linux/fs/ext4/extents.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
   3 * Written by Alex Tomas <alex@clusterfs.com>
   4 *
   5 * Architecture independence:
   6 *   Copyright (c) 2005, Bull S.A.
   7 *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License version 2 as
  11 * published by the Free Software Foundation.
  12 *
  13 * This program is distributed in the hope that it will be useful,
  14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16 * GNU General Public License for more details.
  17 *
  18 * You should have received a copy of the GNU General Public Licens
  19 * along with this program; if not, write to the Free Software
  20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
  21 */
  22
  23/*
  24 * Extents support for EXT4
  25 *
  26 * TODO:
  27 *   - ext4*_error() should be used in some situations
  28 *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
  29 *   - smart tree reduction
  30 */
  31
  32#include <linux/fs.h>
  33#include <linux/time.h>
  34#include <linux/jbd2.h>
  35#include <linux/highuid.h>
  36#include <linux/pagemap.h>
  37#include <linux/quotaops.h>
  38#include <linux/string.h>
  39#include <linux/slab.h>
  40#include <linux/falloc.h>
  41#include <asm/uaccess.h>
  42#include <linux/fiemap.h>
  43#include "ext4_jbd2.h"
  44
  45#include <trace/events/ext4.h>
  46
  47static int ext4_split_extent(handle_t *handle,
  48                                struct inode *inode,
  49                                struct ext4_ext_path *path,
  50                                struct ext4_map_blocks *map,
  51                                int split_flag,
  52                                int flags);
  53
  54static int ext4_ext_truncate_extend_restart(handle_t *handle,
  55                                            struct inode *inode,
  56                                            int needed)
  57{
  58        int err;
  59
  60        if (!ext4_handle_valid(handle))
  61                return 0;
  62        if (handle->h_buffer_credits > needed)
  63                return 0;
  64        err = ext4_journal_extend(handle, needed);
  65        if (err <= 0)
  66                return err;
  67        err = ext4_truncate_restart_trans(handle, inode, needed);
  68        if (err == 0)
  69                err = -EAGAIN;
  70
  71        return err;
  72}
  73
  74/*
  75 * could return:
  76 *  - EROFS
  77 *  - ENOMEM
  78 */
  79static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
  80                                struct ext4_ext_path *path)
  81{
  82        if (path->p_bh) {
  83                /* path points to block */
  84                return ext4_journal_get_write_access(handle, path->p_bh);
  85        }
  86        /* path points to leaf/index in inode body */
  87        /* we use in-core data, no need to protect them */
  88        return 0;
  89}
  90
  91/*
  92 * could return:
  93 *  - EROFS
  94 *  - ENOMEM
  95 *  - EIO
  96 */
  97#define ext4_ext_dirty(handle, inode, path) \
  98                __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
  99static int __ext4_ext_dirty(const char *where, unsigned int line,
 100                            handle_t *handle, struct inode *inode,
 101                            struct ext4_ext_path *path)
 102{
 103        int err;
 104        if (path->p_bh) {
 105                /* path points to block */
 106                err = __ext4_handle_dirty_metadata(where, line, handle,
 107                                                   inode, path->p_bh);
 108        } else {
 109                /* path points to leaf/index in inode body */
 110                err = ext4_mark_inode_dirty(handle, inode);
 111        }
 112        return err;
 113}
 114
 115static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
 116                              struct ext4_ext_path *path,
 117                              ext4_lblk_t block)
 118{
 119        if (path) {
 120                int depth = path->p_depth;
 121                struct ext4_extent *ex;
 122
 123                /*
 124                 * Try to predict block placement assuming that we are
 125                 * filling in a file which will eventually be
 126                 * non-sparse --- i.e., in the case of libbfd writing
 127                 * an ELF object sections out-of-order but in a way
 128                 * the eventually results in a contiguous object or
 129                 * executable file, or some database extending a table
 130                 * space file.  However, this is actually somewhat
 131                 * non-ideal if we are writing a sparse file such as
 132                 * qemu or KVM writing a raw image file that is going
 133                 * to stay fairly sparse, since it will end up
 134                 * fragmenting the file system's free space.  Maybe we
 135                 * should have some hueristics or some way to allow
 136                 * userspace to pass a hint to file system,
 137                 * especially if the latter case turns out to be
 138                 * common.
 139                 */
 140                ex = path[depth].p_ext;
 141                if (ex) {
 142                        ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
 143                        ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
 144
 145                        if (block > ext_block)
 146                                return ext_pblk + (block - ext_block);
 147                        else
 148                                return ext_pblk - (ext_block - block);
 149                }
 150
 151                /* it looks like index is empty;
 152                 * try to find starting block from index itself */
 153                if (path[depth].p_bh)
 154                        return path[depth].p_bh->b_blocknr;
 155        }
 156
 157        /* OK. use inode's group */
 158        return ext4_inode_to_goal_block(inode);
 159}
 160
 161/*
 162 * Allocation for a meta data block
 163 */
 164static ext4_fsblk_t
 165ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
 166                        struct ext4_ext_path *path,
 167                        struct ext4_extent *ex, int *err, unsigned int flags)
 168{
 169        ext4_fsblk_t goal, newblock;
 170
 171        goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
 172        newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
 173                                        NULL, err);
 174        return newblock;
 175}
 176
 177static inline int ext4_ext_space_block(struct inode *inode, int check)
 178{
 179        int size;
 180
 181        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 182                        / sizeof(struct ext4_extent);
 183#ifdef AGGRESSIVE_TEST
 184        if (!check && size > 6)
 185                size = 6;
 186#endif
 187        return size;
 188}
 189
 190static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
 191{
 192        int size;
 193
 194        size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 195                        / sizeof(struct ext4_extent_idx);
 196#ifdef AGGRESSIVE_TEST
 197        if (!check && size > 5)
 198                size = 5;
 199#endif
 200        return size;
 201}
 202
 203static inline int ext4_ext_space_root(struct inode *inode, int check)
 204{
 205        int size;
 206
 207        size = sizeof(EXT4_I(inode)->i_data);
 208        size -= sizeof(struct ext4_extent_header);
 209        size /= sizeof(struct ext4_extent);
 210#ifdef AGGRESSIVE_TEST
 211        if (!check && size > 3)
 212                size = 3;
 213#endif
 214        return size;
 215}
 216
 217static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
 218{
 219        int size;
 220
 221        size = sizeof(EXT4_I(inode)->i_data);
 222        size -= sizeof(struct ext4_extent_header);
 223        size /= sizeof(struct ext4_extent_idx);
 224#ifdef AGGRESSIVE_TEST
 225        if (!check && size > 4)
 226                size = 4;
 227#endif
 228        return size;
 229}
 230
 231/*
 232 * Calculate the number of metadata blocks needed
 233 * to allocate @blocks
 234 * Worse case is one block per extent
 235 */
 236int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 237{
 238        struct ext4_inode_info *ei = EXT4_I(inode);
 239        int idxs;
 240
 241        idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
 242                / sizeof(struct ext4_extent_idx));
 243
 244        /*
 245         * If the new delayed allocation block is contiguous with the
 246         * previous da block, it can share index blocks with the
 247         * previous block, so we only need to allocate a new index
 248         * block every idxs leaf blocks.  At ldxs**2 blocks, we need
 249         * an additional index block, and at ldxs**3 blocks, yet
 250         * another index blocks.
 251         */
 252        if (ei->i_da_metadata_calc_len &&
 253            ei->i_da_metadata_calc_last_lblock+1 == lblock) {
 254                int num = 0;
 255
 256                if ((ei->i_da_metadata_calc_len % idxs) == 0)
 257                        num++;
 258                if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0)
 259                        num++;
 260                if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) {
 261                        num++;
 262                        ei->i_da_metadata_calc_len = 0;
 263                } else
 264                        ei->i_da_metadata_calc_len++;
 265                ei->i_da_metadata_calc_last_lblock++;
 266                return num;
 267        }
 268
 269        /*
 270         * In the worst case we need a new set of index blocks at
 271         * every level of the inode's extent tree.
 272         */
 273        ei->i_da_metadata_calc_len = 1;
 274        ei->i_da_metadata_calc_last_lblock = lblock;
 275        return ext_depth(inode) + 1;
 276}
 277
 278static int
 279ext4_ext_max_entries(struct inode *inode, int depth)
 280{
 281        int max;
 282
 283        if (depth == ext_depth(inode)) {
 284                if (depth == 0)
 285                        max = ext4_ext_space_root(inode, 1);
 286                else
 287                        max = ext4_ext_space_root_idx(inode, 1);
 288        } else {
 289                if (depth == 0)
 290                        max = ext4_ext_space_block(inode, 1);
 291                else
 292                        max = ext4_ext_space_block_idx(inode, 1);
 293        }
 294
 295        return max;
 296}
 297
 298static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
 299{
 300        ext4_fsblk_t block = ext4_ext_pblock(ext);
 301        int len = ext4_ext_get_actual_len(ext);
 302
 303        if (len == 0)
 304                return 0;
 305        return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
 306}
 307
 308static int ext4_valid_extent_idx(struct inode *inode,
 309                                struct ext4_extent_idx *ext_idx)
 310{
 311        ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
 312
 313        return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
 314}
 315
 316static int ext4_valid_extent_entries(struct inode *inode,
 317                                struct ext4_extent_header *eh,
 318                                int depth)
 319{
 320        unsigned short entries;
 321        if (eh->eh_entries == 0)
 322                return 1;
 323
 324        entries = le16_to_cpu(eh->eh_entries);
 325
 326        if (depth == 0) {
 327                /* leaf entries */
 328                struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
 329                while (entries) {
 330                        if (!ext4_valid_extent(inode, ext))
 331                                return 0;
 332                        ext++;
 333                        entries--;
 334                }
 335        } else {
 336                struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
 337                while (entries) {
 338                        if (!ext4_valid_extent_idx(inode, ext_idx))
 339                                return 0;
 340                        ext_idx++;
 341                        entries--;
 342                }
 343        }
 344        return 1;
 345}
 346
 347static int __ext4_ext_check(const char *function, unsigned int line,
 348                            struct inode *inode, struct ext4_extent_header *eh,
 349                            int depth)
 350{
 351        const char *error_msg;
 352        int max = 0;
 353
 354        if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
 355                error_msg = "invalid magic";
 356                goto corrupted;
 357        }
 358        if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
 359                error_msg = "unexpected eh_depth";
 360                goto corrupted;
 361        }
 362        if (unlikely(eh->eh_max == 0)) {
 363                error_msg = "invalid eh_max";
 364                goto corrupted;
 365        }
 366        max = ext4_ext_max_entries(inode, depth);
 367        if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
 368                error_msg = "too large eh_max";
 369                goto corrupted;
 370        }
 371        if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
 372                error_msg = "invalid eh_entries";
 373                goto corrupted;
 374        }
 375        if (!ext4_valid_extent_entries(inode, eh, depth)) {
 376                error_msg = "invalid extent entries";
 377                goto corrupted;
 378        }
 379        return 0;
 380
 381corrupted:
 382        ext4_error_inode(inode, function, line, 0,
 383                        "bad header/extent: %s - magic %x, "
 384                        "entries %u, max %u(%u), depth %u(%u)",
 385                        error_msg, le16_to_cpu(eh->eh_magic),
 386                        le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
 387                        max, le16_to_cpu(eh->eh_depth), depth);
 388
 389        return -EIO;
 390}
 391
 392#define ext4_ext_check(inode, eh, depth)        \
 393        __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
 394
 395int ext4_ext_check_inode(struct inode *inode)
 396{
 397        return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
 398}
 399
 400#ifdef EXT_DEBUG
 401static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 402{
 403        int k, l = path->p_depth;
 404
 405        ext_debug("path:");
 406        for (k = 0; k <= l; k++, path++) {
 407                if (path->p_idx) {
 408                  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
 409                            ext4_idx_pblock(path->p_idx));
 410                } else if (path->p_ext) {
 411                        ext_debug("  %d:[%d]%d:%llu ",
 412                                  le32_to_cpu(path->p_ext->ee_block),
 413                                  ext4_ext_is_uninitialized(path->p_ext),
 414                                  ext4_ext_get_actual_len(path->p_ext),
 415                                  ext4_ext_pblock(path->p_ext));
 416                } else
 417                        ext_debug("  []");
 418        }
 419        ext_debug("\n");
 420}
 421
 422static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
 423{
 424        int depth = ext_depth(inode);
 425        struct ext4_extent_header *eh;
 426        struct ext4_extent *ex;
 427        int i;
 428
 429        if (!path)
 430                return;
 431
 432        eh = path[depth].p_hdr;
 433        ex = EXT_FIRST_EXTENT(eh);
 434
 435        ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
 436
 437        for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
 438                ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
 439                          ext4_ext_is_uninitialized(ex),
 440                          ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
 441        }
 442        ext_debug("\n");
 443}
 444
 445static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
 446                        ext4_fsblk_t newblock, int level)
 447{
 448        int depth = ext_depth(inode);
 449        struct ext4_extent *ex;
 450
 451        if (depth != level) {
 452                struct ext4_extent_idx *idx;
 453                idx = path[level].p_idx;
 454                while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
 455                        ext_debug("%d: move %d:%llu in new index %llu\n", level,
 456                                        le32_to_cpu(idx->ei_block),
 457                                        ext4_idx_pblock(idx),
 458                                        newblock);
 459                        idx++;
 460                }
 461
 462                return;
 463        }
 464
 465        ex = path[depth].p_ext;
 466        while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
 467                ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
 468                                le32_to_cpu(ex->ee_block),
 469                                ext4_ext_pblock(ex),
 470                                ext4_ext_is_uninitialized(ex),
 471                                ext4_ext_get_actual_len(ex),
 472                                newblock);
 473                ex++;
 474        }
 475}
 476
 477#else
 478#define ext4_ext_show_path(inode, path)
 479#define ext4_ext_show_leaf(inode, path)
 480#define ext4_ext_show_move(inode, path, newblock, level)
 481#endif
 482
 483void ext4_ext_drop_refs(struct ext4_ext_path *path)
 484{
 485        int depth = path->p_depth;
 486        int i;
 487
 488        for (i = 0; i <= depth; i++, path++)
 489                if (path->p_bh) {
 490                        brelse(path->p_bh);
 491                        path->p_bh = NULL;
 492                }
 493}
 494
 495/*
 496 * ext4_ext_binsearch_idx:
 497 * binary search for the closest index of the given block
 498 * the header must be checked before calling this
 499 */
 500static void
 501ext4_ext_binsearch_idx(struct inode *inode,
 502                        struct ext4_ext_path *path, ext4_lblk_t block)
 503{
 504        struct ext4_extent_header *eh = path->p_hdr;
 505        struct ext4_extent_idx *r, *l, *m;
 506
 507
 508        ext_debug("binsearch for %u(idx):  ", block);
 509
 510        l = EXT_FIRST_INDEX(eh) + 1;
 511        r = EXT_LAST_INDEX(eh);
 512        while (l <= r) {
 513                m = l + (r - l) / 2;
 514                if (block < le32_to_cpu(m->ei_block))
 515                        r = m - 1;
 516                else
 517                        l = m + 1;
 518                ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
 519                                m, le32_to_cpu(m->ei_block),
 520                                r, le32_to_cpu(r->ei_block));
 521        }
 522
 523        path->p_idx = l - 1;
 524        ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
 525                  ext4_idx_pblock(path->p_idx));
 526
 527#ifdef CHECK_BINSEARCH
 528        {
 529                struct ext4_extent_idx *chix, *ix;
 530                int k;
 531
 532                chix = ix = EXT_FIRST_INDEX(eh);
 533                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
 534                  if (k != 0 &&
 535                      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
 536                                printk(KERN_DEBUG "k=%d, ix=0x%p, "
 537                                       "first=0x%p\n", k,
 538                                       ix, EXT_FIRST_INDEX(eh));
 539                                printk(KERN_DEBUG "%u <= %u\n",
 540                                       le32_to_cpu(ix->ei_block),
 541                                       le32_to_cpu(ix[-1].ei_block));
 542                        }
 543                        BUG_ON(k && le32_to_cpu(ix->ei_block)
 544                                           <= le32_to_cpu(ix[-1].ei_block));
 545                        if (block < le32_to_cpu(ix->ei_block))
 546                                break;
 547                        chix = ix;
 548                }
 549                BUG_ON(chix != path->p_idx);
 550        }
 551#endif
 552
 553}
 554
 555/*
 556 * ext4_ext_binsearch:
 557 * binary search for closest extent of the given block
 558 * the header must be checked before calling this
 559 */
 560static void
 561ext4_ext_binsearch(struct inode *inode,
 562                struct ext4_ext_path *path, ext4_lblk_t block)
 563{
 564        struct ext4_extent_header *eh = path->p_hdr;
 565        struct ext4_extent *r, *l, *m;
 566
 567        if (eh->eh_entries == 0) {
 568                /*
 569                 * this leaf is empty:
 570                 * we get such a leaf in split/add case
 571                 */
 572                return;
 573        }
 574
 575        ext_debug("binsearch for %u:  ", block);
 576
 577        l = EXT_FIRST_EXTENT(eh) + 1;
 578        r = EXT_LAST_EXTENT(eh);
 579
 580        while (l <= r) {
 581                m = l + (r - l) / 2;
 582                if (block < le32_to_cpu(m->ee_block))
 583                        r = m - 1;
 584                else
 585                        l = m + 1;
 586                ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
 587                                m, le32_to_cpu(m->ee_block),
 588                                r, le32_to_cpu(r->ee_block));
 589        }
 590
 591        path->p_ext = l - 1;
 592        ext_debug("  -> %d:%llu:[%d]%d ",
 593                        le32_to_cpu(path->p_ext->ee_block),
 594                        ext4_ext_pblock(path->p_ext),
 595                        ext4_ext_is_uninitialized(path->p_ext),
 596                        ext4_ext_get_actual_len(path->p_ext));
 597
 598#ifdef CHECK_BINSEARCH
 599        {
 600                struct ext4_extent *chex, *ex;
 601                int k;
 602
 603                chex = ex = EXT_FIRST_EXTENT(eh);
 604                for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
 605                        BUG_ON(k && le32_to_cpu(ex->ee_block)
 606                                          <= le32_to_cpu(ex[-1].ee_block));
 607                        if (block < le32_to_cpu(ex->ee_block))
 608                                break;
 609                        chex = ex;
 610                }
 611                BUG_ON(chex != path->p_ext);
 612        }
 613#endif
 614
 615}
 616
 617int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
 618{
 619        struct ext4_extent_header *eh;
 620
 621        eh = ext_inode_hdr(inode);
 622        eh->eh_depth = 0;
 623        eh->eh_entries = 0;
 624        eh->eh_magic = EXT4_EXT_MAGIC;
 625        eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
 626        ext4_mark_inode_dirty(handle, inode);
 627        ext4_ext_invalidate_cache(inode);
 628        return 0;
 629}
 630
 631struct ext4_ext_path *
 632ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
 633                                        struct ext4_ext_path *path)
 634{
 635        struct ext4_extent_header *eh;
 636        struct buffer_head *bh;
 637        short int depth, i, ppos = 0, alloc = 0;
 638
 639        eh = ext_inode_hdr(inode);
 640        depth = ext_depth(inode);
 641
 642        /* account possible depth increase */
 643        if (!path) {
 644                path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
 645                                GFP_NOFS);
 646                if (!path)
 647                        return ERR_PTR(-ENOMEM);
 648                alloc = 1;
 649        }
 650        path[0].p_hdr = eh;
 651        path[0].p_bh = NULL;
 652
 653        i = depth;
 654        /* walk through the tree */
 655        while (i) {
 656                int need_to_validate = 0;
 657
 658                ext_debug("depth %d: num %d, max %d\n",
 659                          ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
 660
 661                ext4_ext_binsearch_idx(inode, path + ppos, block);
 662                path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 663                path[ppos].p_depth = i;
 664                path[ppos].p_ext = NULL;
 665
 666                bh = sb_getblk(inode->i_sb, path[ppos].p_block);
 667                if (unlikely(!bh))
 668                        goto err;
 669                if (!bh_uptodate_or_lock(bh)) {
 670                        trace_ext4_ext_load_extent(inode, block,
 671                                                path[ppos].p_block);
 672                        if (bh_submit_read(bh) < 0) {
 673                                put_bh(bh);
 674                                goto err;
 675                        }
 676                        /* validate the extent entries */
 677                        need_to_validate = 1;
 678                }
 679                eh = ext_block_hdr(bh);
 680                ppos++;
 681                if (unlikely(ppos > depth)) {
 682                        put_bh(bh);
 683                        EXT4_ERROR_INODE(inode,
 684                                         "ppos %d > depth %d", ppos, depth);
 685                        goto err;
 686                }
 687                path[ppos].p_bh = bh;
 688                path[ppos].p_hdr = eh;
 689                i--;
 690
 691                if (need_to_validate && ext4_ext_check(inode, eh, i))
 692                        goto err;
 693        }
 694
 695        path[ppos].p_depth = i;
 696        path[ppos].p_ext = NULL;
 697        path[ppos].p_idx = NULL;
 698
 699        /* find extent */
 700        ext4_ext_binsearch(inode, path + ppos, block);
 701        /* if not an empty leaf */
 702        if (path[ppos].p_ext)
 703                path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
 704
 705        ext4_ext_show_path(inode, path);
 706
 707        return path;
 708
 709err:
 710        ext4_ext_drop_refs(path);
 711        if (alloc)
 712                kfree(path);
 713        return ERR_PTR(-EIO);
 714}
 715
 716/*
 717 * ext4_ext_insert_index:
 718 * insert new index [@logical;@ptr] into the block at @curp;
 719 * check where to insert: before @curp or after @curp
 720 */
 721static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
 722                                 struct ext4_ext_path *curp,
 723                                 int logical, ext4_fsblk_t ptr)
 724{
 725        struct ext4_extent_idx *ix;
 726        int len, err;
 727
 728        err = ext4_ext_get_access(handle, inode, curp);
 729        if (err)
 730                return err;
 731
 732        if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
 733                EXT4_ERROR_INODE(inode,
 734                                 "logical %d == ei_block %d!",
 735                                 logical, le32_to_cpu(curp->p_idx->ei_block));
 736                return -EIO;
 737        }
 738
 739        if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
 740                             >= le16_to_cpu(curp->p_hdr->eh_max))) {
 741                EXT4_ERROR_INODE(inode,
 742                                 "eh_entries %d >= eh_max %d!",
 743                                 le16_to_cpu(curp->p_hdr->eh_entries),
 744                                 le16_to_cpu(curp->p_hdr->eh_max));
 745                return -EIO;
 746        }
 747
 748        if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
 749                /* insert after */
 750                ext_debug("insert new index %d after: %llu\n", logical, ptr);
 751                ix = curp->p_idx + 1;
 752        } else {
 753                /* insert before */
 754                ext_debug("insert new index %d before: %llu\n", logical, ptr);
 755                ix = curp->p_idx;
 756        }
 757
 758        len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
 759        BUG_ON(len < 0);
 760        if (len > 0) {
 761                ext_debug("insert new index %d: "
 762                                "move %d indices from 0x%p to 0x%p\n",
 763                                logical, len, ix, ix + 1);
 764                memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
 765        }
 766
 767        if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
 768                EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
 769                return -EIO;
 770        }
 771
 772        ix->ei_block = cpu_to_le32(logical);
 773        ext4_idx_store_pblock(ix, ptr);
 774        le16_add_cpu(&curp->p_hdr->eh_entries, 1);
 775
 776        if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
 777                EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
 778                return -EIO;
 779        }
 780
 781        err = ext4_ext_dirty(handle, inode, curp);
 782        ext4_std_error(inode->i_sb, err);
 783
 784        return err;
 785}
 786
 787/*
 788 * ext4_ext_split:
 789 * inserts new subtree into the path, using free index entry
 790 * at depth @at:
 791 * - allocates all needed blocks (new leaf and all intermediate index blocks)
 792 * - makes decision where to split
 793 * - moves remaining extents and index entries (right to the split point)
 794 *   into the newly allocated blocks
 795 * - initializes subtree
 796 */
 797static int ext4_ext_split(handle_t *handle, struct inode *inode,
 798                          unsigned int flags,
 799                          struct ext4_ext_path *path,
 800                          struct ext4_extent *newext, int at)
 801{
 802        struct buffer_head *bh = NULL;
 803        int depth = ext_depth(inode);
 804        struct ext4_extent_header *neh;
 805        struct ext4_extent_idx *fidx;
 806        int i = at, k, m, a;
 807        ext4_fsblk_t newblock, oldblock;
 808        __le32 border;
 809        ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
 810        int err = 0;
 811
 812        /* make decision: where to split? */
 813        /* FIXME: now decision is simplest: at current extent */
 814
 815        /* if current leaf will be split, then we should use
 816         * border from split point */
 817        if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
 818                EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
 819                return -EIO;
 820        }
 821        if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
 822                border = path[depth].p_ext[1].ee_block;
 823                ext_debug("leaf will be split."
