342        ext4_lblk_t last_block;
4343        int err = 0;
4344
4345        /*
4346         * TODO: optimization is possible here.
4347         * Probably we need not scan at all,
4348         * because page truncation is enough.
4349         */
4350
4351        /* we have to know where to truncate from in crash case */
4352        EXT4_I(inode)->i_disksize = inode->i_size;
4353        err = ext4_mark_inode_dirty(handle, inode);
4354        if (err)
4355                return err;
4356
4357        last_block = (inode->i_size + sb->s_blocksize - 1)
4358                        >> EXT4_BLOCK_SIZE_BITS(sb);
4359retry:
4360        err = ext4_es_remove_extent(inode, last_block,
4361                                    EXT_MAX_BLOCKS - last_block);
4362        if (err == -ENOMEM) {
4363                cond_resched();
4364                congestion_wait(BLK_RW_ASYNC, HZ/50);
4365                goto retry;
4366        }
4367        if (err)
4368                return err;
4369retry_remove_space:
4370        err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4371        if (err == -ENOMEM) {
4372                cond_resched();
4373                congestion_wait(BLK_RW_ASYNC, HZ/50);
4374                goto retry_remove_space;
4375        }
4376        return err;
4377}
4378
4379static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4380                                  ext4_lblk_t len, loff_t new_size,
4381                                  int flags)
4382{
4383        struct inode *inode = file_inode(file);
4384        handle_t *handle;
4385        int ret = 0;
4386        int ret2 = 0, ret3 = 0;
4387        int retries = 0;
4388        int depth = 0;
4389        struct ext4_map_blocks map;
4390        unsigned int credits;
4391        loff_t epos;
4392
4393        BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4394        map.m_lblk = offset;
4395        map.m_len = len;
4396        /*
4397         * Don't normalize the request if it can fit in one extent so
4398         * that it doesn't get unnecessarily split into multiple
4399         * extents.
4400         */
4401        if (len <= EXT_UNWRITTEN_MAX_LEN)
4402                flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4403
4404        /*
4405         * credits to insert 1 extent into extent tree
4406         */
4407        credits = ext4_chunk_trans_blocks(inode, len);
4408        depth = ext_depth(inode);
4409
4410retry:
4411        while (ret >= 0 && len) {
4412                /*
4413                 * Recalculate credits when extent tree depth changes.
4414                 */
4415                if (depth != ext_depth(inode)) {
4416                        credits = ext4_chunk_trans_blocks(inode, len);
4417                        depth = ext_depth(inode);
4418                }
4419
4420                handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4421                                            credits);
4422                if (IS_ERR(handle)) {
4423                        ret = PTR_ERR(handle);
4424                        break;
4425                }
4426                ret = ext4_map_blocks(handle, inode, &map, flags);
4427                if (ret <= 0) {
4428                        ext4_debug("inode #%lu: block %u: len %u: "
4429                                   "ext4_ext_map_blocks returned %d",
4430                                   inode->i_ino, map.m_lblk,
4431                                   map.m_len, ret);
4432                        ext4_mark_inode_dirty(handle, inode);
4433                        ret2 = ext4_journal_stop(handle);
4434                        break;
4435                }
4436                map.m_lblk += ret;
4437                map.m_len = len = len - ret;
4438                epos = (loff_t)map.m_lblk << inode->i_blkbits;
4439                inode->i_ctime = current_time(inode);
4440                if (new_size) {
4441                        if (epos > new_size)
4442                                epos = new_size;
4443                        if (ext4_update_inode_size(inode, epos) & 0x1)
4444                                inode->i_mtime = inode->i_ctime;
4445                }
4446                ret2 = ext4_mark_inode_dirty(handle, inode);
4447                ext4_update_inode_fsync_trans(handle, inode, 1);
4448                ret3 = ext4_journal_stop(handle);
4449                ret2 = ret3 ? ret3 : ret2;
4450                if (unlikely(ret2))
4451                        break;
4452        }
4453        if (ret == -ENOSPC &&
4454                        ext4_should_retry_alloc(inode->i_sb, &retries)) {
4455                ret = 0;
4456                goto retry;
4457        }
4458
4459        return ret > 0 ? ret2 : ret;
4460}
4461
4462static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4463
4464static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4465
4466static long ext4_zero_range(struct file *file, loff_t offset,
4467                            loff_t len, int mode)
4468{
4469        struct inode *inode = file_inode(file);
4470        handle_t *handle = NULL;
4471        unsigned int max_blocks;
4472        loff_t new_size = 0;
4473        int ret = 0;
4474        int flags;
4475        int credits;
4476        int partial_begin, partial_end;
4477        loff_t start, end;
4478        ext4_lblk_t lblk;
4479        unsigned int blkbits = inode->i_blkbits;
4480
4481        trace_ext4_zero_range(inode, offset, len, mode);
4482
4483        /* Call ext4_force_commit to flush all data in case of data=journal. */
4484        if (ext4_should_journal_data(inode)) {
4485                ret = ext4_force_commit(inode->i_sb);
4486                if (ret)
4487                        return ret;
4488        }
4489
4490        /*
4491         * Round up offset. This is not fallocate, we need to zero out
4492         * blocks, so convert interior block aligned part of the range to
4493         * unwritten and possibly manually zero out unaligned parts of the
4494         * range.
4495         */
4496        start = round_up(offset, 1 << blkbits);
4497        end = round_down((offset + len), 1 << blkbits);
4498
4499        if (start < offset || end > offset + len)
4500                return -EINVAL;
4501        partial_begin = offset & ((1 << blkbits) - 1);
4502        partial_end = (offset + len) & ((1 << blkbits) - 1);
4503
4504        lblk = start >> blkbits;
4505        max_blocks = (end >> blkbits);
4506        if (max_blocks < lblk)
4507                max_blocks = 0;
4508        else
4509                max_blocks -= lblk;
4510
4511        inode_lock(inode);
4512
4513        /*
4514         * Indirect files do not support unwritten extents
4515         */
4516        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4517                ret = -EOPNOTSUPP;
4518                goto out_mutex;
4519        }
4520
4521        if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4522            (offset + len > inode->i_size ||
4523             offset + len > EXT4_I(inode)->i_disksize)) {
4524                new_size = offset + len;
4525                ret = inode_newsize_ok(inode, new_size);
4526                if (ret)
4527                        goto out_mutex;
4528        }
4529
4530        flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4531
4532        /* Wait all existing dio workers, newcomers will block on i_mutex */
4533        inode_dio_wait(inode);
4534
4535        /* Preallocate the range including the unaligned edges */
4536        if (partial_begin || partial_end) {
4537                ret = ext4_alloc_file_blocks(file,
4538                                round_down(offset, 1 << blkbits) >> blkbits,
4539                                (round_up((offset + len), 1 << blkbits) -
4540                                 round_down(offset, 1 << blkbits)) >> blkbits,
4541                                new_size, flags);
4542                if (ret)
4543                        goto out_mutex;
4544
4545        }
4546
4547        /* Zero range excluding the unaligned edges */
4548        if (max_blocks > 0) {
4549                flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4550                          EXT4_EX_NOCACHE);
4551
4552                /*
4553                 * Prevent page faults from reinstantiating pages we have
4554                 * released from page cache.
