linux/fs/ext4/indirect.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/indirect.c
   4 *
   5 *  from
   6 *
   7 *  linux/fs/ext4/inode.c
   8 *
   9 * Copyright (C) 1992, 1993, 1994, 1995
  10 * Remy Card (card@masi.ibp.fr)
  11 * Laboratoire MASI - Institut Blaise Pascal
  12 * Universite Pierre et Marie Curie (Paris VI)
  13 *
  14 *  from
  15 *
  16 *  linux/fs/minix/inode.c
  17 *
  18 *  Copyright (C) 1991, 1992  Linus Torvalds
  19 *
  20 *  Goal-directed block allocation by Stephen Tweedie
  21 *      (sct@redhat.com), 1993, 1998
  22 */
  23
  24#include "ext4_jbd2.h"
  25#include "truncate.h"
  26#include <linux/dax.h>
  27#include <linux/uio.h>
  28
  29#include <trace/events/ext4.h>
  30
  31typedef struct {
  32        __le32  *p;
  33        __le32  key;
  34        struct buffer_head *bh;
  35} Indirect;
  36
  37static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
  38{
  39        p->key = *(p->p = v);
  40        p->bh = bh;
  41}
  42
  43/**
  44 *      ext4_block_to_path - parse the block number into array of offsets
  45 *      @inode: inode in question (we are only interested in its superblock)
  46 *      @i_block: block number to be parsed
  47 *      @offsets: array to store the offsets in
  48 *      @boundary: set this non-zero if the referred-to block is likely to be
  49 *             followed (on disk) by an indirect block.
  50 *
  51 *      To store the locations of file's data ext4 uses a data structure common
  52 *      for UNIX filesystems - tree of pointers anchored in the inode, with
  53 *      data blocks at leaves and indirect blocks in intermediate nodes.
  54 *      This function translates the block number into path in that tree -
  55 *      return value is the path length and @offsets[n] is the offset of
  56 *      pointer to (n+1)th node in the nth one. If @block is out of range
  57 *      (negative or too large) warning is printed and zero returned.
  58 *
  59 *      Note: function doesn't find node addresses, so no IO is needed. All
  60 *      we need to know is the capacity of indirect blocks (taken from the
  61 *      inode->i_sb).
  62 */
  63
  64/*
  65 * Portability note: the last comparison (check that we fit into triple
  66 * indirect block) is spelled differently, because otherwise on an
  67 * architecture with 32-bit longs and 8Kb pages we might get into trouble
  68 * if our filesystem had 8Kb blocks. We might use long long, but that would
  69 * kill us on x86. Oh, well, at least the sign propagation does not matter -
  70 * i_block would have to be negative in the very beginning, so we would not
  71 * get there at all.
  72 */
  73
  74static int ext4_block_to_path(struct inode *inode,
  75                              ext4_lblk_t i_block,
  76                              ext4_lblk_t offsets[4], int *boundary)
  77{
  78        int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
  79        int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
  80        const long direct_blocks = EXT4_NDIR_BLOCKS,
  81                indirect_blocks = ptrs,
  82                double_blocks = (1 << (ptrs_bits * 2));
  83        int n = 0;
  84        int final = 0;
  85
  86        if (i_block < direct_blocks) {
  87                offsets[n++] = i_block;
  88                final = direct_blocks;
  89        } else if ((i_block -= direct_blocks) < indirect_blocks) {
  90                offsets[n++] = EXT4_IND_BLOCK;
  91                offsets[n++] = i_block;
  92                final = ptrs;
  93        } else if ((i_block -= indirect_blocks) < double_blocks) {
  94                offsets[n++] = EXT4_DIND_BLOCK;
  95                offsets[n++] = i_block >> ptrs_bits;
  96                offsets[n++] = i_block & (ptrs - 1);
  97                final = ptrs;
  98        } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
  99                offsets[n++] = EXT4_TIND_BLOCK;
 100                offsets[n++] = i_block >> (ptrs_bits * 2);
 101                offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
 102                offsets[n++] = i_block & (ptrs - 1);
 103                final = ptrs;
 104        } else {
 105                ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
 106                             i_block + direct_blocks +
 107                             indirect_blocks + double_blocks, inode->i_ino);
 108        }
 109        if (boundary)
 110                *boundary = final - 1 - (i_block & (ptrs - 1));
 111        return n;
 112}
 113
 114/**
 115 *      ext4_get_branch - read the chain of indirect blocks leading to data
 116 *      @inode: inode in question
 117 *      @depth: depth of the chain (1 - direct pointer, etc.)