 824                                " next leaf starts at %d\n",
 825                                  le32_to_cpu(border));
 826        } else {
 827                border = newext->ee_block;
 828                ext_debug("leaf will be added."
 829                                " next leaf starts at %d\n",
 830                                le32_to_cpu(border));
 831        }
 832
 833        /*
 834         * If error occurs, then we break processing
 835         * and mark filesystem read-only. index won't
 836         * be inserted and tree will be in consistent
 837         * state. Next mount will repair buffers too.
 838         */
 839
 840        /*
 841         * Get array to track all allocated blocks.
 842         * We need this to handle errors and free blocks
 843         * upon them.
 844         */
 845        ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
 846        if (!ablocks)
 847                return -ENOMEM;
 848
 849        /* allocate all needed blocks */
 850        ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
 851        for (a = 0; a < depth - at; a++) {
 852                newblock = ext4_ext_new_meta_block(handle, inode, path,
 853                                                   newext, &err, flags);
 854                if (newblock == 0)
 855                        goto cleanup;
 856                ablocks[a] = newblock;
 857        }
 858
 859        /* initialize new leaf */
 860        newblock = ablocks[--a];
 861        if (unlikely(newblock == 0)) {
 862                EXT4_ERROR_INODE(inode, "newblock == 0!");
 863                err = -EIO;
 864                goto cleanup;
 865        }
 866        bh = sb_getblk(inode->i_sb, newblock);
 867        if (!bh) {
 868                err = -EIO;
 869                goto cleanup;
 870        }
 871        lock_buffer(bh);
 872
 873        err = ext4_journal_get_create_access(handle, bh);
 874        if (err)
 875                goto cleanup;
 876
 877        neh = ext_block_hdr(bh);
 878        neh->eh_entries = 0;
 879        neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
 880        neh->eh_magic = EXT4_EXT_MAGIC;
 881        neh->eh_depth = 0;
 882
 883        /* move remainder of path[depth] to the new leaf */
 884        if (unlikely(path[depth].p_hdr->eh_entries !=
 885                     path[depth].p_hdr->eh_max)) {
 886                EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
 887                                 path[depth].p_hdr->eh_entries,
 888                                 path[depth].p_hdr->eh_max);
 889                err = -EIO;
 890                goto cleanup;
 891        }
 892        /* start copy from next extent */
 893        m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
 894        ext4_ext_show_move(inode, path, newblock, depth);
 895        if (m) {
 896                struct ext4_extent *ex;
 897                ex = EXT_FIRST_EXTENT(neh);
 898                memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
 899                le16_add_cpu(&neh->eh_entries, m);
 900        }
 901
 902        set_buffer_uptodate(bh);
 903        unlock_buffer(bh);
 904
 905        err = ext4_handle_dirty_metadata(handle, inode, bh);
 906        if (err)
 907                goto cleanup;
 908        brelse(bh);
 909        bh = NULL;
 910
 911        /* correct old leaf */
 912        if (m) {
 913                err = ext4_ext_get_access(handle, inode, path + depth);
 914                if (err)
 915                        goto cleanup;
 916                le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
 917                err = ext4_ext_dirty(handle, inode, path + depth);
 918                if (err)
 919                        goto cleanup;
 920
 921        }
 922
 923        /* create intermediate indexes */
 924        k = depth - at - 1;
 925        if (unlikely(k < 0)) {
 926                EXT4_ERROR_INODE(inode, "k %d < 0!", k);
 927                err = -EIO;
 928                goto cleanup;
 929        }
 930        if (k)
 931                ext_debug("create %d intermediate indices\n", k);
 932        /* insert new index into current index block */
 933        /* current depth stored in i var */
 934        i = depth - 1;
 935        while (k--) {
 936                oldblock = newblock;
 937                newblock = ablocks[--a];
 938                bh = sb_getblk(inode->i_sb, newblock);
 939                if (!bh) {
 940                        err = -EIO;
 941                        goto cleanup;
 942                }
 943                lock_buffer(bh);
 944
 945                err = ext4_journal_get_create_access(handle, bh);
 946                if (err)
 947                        goto cleanup;
 948
 949                neh = ext_block_hdr(bh);
 950                neh->eh_entries = cpu_to_le16(1);
 951                neh->eh_magic = EXT4_EXT_MAGIC;
 952                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
 953                neh->eh_depth = cpu_to_le16(depth - i);
 954                fidx = EXT_FIRST_INDEX(neh);
 955                fidx->ei_block = border;
 956                ext4_idx_store_pblock(fidx, oldblock);
 957
 958                ext_debug("int.index at %d (block %llu): %u -> %llu\n",
 959                                i, newblock, le32_to_cpu(border), oldblock);
 960
 961                /* move remainder of path[i] to the new index block */
 962                if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
 963                                        EXT_LAST_INDEX(path[i].p_hdr))) {
 964                        EXT4_ERROR_INODE(inode,
 965                                         "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
 966                                         le32_to_cpu(path[i].p_ext->ee_block));
 967                        err = -EIO;
 968                        goto cleanup;
 969                }
 970                /* start copy indexes */
 971                m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
 972                ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
 973                                EXT_MAX_INDEX(path[i].p_hdr));
 974                ext4_ext_show_move(inode, path, newblock, i);
 975                if (m) {
 976                        memmove(++fidx, path[i].p_idx,
 977                                sizeof(struct ext4_extent_idx) * m);
 978                        le16_add_cpu(&neh->eh_entries, m);
 979                }
 980                set_buffer_uptodate(bh);
 981                unlock_buffer(bh);
 982
 983                err = ext4_handle_dirty_metadata(handle, inode, bh);
 984                if (err)
 985                        goto cleanup;
 986                brelse(bh);
 987                bh = NULL;
 988
 989                /* correct old index */
 990                if (m) {
 991                        err = ext4_ext_get_access(handle, inode, path + i);
 992                        if (err)
 993                                goto cleanup;
 994                        le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
 995                        err = ext4_ext_dirty(handle, inode, path + i);
 996                        if (err)
 997                                goto cleanup;
 998                }
 999
1000                i--;
1001        }
1002
1003        /* insert new index */
1004        err = ext4_ext_insert_index(handle, inode, path + at,
1005                                    le32_to_cpu(border), newblock);
1006
1007cleanup:
1008        if (bh) {
1009                if (buffer_locked(bh))
1010                        unlock_buffer(bh);
1011                brelse(bh);
1012        }
1013
1014        if (err) {
1015                /* free all allocated blocks in error case */
1016                for (i = 0; i < depth; i++) {
1017                        if (!ablocks[i])
1018                                continue;
1019                        ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1020                                         EXT4_FREE_BLOCKS_METADATA);
1021                }
1022        }
1023        kfree(ablocks);
1024
1025        return err;
1026}
1027
1028/*
1029 * ext4_ext_grow_indepth:
1030 * implements tree growing procedure:
1031 * - allocates new block
1032 * - moves top-level data (index block or leaf) into the new block
1033 * - initializes new top-level, creating index that points to the
1034 *   just created block
1035 */
1036static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1037                                 unsigned int flags,
1038                                 struct ext4_extent *newext)
1039{
1040        struct ext4_extent_header *neh;
1041        struct buffer_head *bh;
1042        ext4_fsblk_t newblock;
1043        int err = 0;
1044
1045        newblock = ext4_ext_new_meta_block(handle, inode, NULL,
1046                newext, &err, flags);
1047        if (newblock == 0)
1048                return err;
1049
1050        bh = sb_getblk(inode->i_sb, newblock);
1051        if (!bh) {
1052                err = -EIO;
1053                ext4_std_error(inode->i_sb, err);
1054                return err;
1055        }
1056        lock_buffer(bh);
1057
1058        err = ext4_journal_get_create_access(handle, bh);
1059        if (err) {
1060                unlock_buffer(bh);
1061                goto out;
1062        }
1063
1064        /* move top-level index/leaf into new block */
1065        memmove(bh->b_data, EXT4_I(inode)->i_data,
1066                sizeof(EXT4_I(inode)->i_data));
1067
1068        /* set size of new block */
1069        neh = ext_block_hdr(bh);
1070        /* old root could have indexes or leaves
1071         * so calculate e_max right way */
1072        if (ext_depth(inode))
1073                neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1074        else
1075                neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1076        neh->eh_magic = EXT4_EXT_MAGIC;
1077        set_buffer_uptodate(bh);
1078        unlock_buffer(bh);
1079
1080        err = ext4_handle_dirty_metadata(handle, inode, bh);
1081        if (err)
1082                goto out;
1083
1084        /* Update top-level index: num,max,pointer */
1085        neh = ext_inode_hdr(inode);
1086        neh->eh_entries = cpu_to_le16(1);
1087        ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1088        if (neh->eh_depth == 0) {
1089                /* Root extent block becomes index block */
1090                neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1091                EXT_FIRST_INDEX(neh)->ei_block =
1092                        EXT_FIRST_EXTENT(neh)->ee_block;
1093        }
1094        ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1095                  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1096                  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1097                  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1098
1099        neh->eh_depth = cpu_to_le16(le16_to_cpu(neh->eh_depth) + 1);
1100        ext4_mark_inode_dirty(handle, inode);
1101out:
1102        brelse(bh);
1103
1104        return err;
1105}
1106
1107/*
1108 * ext4_ext_create_new_leaf:
1109 * finds empty index and adds new leaf.
1110 * if no free index is found, then it requests in-depth growing.
1111 */
1112static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1113                                    unsigned int flags,
1114                                    struct ext4_ext_path *path,
1115                                    struct ext4_extent *newext)
1116{
1117        struct ext4_ext_path *curp;
1118        int depth, i, err = 0;
1119
1120repeat:
1121        i = depth = ext_depth(inode);
1122
1123        /* walk up to the tree and look for free index entry */
1124        curp = path + depth;
1125        while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1126                i--;
1127                curp--;
1128        }
1129
1130        /* we use already allocated block for index block,
1131         * so subsequent data blocks should be contiguous */
1132        if (EXT_HAS_FREE_INDEX(curp)) {
1133                /* if we found index with free entry, then use that
1134                 * entry: create all needed subtree and add new leaf */
1135                err = ext4_ext_split(handle, inode, flags, path, newext, i);
1136                if (err)
1137                        goto out;
1138
1139                /* refill path */
1140                ext4_ext_drop_refs(path);
1141                path = ext4_ext_find_extent(inode,
1142                                    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1143                                    path);
1144                if (IS_ERR(path))
1145                        err = PTR_ERR(path);
1146        } else {
1147                /* tree is full, time to grow in depth */
1148                err = ext4_ext_grow_indepth(handle, inode, flags, newext);
1149                if (err)
1150                        goto out;
1151
1152                /* refill path */
1153                ext4_ext_drop_refs(path);
1154                path = ext4_ext_find_extent(inode,
1155                                   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1156                                    path);
1157                if (IS_ERR(path)) {
1158                        err = PTR_ERR(path);
1159                        goto out;
1160                }
1161
1162                /*
1163                 * only first (depth 0 -> 1) produces free space;
1164                 * in all other cases we have to split the grown tree
1165                 */
1166                depth = ext_depth(inode);
1167                if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1168                        /* now we need to split */
1169                        goto repeat;
1170                }
1171        }
1172
1173out:
1174        return err;
1175}
1176
1177/*
1178 * search the closest allocated block to the left for *logical
1179 * and returns it at @logical + it's physical address at @phys
1180 * if *logical is the smallest allocated block, the function
1181 * returns 0 at @phys
1182 * return value contains 0 (success) or error code
1183 */
1184static int ext4_ext_search_left(struct inode *inode,
1185                                struct ext4_ext_path *path,
1186                                ext4_lblk_t *logical, ext4_fsblk_t *phys)
1187{
1188        struct ext4_extent_idx *ix;
1189        struct ext4_extent *ex;
1190        int depth, ee_len;
1191
1192        if (unlikely(path == NULL)) {
1193                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1194                return -EIO;
1195        }
1196        depth = path->p_depth;
1197        *phys = 0;
1198
1199        if (depth == 0 && path->p_ext == NULL)
1200                return 0;
1201
1202        /* usually extent in the path covers blocks smaller
1203         * then *logical, but it can be that extent is the
1204         * first one in the file */
1205
1206        ex = path[depth].p_ext;
1207        ee_len = ext4_ext_get_actual_len(ex);
1208        if (*logical < le32_to_cpu(ex->ee_block)) {
1209                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1210                        EXT4_ERROR_INODE(inode,
1211                                         "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1212                                         *logical, le32_to_cpu(ex->ee_block));
1213                        return -EIO;
1214                }
1215                while (--depth >= 0) {
1216                        ix = path[depth].p_idx;
1217                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1218                                EXT4_ERROR_INODE(inode,
1219                                  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1220                                  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1221                                  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1222                le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1223                                  depth);
1224                                return -EIO;
1225                        }
1226                }
1227                return 0;
1228        }
1229
1230        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1231                EXT4_ERROR_INODE(inode,
1232                                 "logical %d < ee_block %d + ee_len %d!",
1233                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1234                return -EIO;
1235        }
1236
1237        *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1238        *phys = ext4_ext_pblock(ex) + ee_len - 1;
1239        return 0;
1240}
1241
1242/*
1243 * search the closest allocated block to the right for *logical
1244 * and returns it at @logical + it's physical address at @phys
1245 * if *logical is the largest allocated block, the function
1246 * returns 0 at @phys
1247 * return value contains 0 (success) or error code
1248 */
1249static int ext4_ext_search_right(struct inode *inode,
1250                                 struct ext4_ext_path *path,
1251                                 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1252                                 struct ext4_extent **ret_ex)
1253{
1254        struct buffer_head *bh = NULL;
1255        struct ext4_extent_header *eh;
1256        struct ext4_extent_idx *ix;
1257        struct ext4_extent *ex;
1258        ext4_fsblk_t block;
1259        int depth;      /* Note, NOT eh_depth; depth from top of tree */
1260        int ee_len;
1261
1262        if (unlikely(path == NULL)) {
1263                EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1264                return -EIO;
1265        }
1266        depth = path->p_depth;
1267        *phys = 0;
1268
1269        if (depth == 0 && path->p_ext == NULL)
1270                return 0;
1271
1272        /* usually extent in the path covers blocks smaller
1273         * then *logical, but it can be that extent is the
1274         * first one in the file */
1275
1276        ex = path[depth].p_ext;
1277        ee_len = ext4_ext_get_actual_len(ex);
1278        if (*logical < le32_to_cpu(ex->ee_block)) {
1279                if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1280                        EXT4_ERROR_INODE(inode,
1281                                         "first_extent(path[%d].p_hdr) != ex",
1282                                         depth);
1283                        return -EIO;
1284                }
1285                while (--depth >= 0) {
1286                        ix = path[depth].p_idx;
1287                        if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1288                                EXT4_ERROR_INODE(inode,
1289                                                 "ix != EXT_FIRST_INDEX *logical %d!",
1290                                                 *logical);
1291                                return -EIO;
1292                        }
1293                }
1294                goto found_extent;
1295        }
1296
1297        if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1298                EXT4_ERROR_INODE(inode,
1299                                 "logical %d < ee_block %d + ee_len %d!",
1300                                 *logical, le32_to_cpu(ex->ee_block), ee_len);
1301                return -EIO;
1302        }
1303
1304        if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1305                /* next allocated block in this leaf */
1306                ex++;
1307                goto found_extent;
1308        }
1309
1310        /* go up and search for index to the right */
1311        while (--depth >= 0) {
1312                ix = path[depth].p_idx;
1313                if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1314                        goto got_index;
1315        }
1316
1317        /* we've gone up to the root and found no index to the right */
1318        return 0;
1319
1320got_index:
1321        /* we've found index to the right, let's
1322         * follow it and find the closest allocated
1323         * block to the right */
1324        ix++;
1325        block = ext4_idx_pblock(ix);
1326        while (++depth < path->p_depth) {
1327                bh = sb_bread(inode->i_sb, block);
1328                if (bh == NULL)
1329                        return -EIO;
1330                eh = ext_block_hdr(bh);
1331                /* subtract from p_depth to get proper eh_depth */
1332                if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1333                        put_bh(bh);
1334                        return -EIO;
1335                }
1336                ix = EXT_FIRST_INDEX(eh);
1337                block = ext4_idx_pblock(ix);
1338                put_bh(bh);
1339        }
1340
1341        bh = sb_bread(inode->i_sb, block);
1342        if (bh == NULL)
1343                return -EIO;
1344        eh = ext_block_hdr(bh);
1345        if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
1346                put_bh(bh);
1347                return -EIO;
1348        }
1349        ex = EXT_FIRST_EXTENT(eh);
1350found_extent:
1351        *logical = le32_to_cpu(ex->ee_block);
1352        *phys = ext4_ext_pblock(ex);
1353        *ret_ex = ex;
1354        if (bh)
1355                put_bh(bh);
1356        return 0;
1357}
1358
1359/*
1360 * ext4_ext_next_allocated_block:
1361 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1362 * NOTE: it considers block number from index entry as
1363 * allocated block. Thus, index entries have to be consistent
1364 * with leaves.