4555                 */
4556                down_write(&EXT4_I(inode)->i_mmap_sem);
4557
4558                ret = ext4_break_layouts(inode);
4559                if (ret) {
4560                        up_write(&EXT4_I(inode)->i_mmap_sem);
4561                        goto out_mutex;
4562                }
4563
4564                ret = ext4_update_disksize_before_punch(inode, offset, len);
4565                if (ret) {
4566                        up_write(&EXT4_I(inode)->i_mmap_sem);
4567                        goto out_mutex;
4568                }
4569                /* Now release the pages and zero block aligned part of pages */
4570                truncate_pagecache_range(inode, start, end - 1);
4571                inode->i_mtime = inode->i_ctime = current_time(inode);
4572
4573                ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4574                                             flags);
4575                up_write(&EXT4_I(inode)->i_mmap_sem);
4576                if (ret)
4577                        goto out_mutex;
4578        }
4579        if (!partial_begin && !partial_end)
4580                goto out_mutex;
4581
4582        /*
4583         * In worst case we have to writeout two nonadjacent unwritten
4584         * blocks and update the inode
4585         */
4586        credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4587        if (ext4_should_journal_data(inode))
4588                credits += 2;
4589        handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4590        if (IS_ERR(handle)) {
4591                ret = PTR_ERR(handle);
4592                ext4_std_error(inode->i_sb, ret);
4593                goto out_mutex;
4594        }
4595
4596        inode->i_mtime = inode->i_ctime = current_time(inode);
4597        if (new_size)
4598                ext4_update_inode_size(inode, new_size);
4599        ret = ext4_mark_inode_dirty(handle, inode);
4600        if (unlikely(ret))
4601                goto out_handle;
4602        ext4_fc_track_range(handle, inode, offset >> inode->i_sb->s_blocksize_bits,
4603                        (offset + len - 1) >> inode->i_sb->s_blocksize_bits);
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        ext4_fc_start_update(inode);
4653
4654        if (mode & FALLOC_FL_PUNCH_HOLE) {
4655                ret = ext4_punch_hole(inode, offset, len);
4656                goto exit;
4657        }
4658
4659        ret = ext4_convert_inline_data(inode);
4660        if (ret)
4661                goto exit;
4662
4663        if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4664                ret = ext4_collapse_range(inode, offset, len);
4665                goto exit;
4666        }
4667
4668        if (mode & FALLOC_FL_INSERT_RANGE) {
4669                ret = ext4_insert_range(inode, offset, len);
4670                goto exit;
4671        }
4672
4673        if (mode & FALLOC_FL_ZERO_RANGE) {
4674                ret = ext4_zero_range(file, offset, len, mode);
4675                goto exit;
4676        }
4677        trace_ext4_fallocate_enter(inode, offset, len, mode);
4678        lblk = offset >> blkbits;
4679
4680        max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4681        flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4682
4683        inode_lock(inode);
4684
4685        /*
4686         * We only support preallocation for extent-based files only
4687         */
4688        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4689                ret = -EOPNOTSUPP;
4690                goto out;
4691        }
4692
4693        if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4694            (offset + len > inode->i_size ||
4695             offset + len > EXT4_I(inode)->i_disksize)) {
4696                new_size = offset + len;
4697                ret = inode_newsize_ok(inode, new_size);
4698                if (ret)
4699                        goto out;
4700        }
4701
4702        /* Wait all existing dio workers, newcomers will block on i_mutex */
4703        inode_dio_wait(inode);
4704
4705        ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4706        if (ret)
4707                goto out;
4708
4709        if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4710                ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4711                                        EXT4_I(inode)->i_sync_tid);
4712        }
4713out:
4714        inode_unlock(inode);
4715        trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4716exit:
4717        ext4_fc_stop_update(inode);
4718        return ret;
4719}
4720
4721/*
4722 * This function convert a range of blocks to written extents
4723 * The caller of this function will pass the start offset and the size.
4724 * all unwritten extents within this range will be converted to
4725 * written extents.
4726 *
4727 * This function is called from the direct IO end io call back
4728 * function, to convert the fallocated extents after IO is completed.
4729 * Returns 0 on success.
4730 */
4731int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4732                                   loff_t offset, ssize_t len)
4733{
4734        unsigned int max_blocks;
4735        int ret = 0, ret2 = 0, ret3 = 0;
4736        struct ext4_map_blocks map;
4737        unsigned int blkbits = inode->i_blkbits;
4738        unsigned int credits = 0;
4739
4740        map.m_lblk = offset >> blkbits;
4741        max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4742
4743        if (!handle) {
4744                /*
4745                 * credits to insert 1 extent into extent tree
4746                 */
4747                credits = ext4_chunk_trans_blocks(inode, max_blocks);
4748        }
4749        while (ret >= 0 && ret < max_blocks) {
4750                map.m_lblk += ret;
4751                map.m_len = (max_blocks -= ret);
4752                if (credits) {
4753                        handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4754                                                    credits);
4755                        if (IS_ERR(handle)) {
4756                                ret = PTR_ERR(handle);
4757                                break;
4758                        }
4759                }
4760                ret = ext4_map_blocks(handle, inode, &map,
4761                                      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4762                if (ret <= 0)
4763                        ext4_warning(inode->i_sb,
4764                                     "inode #%lu: block %u: len %u: "
4765                                     "ext4_ext_map_blocks returned %d",
4766                                     inode->i_ino, map.m_lblk,
4767                                     map.m_len, ret);
4768                ret2 = ext4_mark_inode_dirty(handle, inode);
4769                if (credits) {
4770                        ret3 = ext4_journal_stop(handle);
4771                        if (unlikely(ret3))
4772                                ret2 = ret3;
4773                }
4774
4775                if (ret <= 0 || ret2)
4776                        break;
4777        }
4778        return ret > 0 ? ret2 : ret;
4779}
4780
4781int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4782{
4783        int ret = 0, err = 0;
4784        struct ext4_io_end_vec *io_end_vec;
4785
4786        /*
4787         * This is somewhat ugly but the idea is clear: When transaction is
4788         * reserved, everything goes into it. Otherwise we rather start several
4789         * smaller transactions for conversion of each extent separately.