 118 *      @offsets: offsets of pointers in inode/indirect blocks
 119 *      @chain: place to store the result
 120 *      @err: here we store the error value
 121 *
 122 *      Function fills the array of triples <key, p, bh> and returns %NULL
 123 *      if everything went OK or the pointer to the last filled triple
 124 *      (incomplete one) otherwise. Upon the return chain[i].key contains
 125 *      the number of (i+1)-th block in the chain (as it is stored in memory,
 126 *      i.e. little-endian 32-bit), chain[i].p contains the address of that
 127 *      number (it points into struct inode for i==0 and into the bh->b_data
 128 *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 129 *      block for i>0 and NULL for i==0. In other words, it holds the block
 130 *      numbers of the chain, addresses they were taken from (and where we can
 131 *      verify that chain did not change) and buffer_heads hosting these
 132 *      numbers.
 133 *
 134 *      Function stops when it stumbles upon zero pointer (absent block)
 135 *              (pointer to last triple returned, *@err == 0)
 136 *      or when it gets an IO error reading an indirect block
 137 *              (ditto, *@err == -EIO)
 138 *      or when it reads all @depth-1 indirect blocks successfully and finds
 139 *      the whole chain, all way to the data (returns %NULL, *err == 0).
 140 *
 141 *      Need to be called with
 142 *      down_read(&EXT4_I(inode)->i_data_sem)
 143 */
 144static Indirect *ext4_get_branch(struct inode *inode, int depth,
 145                                 ext4_lblk_t  *offsets,
 146                                 Indirect chain[4], int *err)
 147{
 148        struct super_block *sb = inode->i_sb;
 149        Indirect *p = chain;
 150        struct buffer_head *bh;
 151        int ret = -EIO;
 152
 153        *err = 0;
 154        /* i_data is not going away, no lock needed */
 155        add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
 156        if (!p->key)
 157                goto no_block;
 158        while (--depth) {
 159                bh = sb_getblk(sb, le32_to_cpu(p->key));
 160                if (unlikely(!bh)) {
 161                        ret = -ENOMEM;
 162                        goto failure;
 163                }
 164
 165                if (!bh_uptodate_or_lock(bh)) {
 166                        if (ext4_read_bh(bh, 0, NULL) < 0) {
 167                                put_bh(bh);
 168                                goto failure;
 169                        }
 170                        /* validate block references */
 171                        if (ext4_check_indirect_blockref(inode, bh)) {
 172                                put_bh(bh);
 173                                goto failure;
 174                        }
 175                }
 176
 177                add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
 178                /* Reader: end */
 179                if (!p->key)
 180                        goto no_block;
 181        }
 182        return NULL;
 183
 184failure:
 185        *err = ret;
 186no_block:
 187        return p;
 188}
 189
 190/**
 191 *      ext4_find_near - find a place for allocation with sufficient locality
 192 *      @inode: owner
 193 *      @ind: descriptor of indirect block.
 194 *
 195 *      This function returns the preferred place for block allocation.
 196 *      It is used when heuristic for sequential allocation fails.
 197 *      Rules are:
 198 *        + if there is a block to the left of our position - allocate near it.
 199 *        + if pointer will live in indirect block - allocate near that block.
 200 *        + if pointer will live in inode - allocate in the same
 201 *          cylinder group.