1365 */
1366static ext4_lblk_t
1367ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1368{
1369        int depth;
1370
1371        BUG_ON(path == NULL);
1372        depth = path->p_depth;
1373
1374        if (depth == 0 && path->p_ext == NULL)
1375                return EXT_MAX_BLOCKS;
1376
1377        while (depth >= 0) {
1378                if (depth == path->p_depth) {
1379                        /* leaf */
1380                        if (path[depth].p_ext &&
1381                                path[depth].p_ext !=
1382                                        EXT_LAST_EXTENT(path[depth].p_hdr))
1383                          return le32_to_cpu(path[depth].p_ext[1].ee_block);
1384                } else {
1385                        /* index */
1386                        if (path[depth].p_idx !=
1387                                        EXT_LAST_INDEX(path[depth].p_hdr))
1388                          return le32_to_cpu(path[depth].p_idx[1].ei_block);
1389                }
1390                depth--;
1391        }
1392
1393        return EXT_MAX_BLOCKS;
1394}
1395
1396/*
1397 * ext4_ext_next_leaf_block:
1398 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1399 */
1400static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1401{
1402        int depth;
1403
1404        BUG_ON(path == NULL);
1405        depth = path->p_depth;
1406
1407        /* zero-tree has no leaf blocks at all */
1408        if (depth == 0)
1409                return EXT_MAX_BLOCKS;
1410
1411        /* go to index block */
1412        depth--;
1413
1414        while (depth >= 0) {
1415                if (path[depth].p_idx !=
1416                                EXT_LAST_INDEX(path[depth].p_hdr))
1417                        return (ext4_lblk_t)
1418                                le32_to_cpu(path[depth].p_idx[1].ei_block);
1419                depth--;
1420        }
1421
1422        return EXT_MAX_BLOCKS;
1423}
1424
1425/*
1426 * ext4_ext_correct_indexes:
1427 * if leaf gets modified and modified extent is first in the leaf,
1428 * then we have to correct all indexes above.
1429 * TODO: do we need to correct tree in all cases?
1430 */
1431static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1432                                struct ext4_ext_path *path)
1433{
1434        struct ext4_extent_header *eh;
1435        int depth = ext_depth(inode);
1436        struct ext4_extent *ex;
1437        __le32 border;
1438        int k, err = 0;
1439
1440        eh = path[depth].p_hdr;
1441        ex = path[depth].p_ext;
1442
1443        if (unlikely(ex == NULL || eh == NULL)) {
1444                EXT4_ERROR_INODE(inode,
1445                                 "ex %p == NULL or eh %p == NULL", ex, eh);
1446                return -EIO;
1447        }
1448
1449        if (depth == 0) {
1450                /* there is no tree at all */
1451                return 0;
1452        }
1453
1454        if (ex != EXT_FIRST_EXTENT(eh)) {
1455                /* we correct tree if first leaf got modified only */
1456                return 0;
1457        }
1458
1459        /*
1460         * TODO: we need correction if border is smaller than current one
1461         */
1462        k = depth - 1;
1463        border = path[depth].p_ext->ee_block;
1464        err = ext4_ext_get_access(handle, inode, path + k);
1465        if (err)
1466                return err;
1467        path[k].p_idx->ei_block = border;
1468        err = ext4_ext_dirty(handle, inode, path + k);
1469        if (err)
1470                return err;
1471
1472        while (k--) {
1473                /* change all left-side indexes */
1474                if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1475                        break;
1476                err = ext4_ext_get_access(handle, inode, path + k);
1477                if (err)
1478                        break;
1479                path[k].p_idx->ei_block = border;
1480                err = ext4_ext_dirty(handle, inode, path + k);
1481                if (err)
1482                        break;
1483        }
1484
1485        return err;
1486}
1487
1488int
1489ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1490                                struct ext4_extent *ex2)
1491{
1492        unsigned short ext1_ee_len, ext2_ee_len, max_len;
1493
1494        /*
1495         * Make sure that either both extents are uninitialized, or
1496         * both are _not_.
1497         */
1498        if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1499                return 0;
1500
1501        if (ext4_ext_is_uninitialized(ex1))
1502                max_len = EXT_UNINIT_MAX_LEN;
1503        else
1504                max_len = EXT_INIT_MAX_LEN;
1505
1506        ext1_ee_len = ext4_ext_get_actual_len(ex1);
1507        ext2_ee_len = ext4_ext_get_actual_len(ex2);
1508
1509        if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1510                        le32_to_cpu(ex2->ee_block))
1511                return 0;
1512
1513        /*
1514         * To allow future support for preallocated extents to be added
1515         * as an RO_COMPAT feature, refuse to merge to extents if
1516         * this can result in the top bit of ee_len being set.
1517         */
1518        if (ext1_ee_len + ext2_ee_len > max_len)
1519                return 0;
1520#ifdef AGGRESSIVE_TEST
1521        if (ext1_ee_len >= 4)
1522                return 0;
1523#endif
1524
1525        if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1526                return 1;
1527        return 0;
1528}
1529
1530/*
1531 * This function tries to merge the "ex" extent to the next extent in the tree.
1532 * It always tries to merge towards right. If you want to merge towards
1533 * left, pass "ex - 1" as argument instead of "ex".
1534 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1535 * 1 if they got merged.
1536 */
1537static int ext4_ext_try_to_merge_right(struct inode *inode,
1538                                 struct ext4_ext_path *path,
1539                                 struct ext4_extent *ex)
1540{
1541        struct ext4_extent_header *eh;
1542        unsigned int depth, len;
1543        int merge_done = 0;
1544        int uninitialized = 0;
1545
1546        depth = ext_depth(inode);
1547        BUG_ON(path[depth].p_hdr == NULL);
1548        eh = path[depth].p_hdr;
1549
1550        while (ex < EXT_LAST_EXTENT(eh)) {
1551                if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1552                        break;
1553                /* merge with next extent! */
1554                if (ext4_ext_is_uninitialized(ex))
1555                        uninitialized = 1;
1556                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1557                                + ext4_ext_get_actual_len(ex + 1));
1558                if (uninitialized)
1559                        ext4_ext_mark_uninitialized(ex);
1560
1561                if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1562                        len = (EXT_LAST_EXTENT(eh) - ex - 1)
1563                                * sizeof(struct ext4_extent);
1564                        memmove(ex + 1, ex + 2, len);
1565                }
1566                le16_add_cpu(&eh->eh_entries, -1);
1567                merge_done = 1;
1568                WARN_ON(eh->eh_entries == 0);
1569                if (!eh->eh_entries)
1570                        EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1571        }
1572
1573        return merge_done;
1574}
1575
1576/*
1577 * This function tries to merge the @ex extent to neighbours in the tree.
1578 * return 1 if merge left else 0.
1579 */
1580static int ext4_ext_try_to_merge(struct inode *inode,
1581                                  struct ext4_ext_path *path,
1582                                  struct ext4_extent *ex) {
1583        struct ext4_extent_header *eh;
1584        unsigned int depth;
1585        int merge_done = 0;
1586        int ret = 0;
1587
1588        depth = ext_depth(inode);
1589        BUG_ON(path[depth].p_hdr == NULL);
1590        eh = path[depth].p_hdr;
1591
1592        if (ex > EXT_FIRST_EXTENT(eh))
1593                merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1594
1595        if (!merge_done)
1596                ret = ext4_ext_try_to_merge_right(inode, path, ex);
1597
1598        return ret;
1599}
1600
1601/*
1602 * check if a portion of the "newext" extent overlaps with an
1603 * existing extent.
1604 *
1605 * If there is an overlap discovered, it updates the length of the newext
1606 * such that there will be no overlap, and then returns 1.
1607 * If there is no overlap found, it returns 0.
1608 */
1609static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1610                                           struct inode *inode,
1611                                           struct ext4_extent *newext,
1612                                           struct ext4_ext_path *path)
1613{
1614        ext4_lblk_t b1, b2;
1615        unsigned int depth, len1;
1616        unsigned int ret = 0;
1617
1618        b1 = le32_to_cpu(newext->ee_block);
1619        len1 = ext4_ext_get_actual_len(newext);
1620        depth = ext_depth(inode);
1621        if (!path[depth].p_ext)
1622                goto out;
1623        b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1624        b2 &= ~(sbi->s_cluster_ratio - 1);
1625
1626        /*
1627         * get the next allocated block if the extent in the path
1628         * is before the requested block(s)
1629         */
1630        if (b2 < b1) {
1631                b2 = ext4_ext_next_allocated_block(path);
1632                if (b2 == EXT_MAX_BLOCKS)
1633                        goto out;
1634                b2 &= ~(sbi->s_cluster_ratio - 1);
1635        }
1636
1637        /* check for wrap through zero on extent logical start block*/
1638        if (b1 + len1 < b1) {
1639                len1 = EXT_MAX_BLOCKS - b1;
1640                newext->ee_len = cpu_to_le16(len1);
1641                ret = 1;
1642        }
1643
1644        /* check for overlap */
1645        if (b1 + len1 > b2) {
1646                newext->ee_len = cpu_to_le16(b2 - b1);
1647                ret = 1;
1648        }
1649out:
1650        return ret;
1651}
1652
1653/*
1654 * ext4_ext_insert_extent:
1655 * tries to merge requsted extent into the existing extent or
1656 * inserts requested extent as new one into the tree,
1657 * creating new leaf in the no-space case.
1658 */
1659int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1660                                struct ext4_ext_path *path,
1661                                struct ext4_extent *newext, int flag)
1662{
1663        struct ext4_extent_header *eh;
1664        struct ext4_extent *ex, *fex;
1665        struct ext4_extent *nearex; /* nearest extent */
1666        struct ext4_ext_path *npath = NULL;
1667        int depth, len, err;
1668        ext4_lblk_t next;
1669        unsigned uninitialized = 0;
1670        int flags = 0;
1671
1672        if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1673                EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1674                return -EIO;
1675        }
1676        depth = ext_depth(inode);
1677        ex = path[depth].p_ext;
1678        if (unlikely(path[depth].p_hdr == NULL)) {
1679                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1680                return -EIO;
1681        }
1682
1683        /* try to insert block into found extent and return */
1684        if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)
1685                && ext4_can_extents_be_merged(inode, ex, newext)) {
1686                ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1687                          ext4_ext_is_uninitialized(newext),
1688                          ext4_ext_get_actual_len(newext),
1689                          le32_to_cpu(ex->ee_block),
1690                          ext4_ext_is_uninitialized(ex),
1691                          ext4_ext_get_actual_len(ex),
1692                          ext4_ext_pblock(ex));
1693                err = ext4_ext_get_access(handle, inode, path + depth);
1694                if (err)
1695                        return err;
1696
1697                /*
1698                 * ext4_can_extents_be_merged should have checked that either
1699                 * both extents are uninitialized, or both aren't. Thus we
1700                 * need to check only one of them here.
1701                 */
1702                if (ext4_ext_is_uninitialized(ex))
1703                        uninitialized = 1;
1704                ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1705                                        + ext4_ext_get_actual_len(newext));
1706                if (uninitialized)
1707                        ext4_ext_mark_uninitialized(ex);
1708                eh = path[depth].p_hdr;
1709                nearex = ex;
1710                goto merge;
1711        }
1712
1713        depth = ext_depth(inode);
1714        eh = path[depth].p_hdr;
1715        if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1716                goto has_space;
1717
1718        /* probably next leaf has space for us? */
1719        fex = EXT_LAST_EXTENT(eh);
1720        next = EXT_MAX_BLOCKS;
1721        if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
1722                next = ext4_ext_next_leaf_block(path);
1723        if (next != EXT_MAX_BLOCKS) {
1724                ext_debug("next leaf block - %u\n", next);
1725                BUG_ON(npath != NULL);
1726                npath = ext4_ext_find_extent(inode, next, NULL);
1727                if (IS_ERR(npath))
1728                        return PTR_ERR(npath);
1729                BUG_ON(npath->p_depth != path->p_depth);
1730                eh = npath[depth].p_hdr;
1731                if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1732                        ext_debug("next leaf isn't full(%d)\n",
1733                                  le16_to_cpu(eh->eh_entries));
1734                        path = npath;
1735                        goto has_space;
1736                }
1737                ext_debug("next leaf has no free space(%d,%d)\n",
1738                          le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1739        }
1740
1741        /*
1742         * There is no free space in the found leaf.
1743         * We're gonna add a new leaf in the tree.
1744         */
1745        if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT)
1746                flags = EXT4_MB_USE_ROOT_BLOCKS;
1747        err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext);
1748        if (err)
1749                goto cleanup;
1750        depth = ext_depth(inode);
1751        eh = path[depth].p_hdr;
1752
1753has_space:
1754        nearex = path[depth].p_ext;
1755
1756        err = ext4_ext_get_access(handle, inode, path + depth);
1757        if (err)
1758                goto cleanup;
1759
1760        if (!nearex) {
1761                /* there is no extent in this leaf, create first one */
1762                ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1763                                le32_to_cpu(newext->ee_block),
1764                                ext4_ext_pblock(newext),
1765                                ext4_ext_is_uninitialized(newext),
1766                                ext4_ext_get_actual_len(newext));
1767                nearex = EXT_FIRST_EXTENT(eh);
1768        } else {
1769                if (le32_to_cpu(newext->ee_block)
1770                           > le32_to_cpu(nearex->ee_block)) {
1771                        /* Insert after */
1772                        ext_debug("insert %u:%llu:[%d]%d before: "
1773                                        "nearest %p\n",
1774                                        le32_to_cpu(newext->ee_block),
1775                                        ext4_ext_pblock(newext),
1776                                        ext4_ext_is_uninitialized(newext),
1777                                        ext4_ext_get_actual_len(newext),
1778                                        nearex);
1779                        nearex++;
1780                } else {
1781                        /* Insert before */
1782                        BUG_ON(newext->ee_block == nearex->ee_block);
1783                        ext_debug("insert %u:%llu:[%d]%d after: "
1784                                        "nearest %p\n",
1785                                        le32_to_cpu(newext->ee_block),
1786                                        ext4_ext_pblock(newext),
1787                                        ext4_ext_is_uninitialized(newext),
1788                                        ext4_ext_get_actual_len(newext),
1789                                        nearex);
1790                }
1791                len = EXT_LAST_EXTENT(eh) - nearex + 1;
1792                if (len > 0) {
1793                        ext_debug("insert %u:%llu:[%d]%d: "
1794                                        "move %d extents from 0x%p to 0x%p\n",
1795                                        le32_to_cpu(newext->ee_block),
1796                                        ext4_ext_pblock(newext),
1797                                        ext4_ext_is_uninitialized(newext),
1798                                        ext4_ext_get_actual_len(newext),
1799                                        len, nearex, nearex + 1);
1800                        memmove(nearex + 1, nearex,
1801                                len * sizeof(struct ext4_extent));
1802                }
1803        }
1804
1805        le16_add_cpu(&eh->eh_entries, 1);
1806        path[depth].p_ext = nearex;
1807        nearex->ee_block = newext->ee_block;
1808        ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
1809        nearex->ee_len = newext->ee_len;
1810
1811merge:
1812        /* try to merge extents to the right */
1813        if (!(flag & EXT4_GET_BLOCKS_PRE_IO))
1814                ext4_ext_try_to_merge(inode, path, nearex);
1815
1816        /* try to merge extents to the left */
1817
1818        /* time to correct all indexes above */
1819        err = ext4_ext_correct_indexes(handle, inode, path);
1820        if (err)
1821                goto cleanup;
1822
1823        err = ext4_ext_dirty(handle, inode, path + depth);
1824
1825cleanup:
1826        if (npath) {
1827                ext4_ext_drop_refs(npath);
1828                kfree(npath);
1829        }
1830        ext4_ext_invalidate_cache(inode);
1831        return err;
1832}
1833
1834static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1835                               ext4_lblk_t num, ext_prepare_callback func,
1836                               void *cbdata)
1837{
1838        struct ext4_ext_path *path = NULL;
1839        struct ext4_ext_cache cbex;
1840        struct ext4_extent *ex;
1841        ext4_lblk_t next, start = 0, end = 0;
1842        ext4_lblk_t last = block + num;
1843        int depth, exists, err = 0;
1844
1845        BUG_ON(func == NULL);
1846        BUG_ON(inode == NULL);
1847
1848        while (block < last && block != EXT_MAX_BLOCKS) {
1849                num = last - block;
1850                /* find extent for this block */
1851                down_read(&EXT4_I(inode)->i_data_sem);
1852                path = ext4_ext_find_extent(inode, block, path);
1853                up_read(&EXT4_I(inode)->i_data_sem);
1854                if (IS_ERR(path)) {
1855                        err = PTR_ERR(path);
1856                        path = NULL;
1857                        break;
1858                }
1859
1860                depth = ext_depth(inode);
1861                if (unlikely(path[depth].p_hdr == NULL)) {
1862                        EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1863                        err = -EIO;
1864                        break;
1865                }
1866                ex = path[depth].p_ext;
1867                next = ext4_ext_next_allocated_block(path);
1868
1869                exists = 0;
1870                if (!ex) {
1871                        /* there is no extent yet, so try to allocate
1872                         * all requested space */
1873                        start = block;
1874                        end = block + num;
1875                } else if (le32_to_cpu(ex->ee_block) > block) {
1876                        /* need to allocate space before found extent */
1877                        start = block;
1878                        end = le32_to_cpu(ex->ee_block);
1879                        if (block + num < end)
1880                                end = block + num;
1881                } else if (block >= le32_to_cpu(ex->ee_block)
1882                                        + ext4_ext_get_actual_len(ex)) {
1883                        /* need to allocate space after found extent */
1884                        start = block;
1885                        end = block + num;
1886                        if (end >= next)
1887                                end = next;
1888                } else if (block >= le32_to_cpu(ex->ee_block)) {
1889                        /*
1890                         * some part of requested space is covered
1891                         * by found extent
1892                         */
1893                        start = block;
1894                        end = le32_to_cpu(ex->ee_block)
1895                                + ext4_ext_get_actual_len(ex);
1896                        if (block + num < end)
1897                                end = block + num;
1898                        exists = 1;
1899                } else {
1900                        BUG();
1901                }
1902                BUG_ON(end <= start);
1903
1904                if (!exists) {
1905                        cbex.ec_block = start;
1906                        cbex.ec_len = end - start;
1907                        cbex.ec_start = 0;
1908                } else {
1909                        cbex.ec_block = le32_to_cpu(ex->ee_block);
1910                        cbex.ec_len = ext4_ext_get_actual_len(ex);
1911                        cbex.ec_start = ext4_ext_pblock(ex);
1912                }
1913
1914                if (unlikely(cbex.ec_len == 0)) {
1915                        EXT4_ERROR_INODE(inode, "cbex.ec_len == 0");
1916                        err = -EIO;
1917                        break;
1918                }
1919                err = func(inode, next, &cbex, ex, cbdata);
1920                ext4_ext_drop_refs(path);
1921
1922                if (err < 0)
1923                        break;
1924
1925                if (err == EXT_REPEAT)
1926                        continue;
1927                else if (err == EXT_BREAK) {
1928                        err = 0;
1929                        break;
1930                }
1931
1932                if (ext_depth(inode) != depth) {
1933                        /* depth was changed. we have to realloc path */
1934                        kfree(path);
1935                        path = NULL;
1936                }
1937
1938                block = cbex.ec_block + cbex.ec_len;
1939        }
1940
1941        if (path) {
1942                ext4_ext_drop_refs(path);
1943                kfree(path);
1944        }
1945
1946        return err;
1947}
1948
1949static void
1950ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
1951                        __u32 len, ext4_fsblk_t start)
1952{
1953        struct ext4_ext_cache *cex;
1954        BUG_ON(len == 0);
1955        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1956        trace_ext4_ext_put_in_cache(inode, block, len, start);
1957        cex = &EXT4_I(inode)->i_cached_extent;
1958        cex->ec_block = block;
1959        cex->ec_len = len;
1960        cex->ec_start = start;
1961        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1962}
1963
1964/*
1965 * ext4_ext_put_gap_in_cache:
1966 * calculate boundaries of the gap that the requested block fits into
1967 * and cache this gap
1968 */
1969static void
1970ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1971                                ext4_lblk_t block)
1972{
1973        int depth = ext_depth(inode);
1974        unsigned long len;
1975        ext4_lblk_t lblock;
1976        struct ext4_extent *ex;
1977
1978        ex = path[depth].p_ext;
1979        if (ex == NULL) {
1980                /* there is no extent yet, so gap is [0;-] */
1981                lblock = 0;
1982                len = EXT_MAX_BLOCKS;
1983                ext_debug("cache gap(whole file):");
1984        } else if (block < le32_to_cpu(ex->ee_block)) {
1985                lblock = block;
1986                len = le32_to_cpu(ex->ee_block) - block;
1987                ext_debug("cache gap(before): %u [%u:%u]",
1988                                block,
1989                                le32_to_cpu(ex->ee_block),
1990                                 ext4_ext_get_actual_len(ex));
1991        } else if (block >= le32_to_cpu(ex->ee_block)
1992                        + ext4_ext_get_actual_len(ex)) {
1993                ext4_lblk_t next;
1994                lblock = le32_to_cpu(ex->ee_block)
1995                        + ext4_ext_get_actual_len(ex);
1996
1997                next = ext4_ext_next_allocated_block(path);
1998                ext_debug("cache gap(after): [%u:%u] %u",
1999                                le32_to_cpu(ex->ee_block),
2000                                ext4_ext_get_actual_len(ex),
2001                                block);
2002                BUG_ON(next == lblock);
2003                len = next - lblock;
2004        } else {
2005                lblock = len = 0;
2006                BUG();
2007        }
2008
2009        ext_debug(" -> %u:%lu\n", lblock, len);
2010        ext4_ext_put_in_cache(inode, lblock, len, 0);
2011}
2012
2013/*
2014 * ext4_ext_check_cache()
2015 * Checks to see if the given block is in the cache.