4790         */
4791        if (handle) {
4792                handle = ext4_journal_start_reserved(handle,
4793                                                     EXT4_HT_EXT_CONVERT);
4794                if (IS_ERR(handle))
4795                        return PTR_ERR(handle);
4796        }
4797
4798        list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4799                ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4800                                                     io_end_vec->offset,
4801                                                     io_end_vec->size);
4802                if (ret)
4803                        break;
4804        }
4805
4806        if (handle)
4807                err = ext4_journal_stop(handle);
4808
4809        return ret < 0 ? ret : err;
4810}
4811
4812static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4813{
4814        __u64 physical = 0;
4815        __u64 length = 0;
4816        int blockbits = inode->i_sb->s_blocksize_bits;
4817        int error = 0;
4818        u16 iomap_type;
4819
4820        /* in-inode? */
4821        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4822                struct ext4_iloc iloc;
4823                int offset;     /* offset of xattr in inode */
4824
4825                error = ext4_get_inode_loc(inode, &iloc);
4826                if (error)
4827                        return error;
4828                physical = (__u64)iloc.bh->b_blocknr << blockbits;
4829                offset = EXT4_GOOD_OLD_INODE_SIZE +
4830                                EXT4_I(inode)->i_extra_isize;
4831                physical += offset;
4832                length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4833                brelse(iloc.bh);
4834                iomap_type = IOMAP_INLINE;
4835        } else if (EXT4_I(inode)->i_file_acl) { /* external block */
4836                physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4837                length = inode->i_sb->s_blocksize;
4838                iomap_type = IOMAP_MAPPED;
4839        } else {
4840                /* no in-inode or external block for xattr, so return -ENOENT */
4841                error = -ENOENT;
4842                goto out;
4843        }
4844
4845        iomap->addr = physical;
4846        iomap->offset = 0;
4847        iomap->length = length;
4848        iomap->type = iomap_type;
4849        iomap->flags = 0;
4850out:
4851        return error;
4852}
4853
4854static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4855                                  loff_t length, unsigned flags,
4856                                  struct iomap *iomap, struct iomap *srcmap)
4857{
4858        int error;
4859
4860        error = ext4_iomap_xattr_fiemap(inode, iomap);
4861        if (error == 0 && (offset >= iomap->length))
4862                error = -ENOENT;
4863        return error;
4864}
4865
4866static const struct iomap_ops ext4_iomap_xattr_ops = {
4867        .iomap_begin            = ext4_iomap_xattr_begin,
4868};
4869
4870static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4871{
4872        u64 maxbytes;
4873
4874        if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4875                maxbytes = inode->i_sb->s_maxbytes;
4876        else
4877                maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4878
4879        if (*len == 0)
4880                return -EINVAL;
4881        if (start > maxbytes)
4882                return -EFBIG;
4883
4884        /*
4885         * Shrink request scope to what the fs can actually handle.
4886         */
4887        if (*len > maxbytes || (maxbytes - *len) < start)
4888                *len = maxbytes - start;
4889        return 0;
4890}
4891
4892int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4893                u64 start, u64 len)
4894{
4895        int error = 0;
4896
4897        if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4898                error = ext4_ext_precache(inode);
4899                if (error)
4900                        return error;
4901                fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4902        }
4903
4904        /*
4905         * For bitmap files the maximum size limit could be smaller than
4906         * s_maxbytes, so check len here manually instead of just relying on the
4907         * generic check.
4908         */
4909        error = ext4_fiemap_check_ranges(inode, start, &len);
4910        if (error)
4911                return error;
4912
4913        if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4914                fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4915                return iomap_fiemap(inode, fieinfo, start, len,
4916                                    &ext4_iomap_xattr_ops);
4917        }
4918
4919        return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4920}
4921
4922int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4923                      __u64 start, __u64 len)
4924{
4925        ext4_lblk_t start_blk, len_blks;
4926        __u64 last_blk;
4927        int error = 0;
4928
4929        if (ext4_has_inline_data(inode)) {
4930                int has_inline;
4931
4932                down_read(&EXT4_I(inode)->xattr_sem);
4933                has_inline = ext4_has_inline_data(inode);
4934                up_read(&EXT4_I(inode)->xattr_sem);
4935                if (has_inline)
4936                        return 0;
4937        }
4938
4939        if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4940                error = ext4_ext_precache(inode);
4941                if (error)
4942                        return error;
4943                fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4944        }
4945
4946        error = fiemap_prep(inode, fieinfo, start, &len, 0);
4947        if (error)
4948                return error;
4949
4950        error = ext4_fiemap_check_ranges(inode, start, &len);
4951        if (error)
4952                return error;
4953
4954        start_blk = start >> inode->i_sb->s_blocksize_bits;
4955        last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4956        if (last_blk >= EXT_MAX_BLOCKS)
4957                last_blk = EXT_MAX_BLOCKS-1;
4958        len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4959
4960        /*
4961         * Walk the extent tree gathering extent information
4962         * and pushing extents back to the user.
4963         */
4964        return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
4965}
4966
4967/*
4968 * ext4_access_path:
4969 * Function to access the path buffer for marking it dirty.
4970 * It also checks if there are sufficient credits left in the journal handle
4971 * to update path.
4972 */
4973static int
4974ext4_access_path(handle_t *handle, struct inode *inode,
4975                struct ext4_ext_path *path)
4976{
4977        int credits, err;
4978
4979        if (!ext4_handle_valid(handle))
4980                return 0;
4981
4982        /*
4983         * Check if need to extend journal credits
4984         * 3 for leaf, sb, and inode plus 2 (bmap and group
4985         * descriptor) for each block group; assume two block
4986         * groups
4987         */
4988        credits = ext4_writepage_trans_blocks(inode);
4989        err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
4990        if (err < 0)
4991                return err;
4992
4993        err = ext4_ext_get_access(handle, inode, path);
4994        return err;
4995}
4996
4997/*
4998 * ext4_ext_shift_path_extents:
4999 * Shift the extents of a path structure lying between path[depth].p_ext
5000 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5001 * if it is right shift or left shift operation.
5002 */
5003static int
5004ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5005                            struct inode *inode, handle_t *handle,
5006                            enum SHIFT_DIRECTION SHIFT)
5007{
5008        int depth, err = 0;
5009        struct ext4_extent *ex_start, *ex_last;
5010        bool update = false;
5011        depth = path->p_depth;
5012
5013        while (depth >= 0) {
5014                if (depth == path->p_depth) {
5015                        ex_start = path[depth].p_ext;
5016                        if (!ex_start)
5017                                return -EFSCORRUPTED;
5018
5019                        ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5020
5021                        err = ext4_access_path(handle, inode, path + depth);
5022                        if (err)
5023                                goto out;
5024
5025                        if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5026                                update = true;
5027
5028                        while (ex_start <= ex_last) {
5029                                if (SHIFT == SHIFT_LEFT) {
5030                                        le32_add_cpu(&ex_start->ee_block,
5031                                                -shift);
5032                                        /* Try to merge to the left. */
5033                                        if ((ex_start >
5034                                            EXT_FIRST_EXTENT(path[depth].p_hdr))
5035                                            &&
5036                                            ext4_ext_try_to_merge_right(inode,
5037                                            path, ex_start - 1))
5038                                                ex_last--;
5039                                        else
5040                                                ex_start++;
5041                                } else {
5042                                        le32_add_cpu(&ex_last->ee_block, shift);
5043                                        ext4_ext_try_to_merge_right(inode, path,
5044                                                ex_last);
5045                                        ex_last--;
5046                                }
5047                        }
5048                        err = ext4_ext_dirty(handle, inode, path + depth);
5049                        if (err)
5050                                goto out;
5051
5052                        if (--depth < 0 || !update)
5053                                break;
5054                }
5055
5056                /* Update index too */
5057                err = ext4_access_path(handle, inode, path + depth);
5058                if (err)
5059                        goto out;
5060
5061                if (SHIFT == SHIFT_LEFT)
5062                        le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5063                else
5064                        le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5065                err = ext4_ext_dirty(handle, inode, path + depth);
5066                if (err)
5067                        goto out;
5068
5069                /* we are done if current index is not a starting index */
5070                if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5071                        break;
5072
5073                depth--;
5074        }
5075
5076out:
5077        return err;
5078}
5079
5080/*
5081 * ext4_ext_shift_extents:
5082 * All the extents which lies in the range from @start to the last allocated
5083 * block for the @inode are shifted either towards left or right (depending
5084 * upon @SHIFT) by @shift blocks.