2016 * If it is, the cached extent is stored in the given
2017 * cache extent pointer.  If the cached extent is a hole,
2018 * this routine should be used instead of
2019 * ext4_ext_in_cache if the calling function needs to
2020 * know the size of the hole.
2021 *
2022 * @inode: The files inode
2023 * @block: The block to look for in the cache
2024 * @ex:    Pointer where the cached extent will be stored
2025 *         if it contains block
2026 *
2027 * Return 0 if cache is invalid; 1 if the cache is valid
2028 */
2029static int ext4_ext_check_cache(struct inode *inode, ext4_lblk_t block,
2030        struct ext4_ext_cache *ex){
2031        struct ext4_ext_cache *cex;
2032        struct ext4_sb_info *sbi;
2033        int ret = 0;
2034
2035        /*
2036         * We borrow i_block_reservation_lock to protect i_cached_extent
2037         */
2038        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2039        cex = &EXT4_I(inode)->i_cached_extent;
2040        sbi = EXT4_SB(inode->i_sb);
2041
2042        /* has cache valid data? */
2043        if (cex->ec_len == 0)
2044                goto errout;
2045
2046        if (in_range(block, cex->ec_block, cex->ec_len)) {
2047                memcpy(ex, cex, sizeof(struct ext4_ext_cache));
2048                ext_debug("%u cached by %u:%u:%llu\n",
2049                                block,
2050                                cex->ec_block, cex->ec_len, cex->ec_start);
2051                ret = 1;
2052        }
2053errout:
2054        if (!ret)
2055                sbi->extent_cache_misses++;
2056        else
2057                sbi->extent_cache_hits++;
2058        trace_ext4_ext_in_cache(inode, block, ret);
2059        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2060        return ret;
2061}
2062
2063/*
2064 * ext4_ext_in_cache()
2065 * Checks to see if the given block is in the cache.
2066 * If it is, the cached extent is stored in the given
2067 * extent pointer.
2068 *
2069 * @inode: The files inode
2070 * @block: The block to look for in the cache
2071 * @ex:    Pointer where the cached extent will be stored
2072 *         if it contains block
2073 *
2074 * Return 0 if cache is invalid; 1 if the cache is valid
2075 */
2076static int
2077ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
2078                        struct ext4_extent *ex)
2079{
2080        struct ext4_ext_cache cex;
2081        int ret = 0;
2082
2083        if (ext4_ext_check_cache(inode, block, &cex)) {
2084                ex->ee_block = cpu_to_le32(cex.ec_block);
2085                ext4_ext_store_pblock(ex, cex.ec_start);
2086                ex->ee_len = cpu_to_le16(cex.ec_len);
2087                ret = 1;
2088        }
2089
2090        return ret;
2091}
2092
2093
2094/*
2095 * ext4_ext_rm_idx:
2096 * removes index from the index block.
2097 */
2098static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2099                        struct ext4_ext_path *path)
2100{
2101        int err;
2102        ext4_fsblk_t leaf;
2103
2104        /* free index block */
2105        path--;
2106        leaf = ext4_idx_pblock(path->p_idx);
2107        if (unlikely(path->p_hdr->eh_entries == 0)) {
2108                EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2109                return -EIO;
2110        }
2111        err = ext4_ext_get_access(handle, inode, path);
2112        if (err)
2113                return err;
2114
2115        if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2116                int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2117                len *= sizeof(struct ext4_extent_idx);
2118                memmove(path->p_idx, path->p_idx + 1, len);
2119        }
2120
2121        le16_add_cpu(&path->p_hdr->eh_entries, -1);
2122        err = ext4_ext_dirty(handle, inode, path);
2123        if (err)
2124                return err;
2125        ext_debug("index is empty, remove it, free block %llu\n", leaf);
2126        trace_ext4_ext_rm_idx(inode, leaf);
2127
2128        ext4_free_blocks(handle, inode, NULL, leaf, 1,
2129                         EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2130        return err;
2131}
2132
2133/*
2134 * ext4_ext_calc_credits_for_single_extent:
2135 * This routine returns max. credits that needed to insert an extent
2136 * to the extent tree.
2137 * When pass the actual path, the caller should calculate credits
2138 * under i_data_sem.
2139 */
2140int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2141                                                struct ext4_ext_path *path)
2142{
2143        if (path) {
2144                int depth = ext_depth(inode);
2145                int ret = 0;
2146
2147                /* probably there is space in leaf? */
2148                if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2149                                < le16_to_cpu(path[depth].p_hdr->eh_max)) {
2150
2151                        /*
2152                         *  There are some space in the leaf tree, no
2153                         *  need to account for leaf block credit
2154                         *
2155                         *  bitmaps and block group descriptor blocks
2156                         *  and other metadata blocks still need to be
2157                         *  accounted.
2158                         */
2159                        /* 1 bitmap, 1 block group descriptor */
2160                        ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2161                        return ret;
2162                }
2163        }
2164
2165        return ext4_chunk_trans_blocks(inode, nrblocks);
2166}
2167
2168/*
2169 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2170 *
2171 * if nrblocks are fit in a single extent (chunk flag is 1), then
2172 * in the worse case, each tree level index/leaf need to be changed
2173 * if the tree split due to insert a new extent, then the old tree
2174 * index/leaf need to be updated too
2175 *
2176 * If the nrblocks are discontiguous, they could cause
2177 * the whole tree split more than once, but this is really rare.
2178 */
2179int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
2180{
2181        int index;
2182        int depth = ext_depth(inode);
2183
2184        if (chunk)
2185                index = depth * 2;
2186        else
2187                index = depth * 3;
2188
2189        return index;
2190}
2191
2192static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2193                              struct ext4_extent *ex,
2194                              ext4_fsblk_t *partial_cluster,
2195                              ext4_lblk_t from, ext4_lblk_t to)
2196{
2197        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2198        unsigned short ee_len =  ext4_ext_get_actual_len(ex);
2199        ext4_fsblk_t pblk;
2200        int flags = EXT4_FREE_BLOCKS_FORGET;
2201
2202        if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2203                flags |= EXT4_FREE_BLOCKS_METADATA;
2204        /*
2205         * For bigalloc file systems, we never free a partial cluster
2206         * at the beginning of the extent.  Instead, we make a note
2207         * that we tried freeing the cluster, and check to see if we
2208         * need to free it on a subsequent call to ext4_remove_blocks,
2209         * or at the end of the ext4_truncate() operation.
2210         */
2211        flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2212
2213        trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster);
2214        /*
2215         * If we have a partial cluster, and it's different from the
2216         * cluster of the last block, we need to explicitly free the
2217         * partial cluster here.
2218         */
2219        pblk = ext4_ext_pblock(ex) + ee_len - 1;
2220        if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) {
2221                ext4_free_blocks(handle, inode, NULL,
2222                                 EXT4_C2B(sbi, *partial_cluster),
2223                                 sbi->s_cluster_ratio, flags);
2224                *partial_cluster = 0;
2225        }
2226
2227#ifdef EXTENTS_STATS
2228        {
2229                struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2230                spin_lock(&sbi->s_ext_stats_lock);
2231                sbi->s_ext_blocks += ee_len;
2232                sbi->s_ext_extents++;
2233                if (ee_len < sbi->s_ext_min)
2234                        sbi->s_ext_min = ee_len;
2235                if (ee_len > sbi->s_ext_max)
2236                        sbi->s_ext_max = ee_len;
2237                if (ext_depth(inode) > sbi->s_depth_max)
2238                        sbi->s_depth_max = ext_depth(inode);
2239                spin_unlock(&sbi->s_ext_stats_lock);
2240        }
2241#endif
2242        if (from >= le32_to_cpu(ex->ee_block)
2243            && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
2244                /* tail removal */
2245                ext4_lblk_t num;
2246
2247                num = le32_to_cpu(ex->ee_block) + ee_len - from;
2248                pblk = ext4_ext_pblock(ex) + ee_len - num;
2249                ext_debug("free last %u blocks starting %llu\n", num, pblk);
2250                ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2251                /*
2252                 * If the block range to be freed didn't start at the
2253                 * beginning of a cluster, and we removed the entire
2254                 * extent, save the partial cluster here, since we
2255                 * might need to delete if we determine that the
2256                 * truncate operation has removed all of the blocks in
2257                 * the cluster.
2258                 */
2259                if (pblk & (sbi->s_cluster_ratio - 1) &&
2260                    (ee_len == num))
2261                        *partial_cluster = EXT4_B2C(sbi, pblk);
2262                else
2263                        *partial_cluster = 0;
2264        } else if (from == le32_to_cpu(ex->ee_block)
2265                   && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
2266                /* head removal */
2267                ext4_lblk_t num;
2268                ext4_fsblk_t start;
2269
2270                num = to - from;
2271                start = ext4_ext_pblock(ex);
2272
2273                ext_debug("free first %u blocks starting %llu\n", num, start);
2274                ext4_free_blocks(handle, inode, NULL, start, num, flags);
2275
2276        } else {
2277                printk(KERN_INFO "strange request: removal(2) "
2278                                "%u-%u from %u:%u\n",
2279                                from, to, le32_to_cpu(ex->ee_block), ee_len);
2280        }
2281        return 0;
2282}
2283
2284
2285/*
2286 * ext4_ext_rm_leaf() Removes the extents associated with the
2287 * blocks appearing between "start" and "end", and splits the extents
2288 * if "start" and "end" appear in the same extent
2289 *
2290 * @handle: The journal handle
2291 * @inode:  The files inode
2292 * @path:   The path to the leaf
2293 * @start:  The first block to remove
2294 * @end:   The last block to remove
2295 */
2296static int
2297ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2298                 struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster,
2299                 ext4_lblk_t start, ext4_lblk_t end)
2300{
2301        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2302        int err = 0, correct_index = 0;
2303        int depth = ext_depth(inode), credits;
2304        struct ext4_extent_header *eh;
2305        ext4_lblk_t a, b;
2306        unsigned num;
2307        ext4_lblk_t ex_ee_block;
2308        unsigned short ex_ee_len;
2309        unsigned uninitialized = 0;
2310        struct ext4_extent *ex;
2311
2312        /* the header must be checked already in ext4_ext_remove_space() */
2313        ext_debug("truncate since %u in leaf\n", start);
2314        if (!path[depth].p_hdr)
2315                path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2316        eh = path[depth].p_hdr;
2317        if (unlikely(path[depth].p_hdr == NULL)) {
2318                EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2319                return -EIO;
2320        }
2321        /* find where to start removing */
2322        ex = EXT_LAST_EXTENT(eh);
2323
2324        ex_ee_block = le32_to_cpu(ex->ee_block);
2325        ex_ee_len = ext4_ext_get_actual_len(ex);
2326
2327        trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster);
2328
2329        while (ex >= EXT_FIRST_EXTENT(eh) &&
2330                        ex_ee_block + ex_ee_len > start) {
2331
2332                if (ext4_ext_is_uninitialized(ex))
2333                        uninitialized = 1;
2334                else
2335                        uninitialized = 0;
2336
2337                ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2338                         uninitialized, ex_ee_len);
2339                path[depth].p_ext = ex;
2340
2341                a = ex_ee_block > start ? ex_ee_block : start;
2342                b = ex_ee_block+ex_ee_len - 1 < end ?
2343                        ex_ee_block+ex_ee_len - 1 : end;
2344
2345                ext_debug("  border %u:%u\n", a, b);
2346
2347                /* If this extent is beyond the end of the hole, skip it */
2348                if (end <= ex_ee_block) {
2349                        ex--;
2350                        ex_ee_block = le32_to_cpu(ex->ee_block);
2351                        ex_ee_len = ext4_ext_get_actual_len(ex);
2352                        continue;
2353                } else if (b != ex_ee_block + ex_ee_len - 1) {
2354                        EXT4_ERROR_INODE(inode,"  bad truncate %u:%u\n",
2355                                         start, end);
2356                        err = -EIO;
2357                        goto out;
2358                } else if (a != ex_ee_block) {
2359                        /* remove tail of the extent */
2360                        num = a - ex_ee_block;
2361                } else {
2362                        /* remove whole extent: excellent! */
2363                        num = 0;
2364                }
2365                /*
2366                 * 3 for leaf, sb, and inode plus 2 (bmap and group
2367                 * descriptor) for each block group; assume two block
2368                 * groups plus ex_ee_len/blocks_per_block_group for
2369                 * the worst case
2370                 */
2371                credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2372                if (ex == EXT_FIRST_EXTENT(eh)) {
2373                        correct_index = 1;
2374                        credits += (ext_depth(inode)) + 1;
2375                }
2376                credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2377
2378                err = ext4_ext_truncate_extend_restart(handle, inode, credits);
2379                if (err)
2380                        goto out;
2381
2382                err = ext4_ext_get_access(handle, inode, path + depth);
2383                if (err)
2384                        goto out;
2385
2386                err = ext4_remove_blocks(handle, inode, ex, partial_cluster,
2387                                         a, b);
2388                if (err)
2389                        goto out;
2390
2391                if (num == 0)
2392                        /* this extent is removed; mark slot entirely unused */
2393                        ext4_ext_store_pblock(ex, 0);
2394
2395                ex->ee_len = cpu_to_le16(num);
2396                /*
2397                 * Do not mark uninitialized if all the blocks in the
2398                 * extent have been removed.
2399                 */
2400                if (uninitialized && num)
2401                        ext4_ext_mark_uninitialized(ex);
2402                /*
2403                 * If the extent was completely released,
2404                 * we need to remove it from the leaf
2405                 */
2406                if (num == 0) {
2407                        if (end != EXT_MAX_BLOCKS - 1) {
2408                                /*
2409                                 * For hole punching, we need to scoot all the
2410                                 * extents up when an extent is removed so that
2411                                 * we dont have blank extents in the middle
2412                                 */
2413                                memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2414                                        sizeof(struct ext4_extent));
2415
2416                                /* Now get rid of the one at the end */
2417                                memset(EXT_LAST_EXTENT(eh), 0,
2418                                        sizeof(struct ext4_extent));
2419                        }
2420                        le16_add_cpu(&eh->eh_entries, -1);
2421                } else
2422                        *partial_cluster = 0;
2423
2424                err = ext4_ext_dirty(handle, inode, path + depth);
2425                if (err)
2426                        goto out;
2427
2428                ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2429                                ext4_ext_pblock(ex));
2430                ex--;
2431                ex_ee_block = le32_to_cpu(ex->ee_block);
2432                ex_ee_len = ext4_ext_get_actual_len(ex);
2433        }
2434
2435        if (correct_index && eh->eh_entries)
2436                err = ext4_ext_correct_indexes(handle, inode, path);
2437
2438        /*
2439         * If there is still a entry in the leaf node, check to see if
2440         * it references the partial cluster.  This is the only place
2441         * where it could; if it doesn't, we can free the cluster.
2442         */
2443        if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) &&
2444            (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) !=
2445             *partial_cluster)) {
2446                int flags = EXT4_FREE_BLOCKS_FORGET;
2447
2448                if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2449                        flags |= EXT4_FREE_BLOCKS_METADATA;
2450
2451                ext4_free_blocks(handle, inode, NULL,
2452                                 EXT4_C2B(sbi, *partial_cluster),
2453                                 sbi->s_cluster_ratio, flags);
2454                *partial_cluster = 0;
2455        }
2456
2457        /* if this leaf is free, then we should
2458         * remove it from index block above */
2459        if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2460                err = ext4_ext_rm_idx(handle, inode, path + depth);
2461
2462out:
2463        return err;
2464}
2465
2466/*
2467 * ext4_ext_more_to_rm:
2468 * returns 1 if current index has to be freed (even partial)
2469 */
2470static int
2471ext4_ext_more_to_rm(struct ext4_ext_path *path)
2472{
2473        BUG_ON(path->p_idx == NULL);
2474
2475        if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2476                return 0;
2477
2478        /*
2479         * if truncate on deeper level happened, it wasn't partial,
2480         * so we have to consider current index for truncation
2481         */
2482        if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2483                return 0;
2484        return 1;
2485}
2486
2487static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
2488{
2489        struct super_block *sb = inode->i_sb;
2490        int depth = ext_depth(inode);
2491        struct ext4_ext_path *path;
2492        ext4_fsblk_t partial_cluster = 0;
2493        handle_t *handle;
2494        int i, err;
2495
2496        ext_debug("truncate since %u\n", start);
2497
2498        /* probably first extent we're gonna free will be last in block */
2499        handle = ext4_journal_start(inode, depth + 1);
2500        if (IS_ERR(handle))
2501                return PTR_ERR(handle);
2502
2503again:
2504        ext4_ext_invalidate_cache(inode);
2505
2506        trace_ext4_ext_remove_space(inode, start, depth);
2507
2508        /*
2509         * We start scanning from right side, freeing all the blocks
2510         * after i_size and walking into the tree depth-wise.