5085 * On success, 0 is returned, error otherwise.
5086 */
5087static int
5088ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5089                       ext4_lblk_t start, ext4_lblk_t shift,
5090                       enum SHIFT_DIRECTION SHIFT)
5091{
5092        struct ext4_ext_path *path;
5093        int ret = 0, depth;
5094        struct ext4_extent *extent;
5095        ext4_lblk_t stop, *iterator, ex_start, ex_end;
5096
5097        /* Let path point to the last extent */
5098        path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5099                                EXT4_EX_NOCACHE);
5100        if (IS_ERR(path))
5101                return PTR_ERR(path);
5102
5103        depth = path->p_depth;
5104        extent = path[depth].p_ext;
5105        if (!extent)
5106                goto out;
5107
5108        stop = le32_to_cpu(extent->ee_block);
5109
5110       /*
5111        * For left shifts, make sure the hole on the left is big enough to
5112        * accommodate the shift.  For right shifts, make sure the last extent
5113        * won't be shifted beyond EXT_MAX_BLOCKS.
5114        */
5115        if (SHIFT == SHIFT_LEFT) {
5116                path = ext4_find_extent(inode, start - 1, &path,
5117                                        EXT4_EX_NOCACHE);
5118                if (IS_ERR(path))
5119                        return PTR_ERR(path);
5120                depth = path->p_depth;
5121                extent =  path[depth].p_ext;
5122                if (extent) {
5123                        ex_start = le32_to_cpu(extent->ee_block);
5124                        ex_end = le32_to_cpu(extent->ee_block) +
5125                                ext4_ext_get_actual_len(extent);
5126                } else {
5127                        ex_start = 0;
5128                        ex_end = 0;
5129                }
5130
5131                if ((start == ex_start && shift > ex_start) ||
5132                    (shift > start - ex_end)) {
5133                        ret = -EINVAL;
5134                        goto out;
5135                }
5136        } else {
5137                if (shift > EXT_MAX_BLOCKS -
5138                    (stop + ext4_ext_get_actual_len(extent))) {
5139                        ret = -EINVAL;
5140                        goto out;
5141                }
5142        }
5143
5144        /*
5145         * In case of left shift, iterator points to start and it is increased
5146         * till we reach stop. In case of right shift, iterator points to stop
5147         * and it is decreased till we reach start.
5148         */
5149        if (SHIFT == SHIFT_LEFT)
5150                iterator = &start;
5151        else
5152                iterator = &stop;
5153
5154        /*
5155         * Its safe to start updating extents.  Start and stop are unsigned, so
5156         * in case of right shift if extent with 0 block is reached, iterator
5157         * becomes NULL to indicate the end of the loop.
5158         */
5159        while (iterator && start <= stop) {
5160                path = ext4_find_extent(inode, *iterator, &path,
5161                                        EXT4_EX_NOCACHE);
5162                if (IS_ERR(path))
5163                        return PTR_ERR(path);
5164                depth = path->p_depth;
5165                extent = path[depth].p_ext;
5166                if (!extent) {
5167                        EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5168                                         (unsigned long) *iterator);
5169                        return -EFSCORRUPTED;
5170                }
5171                if (SHIFT == SHIFT_LEFT && *iterator >
5172                    le32_to_cpu(extent->ee_block)) {
5173                        /* Hole, move to the next extent */
5174                        if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5175                                path[depth].p_ext++;
5176                        } else {
5177                                *iterator = ext4_ext_next_allocated_block(path);
5178                                continue;
5179                        }
5180                }
5181
5182                if (SHIFT == SHIFT_LEFT) {
5183                        extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5184                        *iterator = le32_to_cpu(extent->ee_block) +
5185                                        ext4_ext_get_actual_len(extent);
5186                } else {
5187                        extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5188                        if (le32_to_cpu(extent->ee_block) > 0)
5189                                *iterator = le32_to_cpu(extent->ee_block) - 1;
5190                        else
5191                                /* Beginning is reached, end of the loop */
5192                                iterator = NULL;
5193                        /* Update path extent in case we need to stop */
5194                        while (le32_to_cpu(extent->ee_block) < start)
5195                                extent++;
5196                        path[depth].p_ext = extent;
5197                }
5198                ret = ext4_ext_shift_path_extents(path, shift, inode,
5199                                handle, SHIFT);
5200                if (ret)
5201                        break;
5202        }
5203out:
5204        ext4_ext_drop_refs(path);
5205        kfree(path);
5206        return ret;
5207}
5208
5209/*
5210 * ext4_collapse_range:
5211 * This implements the fallocate's collapse range functionality for ext4
5212 * Returns: 0 and non-zero on error.
5213 */
5214static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5215{
5216        struct super_block *sb = inode->i_sb;
5217        ext4_lblk_t punch_start, punch_stop;
5218        handle_t *handle;
5219        unsigned int credits;
5220        loff_t new_size, ioffset;
5221        int ret;
5222
5223        /*
5224         * We need to test this early because xfstests assumes that a
5225         * collapse range of (0, 1) will return EOPNOTSUPP if the file
5226         * system does not support collapse range.
5227         */
5228        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5229                return -EOPNOTSUPP;
5230
5231        /* Collapse range works only on fs cluster size aligned regions. */
5232        if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5233                return -EINVAL;
5234
5235        trace_ext4_collapse_range(inode, offset, len);
5236
5237        punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5238        punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5239
5240        /* Call ext4_force_commit to flush all data in case of data=journal. */
5241        if (ext4_should_journal_data(inode)) {
5242                ret = ext4_force_commit(inode->i_sb);
5243                if (ret)
5244                        return ret;
5245        }
5246
5247        inode_lock(inode);
5248        /*
5249         * There is no need to overlap collapse range with EOF, in which case
5250         * it is effectively a truncate operation
5251         */
5252        if (offset + len >= inode->i_size) {
5253                ret = -EINVAL;
5254                goto out_mutex;
5255        }
5256
5257        /* Currently just for extent based files */
5258        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5259                ret = -EOPNOTSUPP;
5260                goto out_mutex;
5261        }
5262
5263        /* Wait for existing dio to complete */
5264        inode_dio_wait(inode);
5265
5266        /*
5267         * Prevent page faults from reinstantiating pages we have released from
5268         * page cache.
5269         */
5270        down_write(&EXT4_I(inode)->i_mmap_sem);
5271
5272        ret = ext4_break_layouts(inode);
5273        if (ret)
5274                goto out_mmap;
5275
5276        /*
5277         * Need to round down offset to be aligned with page size boundary
5278         * for page size > block size.
5279         */
5280        ioffset = round_down(offset, PAGE_SIZE);
5281        /*
5282         * Write tail of the last page before removed range since it will get
5283         * removed from the page cache below.
5284         */
5285        ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5286        if (ret)
5287                goto out_mmap;
5288        /*
5289         * Write data that will be shifted to preserve them when discarding
5290         * page cache below. We are also protected from pages becoming dirty
5291         * by i_mmap_sem.
5292         */
5293        ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5294                                           LLONG_MAX);
5295        if (ret)
5296                goto out_mmap;
5297        truncate_pagecache(inode, ioffset);
5298
5299        credits = ext4_writepage_trans_blocks(inode);
5300        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5301        if (IS_ERR(handle)) {
5302                ret = PTR_ERR(handle);
5303                goto out_mmap;
5304        }
5305        ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5306
5307        down_write(&EXT4_I(inode)->i_data_sem);
5308        ext4_discard_preallocations(inode, 0);
5309
5310        ret = ext4_es_remove_extent(inode, punch_start,
5311                                    EXT_MAX_BLOCKS - punch_start);
5312        if (ret) {
5313                up_write(&EXT4_I(inode)->i_data_sem);
5314                goto out_stop;
5315        }
5316
5317        ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5318        if (ret) {
5319                up_write(&EXT4_I(inode)->i_data_sem);
5320                goto out_stop;
5321        }
5322        ext4_discard_preallocations(inode, 0);
5323
5324        ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5325                                     punch_stop - punch_start, SHIFT_LEFT);
5326        if (ret) {
5327                up_write(&EXT4_I(inode)->i_data_sem);
5328                goto out_stop;
5329        }
5330
5331        new_size = inode->i_size - len;
5332        i_size_write(inode, new_size);
5333        EXT4_I(inode)->i_disksize = new_size;
5334
5335        up_write(&EXT4_I(inode)->i_data_sem);
5336        if (IS_SYNC(inode))
5337                ext4_handle_sync(handle);
5338        inode->i_mtime = inode->i_ctime = current_time(inode);
5339        ret = ext4_mark_inode_dirty(handle, inode);
5340        ext4_update_inode_fsync_trans(handle, inode, 1);
5341
5342out_stop:
5343        ext4_journal_stop(handle);
5344        ext4_fc_stop_ineligible(sb);
5345out_mmap:
5346        up_write(&EXT4_I(inode)->i_mmap_sem);
5347out_mutex:
5348        inode_unlock(inode);
5349        return ret;
5350}
5351
5352/*
5353 * ext4_insert_range:
5354 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5355 * The data blocks starting from @offset to the EOF are shifted by @len
5356 * towards right to create a hole in the @inode. Inode size is increased
5357 * by len bytes.
5358 * Returns 0 on success, error otherwise.
5359 */
5360static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5361{
5362        struct super_block *sb = inode->i_sb;
5363        handle_t *handle;
5364        struct ext4_ext_path *path;
5365        struct ext4_extent *extent;
5366        ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5367        unsigned int credits, ee_len;
5368        int ret = 0, depth, split_flag = 0;
5369        loff_t ioffset;
5370
5371        /*
5372         * We need to test this early because xfstests assumes that an
5373         * insert range of (0, 1) will return EOPNOTSUPP if the file
5374         * system does not support insert range.
5375         */
5376        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5377                return -EOPNOTSUPP;
5378
5379        /* Insert range works only on fs cluster size aligned regions. */
5380        if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5381                return -EINVAL;
5382
5383        trace_ext4_insert_range(inode, offset, len);
5384
5385        offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5386        len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5387
5388        /* Call ext4_force_commit to flush all data in case of data=journal */
5389        if (ext4_should_journal_data(inode)) {
5390                ret = ext4_force_commit(inode->i_sb);
5391                if (ret)
5392                        return ret;
5393        }
5394
5395        inode_lock(inode);
5396        /* Currently just for extent based files */
5397        if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5398                ret = -EOPNOTSUPP;
5399                goto out_mutex;
5400        }
5401
5402        /* Check whether the maximum file size would be exceeded */
5403        if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5404                ret = -EFBIG;
5405                goto out_mutex;
5406        }
5407
5408        /* Offset must be less than i_size */
5409        if (offset >= inode->i_size) {
5410                ret = -EINVAL;
5411                goto out_mutex;
5412        }
5413
5414        /* Wait for existing dio to complete */
5415        inode_dio_wait(inode);
5416
5417        /*
5418         * Prevent page faults from reinstantiating pages we have released from
5419         * page cache.
5420         */
5421        down_write(&EXT4_I(inode)->i_mmap_sem);
5422
5423        ret = ext4_break_layouts(inode);
5424        if (ret)
5425                goto out_mmap;
5426
5427        /*
5428         * Need to round down to align start offset to page size boundary
5429         * for page size > block size.