2511         */
2512        depth = ext_depth(inode);
2513        path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
2514        if (path == NULL) {
2515                ext4_journal_stop(handle);
2516                return -ENOMEM;
2517        }
2518        path[0].p_depth = depth;
2519        path[0].p_hdr = ext_inode_hdr(inode);
2520        if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
2521                err = -EIO;
2522                goto out;
2523        }
2524        i = err = 0;
2525
2526        while (i >= 0 && err == 0) {
2527                if (i == depth) {
2528                        /* this is leaf block */
2529                        err = ext4_ext_rm_leaf(handle, inode, path,
2530                                               &partial_cluster, start,
2531                                               EXT_MAX_BLOCKS - 1);
2532                        /* root level has p_bh == NULL, brelse() eats this */
2533                        brelse(path[i].p_bh);
2534                        path[i].p_bh = NULL;
2535                        i--;
2536                        continue;
2537                }
2538
2539                /* this is index block */
2540                if (!path[i].p_hdr) {
2541                        ext_debug("initialize header\n");
2542                        path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2543                }
2544
2545                if (!path[i].p_idx) {
2546                        /* this level hasn't been touched yet */
2547                        path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2548                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2549                        ext_debug("init index ptr: hdr 0x%p, num %d\n",
2550                                  path[i].p_hdr,
2551                                  le16_to_cpu(path[i].p_hdr->eh_entries));
2552                } else {
2553                        /* we were already here, see at next index */
2554                        path[i].p_idx--;
2555                }
2556
2557                ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2558                                i, EXT_FIRST_INDEX(path[i].p_hdr),
2559                                path[i].p_idx);
2560                if (ext4_ext_more_to_rm(path + i)) {
2561                        struct buffer_head *bh;
2562                        /* go to the next level */
2563                        ext_debug("move to level %d (block %llu)\n",
2564                                  i + 1, ext4_idx_pblock(path[i].p_idx));
2565                        memset(path + i + 1, 0, sizeof(*path));
2566                        bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx));
2567                        if (!bh) {
2568                                /* should we reset i_size? */
2569                                err = -EIO;
2570                                break;
2571                        }
2572                        if (WARN_ON(i + 1 > depth)) {
2573                                err = -EIO;
2574                                break;
2575                        }
2576                        if (ext4_ext_check(inode, ext_block_hdr(bh),
2577                                                        depth - i - 1)) {
2578                                err = -EIO;
2579                                break;
2580                        }
2581                        path[i + 1].p_bh = bh;
2582
2583                        /* save actual number of indexes since this
2584                         * number is changed at the next iteration */
2585                        path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2586                        i++;
2587                } else {
2588                        /* we finished processing this index, go up */
2589                        if (path[i].p_hdr->eh_entries == 0 && i > 0) {
2590                                /* index is empty, remove it;
2591                                 * handle must be already prepared by the
2592                                 * truncatei_leaf() */
2593                                err = ext4_ext_rm_idx(handle, inode, path + i);
2594                        }
2595                        /* root level has p_bh == NULL, brelse() eats this */
2596                        brelse(path[i].p_bh);
2597                        path[i].p_bh = NULL;
2598                        i--;
2599                        ext_debug("return to level %d\n", i);
2600                }
2601        }
2602
2603        trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster,
2604                        path->p_hdr->eh_entries);
2605
2606        /* If we still have something in the partial cluster and we have removed
2607         * even the first extent, then we should free the blocks in the partial
2608         * cluster as well. */
2609        if (partial_cluster && path->p_hdr->eh_entries == 0) {
2610                int flags = EXT4_FREE_BLOCKS_FORGET;
2611
2612                if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2613                        flags |= EXT4_FREE_BLOCKS_METADATA;
2614
2615                ext4_free_blocks(handle, inode, NULL,
2616                                 EXT4_C2B(EXT4_SB(sb), partial_cluster),
2617                                 EXT4_SB(sb)->s_cluster_ratio, flags);
2618                partial_cluster = 0;
2619        }
2620
2621        /* TODO: flexible tree reduction should be here */
2622        if (path->p_hdr->eh_entries == 0) {
2623                /*
2624                 * truncate to zero freed all the tree,
2625                 * so we need to correct eh_depth
2626                 */
2627                err = ext4_ext_get_access(handle, inode, path);
2628                if (err == 0) {
2629                        ext_inode_hdr(inode)->eh_depth = 0;
2630                        ext_inode_hdr(inode)->eh_max =
2631                                cpu_to_le16(ext4_ext_space_root(inode, 0));
2632                        err = ext4_ext_dirty(handle, inode, path);
2633                }
2634        }
2635out:
2636        ext4_ext_drop_refs(path);
2637        kfree(path);
2638        if (err == -EAGAIN)
2639                goto again;
2640        ext4_journal_stop(handle);
2641
2642        return err;
2643}
2644
2645/*
2646 * called at mount time
2647 */
2648void ext4_ext_init(struct super_block *sb)
2649{
2650        /*
2651         * possible initialization would be here
2652         */
2653
2654        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2655#if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2656                printk(KERN_INFO "EXT4-fs: file extents enabled");
2657#ifdef AGGRESSIVE_TEST
2658                printk(", aggressive tests");
2659#endif
2660#ifdef CHECK_BINSEARCH
2661                printk(", check binsearch");
2662#endif
2663#ifdef EXTENTS_STATS
2664                printk(", stats");
2665#endif
2666                printk("\n");
2667#endif
2668#ifdef EXTENTS_STATS
2669                spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2670                EXT4_SB(sb)->s_ext_min = 1 << 30;
2671                EXT4_SB(sb)->s_ext_max = 0;
2672#endif
2673        }
2674}
2675
2676/*
2677 * called at umount time
2678 */
2679void ext4_ext_release(struct super_block *sb)
2680{
2681        if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2682                return;
2683
2684#ifdef EXTENTS_STATS
2685        if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2686                struct ext4_sb_info *sbi = EXT4_SB(sb);
2687                printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2688                        sbi->s_ext_blocks, sbi->s_ext_extents,
2689                        sbi->s_ext_blocks / sbi->s_ext_extents);
2690                printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2691                        sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2692        }
2693#endif
2694}
2695
2696/* FIXME!! we need to try to merge to left or right after zero-out  */
2697static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2698{
2699        ext4_fsblk_t ee_pblock;
2700        unsigned int ee_len;
2701        int ret;
2702
2703        ee_len    = ext4_ext_get_actual_len(ex);
2704        ee_pblock = ext4_ext_pblock(ex);
2705
2706        ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS);
2707        if (ret > 0)
2708                ret = 0;
2709
2710        return ret;
2711}
2712
2713/*
2714 * used by extent splitting.
2715 */
2716#define EXT4_EXT_MAY_ZEROOUT    0x1  /* safe to zeroout if split fails \
2717                                        due to ENOSPC */
2718#define EXT4_EXT_MARK_UNINIT1   0x2  /* mark first half uninitialized */
2719#define EXT4_EXT_MARK_UNINIT2   0x4  /* mark second half uninitialized */
2720
2721/*
2722 * ext4_split_extent_at() splits an extent at given block.
2723 *
2724 * @handle: the journal handle
2725 * @inode: the file inode
2726 * @path: the path to the extent
2727 * @split: the logical block where the extent is splitted.
2728 * @split_flags: indicates if the extent could be zeroout if split fails, and
2729 *               the states(init or uninit) of new extents.
2730 * @flags: flags used to insert new extent to extent tree.
2731 *
2732 *
2733 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2734 * of which are deterimined by split_flag.
2735 *
2736 * There are two cases:
2737 *  a> the extent are splitted into two extent.
2738 *  b> split is not needed, and just mark the extent.
2739 *
2740 * return 0 on success.
2741 */
2742static int ext4_split_extent_at(handle_t *handle,
2743                             struct inode *inode,
2744                             struct ext4_ext_path *path,
2745                             ext4_lblk_t split,
2746                             int split_flag,
2747                             int flags)
2748{
2749        ext4_fsblk_t newblock;
2750        ext4_lblk_t ee_block;
2751        struct ext4_extent *ex, newex, orig_ex;
2752        struct ext4_extent *ex2 = NULL;
2753        unsigned int ee_len, depth;
2754        int err = 0;
2755
2756        ext_debug("ext4_split_extents_at: inode %lu, logical"
2757                "block %llu\n", inode->i_ino, (unsigned long long)split);
2758
2759        ext4_ext_show_leaf(inode, path);
2760
2761        depth = ext_depth(inode);
2762        ex = path[depth].p_ext;
2763        ee_block = le32_to_cpu(ex->ee_block);
2764        ee_len = ext4_ext_get_actual_len(ex);
2765        newblock = split - ee_block + ext4_ext_pblock(ex);
2766
2767        BUG_ON(split < ee_block || split >= (ee_block + ee_len));
2768
2769        err = ext4_ext_get_access(handle, inode, path + depth);
2770        if (err)
2771                goto out;
2772
2773        if (split == ee_block) {
2774                /*
2775                 * case b: block @split is the block that the extent begins with
2776                 * then we just change the state of the extent, and splitting
2777                 * is not needed.
2778                 */
2779                if (split_flag & EXT4_EXT_MARK_UNINIT2)
2780                        ext4_ext_mark_uninitialized(ex);
2781                else
2782                        ext4_ext_mark_initialized(ex);
2783
2784                if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
2785                        ext4_ext_try_to_merge(inode, path, ex);
2786
2787                err = ext4_ext_dirty(handle, inode, path + depth);
2788                goto out;
2789        }
2790
2791        /* case a */
2792        memcpy(&orig_ex, ex, sizeof(orig_ex));
2793        ex->ee_len = cpu_to_le16(split - ee_block);
2794        if (split_flag & EXT4_EXT_MARK_UNINIT1)
2795                ext4_ext_mark_uninitialized(ex);
2796
2797        /*
2798         * path may lead to new leaf, not to original leaf any more
2799         * after ext4_ext_insert_extent() returns,
2800         */
2801        err = ext4_ext_dirty(handle, inode, path + depth);
2802        if (err)
2803                goto fix_extent_len;
2804
2805        ex2 = &newex;
2806        ex2->ee_block = cpu_to_le32(split);
2807        ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
2808        ext4_ext_store_pblock(ex2, newblock);
2809        if (split_flag & EXT4_EXT_MARK_UNINIT2)
2810                ext4_ext_mark_uninitialized(ex2);
2811
2812        err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
2813        if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
2814                err = ext4_ext_zeroout(inode, &orig_ex);
2815                if (err)
2816                        goto fix_extent_len;
2817                /* update the extent length and mark as initialized */
2818                ex->ee_len = cpu_to_le32(ee_len);
2819                ext4_ext_try_to_merge(inode, path, ex);
2820                err = ext4_ext_dirty(handle, inode, path + depth);
2821                goto out;
2822        } else if (err)
2823                goto fix_extent_len;
2824
2825out:
2826        ext4_ext_show_leaf(inode, path);
2827        return err;
2828
2829fix_extent_len:
2830        ex->ee_len = orig_ex.ee_len;
2831        ext4_ext_dirty(handle, inode, path + depth);
2832        return err;
2833}
2834
2835/*
2836 * ext4_split_extents() splits an extent and mark extent which is covered
2837 * by @map as split_flags indicates
2838 *
2839 * It may result in splitting the extent into multiple extents (upto three)
2840 * There are three possibilities:
2841 *   a> There is no split required
2842 *   b> Splits in two extents: Split is happening at either end of the extent
2843 *   c> Splits in three extents: Somone is splitting in middle of the extent
2844 *
2845 */
2846static int ext4_split_extent(handle_t *handle,
2847                              struct inode *inode,
2848                              struct ext4_ext_path *path,
2849                              struct ext4_map_blocks *map,
2850                              int split_flag,
2851                              int flags)
2852{
2853        ext4_lblk_t ee_block;
2854        struct ext4_extent *ex;
2855        unsigned int ee_len, depth;
2856        int err = 0;
2857        int uninitialized;
2858        int split_flag1, flags1;
2859
2860        depth = ext_depth(inode);
2861        ex = path[depth].p_ext;
2862        ee_block = le32_to_cpu(ex->ee_block);
2863        ee_len = ext4_ext_get_actual_len(ex);
2864        uninitialized = ext4_ext_is_uninitialized(ex);
2865
2866        if (map->m_lblk + map->m_len < ee_block + ee_len) {
2867                split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2868                              EXT4_EXT_MAY_ZEROOUT : 0;
2869                flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
2870                if (uninitialized)
2871                        split_flag1 |= EXT4_EXT_MARK_UNINIT1 |
2872                                       EXT4_EXT_MARK_UNINIT2;
2873                err = ext4_split_extent_at(handle, inode, path,
2874                                map->m_lblk + map->m_len, split_flag1, flags1);
2875                if (err)
2876                        goto out;
2877        }
2878
2879        ext4_ext_drop_refs(path);
2880        path = ext4_ext_find_extent(inode, map->m_lblk, path);
2881        if (IS_ERR(path))
2882                return PTR_ERR(path);
2883
2884        if (map->m_lblk >= ee_block) {
2885                split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT ?
2886                              EXT4_EXT_MAY_ZEROOUT : 0;
2887                if (uninitialized)
2888                        split_flag1 |= EXT4_EXT_MARK_UNINIT1;
2889                if (split_flag & EXT4_EXT_MARK_UNINIT2)
2890                        split_flag1 |= EXT4_EXT_MARK_UNINIT2;
2891                err = ext4_split_extent_at(handle, inode, path,
2892                                map->m_lblk, split_flag1, flags);
2893                if (err)
2894                        goto out;
2895        }
2896
2897        ext4_ext_show_leaf(inode, path);
2898out:
2899        return err ? err : map->m_len;
2900}
2901
2902#define EXT4_EXT_ZERO_LEN 7
2903/*
2904 * This function is called by ext4_ext_map_blocks() if someone tries to write
2905 * to an uninitialized extent. It may result in splitting the uninitialized
2906 * extent into multiple extents (up to three - one initialized and two
2907 * uninitialized).
2908 * There are three possibilities:
2909 *   a> There is no split required: Entire extent should be initialized
2910 *   b> Splits in two extents: Write is happening at either end of the extent
2911 *   c> Splits in three extents: Somone is writing in middle of the extent
2912 *
2913 * Pre-conditions:
2914 *  - The extent pointed to by 'path' is uninitialized.
2915 *  - The extent pointed to by 'path' contains a superset
2916 *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
2917 *
2918 * Post-conditions on success:
2919 *  - the returned value is the number of blocks beyond map->l_lblk
2920 *    that are allocated and initialized.
2921 *    It is guaranteed to be >= map->m_len.
2922 */
2923static int ext4_ext_convert_to_initialized(handle_t *handle,
2924                                           struct inode *inode,
2925                                           struct ext4_map_blocks *map,
2926                                           struct ext4_ext_path *path)
2927{
2928        struct ext4_extent_header *eh;
2929        struct ext4_map_blocks split_map;
2930        struct ext4_extent zero_ex;
2931        struct ext4_extent *ex;
2932        ext4_lblk_t ee_block, eof_block;
2933        unsigned int ee_len, depth;
2934        int allocated;
2935        int err = 0;
2936        int split_flag = 0;
2937
2938        ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2939                "block %llu, max_blocks %u\n", inode->i_ino,
2940                (unsigned long long)map->m_lblk, map->m_len);
2941
2942        eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
2943                inode->i_sb->s_blocksize_bits;
2944        if (eof_block < map->m_lblk + map->m_len)
2945                eof_block = map->m_lblk + map->m_len;
2946
2947        depth = ext_depth(inode);
2948        eh = path[depth].p_hdr;
2949        ex = path[depth].p_ext;
2950        ee_block = le32_to_cpu(ex->ee_block);
2951        ee_len = ext4_ext_get_actual_len(ex);
2952        allocated = ee_len - (map->m_lblk - ee_block);
2953
2954        trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
2955
2956        /* Pre-conditions */
2957        BUG_ON(!ext4_ext_is_uninitialized(ex));
2958        BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
2959
2960        /*
2961         * Attempt to transfer newly initialized blocks from the currently
2962         * uninitialized extent to its left neighbor. This is much cheaper
2963         * than an insertion followed by a merge as those involve costly
2964         * memmove() calls. This is the common case in steady state for
2965         * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
2966         * writes.
2967         *
2968         * Limitations of the current logic:
2969         *  - L1: we only deal with writes at the start of the extent.
2970         *    The approach could be extended to writes at the end
2971         *    of the extent but this scenario was deemed less common.
2972         *  - L2: we do not deal with writes covering the whole extent.
2973         *    This would require removing the extent if the transfer
2974         *    is possible.
2975         *  - L3: we only attempt to merge with an extent stored in the
2976         *    same extent tree node.
2977         */
2978        if ((map->m_lblk == ee_block) &&        /*L1*/
2979                (map->m_len < ee_len) &&        /*L2*/
2980                (ex > EXT_FIRST_EXTENT(eh))) {  /*L3*/
2981                struct ext4_extent *prev_ex;
2982                ext4_lblk_t prev_lblk;
2983                ext4_fsblk_t prev_pblk, ee_pblk;
2984                unsigned int prev_len, write_len;
2985
2986                prev_ex = ex - 1;
2987                prev_lblk = le32_to_cpu(prev_ex->ee_block);
2988                prev_len = ext4_ext_get_actual_len(prev_ex);
2989                prev_pblk = ext4_ext_pblock(prev_ex);
2990                ee_pblk = ext4_ext_pblock(ex);
2991                write_len = map->m_len;
2992
2993                /*
2994                 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
2995                 * upon those conditions:
2996                 * - C1: prev_ex is initialized,
2997                 * - C2: prev_ex is logically abutting ex,
2998                 * - C3: prev_ex is physically abutting ex,
2999                 * - C4: prev_ex can receive the additional blocks without
3000                 *   overflowing the (initialized) length limit.
3001                 */
3002                if ((!ext4_ext_is_uninitialized(prev_ex)) &&            /*C1*/
3003                        ((prev_lblk + prev_len) == ee_block) &&         /*C2*/
3004                        ((prev_pblk + prev_len) == ee_pblk) &&          /*C3*/
3005                        (prev_len < (EXT_INIT_MAX_LEN - write_len))) {  /*C4*/
3006                        err = ext4_ext_get_access(handle, inode, path + depth);
3007                        if (err)
3008                                goto out;
3009
3010                        trace_ext4_ext_convert_to_initialized_fastpath(inode,
3011                                map, ex, prev_ex);
3012
3013                        /* Shift the start of ex by 'write_len' blocks */
3014                        ex->ee_block = cpu_to_le32(ee_block + write_len);
3015                        ext4_ext_store_pblock(ex, ee_pblk + write_len);
3016                        ex->ee_len = cpu_to_le16(ee_len - write_len);
3017                        ext4_ext_mark_uninitialized(ex); /* Restore the flag */
3018
3019                        /* Extend prev_ex by 'write_len' blocks */
3020                        prev_ex->ee_len = cpu_to_le16(prev_len + write_len);
3021
3022                        /* Mark the block containing both extents as dirty */
3023                        ext4_ext_dirty(handle, inode, path + depth);
3024
3025                        /* Update path to point to the right extent */
3026                        path[depth].p_ext = prev_ex;
3027
3028                        /* Result: number of initialized blocks past m_lblk */
3029                        allocated = write_len;
3030                        goto out;
3031                }
3032        }
3033
3034        WARN_ON(map->m_lblk < ee_block);
3035        /*
3036         * It is safe to convert extent to initialized via explicit
3037         * zeroout only if extent is fully insde i_size or new_size.
3038         */
3039        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3040
3041        /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3042        if (ee_len <= 2*EXT4_EXT_ZERO_LEN &&
3043            (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3044                err = ext4_ext_zeroout(inode, ex);
3045                if (err)
3046                        goto out;
3047
3048                err = ext4_ext_get_access(handle, inode, path + depth);
3049                if (err)
3050                        goto out;
3051                ext4_ext_mark_initialized(ex);
3052                ext4_ext_try_to_merge(inode, path, ex);
3053                err = ext4_ext_dirty(handle, inode, path + depth);
3054                goto out;
3055        }
3056
3057        /*
3058         * four cases:
3059         * 1. split the extent into three extents.
3060         * 2. split the extent into two extents, zeroout the first half.
3061         * 3. split the extent into two extents, zeroout the second half.
3062         * 4. split the extent into two extents with out zeroout.
3063         */
3064        split_map.m_lblk = map->m_lblk;
3065        split_map.m_len = map->m_len;
3066
3067        if (allocated > map->m_len) {
3068                if (allocated <= EXT4_EXT_ZERO_LEN &&
3069                    (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3070                        /* case 3 */
3071                        zero_ex.ee_block =
3072                                         cpu_to_le32(map->m_lblk);
3073                        zero_ex.ee_len = cpu_to_le16(allocated);
3074                        ext4_ext_store_pblock(&zero_ex,
3075                                ext4_ext_pblock(ex) + map->m_lblk - ee_block);
3076                        err = ext4_ext_zeroout(inode, &zero_ex);
3077                        if (err)
3078                                goto out;
3079                        split_map.m_lblk = map->m_lblk;
3080                        split_map.m_len = allocated;
3081                } else if ((map->m_lblk - ee_block + map->m_len <
3082                           EXT4_EXT_ZERO_LEN) &&
3083                           (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3084                        /* case 2 */
3085                        if (map->m_lblk != ee_block) {
3086                                zero_ex.ee_block = ex->ee_block;
3087                                zero_ex.ee_len = cpu_to_le16(map->m_lblk -
3088                                                        ee_block);
3089                                ext4_ext_store_pblock(&zero_ex,
3090                                                      ext4_ext_pblock(ex));
3091                                err = ext4_ext_zeroout(inode, &zero_ex);
3092                                if (err)
3093                                        goto out;
3094                        }
3095
3096                        split_map.m_lblk = ee_block;
3097                        split_map.m_len = map->m_lblk - ee_block + map->m_len;
3098                        allocated = map->m_len;
3099                }
3100        }
3101
3102        allocated = ext4_split_extent(handle, inode, path,
3103                                       &split_map, split_flag, 0);
3104        if (allocated < 0)
3105                err = allocated;
3106
3107out:
3108        return err ? err : allocated;
3109}
3110
3111/*
3112 * This function is called by ext4_ext_map_blocks() from
3113 * ext4_get_blocks_dio_write() when DIO to write
3114 * to an uninitialized extent.
3115 *
3116 * Writing to an uninitialized extent may result in splitting the uninitialized
3117 * extent into multiple /initialized uninitialized extents (up to three)
3118 * There are three possibilities:
3119 *   a> There is no split required: Entire extent should be uninitialized
3120 *   b> Splits in two extents: Write is happening at either end of the extent
3121 *   c> Splits in three extents: Somone is writing in middle of the extent
3122 *
3123 * One of more index blocks maybe needed if the extent tree grow after
3124 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3125 * complete, we need to split the uninitialized extent before DIO submit
3126 * the IO. The uninitialized extent called at this time will be split
3127 * into three uninitialized extent(at most). After IO complete, the part
3128 * being filled will be convert to initialized by the end_io callback function
3129 * via ext4_convert_unwritten_extents().
3130 *
3131 * Returns the size of uninitialized extent to be written on success.
3132 */
3133static int ext4_split_unwritten_extents(handle_t *handle,
3134                                        struct inode *inode,
3135                                        struct ext4_map_blocks *map,
3136                                        struct ext4_ext_path *path,
3137                                        int flags)
3138{
3139        ext4_lblk_t eof_block;
3140        ext4_lblk_t ee_block;
3141        struct ext4_extent *ex;
3142        unsigned int ee_len;
3143        int split_flag = 0, depth;
3144
3145        ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3146                "block %llu, max_blocks %u\n", inode->i_ino,
3147                (unsigned long long)map->m_lblk, map->m_len);
3148
3149        eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3150                inode->i_sb->s_blocksize_bits;
3151        if (eof_block < map->m_lblk + map->m_len)
3152                eof_block = map->m_lblk + map->m_len;
3153        /*
3154         * It is safe to convert extent to initialized via explicit
3155         * zeroout only if extent is fully insde i_size or new_size.
3156         */
3157        depth = ext_depth(inode);
3158        ex = path[depth].p_ext;
3159        ee_block = le32_to_cpu(ex->ee_block);
3160        ee_len = ext4_ext_get_actual_len(ex);
3161
3162        split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3163        split_flag |= EXT4_EXT_MARK_UNINIT2;
3164
3165        flags |= EXT4_GET_BLOCKS_PRE_IO;
3166        return ext4_split_extent(handle, inode, path, map, split_flag, flags);
3167}
3168
3169static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3170                                              struct inode *inode,
3171                                              struct ext4_ext_path *path)
3172{
3173        struct ext4_extent *ex;
3174        int depth;
3175        int err = 0;
3176
3177        depth = ext_depth(inode);
3178        ex = path[depth].p_ext;
3179
3180        ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3181                "block %llu, max_blocks %u\n", inode->i_ino,
3182                (unsigned long long)le32_to_cpu(ex->ee_block),
3183                ext4_ext_get_actual_len(ex));
3184
3185        err = ext4_ext_get_access(handle, inode, path + depth);
3186        if (err)
3187                goto out;
3188        /* first mark the extent as initialized */
3189        ext4_ext_mark_initialized(ex);
3190
3191        /* note: ext4_ext_correct_indexes() isn't needed here because
3192         * borders are not changed
3193         */
3194        ext4_ext_try_to_merge(inode, path, ex);
3195
3196        /* Mark modified extent as dirty */
3197        err = ext4_ext_dirty(handle, inode, path + depth);
3198out:
3199        ext4_ext_show_leaf(inode, path);
3200        return err;
3201}
3202
3203static void unmap_underlying_metadata_blocks(struct block_device *bdev,
3204                        sector_t block, int count)
3205{
3206        int i;
3207        for (i = 0; i < count; i++)
3208                unmap_underlying_metadata(bdev, block + i);
3209}
3210
3211/*
3212 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3213 */
3214static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3215                              ext4_lblk_t lblk,
3216                              struct ext4_ext_path *path,
3217                              unsigned int len)
3218{
3219        int i, depth;
3220        struct ext4_extent_header *eh;
3221        struct ext4_extent *last_ex;
3222
3223        if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3224                return 0;
3225
3226        depth = ext_depth(inode);
3227        eh = path[depth].p_hdr;
3228
3229        if (unlikely(!eh->eh_entries)) {
3230                EXT4_ERROR_INODE(inode, "eh->eh_entries == 0 and "
3231                                 "EOFBLOCKS_FL set");
3232                return -EIO;
3233        }
3234        last_ex = EXT_LAST_EXTENT(eh);
3235        /*
3236         * We should clear the EOFBLOCKS_FL flag if we are writing the
3237         * last block in the last extent in the file.  We test this by
3238         * first checking to see if the caller to
3239         * ext4_ext_get_blocks() was interested in the last block (or
3240         * a block beyond the last block) in the current extent.  If
3241         * this turns out to be false, we can bail out from this
3242         * function immediately.
3243         */
3244        if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3245            ext4_ext_get_actual_len(last_ex))
3246                return 0;
3247        /*
3248         * If the caller does appear to be planning to write at or
3249         * beyond the end of the current extent, we then test to see
3250         * if the current extent is the last extent in the file, by
3251         * checking to make sure it was reached via the rightmost node
3252         * at each level of the tree.
3253         */
3254        for (i = depth-1; i >= 0; i--)
3255                if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3256                        return 0;
3257        ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3258        return ext4_mark_inode_dirty(handle, inode);
3259}
3260
3261/**
3262 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3263 *
3264 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3265 * whether there are any buffers marked for delayed allocation. It returns '1'
3266 * on the first delalloc'ed buffer head found. If no buffer head in the given
3267 * range is marked for delalloc, it returns 0.
3268 * lblk_start should always be <= lblk_end.
3269 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3270 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3271 * block sooner). This is useful when blocks are truncated sequentially from
3272 * lblk_start towards lblk_end.
3273 */
3274static int ext4_find_delalloc_range(struct inode *inode,
3275                                    ext4_lblk_t lblk_start,
3276                                    ext4_lblk_t lblk_end,
3277                                    int search_hint_reverse)
3278{
3279        struct address_space *mapping = inode->i_mapping;
3280        struct buffer_head *head, *bh = NULL;
3281        struct page *page;
3282        ext4_lblk_t i, pg_lblk;
3283        pgoff_t index;
3284
3285        if (!test_opt(inode->i_sb, DELALLOC))
3286                return 0;
3287
3288        /* reverse search wont work if fs block size is less than page size */
3289        if (inode->i_blkbits < PAGE_CACHE_SHIFT)
3290                search_hint_reverse = 0;
3291
3292        if (search_hint_reverse)
3293                i = lblk_end;
3294        else
3295                i = lblk_start;
3296
3297        index = i >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
3298
3299        while ((i >= lblk_start) && (i <= lblk_end)) {
3300                page = find_get_page(mapping, index);
3301                if (!page)
3302                        goto nextpage;
3303
3304                if (!page_has_buffers(page))
3305                        goto nextpage;
3306
3307                head = page_buffers(page);
3308                if (!head)
3309                        goto nextpage;
3310
3311                bh = head;
3312                pg_lblk = index << (PAGE_CACHE_SHIFT -
3313                                                inode->i_blkbits);
3314                do {
3315                        if (unlikely(pg_lblk < lblk_start)) {
3316                                /*
3317                                 * This is possible when fs block size is less
3318                                 * than page size and our cluster starts/ends in
3319                                 * middle of the page. So we need to skip the
3320                                 * initial few blocks till we reach the 'lblk'
3321                                 */
3322                                pg_lblk++;
3323                                continue;
3324                        }
3325
3326                        /* Check if the buffer is delayed allocated and that it
3327                         * is not yet mapped. (when da-buffers are mapped during
3328                         * their writeout, their da_mapped bit is set.)
3329                         */
3330                        if (buffer_delay(bh) && !buffer_da_mapped(bh)) {
3331                                page_cache_release(page);
3332                                trace_ext4_find_delalloc_range(inode,
3333                                                lblk_start, lblk_end,
3334                                                search_hint_reverse,
3335                                                1, i);
3336                                return 1;
3337                        }
3338                        if (search_hint_reverse)
3339                                i--;
3340                        else
3341                                i++;
3342                } while ((i >= lblk_start) && (i <= lblk_end) &&
3343                                ((bh = bh->b_this_page) != head));
3344nextpage:
3345                if (page)
3346                        page_cache_release(page);
3347                /*
3348                 * Move to next page. 'i' will be the first lblk in the next
3349                 * page.
3350                 */
3351                if (search_hint_reverse)
3352                        index--;
3353                else
3354                        index++;
3355                i = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
3356        }
3357
3358        trace_ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3359                                        search_hint_reverse, 0, 0);
3360        return 0;
3361}
3362
3363int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk,
3364                               int search_hint_reverse)
3365{
3366        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3367        ext4_lblk_t lblk_start, lblk_end;
3368        lblk_start = lblk & (~(sbi->s_cluster_ratio - 1));
3369        lblk_end = lblk_start + sbi->s_cluster_ratio - 1;
3370
3371        return ext4_find_delalloc_range(inode, lblk_start, lblk_end,
3372                                        search_hint_reverse);
3373}
3374
3375/**
3376 * Determines how many complete clusters (out of those specified by the 'map')
3377 * are under delalloc and were reserved quota for.
3378 * This function is called when we are writing out the blocks that were
3379 * originally written with their allocation delayed, but then the space was
3380 * allocated using fallocate() before the delayed allocation could be resolved.
3381 * The cases to look for are:
3382 * ('=' indicated delayed allocated blocks
3383 *  '-' indicates non-delayed allocated blocks)
3384 * (a) partial clusters towards beginning and/or end outside of allocated range
3385 *     are not delalloc'ed.
3386 *      Ex:
3387 *      |----c---=|====c====|====c====|===-c----|
3388 *               |++++++ allocated ++++++|
3389 *      ==> 4 complete clusters in above example
3390 *
3391 * (b) partial cluster (outside of allocated range) towards either end is
3392 *     marked for delayed allocation. In this case, we will exclude that
3393 *     cluster.
3394 *      Ex:
3395 *      |----====c========|========c========|
3396 *           |++++++ allocated ++++++|
3397 *      ==> 1 complete clusters in above example
3398 *
3399 *      Ex:
3400 *      |================c================|
3401 *            |++++++ allocated ++++++|
3402 *      ==> 0 complete clusters in above example
3403 *
3404 * The ext4_da_update_reserve_space will be called only if we
3405 * determine here that there were some "entire" clusters that span
3406 * this 'allocated' range.
3407 * In the non-bigalloc case, this function will just end up returning num_blks
3408 * without ever calling ext4_find_delalloc_range.
3409 */
3410static unsigned int
3411get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start,
3412                           unsigned int num_blks)
3413{
3414        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3415        ext4_lblk_t alloc_cluster_start, alloc_cluster_end;
3416        ext4_lblk_t lblk_from, lblk_to, c_offset;
3417        unsigned int allocated_clusters = 0;
3418
3419        alloc_cluster_start = EXT4_B2C(sbi, lblk_start);
3420        alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1);
3421
3422        /* max possible clusters for this allocation */
3423        allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1;
3424
3425        trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks);
3426
3427        /* Check towards left side */
3428        c_offset = lblk_start & (sbi->s_cluster_ratio - 1);
3429        if (c_offset) {
3430                lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1));
3431                lblk_to = lblk_from + c_offset - 1;
3432
3433                if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3434                        allocated_clusters--;
3435        }
3436
3437        /* Now check towards right. */
3438        c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1);
3439        if (allocated_clusters && c_offset) {
3440                lblk_from = lblk_start + num_blks;
3441                lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1;
3442
3443                if (ext4_find_delalloc_range(inode, lblk_from, lblk_to, 0))
3444                        allocated_clusters--;
3445        }
3446
3447        return allocated_clusters;
3448}
3449
3450static int
3451ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
3452                        struct ext4_map_blocks *map,
3453                        struct ext4_ext_path *path, int flags,
3454                        unsigned int allocated, ext4_fsblk_t newblock)
3455{
3456        int ret = 0;
3457        int err = 0;
3458        ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3459
3460        ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3461                  "block %llu, max_blocks %u, flags %x, allocated %u\n",
3462                  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
3463                  flags, allocated);
3464        ext4_ext_show_leaf(inode, path);
3465
3466        trace_ext4_ext_handle_uninitialized_extents(inode, map, allocated,
3467                                                    newblock);
3468
3469        /* get_block() before submit the IO, split the extent */
3470        if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
3471                ret = ext4_split_unwritten_extents(handle, inode, map,
3472                                                   path, flags);
3473                /*
3474                 * Flag the inode(non aio case) or end_io struct (aio case)
3475                 * that this IO needs to conversion to written when IO is
3476                 * completed
3477                 */
3478                if (io)
3479                        ext4_set_io_unwritten_flag(inode, io);
3480                else
3481                        ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3482                if (ext4_should_dioread_nolock(inode))
3483                        map->m_flags |= EXT4_MAP_UNINIT;
3484                goto out;
3485        }
3486        /* IO end_io complete, convert the filled extent to written */
3487        if ((flags & EXT4_GET_BLOCKS_CONVERT)) {
3488                ret = ext4_convert_unwritten_extents_endio(handle, inode,
3489                                                        path);
3490                if (ret >= 0) {
3491                        ext4_update_inode_fsync_trans(handle, inode, 1);
3492                        err = check_eofblocks_fl(handle, inode, map->m_lblk,
3493                                                 path, map->m_len);
3494                } else
3495                        err = ret;
3496                goto out2;
3497        }
3498        /* buffered IO case */
3499        /*
3500         * repeat fallocate creation request
3501         * we already have an unwritten extent
3502         */
3503        if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
3504                goto map_out;
3505
3506        /* buffered READ or buffered write_begin() lookup */
3507        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3508                /*
3509                 * We have blocks reserved already.  We
3510                 * return allocated blocks so that delalloc
3511                 * won't do block reservation for us.  But
3512                 * the buffer head will be unmapped so that
3513                 * a read from the block returns 0s.
3514                 */
3515                map->m_flags |= EXT4_MAP_UNWRITTEN;
3516                goto out1;
3517        }
3518
3519        /* buffered write, writepage time, convert*/
3520        ret = ext4_ext_convert_to_initialized(handle, inode, map, path);
3521        if (ret >= 0)
3522                ext4_update_inode_fsync_trans(handle, inode, 1);
3523out:
3524        if (ret <= 0) {
3525                err = ret;
3526                goto out2;
3527        } else
3528                allocated = ret;
3529        map->m_flags |= EXT4_MAP_NEW;
3530        /*
3531         * if we allocated more blocks than requested
3532         * we need to make sure we unmap the extra block
3533         * allocated. The actual needed block will get
3534         * unmapped later when we find the buffer_head marked
3535         * new.
3536         */
3537        if (allocated > map->m_len) {
3538                unmap_underlying_metadata_blocks(inode->i_sb->s_bdev,
3539                                        newblock + map->m_len,
3540                                        allocated - map->m_len);
3541                allocated = map->m_len;
3542        }
3543
3544        /*
3545         * If we have done fallocate with the offset that is already
3546         * delayed allocated, we would have block reservation
3547         * and quota reservation done in the delayed write path.
3548         * But fallocate would have already updated quota and block
3549         * count for this offset. So cancel these reservation
3550         */
3551        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
3552                unsigned int reserved_clusters;
3553                reserved_clusters = get_reserved_cluster_alloc(inode,
3554                                map->m_lblk, map->m_len);
3555                if (reserved_clusters)
3556                        ext4_da_update_reserve_space(inode,
3557                                                     reserved_clusters,
3558                                                     0);
3559        }
3560
3561map_out:
3562        map->m_flags |= EXT4_MAP_MAPPED;
3563        if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
3564                err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
3565                                         map->m_len);
3566                if (err < 0)
3567                        goto out2;
3568        }
3569out1:
3570        if (allocated > map->m_len)
3571                allocated = map->m_len;
3572        ext4_ext_show_leaf(inode, path);
3573        map->m_pblk = newblock;
3574        map->m_len = allocated;
3575out2:
3576        if (path) {
3577                ext4_ext_drop_refs(path);
3578                kfree(path);
3579        }
3580        return err ? err : allocated;
3581}
3582
3583/*
3584 * get_implied_cluster_alloc - check to see if the requested
3585 * allocation (in the map structure) overlaps with a cluster already
3586 * allocated in an extent.
3587 *      @sb     The filesystem superblock structure
3588 *      @map    The requested lblk->pblk mapping
3589 *      @ex     The extent structure which might contain an implied
3590 *                      cluster allocation
3591 *
3592 * This function is called by ext4_ext_map_blocks() after we failed to
3593 * find blocks that were already in the inode's extent tree.  Hence,
3594 * we know that the beginning of the requested region cannot overlap
3595 * the extent from the inode's extent tree.  There are three cases we
3596 * want to catch.  The first is this case:
3597 *
3598 *               |--- cluster # N--|
3599 *    |--- extent ---|  |---- requested region ---|
3600 *                      |==========|
3601 *
3602 * The second case that we need to test for is this one:
3603 *
3604 *   |--------- cluster # N ----------------|
3605 *         |--- requested region --|   |------- extent ----|
3606 *         |=======================|
3607 *
3608 * The third case is when the requested region lies between two extents
3609 * within the same cluster:
3610 *          |------------- cluster # N-------------|
3611 * |----- ex -----|                  |---- ex_right ----|
3612 *                  |------ requested region ------|
3613 *                  |================|
3614 *
3615 * In each of the above cases, we need to set the map->m_pblk and
3616 * map->m_len so it corresponds to the return the extent labelled as
3617 * "|====|" from cluster #N, since it is already in use for data in
3618 * cluster EXT4_B2C(sbi, map->m_lblk).  We will then return 1 to
3619 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3620 * as a new "allocated" block region.  Otherwise, we will return 0 and
3621 * ext4_ext_map_blocks() will then allocate one or more new clusters
3622 * by calling ext4_mb_new_blocks().
3623 */
3624static int get_implied_cluster_alloc(struct super_block *sb,
3625                                     struct ext4_map_blocks *map,
3626                                     struct ext4_extent *ex,
3627                                     struct ext4_ext_path *path)
3628{
3629        struct ext4_sb_info *sbi = EXT4_SB(sb);
3630        ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3631        ext4_lblk_t ex_cluster_start, ex_cluster_end;
3632        ext4_lblk_t rr_cluster_start;
3633        ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3634        ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3635        unsigned short ee_len = ext4_ext_get_actual_len(ex);
3636
3637        /* The extent passed in that we are trying to match */
3638        ex_cluster_start = EXT4_B2C(sbi, ee_block);
3639        ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
3640
3641        /* The requested region passed into ext4_map_blocks() */
3642        rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
3643
3644        if ((rr_cluster_start == ex_cluster_end) ||
3645            (rr_cluster_start == ex_cluster_start)) {
3646                if (rr_cluster_start == ex_cluster_end)
3647                        ee_start += ee_len - 1;
3648                map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) +
3649                        c_offset;
3650                map->m_len = min(map->m_len,
3651                                 (unsigned) sbi->s_cluster_ratio - c_offset);
3652                /*
3653                 * Check for and handle this case:
3654                 *
3655                 *   |--------- cluster # N-------------|
3656                 *                     |------- extent ----|
3657                 *         |--- requested region ---|
3658                 *         |===========|
3659                 */
3660
3661                if (map->m_lblk < ee_block)
3662                        map->m_len = min(map->m_len, ee_block - map->m_lblk);
3663
3664                /*
3665                 * Check for the case where there is already another allocated
3666                 * block to the right of 'ex' but before the end of the cluster.
3667                 *
3668                 *          |------------- cluster # N-------------|
3669                 * |----- ex -----|                  |---- ex_right ----|
3670                 *                  |------ requested region ------|
3671                 *                  |================|
3672                 */
3673                if (map->m_lblk > ee_block) {
3674                        ext4_lblk_t next = ext4_ext_next_allocated_block(path);
3675                        map->m_len = min(map->m_len, next - map->m_lblk);
3676                }
3677
3678                trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
3679                return 1;
3680        }
3681
3682        trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
3683        return 0;
3684}
3685
3686
3687/*
3688 * Block allocation/map/preallocation routine for extents based files
3689 *
3690 *
3691 * Need to be called with
3692 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3693 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3694 *
3695 * return > 0, number of of blocks already mapped/allocated
3696 *          if create == 0 and these are pre-allocated blocks
3697 *              buffer head is unmapped
3698 *          otherwise blocks are mapped
3699 *
3700 * return = 0, if plain look up failed (blocks have not been allocated)
3701 *          buffer head is unmapped
3702 *
3703 * return < 0, error case.
3704 */
3705int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
3706                        struct ext4_map_blocks *map, int flags)
3707{
3708        struct ext4_ext_path *path = NULL;
3709        struct ext4_extent newex, *ex, *ex2;
3710        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3711        ext4_fsblk_t newblock = 0;
3712        int free_on_err = 0, err = 0, depth, ret;
3713        unsigned int allocated = 0, offset = 0;
3714        unsigned int allocated_clusters = 0;
3715        unsigned int punched_out = 0;
3716        unsigned int result = 0;
3717        struct ext4_allocation_request ar;
3718        ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
3719        ext4_lblk_t cluster_offset;
3720
3721        ext_debug("blocks %u/%u requested for inode %lu\n",
3722                  map->m_lblk, map->m_len, inode->i_ino);
3723        trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
3724
3725        /* check in cache */
3726        if (!(flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) &&
3727                ext4_ext_in_cache(inode, map->m_lblk, &newex)) {
3728                if (!newex.ee_start_lo && !newex.ee_start_hi) {
3729                        if ((sbi->s_cluster_ratio > 1) &&
3730                            ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3731                                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3732
3733                        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3734                                /*
3735                                 * block isn't allocated yet and
3736                                 * user doesn't want to allocate it
3737                                 */
3738                                goto out2;
3739                        }
3740                        /* we should allocate requested block */
3741                } else {
3742                        /* block is already allocated */
3743                        if (sbi->s_cluster_ratio > 1)
3744                                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3745                        newblock = map->m_lblk
3746                                   - le32_to_cpu(newex.ee_block)
3747                                   + ext4_ext_pblock(&newex);
3748                        /* number of remaining blocks in the extent */
3749                        allocated = ext4_ext_get_actual_len(&newex) -
3750                                (map->m_lblk - le32_to_cpu(newex.ee_block));
3751                        goto out;
3752                }
3753        }
3754
3755        /* find extent for this block */
3756        path = ext4_ext_find_extent(inode, map->m_lblk, NULL);
3757        if (IS_ERR(path)) {
3758                err = PTR_ERR(path);
3759                path = NULL;
3760                goto out2;
3761        }
3762
3763        depth = ext_depth(inode);
3764
3765        /*
3766         * consistent leaf must not be empty;
3767         * this situation is possible, though, _during_ tree modification;
3768         * this is why assert can't be put in ext4_ext_find_extent()
3769         */
3770        if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
3771                EXT4_ERROR_INODE(inode, "bad extent address "
3772                                 "lblock: %lu, depth: %d pblock %lld",
3773                                 (unsigned long) map->m_lblk, depth,
3774                                 path[depth].p_block);
3775                err = -EIO;
3776                goto out2;
3777        }
3778
3779        ex = path[depth].p_ext;
3780        if (ex) {
3781                ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
3782                ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
3783                unsigned short ee_len;
3784
3785                /*
3786                 * Uninitialized extents are treated as holes, except that
3787                 * we split out initialized portions during a write.