5430         */
5431        ioffset = round_down(offset, PAGE_SIZE);
5432        /* Write out all dirty pages */
5433        ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5434                        LLONG_MAX);
5435        if (ret)
5436                goto out_mmap;
5437        truncate_pagecache(inode, ioffset);
5438
5439        credits = ext4_writepage_trans_blocks(inode);
5440        handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5441        if (IS_ERR(handle)) {
5442                ret = PTR_ERR(handle);
5443                goto out_mmap;
5444        }
5445        ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5446
5447        /* Expand file to avoid data loss if there is error while shifting */
5448        inode->i_size += len;
5449        EXT4_I(inode)->i_disksize += len;
5450        inode->i_mtime = inode->i_ctime = current_time(inode);
5451        ret = ext4_mark_inode_dirty(handle, inode);
5452        if (ret)
5453                goto out_stop;
5454
5455        down_write(&EXT4_I(inode)->i_data_sem);
5456        ext4_discard_preallocations(inode, 0);
5457
5458        path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5459        if (IS_ERR(path)) {
5460                up_write(&EXT4_I(inode)->i_data_sem);
5461                goto out_stop;
5462        }
5463
5464        depth = ext_depth(inode);
5465        extent = path[depth].p_ext;
5466        if (extent) {
5467                ee_start_lblk = le32_to_cpu(extent->ee_block);
5468                ee_len = ext4_ext_get_actual_len(extent);
5469
5470                /*
5471                 * If offset_lblk is not the starting block of extent, split
5472                 * the extent @offset_lblk
5473                 */
5474                if ((offset_lblk > ee_start_lblk) &&
5475                                (offset_lblk < (ee_start_lblk + ee_len))) {
5476                        if (ext4_ext_is_unwritten(extent))
5477                                split_flag = EXT4_EXT_MARK_UNWRIT1 |
5478                                        EXT4_EXT_MARK_UNWRIT2;
5479                        ret = ext4_split_extent_at(handle, inode, &path,
5480                                        offset_lblk, split_flag,
5481                                        EXT4_EX_NOCACHE |
5482                                        EXT4_GET_BLOCKS_PRE_IO |
5483                                        EXT4_GET_BLOCKS_METADATA_NOFAIL);
5484                }
5485
5486                ext4_ext_drop_refs(path);
5487                kfree(path);
5488                if (ret < 0) {
5489                        up_write(&EXT4_I(inode)->i_data_sem);
5490                        goto out_stop;
5491                }
5492        } else {
5493                ext4_ext_drop_refs(path);
5494                kfree(path);
5495        }
5496
5497        ret = ext4_es_remove_extent(inode, offset_lblk,
5498                        EXT_MAX_BLOCKS - offset_lblk);
5499        if (ret) {
5500                up_write(&EXT4_I(inode)->i_data_sem);
5501                goto out_stop;
5502        }
5503
5504        /*
5505         * if offset_lblk lies in a hole which is at start of file, use
5506         * ee_start_lblk to shift extents
5507         */
5508        ret = ext4_ext_shift_extents(inode, handle,
5509                ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5510                len_lblk, SHIFT_RIGHT);
5511
5512        up_write(&EXT4_I(inode)->i_data_sem);
5513        if (IS_SYNC(inode))
5514                ext4_handle_sync(handle);
5515        if (ret >= 0)
5516                ext4_update_inode_fsync_trans(handle, inode, 1);
5517
5518out_stop:
5519        ext4_journal_stop(handle);
5520        ext4_fc_stop_ineligible(sb);
5521out_mmap:
5522        up_write(&EXT4_I(inode)->i_mmap_sem);
5523out_mutex:
5524        inode_unlock(inode);
5525        return ret;
5526}
5527
5528/**
5529 * ext4_swap_extents() - Swap extents between two inodes
5530 * @handle: handle for this transaction
5531 * @inode1:     First inode
5532 * @inode2:     Second inode
5533 * @lblk1:      Start block for first inode
5534 * @lblk2:      Start block for second inode
5535 * @count:      Number of blocks to swap
5536 * @unwritten: Mark second inode's extents as unwritten after swap
5537 * @erp:        Pointer to save error value
5538 *
5539 * This helper routine does exactly what is promise "swap extents". All other
5540 * stuff such as page-cache locking consistency, bh mapping consistency or
5541 * extent's data copying must be performed by caller.
5542 * Locking:
5543 *              i_mutex is held for both inodes
5544 *              i_data_sem is locked for write for both inodes
5545 * Assumptions:
5546 *              All pages from requested range are locked for both inodes
5547 */
5548int
5549ext4_swap_extents(handle_t *handle, struct inode *inode1,
5550                  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5551                  ext4_lblk_t count, int unwritten, int *erp)
5552{
5553        struct ext4_ext_path *path1 = NULL;
5554        struct ext4_ext_path *path2 = NULL;
5555        int replaced_count = 0;
5556
5557        BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5558        BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5559        BUG_ON(!inode_is_locked(inode1));
5560        BUG_ON(!inode_is_locked(inode2));
5561
5562        *erp = ext4_es_remove_extent(inode1, lblk1, count);
5563        if (unlikely(*erp))
5564                return 0;
5565        *erp = ext4_es_remove_extent(inode2, lblk2, count);
5566        if (unlikely(*erp))
5567                return 0;
5568
5569        while (count) {
5570                struct ext4_extent *ex1, *ex2, tmp_ex;
5571                ext4_lblk_t e1_blk, e2_blk;
5572                int e1_len, e2_len, len;
5573                int split = 0;
5574
5575                path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5576                if (IS_ERR(path1)) {
5577                        *erp = PTR_ERR(path1);
5578                        path1 = NULL;
5579                finish:
5580                        count = 0;
5581                        goto repeat;
5582                }
5583                path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5584                if (IS_ERR(path2)) {
5585                        *erp = PTR_ERR(path2);
5586                        path2 = NULL;
5587                        goto finish;
5588                }
5589                ex1 = path1[path1->p_depth].p_ext;
5590                ex2 = path2[path2->p_depth].p_ext;
5591                /* Do we have something to swap ? */
5592                if (unlikely(!ex2 || !ex1))
5593                        goto finish;
5594
5595                e1_blk = le32_to_cpu(ex1->ee_block);
5596                e2_blk = le32_to_cpu(ex2->ee_block);
5597                e1_len = ext4_ext_get_actual_len(ex1);
5598                e2_len = ext4_ext_get_actual_len(ex2);
5599
5600                /* Hole handling */
5601                if (!in_range(lblk1, e1_blk, e1_len) ||
5602                    !in_range(lblk2, e2_blk, e2_len)) {
5603                        ext4_lblk_t next1, next2;
5604
5605                        /* if hole after extent, then go to next extent */
5606                        next1 = ext4_ext_next_allocated_block(path1);
5607                        next2 = ext4_ext_next_allocated_block(path2);
5608                        /* If hole before extent, then shift to that extent */
5609                        if (e1_blk > lblk1)
5610                                next1 = e1_blk;
5611                        if (e2_blk > lblk2)
5612                                next2 = e2_blk;
5613                        /* Do we have something to swap */
5614                        if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5615                                goto finish;
5616                        /* Move to the rightest boundary */
5617                        len = next1 - lblk1;
5618                        if (len < next2 - lblk2)
5619                                len = next2 - lblk2;
5620                        if (len > count)
5621                                len = count;
5622                        lblk1 += len;
5623                        lblk2 += len;
5624                        count -= len;
5625                        goto repeat;
5626                }
5627
5628                /* Prepare left boundary */
5629                if (e1_blk < lblk1) {
5630                        split = 1;
5631                        *erp = ext4_force_split_extent_at(handle, inode1,
5632                                                &path1, lblk1, 0);
5633                        if (unlikely(*erp))
5634                                goto finish;
5635                }
5636                if (e2_blk < lblk2) {
5637                        split = 1;
5638                        *erp = ext4_force_split_extent_at(handle, inode2,
5639                                                &path2,  lblk2, 0);
5640                        if (unlikely(*erp))
5641                                goto finish;
5642                }
5643                /* ext4_split_extent_at() may result in leaf extent split,
5644                 * path must to be revalidated. */
5645                if (split)
5646                        goto repeat;
5647
5648                /* Prepare right boundary */
5649                len = count;
5650                if (len > e1_blk + e1_len - lblk1)
5651                        len = e1_blk + e1_len - lblk1;
5652                if (len > e2_blk + e2_len - lblk2)
5653                        len = e2_blk + e2_len - lblk2;
5654
5655                if (len != e1_len) {
5656                        split = 1;
5657                        *erp = ext4_force_split_extent_at(handle, inode1,
5658                                                &path1, lblk1 + len, 0);
5659                        if (unlikely(*erp))
5660                                goto finish;
5661                }
5662                if (len != e2_len) {
5663                        split = 1;
5664                        *erp = ext4_force_split_extent_at(handle, inode2,
5665                                                &path2, lblk2 + len, 0);
5666                        if (*erp)
5667                                goto finish;
5668                }
5669                /* ext4_split_extent_at() may result in leaf extent split,
5670                 * path must to be revalidated. */
5671                if (split)
5672                        goto repeat;
5673
5674                BUG_ON(e2_len != e1_len);
5675                *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5676                if (unlikely(*erp))
5677                        goto finish;
5678                *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5679                if (unlikely(*erp))
5680                        goto finish;
5681
5682                /* Both extents are fully inside boundaries. Swap it now */
5683                tmp_ex = *ex1;
5684                ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5685                ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5686                ex1->ee_len = cpu_to_le16(e2_len);
5687                ex2->ee_len = cpu_to_le16(e1_len);
5688                if (unwritten)
5689                        ext4_ext_mark_unwritten(ex2);
5690                if (ext4_ext_is_unwritten(&tmp_ex))
5691                        ext4_ext_mark_unwritten(ex1);
5692
5693                ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5694                ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5695                *erp = ext4_ext_dirty(handle, inode2, path2 +
5696                                      path2->p_depth);
5697                if (unlikely(*erp))
5698                        goto finish;
5699                *erp = ext4_ext_dirty(handle, inode1, path1 +
5700                                      path1->p_depth);
5701                /*
5702                 * Looks scarry ah..? second inode already points to new blocks,
5703                 * and it was successfully dirtied. But luckily error may happen
5704                 * only due to journal error, so full transaction will be
5705                 * aborted anyway.