3788                 */
3789                ee_len = ext4_ext_get_actual_len(ex);
3790
3791                trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
3792
3793                /* if found extent covers block, simply return it */
3794                if (in_range(map->m_lblk, ee_block, ee_len)) {
3795                        struct ext4_map_blocks punch_map;
3796                        ext4_fsblk_t partial_cluster = 0;
3797
3798                        newblock = map->m_lblk - ee_block + ee_start;
3799                        /* number of remaining blocks in the extent */
3800                        allocated = ee_len - (map->m_lblk - ee_block);
3801                        ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
3802                                  ee_block, ee_len, newblock);
3803
3804                        if ((flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) == 0) {
3805                                /*
3806                                 * Do not put uninitialized extent
3807                                 * in the cache
3808                                 */
3809                                if (!ext4_ext_is_uninitialized(ex)) {
3810                                        ext4_ext_put_in_cache(inode, ee_block,
3811                                                ee_len, ee_start);
3812                                        goto out;
3813                                }
3814                                ret = ext4_ext_handle_uninitialized_extents(
3815                                        handle, inode, map, path, flags,
3816                                        allocated, newblock);
3817                                return ret;
3818                        }
3819
3820                        /*
3821                         * Punch out the map length, but only to the
3822                         * end of the extent
3823                         */
3824                        punched_out = allocated < map->m_len ?
3825                                allocated : map->m_len;
3826
3827                        /*
3828                         * Sense extents need to be converted to
3829                         * uninitialized, they must fit in an
3830                         * uninitialized extent
3831                         */
3832                        if (punched_out > EXT_UNINIT_MAX_LEN)
3833                                punched_out = EXT_UNINIT_MAX_LEN;
3834
3835                        punch_map.m_lblk = map->m_lblk;
3836                        punch_map.m_pblk = newblock;
3837                        punch_map.m_len = punched_out;
3838                        punch_map.m_flags = 0;
3839
3840                        /* Check to see if the extent needs to be split */
3841                        if (punch_map.m_len != ee_len ||
3842                                punch_map.m_lblk != ee_block) {
3843
3844                                ret = ext4_split_extent(handle, inode,
3845                                path, &punch_map, 0,
3846                                EXT4_GET_BLOCKS_PUNCH_OUT_EXT |
3847                                EXT4_GET_BLOCKS_PRE_IO);
3848
3849                                if (ret < 0) {
3850                                        err = ret;
3851                                        goto out2;
3852                                }
3853                                /*
3854                                 * find extent for the block at
3855                                 * the start of the hole
3856                                 */
3857                                ext4_ext_drop_refs(path);
3858                                kfree(path);
3859
3860                                path = ext4_ext_find_extent(inode,
3861                                map->m_lblk, NULL);
3862                                if (IS_ERR(path)) {
3863                                        err = PTR_ERR(path);
3864                                        path = NULL;
3865                                        goto out2;
3866                                }
3867
3868                                depth = ext_depth(inode);
3869                                ex = path[depth].p_ext;
3870                                ee_len = ext4_ext_get_actual_len(ex);
3871                                ee_block = le32_to_cpu(ex->ee_block);
3872                                ee_start = ext4_ext_pblock(ex);
3873
3874                        }
3875
3876                        ext4_ext_mark_uninitialized(ex);
3877
3878                        ext4_ext_invalidate_cache(inode);
3879
3880                        err = ext4_ext_rm_leaf(handle, inode, path,
3881                                               &partial_cluster, map->m_lblk,
3882                                               map->m_lblk + punched_out);
3883
3884                        if (!err && path->p_hdr->eh_entries == 0) {
3885                                /*
3886                                 * Punch hole freed all of this sub tree,
3887                                 * so we need to correct eh_depth
3888                                 */
3889                                err = ext4_ext_get_access(handle, inode, path);
3890                                if (err == 0) {
3891                                        ext_inode_hdr(inode)->eh_depth = 0;
3892                                        ext_inode_hdr(inode)->eh_max =
3893                                        cpu_to_le16(ext4_ext_space_root(
3894                                                inode, 0));
3895
3896                                        err = ext4_ext_dirty(
3897                                                handle, inode, path);
3898                                }
3899                        }
3900
3901                        goto out2;
3902                }
3903        }
3904
3905        if ((sbi->s_cluster_ratio > 1) &&
3906            ext4_find_delalloc_cluster(inode, map->m_lblk, 0))
3907                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3908
3909        /*
3910         * requested block isn't allocated yet;
3911         * we couldn't try to create block if create flag is zero
3912         */
3913        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3914                /*
3915                 * put just found gap into cache to speed up
3916                 * subsequent requests
3917                 */
3918                ext4_ext_put_gap_in_cache(inode, path, map->m_lblk);
3919                goto out2;
3920        }
3921
3922        /*
3923         * Okay, we need to do block allocation.
3924         */
3925        map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
3926        newex.ee_block = cpu_to_le32(map->m_lblk);
3927        cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1);
3928
3929        /*
3930         * If we are doing bigalloc, check to see if the extent returned
3931         * by ext4_ext_find_extent() implies a cluster we can use.
3932         */
3933        if (cluster_offset && ex &&
3934            get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
3935                ar.len = allocated = map->m_len;
3936                newblock = map->m_pblk;
3937                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3938                goto got_allocated_blocks;
3939        }
3940
3941        /* find neighbour allocated blocks */
3942        ar.lleft = map->m_lblk;
3943        err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
3944        if (err)
3945                goto out2;
3946        ar.lright = map->m_lblk;
3947        ex2 = NULL;
3948        err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
3949        if (err)
3950                goto out2;
3951
3952        /* Check if the extent after searching to the right implies a
3953         * cluster we can use. */
3954        if ((sbi->s_cluster_ratio > 1) && ex2 &&
3955            get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
3956                ar.len = allocated = map->m_len;
3957                newblock = map->m_pblk;
3958                map->m_flags |= EXT4_MAP_FROM_CLUSTER;
3959                goto got_allocated_blocks;
3960        }
3961
3962        /*
3963         * See if request is beyond maximum number of blocks we can have in
3964         * a single extent. For an initialized extent this limit is
3965         * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3966         * EXT_UNINIT_MAX_LEN.
3967         */
3968        if (map->m_len > EXT_INIT_MAX_LEN &&
3969            !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3970                map->m_len = EXT_INIT_MAX_LEN;
3971        else if (map->m_len > EXT_UNINIT_MAX_LEN &&
3972                 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
3973                map->m_len = EXT_UNINIT_MAX_LEN;
3974
3975        /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3976        newex.ee_len = cpu_to_le16(map->m_len);
3977        err = ext4_ext_check_overlap(sbi, inode, &newex, path);
3978        if (err)
3979                allocated = ext4_ext_get_actual_len(&newex);
3980        else
3981                allocated = map->m_len;
3982
3983        /* allocate new block */
3984        ar.inode = inode;
3985        ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
3986        ar.logical = map->m_lblk;
3987        /*
3988         * We calculate the offset from the beginning of the cluster
3989         * for the logical block number, since when we allocate a
3990         * physical cluster, the physical block should start at the
3991         * same offset from the beginning of the cluster.  This is
3992         * needed so that future calls to get_implied_cluster_alloc()
3993         * work correctly.
3994         */
3995        offset = map->m_lblk & (sbi->s_cluster_ratio - 1);
3996        ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
3997        ar.goal -= offset;
3998        ar.logical -= offset;
3999        if (S_ISREG(inode->i_mode))
4000                ar.flags = EXT4_MB_HINT_DATA;
4001        else
4002                /* disable in-core preallocation for non-regular files */
4003                ar.flags = 0;
4004        if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4005                ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4006        newblock = ext4_mb_new_blocks(handle, &ar, &err);
4007        if (!newblock)
4008                goto out2;
4009        ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4010                  ar.goal, newblock, allocated);
4011        free_on_err = 1;
4012        allocated_clusters = ar.len;
4013        ar.len = EXT4_C2B(sbi, ar.len) - offset;
4014        if (ar.len > allocated)
4015                ar.len = allocated;
4016
4017got_allocated_blocks:
4018        /* try to insert new extent into found leaf and return */
4019        ext4_ext_store_pblock(&newex, newblock + offset);
4020        newex.ee_len = cpu_to_le16(ar.len);
4021        /* Mark uninitialized */
4022        if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
4023                ext4_ext_mark_uninitialized(&newex);
4024                /*
4025                 * io_end structure was created for every IO write to an
4026                 * uninitialized extent. To avoid unnecessary conversion,
4027                 * here we flag the IO that really needs the conversion.
4028                 * For non asycn direct IO case, flag the inode state
4029                 * that we need to perform conversion when IO is done.
4030                 */
4031                if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
4032                        if (io)
4033                                ext4_set_io_unwritten_flag(inode, io);
4034                        else
4035                                ext4_set_inode_state(inode,
4036                                                     EXT4_STATE_DIO_UNWRITTEN);
4037                }
4038                if (ext4_should_dioread_nolock(inode))
4039                        map->m_flags |= EXT4_MAP_UNINIT;
4040        }
4041
4042        err = 0;
4043        if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4044                err = check_eofblocks_fl(handle, inode, map->m_lblk,
4045                                         path, ar.len);
4046        if (!err)
4047                err = ext4_ext_insert_extent(handle, inode, path,
4048                                             &newex, flags);
4049        if (err && free_on_err) {
4050                int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4051                        EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4052                /* free data blocks we just allocated */
4053                /* not a good idea to call discard here directly,
4054                 * but otherwise we'd need to call it every free() */
4055                ext4_discard_preallocations(inode);
4056                ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex),
4057                                 ext4_ext_get_actual_len(&newex), fb_flags);
4058                goto out2;
4059        }
4060
4061        /* previous routine could use block we allocated */
4062        newblock = ext4_ext_pblock(&newex);
4063        allocated = ext4_ext_get_actual_len(&newex);
4064        if (allocated > map->m_len)
4065                allocated = map->m_len;
4066        map->m_flags |= EXT4_MAP_NEW;
4067
4068        /*
4069         * Update reserved blocks/metadata blocks after successful
4070         * block allocation which had been deferred till now.
4071         */
4072        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4073                unsigned int reserved_clusters;
4074                /*
4075                 * Check how many clusters we had reserved this allocated range
4076                 */
4077                reserved_clusters = get_reserved_cluster_alloc(inode,
4078                                                map->m_lblk, allocated);
4079                if (map->m_flags & EXT4_MAP_FROM_CLUSTER) {
4080                        if (reserved_clusters) {
4081                                /*
4082                                 * We have clusters reserved for this range.
4083                                 * But since we are not doing actual allocation
4084                                 * and are simply using blocks from previously
4085                                 * allocated cluster, we should release the
4086                                 * reservation and not claim quota.
4087                                 */
4088                                ext4_da_update_reserve_space(inode,
4089                                                reserved_clusters, 0);
4090                        }
4091                } else {
4092                        BUG_ON(allocated_clusters < reserved_clusters);
4093                        /* We will claim quota for all newly allocated blocks.*/
4094                        ext4_da_update_reserve_space(inode, allocated_clusters,
4095                                                        1);
4096                        if (reserved_clusters < allocated_clusters) {
4097                                struct ext4_inode_info *ei = EXT4_I(inode);
4098                                int reservation = allocated_clusters -
4099                                                  reserved_clusters;
4100                                /*
4101                                 * It seems we claimed few clusters outside of
4102                                 * the range of this allocation. We should give
4103                                 * it back to the reservation pool. This can
4104                                 * happen in the following case:
4105                                 *
4106                                 * * Suppose s_cluster_ratio is 4 (i.e., each
4107                                 *   cluster has 4 blocks. Thus, the clusters
4108                                 *   are [0-3],[4-7],[8-11]...
4109                                 * * First comes delayed allocation write for
4110                                 *   logical blocks 10 & 11. Since there were no
4111                                 *   previous delayed allocated blocks in the
4112                                 *   range [8-11], we would reserve 1 cluster
4113                                 *   for this write.
4114                                 * * Next comes write for logical blocks 3 to 8.
4115                                 *   In this case, we will reserve 2 clusters
4116                                 *   (for [0-3] and [4-7]; and not for [8-11] as
4117                                 *   that range has a delayed allocated blocks.
4118                                 *   Thus total reserved clusters now becomes 3.
4119                                 * * Now, during the delayed allocation writeout
4120                                 *   time, we will first write blocks [3-8] and
4121                                 *   allocate 3 clusters for writing these
4122                                 *   blocks. Also, we would claim all these
4123                                 *   three clusters above.
4124                                 * * Now when we come here to writeout the
4125                                 *   blocks [10-11], we would expect to claim
4126                                 *   the reservation of 1 cluster we had made
4127                                 *   (and we would claim it since there are no
4128                                 *   more delayed allocated blocks in the range
4129                                 *   [8-11]. But our reserved cluster count had
4130                                 *   already gone to 0.
4131                                 *
4132                                 *   Thus, at the step 4 above when we determine
4133                                 *   that there are still some unwritten delayed
4134                                 *   allocated blocks outside of our current
4135                                 *   block range, we should increment the
4136                                 *   reserved clusters count so that when the
4137                                 *   remaining blocks finally gets written, we
4138                                 *   could claim them.
4139                                 */
4140                                dquot_reserve_block(inode,
4141                                                EXT4_C2B(sbi, reservation));
4142                                spin_lock(&ei->i_block_reservation_lock);
4143                                ei->i_reserved_data_blocks += reservation;
4144                                spin_unlock(&ei->i_block_reservation_lock);
4145                        }
4146                }
4147        }
4148
4149        /*
4150         * Cache the extent and update transaction to commit on fdatasync only
4151         * when it is _not_ an uninitialized extent.
4152         */
4153        if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) {
4154                ext4_ext_put_in_cache(inode, map->m_lblk, allocated, newblock);
4155                ext4_update_inode_fsync_trans(handle, inode, 1);
4156        } else
4157                ext4_update_inode_fsync_trans(handle, inode, 0);
4158out:
4159        if (allocated > map->m_len)
4160                allocated = map->m_len;
4161        ext4_ext_show_leaf(inode, path);
4162        map->m_flags |= EXT4_MAP_MAPPED;
4163        map->m_pblk = newblock;
4164        map->m_len = allocated;
4165out2:
4166        if (path) {
4167                ext4_ext_drop_refs(path);
4168                kfree(path);
4169        }
4170        result = (flags & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) ?
4171                        punched_out : allocated;
4172
4173        trace_ext4_ext_map_blocks_exit(inode, map->m_lblk,
4174                newblock, map->m_len, err ? err : result);
4175
4176        return err ? err : result;
4177}
4178
4179void ext4_ext_truncate(struct inode *inode)
4180{
4181        struct address_space *mapping = inode->i_mapping;
4182        struct super_block *sb = inode->i_sb;
4183        ext4_lblk_t last_block;
4184        handle_t *handle;
4185        loff_t page_len;
4186        int err = 0;
4187
4188        /*
4189         * finish any pending end_io work so we won't run the risk of
4190         * converting any truncated blocks to initialized later
4191         */
4192        ext4_flush_completed_IO(inode);
4193
4194        /*
4195         * probably first extent we're gonna free will be last in block
4196         */
4197        err = ext4_writepage_trans_blocks(inode);
4198        handle = ext4_journal_start(inode, err);
4199        if (IS_ERR(handle))
4200                return;
4201
4202        if (inode->i_size % PAGE_CACHE_SIZE != 0) {
4203                page_len = PAGE_CACHE_SIZE -
4204                        (inode->i_size & (PAGE_CACHE_SIZE - 1));
4205
4206                err = ext4_discard_partial_page_buffers(handle,
4207                        mapping, inode->i_size, page_len, 0);
4208
4209                if (err)
4210                        goto out_stop;
4211        }
4212
4213        if (ext4_orphan_add(handle, inode))
4214                goto out_stop;
4215
4216        down_write(&EXT4_I(inode)->i_data_sem);
4217        ext4_ext_invalidate_cache(inode);
4218
4219        ext4_discard_preallocations(inode);
4220
4221        /*
4222         * TODO: optimization is possible here.
4223         * Probably we need not scan at all,
4224         * because page truncation is enough.
4225         */
4226
4227        /* we have to know where to truncate from in crash case */
4228        EXT4_I(inode)->i_disksize = inode->i_size;
4229        ext4_mark_inode_dirty(handle, inode);
4230
4231        last_block = (inode->i_size + sb->s_blocksize - 1)
4232                        >> EXT4_BLOCK_SIZE_BITS(sb);
4233        err = ext4_ext_remove_space(inode, last_block);
4234
4235        /* In a multi-transaction truncate, we only make the final
4236         * transaction synchronous.
4237         */
4238        if (IS_SYNC(inode))
4239                ext4_handle_sync(handle);
4240
4241        up_write(&EXT4_I(inode)->i_data_sem);
4242
4243out_stop:
4244        /*
4245         * If this was a simple ftruncate() and the file will remain alive,
4246         * then we need to clear up the orphan record which we created above.
4247         * However, if this was a real unlink then we were called by
4248         * ext4_delete_inode(), and we allow that function to clean up the
4249         * orphan info for us.
4250         */
4251        if (inode->i_nlink)
4252                ext4_orphan_del(handle, inode);
4253
4254        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4255        ext4_mark_inode_dirty(handle, inode);
4256        ext4_journal_stop(handle);
4257}
4258
4259static void ext4_falloc_update_inode(struct inode *inode,
4260                                int mode, loff_t new_size, int update_ctime)
4261{
4262        struct timespec now;
4263
4264        if (update_ctime) {
4265                now = current_fs_time(inode->i_sb);
4266                if (!timespec_equal(&inode->i_ctime, &now))
4267                        inode->i_ctime = now;
4268        }
4269        /*
4270         * Update only when preallocation was requested beyond
4271         * the file size.
4272         */
4273        if (!(mode & FALLOC_FL_KEEP_SIZE)) {
4274                if (new_size > i_size_read(inode))
4275                        i_size_write(inode, new_size);
4276                if (new_size > EXT4_I(inode)->i_disksize)
4277                        ext4_update_i_disksize(inode, new_size);
4278        } else {
4279                /*
4280                 * Mark that we allocate beyond EOF so the subsequent truncate
4281                 * can proceed even if the new size is the same as i_size.
4282                 */
4283                if (new_size > i_size_read(inode))
4284                        ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4285        }
4286
4287}
4288
4289/*
4290 * preallocate space for a file. This implements ext4's fallocate file
4291 * operation, which gets called from sys_fallocate system call.
4292 * For block-mapped files, posix_fallocate should fall back to the method
4293 * of writing zeroes to the required new blocks (the same behavior which is
4294 * expected for file systems which do not support fallocate() system call).
4295 */
4296long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4297{
4298        struct inode *inode = file->f_path.dentry->d_inode;
4299        handle_t *handle;
4300        loff_t new_size;
4301        unsigned int max_blocks;
4302        int ret = 0;
4303        int ret2 = 0;
4304        int retries = 0;
4305        int flags;
4306        struct ext4_map_blocks map;
4307        unsigned int credits, blkbits = inode->i_blkbits;
4308
4309        /*
4310         * currently supporting (pre)allocate mode for extent-based
4311         * files _only_
4312         */
4313        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4314                return -EOPNOTSUPP;
4315
4316        /* Return error if mode is not supported */
4317        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
4318                return -EOPNOTSUPP;
4319
4320        if (mode & FALLOC_FL_PUNCH_HOLE)
4321                return ext4_punch_hole(file, offset, len);
4322
4323        trace_ext4_fallocate_enter(inode, offset, len, mode);
4324        map.m_lblk = offset >> blkbits;
4325        /*
4326         * We can't just convert len to max_blocks because
4327         * If blocksize = 4096 offset = 3072 and len = 2048
4328         */
4329        max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
4330                - map.m_lblk;
4331        /*
4332         * credits to insert 1 extent into extent tree
4333         */
4334        credits = ext4_chunk_trans_blocks(inode, max_blocks);
4335        mutex_lock(&inode->i_mutex);
4336        ret = inode_newsize_ok(inode, (len + offset));
4337        if (ret) {
4338                mutex_unlock(&inode->i_mutex);
4339                trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4340                return ret;
4341        }
4342        flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT;
4343        if (mode & FALLOC_FL_KEEP_SIZE)
4344                flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4345        /*
4346         * Don't normalize the request if it can fit in one extent so
4347         * that it doesn't get unnecessarily split into multiple
4348         * extents.