5706                 */
5707                if (unlikely(*erp))
5708                        goto finish;
5709                lblk1 += len;
5710                lblk2 += len;
5711                replaced_count += len;
5712                count -= len;
5713
5714        repeat:
5715                ext4_ext_drop_refs(path1);
5716                kfree(path1);
5717                ext4_ext_drop_refs(path2);
5718                kfree(path2);
5719                path1 = path2 = NULL;
5720        }
5721        return replaced_count;
5722}
5723
5724/*
5725 * ext4_clu_mapped - determine whether any block in a logical cluster has
5726 *                   been mapped to a physical cluster
5727 *
5728 * @inode - file containing the logical cluster
5729 * @lclu - logical cluster of interest
5730 *
5731 * Returns 1 if any block in the logical cluster is mapped, signifying
5732 * that a physical cluster has been allocated for it.  Otherwise,
5733 * returns 0.  Can also return negative error codes.  Derived from
5734 * ext4_ext_map_blocks().
5735 */
5736int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5737{
5738        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5739        struct ext4_ext_path *path;
5740        int depth, mapped = 0, err = 0;
5741        struct ext4_extent *extent;
5742        ext4_lblk_t first_lblk, first_lclu, last_lclu;
5743
5744        /* search for the extent closest to the first block in the cluster */
5745        path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5746        if (IS_ERR(path)) {
5747                err = PTR_ERR(path);
5748                path = NULL;
5749                goto out;
5750        }
5751
5752        depth = ext_depth(inode);
5753
5754        /*
5755         * A consistent leaf must not be empty.  This situation is possible,
5756         * though, _during_ tree modification, and it's why an assert can't
5757         * be put in ext4_find_extent().
5758         */
5759        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5760                EXT4_ERROR_INODE(inode,
5761                    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5762                                 (unsigned long) EXT4_C2B(sbi, lclu),
5763                                 depth, path[depth].p_block);
5764                err = -EFSCORRUPTED;
5765                goto out;
5766        }
5767
5768        extent = path[depth].p_ext;
5769
5770        /* can't be mapped if the extent tree is empty */
5771        if (extent == NULL)
5772                goto out;
5773
5774        first_lblk = le32_to_cpu(extent->ee_block);
5775        first_lclu = EXT4_B2C(sbi, first_lblk);
5776
5777        /*
5778         * Three possible outcomes at this point - found extent spanning
5779         * the target cluster, to the left of the target cluster, or to the
5780         * right of the target cluster.  The first two cases are handled here.
5781         * The last case indicates the target cluster is not mapped.
5782         */
5783        if (lclu >= first_lclu) {
5784                last_lclu = EXT4_B2C(sbi, first_lblk +
5785                                     ext4_ext_get_actual_len(extent) - 1);
5786                if (lclu <= last_lclu) {
5787                        mapped = 1;
5788                } else {
5789                        first_lblk = ext4_ext_next_allocated_block(path);
5790                        first_lclu = EXT4_B2C(sbi, first_lblk);
5791                        if (lclu == first_lclu)
5792                                mapped = 1;
5793                }
5794        }
5795
5796out:
5797        ext4_ext_drop_refs(path);
5798        kfree(path);
5799
5800        return err ? err : mapped;
5801}
5802
5803/*
5804 * Updates physical block address and unwritten status of extent
5805 * starting at lblk start and of len. If such an extent doesn't exist,
5806 * this function splits the extent tree appropriately to create an
5807 * extent like this.  This function is called in the fast commit
5808 * replay path.  Returns 0 on success and error on failure.