4349         */
4350        if (len <= EXT_UNINIT_MAX_LEN << blkbits)
4351                flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4352retry:
4353        while (ret >= 0 && ret < max_blocks) {
4354                map.m_lblk = map.m_lblk + ret;
4355                map.m_len = max_blocks = max_blocks - ret;
4356                handle = ext4_journal_start(inode, credits);
4357                if (IS_ERR(handle)) {
4358                        ret = PTR_ERR(handle);
4359                        break;
4360                }
4361                ret = ext4_map_blocks(handle, inode, &map, flags);
4362                if (ret <= 0) {
4363#ifdef EXT4FS_DEBUG
4364                        WARN_ON(ret <= 0);
4365                        printk(KERN_ERR "%s: ext4_ext_map_blocks "
4366                                    "returned error inode#%lu, block=%u, "
4367                                    "max_blocks=%u", __func__,
4368                                    inode->i_ino, map.m_lblk, max_blocks);
4369#endif
4370                        ext4_mark_inode_dirty(handle, inode);
4371                        ret2 = ext4_journal_stop(handle);
4372                        break;
4373                }
4374                if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
4375                                                blkbits) >> blkbits))
4376                        new_size = offset + len;
4377                else
4378                        new_size = ((loff_t) map.m_lblk + ret) << blkbits;
4379
4380                ext4_falloc_update_inode(inode, mode, new_size,
4381                                         (map.m_flags & EXT4_MAP_NEW));
4382                ext4_mark_inode_dirty(handle, inode);
4383                ret2 = ext4_journal_stop(handle);
4384                if (ret2)
4385                        break;
4386        }
4387        if (ret == -ENOSPC &&
4388                        ext4_should_retry_alloc(inode->i_sb, &retries)) {
4389                ret = 0;
4390                goto retry;
4391        }
4392        mutex_unlock(&inode->i_mutex);
4393        trace_ext4_fallocate_exit(inode, offset, max_blocks,
4394                                ret > 0 ? ret2 : ret);
4395        return ret > 0 ? ret2 : ret;
4396}
4397
4398/*
4399 * This function convert a range of blocks to written extents
4400 * The caller of this function will pass the start offset and the size.
4401 * all unwritten extents within this range will be converted to
4402 * written extents.
4403 *
4404 * This function is called from the direct IO end io call back
4405 * function, to convert the fallocated extents after IO is completed.
4406 * Returns 0 on success.
4407 */
4408int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset,
4409                                    ssize_t len)
4410{
4411        handle_t *handle;
4412        unsigned int max_blocks;
4413        int ret = 0;
4414        int ret2 = 0;
4415        struct ext4_map_blocks map;
4416        unsigned int credits, blkbits = inode->i_blkbits;
4417
4418        map.m_lblk = offset >> blkbits;
4419        /*
4420         * We can't just convert len to max_blocks because
4421         * If blocksize = 4096 offset = 3072 and len = 2048
4422         */
4423        max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) -
4424                      map.m_lblk);
4425        /*
4426         * credits to insert 1 extent into extent tree
4427         */
4428        credits = ext4_chunk_trans_blocks(inode, max_blocks);
4429        while (ret >= 0 && ret < max_blocks) {
4430                map.m_lblk += ret;
4431                map.m_len = (max_blocks -= ret);
4432                handle = ext4_journal_start(inode, credits);
4433                if (IS_ERR(handle)) {
4434                        ret = PTR_ERR(handle);
4435                        break;
4436                }
4437                ret = ext4_map_blocks(handle, inode, &map,
4438                                      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4439                if (ret <= 0) {
4440                        WARN_ON(ret <= 0);
4441                        printk(KERN_ERR "%s: ext4_ext_map_blocks "
4442                                    "returned error inode#%lu, block=%u, "
4443                                    "max_blocks=%u", __func__,
4444                                    inode->i_ino, map.m_lblk, map.m_len);
4445                }
4446                ext4_mark_inode_dirty(handle, inode);
4447                ret2 = ext4_journal_stop(handle);
4448                if (ret <= 0 || ret2 )
4449                        break;
4450        }
4451        return ret > 0 ? ret2 : ret;
4452}
4453
4454/*
4455 * Callback function called for each extent to gather FIEMAP information.
4456 */
4457static int ext4_ext_fiemap_cb(struct inode *inode, ext4_lblk_t next,
4458                       struct ext4_ext_cache *newex, struct ext4_extent *ex,
4459                       void *data)
4460{
4461        __u64   logical;
4462        __u64   physical;
4463        __u64   length;
4464        __u32   flags = 0;
4465        int             ret = 0;
4466        struct fiemap_extent_info *fieinfo = data;
4467        unsigned char blksize_bits;
4468
4469        blksize_bits = inode->i_sb->s_blocksize_bits;
4470        logical = (__u64)newex->ec_block << blksize_bits;
4471
4472        if (newex->ec_start == 0) {
4473                /*
4474                 * No extent in extent-tree contains block @newex->ec_start,
4475                 * then the block may stay in 1)a hole or 2)delayed-extent.
4476                 *
4477                 * Holes or delayed-extents are processed as follows.
4478                 * 1. lookup dirty pages with specified range in pagecache.
4479                 *    If no page is got, then there is no delayed-extent and
4480                 *    return with EXT_CONTINUE.
4481                 * 2. find the 1st mapped buffer,
4482                 * 3. check if the mapped buffer is both in the request range
4483                 *    and a delayed buffer. If not, there is no delayed-extent,
4484                 *    then return.
4485                 * 4. a delayed-extent is found, the extent will be collected.
4486                 */
4487                ext4_lblk_t     end = 0;
4488                pgoff_t         last_offset;
4489                pgoff_t         offset;
4490                pgoff_t         index;
4491                pgoff_t         start_index = 0;
4492                struct page     **pages = NULL;
4493                struct buffer_head *bh = NULL;
4494                struct buffer_head *head = NULL;
4495                unsigned int nr_pages = PAGE_SIZE / sizeof(struct page *);
4496
4497                pages = kmalloc(PAGE_SIZE, GFP_KERNEL);
4498                if (pages == NULL)
4499                        return -ENOMEM;
4500
4501                offset = logical >> PAGE_SHIFT;
4502repeat:
4503                last_offset = offset;
4504                head = NULL;
4505                ret = find_get_pages_tag(inode->i_mapping, &offset,
4506                                        PAGECACHE_TAG_DIRTY, nr_pages, pages);
4507
4508                if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4509                        /* First time, try to find a mapped buffer. */
4510                        if (ret == 0) {
4511out:
4512                                for (index = 0; index < ret; index++)
4513                                        page_cache_release(pages[index]);
4514                                /* just a hole. */
4515                                kfree(pages);
4516                                return EXT_CONTINUE;
4517                        }
4518                        index = 0;
4519
4520next_page:
4521                        /* Try to find the 1st mapped buffer. */
4522                        end = ((__u64)pages[index]->index << PAGE_SHIFT) >>
4523                                  blksize_bits;
4524                        if (!page_has_buffers(pages[index]))
4525                                goto out;
4526                        head = page_buffers(pages[index]);
4527                        if (!head)
4528                                goto out;
4529
4530                        index++;
4531                        bh = head;
4532                        do {
4533                                if (end >= newex->ec_block +
4534                                        newex->ec_len)
4535                                        /* The buffer is out of
4536                                         * the request range.
4537                                         */
4538                                        goto out;
4539
4540                                if (buffer_mapped(bh) &&
4541                                    end >= newex->ec_block) {
4542                                        start_index = index - 1;
4543                                        /* get the 1st mapped buffer. */
4544                                        goto found_mapped_buffer;
4545                                }
4546
4547                                bh = bh->b_this_page;
4548                                end++;
4549                        } while (bh != head);
4550
4551                        /* No mapped buffer in the range found in this page,
4552                         * We need to look up next page.
4553                         */
4554                        if (index >= ret) {
4555                                /* There is no page left, but we need to limit
4556                                 * newex->ec_len.
4557                                 */
4558                                newex->ec_len = end - newex->ec_block;
4559                                goto out;
4560                        }
4561                        goto next_page;
4562                } else {
4563                        /*Find contiguous delayed buffers. */
4564                        if (ret > 0 && pages[0]->index == last_offset)
4565                                head = page_buffers(pages[0]);
4566                        bh = head;
4567                        index = 1;
4568                        start_index = 0;
4569                }
4570
4571found_mapped_buffer:
4572                if (bh != NULL && buffer_delay(bh)) {
4573                        /* 1st or contiguous delayed buffer found. */
4574                        if (!(flags & FIEMAP_EXTENT_DELALLOC)) {
4575                                /*
4576                                 * 1st delayed buffer found, record
4577                                 * the start of extent.
4578                                 */
4579                                flags |= FIEMAP_EXTENT_DELALLOC;
4580                                newex->ec_block = end;
4581                                logical = (__u64)end << blksize_bits;
4582                        }
4583                        /* Find contiguous delayed buffers. */
4584                        do {
4585                                if (!buffer_delay(bh))
4586                                        goto found_delayed_extent;
4587                                bh = bh->b_this_page;
4588                                end++;
4589                        } while (bh != head);
4590
4591                        for (; index < ret; index++) {
4592                                if (!page_has_buffers(pages[index])) {
4593                                        bh = NULL;
4594                                        break;
4595                                }
4596                                head = page_buffers(pages[index]);
4597                                if (!head) {
4598                                        bh = NULL;
4599                                        break;
4600                                }
4601
4602                                if (pages[index]->index !=
4603                                    pages[start_index]->index + index
4604                                    - start_index) {
4605                                        /* Blocks are not contiguous. */
4606                                        bh = NULL;
4607                                        break;
4608                                }
4609                                bh = head;
4610                                do {
4611                                        if (!buffer_delay(bh))
4612                                                /* Delayed-extent ends. */
4613                                                goto found_delayed_extent;
4614                                        bh = bh->b_this_page;
4615                                        end++;
4616                                } while (bh != head);
4617                        }
4618                } else if (!(flags & FIEMAP_EXTENT_DELALLOC))
4619                        /* a hole found. */
4620                        goto out;
4621
4622found_delayed_extent:
4623                newex->ec_len = min(end - newex->ec_block,
4624                                                (ext4_lblk_t)EXT_INIT_MAX_LEN);
4625                if (ret == nr_pages && bh != NULL &&
4626                        newex->ec_len < EXT_INIT_MAX_LEN &&
4627                        buffer_delay(bh)) {
4628                        /* Have not collected an extent and continue. */
4629                        for (index = 0; index < ret; index++)
4630                                page_cache_release(pages[index]);
4631                        goto repeat;
4632                }
4633
4634                for (index = 0; index < ret; index++)
4635                        page_cache_release(pages[index]);
4636                kfree(pages);
4637        }
4638
4639        physical = (__u64)newex->ec_start << blksize_bits;
4640        length =   (__u64)newex->ec_len << blksize_bits;
4641
4642        if (ex && ext4_ext_is_uninitialized(ex))
4643                flags |= FIEMAP_EXTENT_UNWRITTEN;
4644
4645        if (next == EXT_MAX_BLOCKS)
4646                flags |= FIEMAP_EXTENT_LAST;
4647
4648        ret = fiemap_fill_next_extent(fieinfo, logical, physical,
4649                                        length, flags);
4650        if (ret < 0)
4651                return ret;
4652        if (ret == 1)
4653                return EXT_BREAK;
4654        return EXT_CONTINUE;
4655}
4656/* fiemap flags we can handle specified here */
4657#define EXT4_FIEMAP_FLAGS       (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4658
4659static int ext4_xattr_fiemap(struct inode *inode,
4660                                struct fiemap_extent_info *fieinfo)
4661{
4662        __u64 physical = 0;
4663        __u64 length;
4664        __u32 flags = FIEMAP_EXTENT_LAST;
4665        int blockbits = inode->i_sb->s_blocksize_bits;
4666        int error = 0;
4667
4668        /* in-inode? */
4669        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4670                struct ext4_iloc iloc;
4671                int offset;     /* offset of xattr in inode */
4672
4673                error = ext4_get_inode_loc(inode, &iloc);
4674                if (error)
4675                        return error;
4676                physical = iloc.bh->b_blocknr << blockbits;
4677                offset = EXT4_GOOD_OLD_INODE_SIZE +
4678                                EXT4_I(inode)->i_extra_isize;
4679                physical += offset;
4680                length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4681                flags |= FIEMAP_EXTENT_DATA_INLINE;
4682                brelse(iloc.bh);
4683        } else { /* external block */
4684                physical = EXT4_I(inode)->i_file_acl << blockbits;
4685                length = inode->i_sb->s_blocksize;
4686        }
4687
4688        if (physical)
4689                error = fiemap_fill_next_extent(fieinfo, 0, physical,
4690                                                length, flags);
4691        return (error < 0 ? error : 0);
4692}
4693
4694/*
4695 * ext4_ext_punch_hole
4696 *
4697 * Punches a hole of "length" bytes in a file starting
4698 * at byte "offset"
4699 *
4700 * @inode:  The inode of the file to punch a hole in
4701 * @offset: The starting byte offset of the hole
4702 * @length: The length of the hole
4703 *
4704 * Returns the number of blocks removed or negative on err
4705 */
4706int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
4707{
4708        struct inode *inode = file->f_path.dentry->d_inode;
4709        struct super_block *sb = inode->i_sb;
4710        struct ext4_ext_cache cache_ex;
4711        ext4_lblk_t first_block, last_block, num_blocks, iblock, max_blocks;
4712        struct address_space *mapping = inode->i_mapping;
4713        struct ext4_map_blocks map;
4714        handle_t *handle;
4715        loff_t first_page, last_page, page_len;
4716        loff_t first_page_offset, last_page_offset;
4717        int ret, credits, blocks_released, err = 0;
4718
4719        /* No need to punch hole beyond i_size */
4720        if (offset >= inode->i_size)
4721                return 0;
4722
4723        /*
4724         * If the hole extends beyond i_size, set the hole
4725         * to end after the page that contains i_size
4726         */
4727        if (offset + length > inode->i_size) {
4728                length = inode->i_size +
4729                   PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
4730                   offset;
4731        }
4732
4733        first_block = (offset + sb->s_blocksize - 1) >>
4734                EXT4_BLOCK_SIZE_BITS(sb);
4735        last_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4736
4737        first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
4738        last_page = (offset + length) >> PAGE_CACHE_SHIFT;
4739
4740        first_page_offset = first_page << PAGE_CACHE_SHIFT;
4741        last_page_offset = last_page << PAGE_CACHE_SHIFT;
4742
4743        /*
4744         * Write out all dirty pages to avoid race conditions
4745         * Then release them.
4746         */
4747        if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4748                err = filemap_write_and_wait_range(mapping,
4749                        offset, offset + length - 1);
4750
4751                if (err)
4752                        return err;
4753        }
4754
4755        /* Now release the pages */
4756        if (last_page_offset > first_page_offset) {
4757                truncate_inode_pages_range(mapping, first_page_offset,
4758                                           last_page_offset-1);
4759        }
4760
4761        /* finish any pending end_io work */
4762        ext4_flush_completed_IO(inode);
4763
4764        credits = ext4_writepage_trans_blocks(inode);
4765        handle = ext4_journal_start(inode, credits);
4766        if (IS_ERR(handle))
4767                return PTR_ERR(handle);
4768
4769        err = ext4_orphan_add(handle, inode);
4770        if (err)
4771                goto out;
4772
4773        /*
4774         * Now we need to zero out the non-page-aligned data in the
4775         * pages at the start and tail of the hole, and unmap the buffer
4776         * heads for the block aligned regions of the page that were
4777         * completely zeroed.
4778         */
4779        if (first_page > last_page) {
4780                /*
4781                 * If the file space being truncated is contained within a page
4782                 * just zero out and unmap the middle of that page
4783                 */
4784                err = ext4_discard_partial_page_buffers(handle,
4785                        mapping, offset, length, 0);
4786
4787                if (err)
4788                        goto out;
4789        } else {
4790                /*
4791                 * zero out and unmap the partial page that contains
4792                 * the start of the hole
4793                 */
4794                page_len  = first_page_offset - offset;
4795                if (page_len > 0) {
4796                        err = ext4_discard_partial_page_buffers(handle, mapping,
4797                                                   offset, page_len, 0);
4798                        if (err)
4799                                goto out;
4800                }
4801
4802                /*
4803                 * zero out and unmap the partial page that contains
4804                 * the end of the hole
4805                 */
4806                page_len = offset + length - last_page_offset;
4807                if (page_len > 0) {
4808                        err = ext4_discard_partial_page_buffers(handle, mapping,
4809                                        last_page_offset, page_len, 0);
4810                        if (err)
4811                                goto out;
4812                }
4813        }
4814
4815
4816        /*
4817         * If i_size is contained in the last page, we need to
4818         * unmap and zero the partial page after i_size
4819         */
4820        if (inode->i_size >> PAGE_CACHE_SHIFT == last_page &&
4821           inode->i_size % PAGE_CACHE_SIZE != 0) {
4822
4823                page_len = PAGE_CACHE_SIZE -
4824                        (inode->i_size & (PAGE_CACHE_SIZE - 1));
4825
4826                if (page_len > 0) {
4827                        err = ext4_discard_partial_page_buffers(handle,
4828                          mapping, inode->i_size, page_len, 0);
4829
4830                        if (err)
4831                                goto out;
4832                }
4833        }
4834
4835        /* If there are no blocks to remove, return now */
4836        if (first_block >= last_block)
4837                goto out;
4838
4839        down_write(&EXT4_I(inode)->i_data_sem);
4840        ext4_ext_invalidate_cache(inode);
4841        ext4_discard_preallocations(inode);
4842
4843        /*
4844         * Loop over all the blocks and identify blocks
4845         * that need to be punched out
4846         */
4847        iblock = first_block;
4848        blocks_released = 0;
4849        while (iblock < last_block) {
4850                max_blocks = last_block - iblock;
4851                num_blocks = 1;
4852                memset(&map, 0, sizeof(map));
4853                map.m_lblk = iblock;
4854                map.m_len = max_blocks;
4855                ret = ext4_ext_map_blocks(handle, inode, &map,
4856                        EXT4_GET_BLOCKS_PUNCH_OUT_EXT);
4857
4858                if (ret > 0) {
4859                        blocks_released += ret;
4860                        num_blocks = ret;
4861                } else if (ret == 0) {
4862                        /*
4863                         * If map blocks could not find the block,
4864                         * then it is in a hole.  If the hole was
4865                         * not already cached, then map blocks should
4866                         * put it in the cache.  So we can get the hole
4867                         * out of the cache
4868                         */
4869                        memset(&cache_ex, 0, sizeof(cache_ex));
4870                        if ((ext4_ext_check_cache(inode, iblock, &cache_ex)) &&
4871                                !cache_ex.ec_start) {
4872
4873                                /* The hole is cached */
4874                                num_blocks = cache_ex.ec_block +
4875                                cache_ex.ec_len - iblock;
4876
4877                        } else {
4878                                /* The block could not be identified */
4879                                err = -EIO;
4880                                break;
4881                        }
4882                } else {
4883                        /* Map blocks error */
4884                        err = ret;
4885                        break;
4886                }
4887
4888                if (num_blocks == 0) {
4889                        /* This condition should never happen */
4890                        ext_debug("Block lookup failed");
4891                        err = -EIO;
4892                        break;
4893                }
4894
4895                iblock += num_blocks;
4896        }
4897
4898        if (blocks_released > 0) {
4899                ext4_ext_invalidate_cache(inode);
4900                ext4_discard_preallocations(inode);
4901        }
4902
4903        if (IS_SYNC(inode))
4904                ext4_handle_sync(handle);
4905
4906        up_write(&EXT4_I(inode)->i_data_sem);
4907
4908out:
4909        ext4_orphan_del(handle, inode);
4910        inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4911        ext4_mark_inode_dirty(handle, inode);
4912        ext4_journal_stop(handle);
4913        return err;
4914}
4915int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4916                __u64 start, __u64 len)
4917{
4918        ext4_lblk_t start_blk;
4919        int error = 0;
4920
4921        /* fallback to generic here if not in extents fmt */
4922        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
4923                return generic_block_fiemap(inode, fieinfo, start, len,
4924                        ext4_get_block);
4925
4926        if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
4927                return -EBADR;
4928
4929        if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4930                error = ext4_xattr_fiemap(inode, fieinfo);
4931        } else {
4932                ext4_lblk_t len_blks;
4933                __u64 last_blk;
4934
4935                start_blk = start >> inode->i_sb->s_blocksize_bits;
4936                last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
4937                if (last_blk >= EXT_MAX_BLOCKS)
4938                        last_blk = EXT_MAX_BLOCKS-1;
4939                len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
4940
4941                /*
4942                 * Walk the extent tree gathering extent information.
4943                 * ext4_ext_fiemap_cb will push extents back to user.
4944                 */
4945                error = ext4_ext_walk_space(inode, start_blk, len_blks,
4946                                          ext4_ext_fiemap_cb, fieinfo);
4947        }
4948
4949        return error;
4950}
4951
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.