5809 */
5810int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5811                              int len, int unwritten, ext4_fsblk_t pblk)
5812{
5813        struct ext4_ext_path *path = NULL, *ppath;
5814        struct ext4_extent *ex;
5815        int ret;
5816
5817        path = ext4_find_extent(inode, start, NULL, 0);
5818        if (!path)
5819                return -EINVAL;
5820        ex = path[path->p_depth].p_ext;
5821        if (!ex) {
5822                ret = -EFSCORRUPTED;
5823                goto out;
5824        }
5825
5826        if (le32_to_cpu(ex->ee_block) != start ||
5827                ext4_ext_get_actual_len(ex) != len) {
5828                /* We need to split this extent to match our extent first */
5829                ppath = path;
5830                down_write(&EXT4_I(inode)->i_data_sem);
5831                ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5832                up_write(&EXT4_I(inode)->i_data_sem);
5833                if (ret)
5834                        goto out;
5835                kfree(path);
5836                path = ext4_find_extent(inode, start, NULL, 0);
5837                if (IS_ERR(path))
5838                        return -1;
5839                ppath = path;
5840                ex = path[path->p_depth].p_ext;
5841                WARN_ON(le32_to_cpu(ex->ee_block) != start);
5842                if (ext4_ext_get_actual_len(ex) != len) {
5843                        down_write(&EXT4_I(inode)->i_data_sem);
5844                        ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5845                                                         start + len, 1);
5846                        up_write(&EXT4_I(inode)->i_data_sem);
5847                        if (ret)
5848                                goto out;
5849                        kfree(path);
5850                        path = ext4_find_extent(inode, start, NULL, 0);
5851                        if (IS_ERR(path))
5852                                return -EINVAL;
5853                        ex = path[path->p_depth].p_ext;
5854                }
5855        }
5856        if (unwritten)
5857                ext4_ext_mark_unwritten(ex);
5858        else
5859                ext4_ext_mark_initialized(ex);
5860        ext4_ext_store_pblock(ex, pblk);
5861        down_write(&EXT4_I(inode)->i_data_sem);
5862        ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5863        up_write(&EXT4_I(inode)->i_data_sem);
5864out:
5865        ext4_ext_drop_refs(path);
5866        kfree(path);
5867        ext4_mark_inode_dirty(NULL, inode);
5868        return ret;
5869}
5870
5871/* Try to shrink the extent tree */
5872void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5873{
5874        struct ext4_ext_path *path = NULL;
5875        struct ext4_extent *ex;
5876        ext4_lblk_t old_cur, cur = 0;
5877
5878        while (cur < end) {
5879                path = ext4_find_extent(inode, cur, NULL, 0);
5880                if (IS_ERR(path))
5881                        return;
5882                ex = path[path->p_depth].p_ext;
5883                if (!ex) {
5884                        ext4_ext_drop_refs(path);
5885                        kfree(path);
5886                        ext4_mark_inode_dirty(NULL, inode);
5887                        return;
5888                }
5889                old_cur = cur;
5890                cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5891                if (cur <= old_cur)
5892                        cur = old_cur + 1;
5893                ext4_ext_try_to_merge(NULL, inode, path, ex);
5894                down_write(&EXT4_I(inode)->i_data_sem);
5895                ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5896                up_write(&EXT4_I(inode)->i_data_sem);
5897                ext4_mark_inode_dirty(NULL, inode);
5898                ext4_ext_drop_refs(path);
5899                kfree(path);
5900        }
5901}
5902
5903/* Check if *cur is a hole and if it is, skip it */
5904static void skip_hole(struct inode *inode, ext4_lblk_t *cur)
5905{
5906        int ret;
5907        struct ext4_map_blocks map;
5908
5909        map.m_lblk = *cur;
5910        map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5911
5912        ret = ext4_map_blocks(NULL, inode, &map, 0);
5913        if (ret != 0)
5914                return;
5915        *cur = *cur + map.m_len;
5916}
5917
5918/* Count number of blocks used by this inode and update i_blocks */
5919int ext4_ext_replay_set_iblocks(struct inode *inode)
5920{
5921        struct ext4_ext_path *path = NULL, *path2 = NULL;
5922        struct ext4_extent *ex;
5923        ext4_lblk_t cur = 0, end;
5924        int numblks = 0, i, ret = 0;
5925        ext4_fsblk_t cmp1, cmp2;
5926        struct ext4_map_blocks map;
5927
5928        /* Determin the size of the file first */
5929        path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5930                                        EXT4_EX_NOCACHE);
5931        if (IS_ERR(path))
5932                return PTR_ERR(path);
5933        ex = path[path->p_depth].p_ext;
5934        if (!ex) {
5935                ext4_ext_drop_refs(path);
5936                kfree(path);
5937                goto out;
5938        }
5939        end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5940        ext4_ext_drop_refs(path);
5941        kfree(path);
5942
5943        /* Count the number of data blocks */
5944        cur = 0;
5945        while (cur < end) {
5946                map.m_lblk = cur;
5947                map.m_len = end - cur;
5948                ret = ext4_map_blocks(NULL, inode, &map, 0);
5949                if (ret < 0)
5950                        break;
5951                if (ret > 0)
5952                        numblks += ret;
5953                cur = cur + map.m_len;
5954        }
5955
5956        /*
5957         * Count the number of extent tree blocks. We do it by looking up
5958         * two successive extents and determining the difference between
5959         * their paths. When path is different for 2 successive extents
5960         * we compare the blocks in the path at each level and increment
5961         * iblocks by total number of differences found.
5962         */
5963        cur = 0;
5964        skip_hole(inode, &cur);
5965        path = ext4_find_extent(inode, cur, NULL, 0);
5966        if (IS_ERR(path))
5967                goto out;
5968        numblks += path->p_depth;
5969        ext4_ext_drop_refs(path);
5970        kfree(path);
5971        while (cur < end) {
5972                path = ext4_find_extent(inode, cur, NULL, 0);
5973                if (IS_ERR(path))
5974                        break;
5975                ex = path[path->p_depth].p_ext;
5976                if (!ex) {
5977                        ext4_ext_drop_refs(path);
5978                        kfree(path);
5979                        return 0;
5980                }
5981                cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
5982                                        ext4_ext_get_actual_len(ex));
5983                skip_hole(inode, &cur);
5984
5985                path2 = ext4_find_extent(inode, cur, NULL, 0);
5986                if (IS_ERR(path2)) {
5987                        ext4_ext_drop_refs(path);
5988                        kfree(path);
5989                        break;
5990                }
5991                ex = path2[path2->p_depth].p_ext;
5992                for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
5993                        cmp1 = cmp2 = 0;
5994                        if (i <= path->p_depth)
5995                                cmp1 = path[i].p_bh ?
5996                                        path[i].p_bh->b_blocknr : 0;
5997                        if (i <= path2->p_depth)
5998                                cmp2 = path2[i].p_bh ?
5999                                        path2[i].p_bh->b_blocknr : 0;
6000                        if (cmp1 != cmp2 && cmp2 != 0)
6001                                numblks++;
6002                }
6003                ext4_ext_drop_refs(path);
6004                ext4_ext_drop_refs(path2);
6005                kfree(path);
6006                kfree(path2);
6007        }
6008
6009out:
6010        inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6011        ext4_mark_inode_dirty(NULL, inode);
6012        return 0;
6013}
6014
6015int ext4_ext_clear_bb(struct inode *inode)
6016{
6017        struct ext4_ext_path *path = NULL;
6018        struct ext4_extent *ex;
6019        ext4_lblk_t cur = 0, end;
6020        int j, ret = 0;
6021        struct ext4_map_blocks map;
6022
6023        /* Determin the size of the file first */
6024        path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6025                                        EXT4_EX_NOCACHE);
6026        if (IS_ERR(path))
6027                return PTR_ERR(path);
6028        ex = path[path->p_depth].p_ext;
6029        if (!ex) {
6030                ext4_ext_drop_refs(path);
6031                kfree(path);
6032                return 0;
6033        }
6034        end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6035        ext4_ext_drop_refs(path);
6036        kfree(path);
6037
6038        cur = 0;
6039        while (cur < end) {
6040                map.m_lblk = cur;
6041                map.m_len = end - cur;
6042                ret = ext4_map_blocks(NULL, inode, &map, 0);
6043                if (ret < 0)
6044                        break;
6045                if (ret > 0) {
6046                        path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6047                        if (!IS_ERR_OR_NULL(path)) {
6048                                for (j = 0; j < path->p_depth; j++) {
6049
6050                                        ext4_mb_mark_bb(inode->i_sb,
6051                                                        path[j].p_block, 1, 0);
6052                                }
6053                                ext4_ext_drop_refs(path);
6054                                kfree(path);
6055                        }
6056                        ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6057                }
6058                cur = cur + map.m_len;
6059        }
6060
6061        return 0;
6062}
6063