 202 *
 203 * In the latter case we colour the starting block by the callers PID to
 204 * prevent it from clashing with concurrent allocations for a different inode
 205 * in the same block group.   The PID is used here so that functionally related
 206 * files will be close-by on-disk.
 207 *
 208 *      Caller must make sure that @ind is valid and will stay that way.
 209 */
 210static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
 211{
 212        struct ext4_inode_info *ei = EXT4_I(inode);
 213        __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
 214        __le32 *p;
 215
 216        /* Try to find previous block */
 217        for (p = ind->p - 1; p >= start; p--) {
 218                if (*p)
 219                        return le32_to_cpu(*p);
 220        }
 221
 222        /* No such thing, so let's try location of indirect block */
 223        if (ind->bh)
 224                return ind->bh->b_blocknr;
 225
 226        /*
 227         * It is going to be referred to from the inode itself? OK, just put it
 228         * into the same cylinder group then.
 229         */
 230        return ext4_inode_to_goal_block(inode);
 231}
 232
 233/**
 234 *      ext4_find_goal - find a preferred place for allocation.
 235 *      @inode: owner
 236 *      @block:  block we want
 237 *      @partial: pointer to the last triple within a chain
 238 *
 239 *      Normally this function find the preferred place for block allocation,
 240 *      returns it.
 241 *      Because this is only used for non-extent files, we limit the block nr
 242 *      to 32 bits.
 243 */
 244static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
 245                                   Indirect *partial)
 246{
 247        ext4_fsblk_t goal;
 248
 249        /*
 250         * XXX need to get goal block from mballoc's data structures
 251         */
 252
 253        goal = ext4_find_near(inode, partial);
 254        goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 255        return goal;
 256}
 257
 258/**
 259 *      ext4_blks_to_allocate - Look up the block map and count the number
 260 *      of direct blocks need to be allocated for the given branch.
 261 *
 262 *      @branch: chain of indirect blocks
 263 *      @k: number of blocks need for indirect blocks
 264 *      @blks: number of data blocks to be mapped.
 265 *      @blocks_to_boundary:  the offset in the indirect block
 266 *
 267 *      return the total number of blocks to be allocate, including the
 268 *      direct and indirect blocks.
 269 */
 270static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
 271                                 int blocks_to_boundary)
 272{
 273        unsigned int count = 0;
 274
 275        /*
 276         * Simple case, [t,d]Indirect block(s) has not allocated yet
 277         * then it's clear blocks on that path have not allocated
 278         */
 279        if (k > 0) {
 280                /* right now we don't handle cross boundary allocation */
 281                if (blks < blocks_to_boundary + 1)
 282                        count += blks;
 283                else
 284                        count += blocks_to_boundary + 1;
 285                return count;
 286        }
 287
 288        count++;
 289        while (count < blks && count <= blocks_to_boundary &&
 290                le32_to_cpu(*(branch[0].p + count)) == 0) {
 291                count++;
 292        }
 293        return count;
 294}
 295
 296/**
 297 * ext4_alloc_branch() - allocate and set up a chain of blocks
 298 * @handle: handle for this transaction
 299 * @ar: structure describing the allocation request
 300 * @indirect_blks: number of allocated indirect blocks
 301 * @offsets: offsets (in the blocks) to store the pointers to next.
 302 * @branch: place to store the chain in.
 303 *
 304 *      This function allocates blocks, zeroes out all but the last one,
 305 *      links them into chain and (if we are synchronous) writes them to disk.
 306 *      In other words, it prepares a branch that can be spliced onto the
 307 *      inode. It stores the information about that chain in the branch[], in
 308 *      the same format as ext4_get_branch() would do. We are calling it after
 309 *      we had read the existing part of chain and partial points to the last
 310 *      triple of that (one with zero ->key). Upon the exit we have the same
 311 *      picture as after the successful ext4_get_block(), except that in one
 312 *      place chain is disconnected - *branch->p is still zero (we did not
 313 *      set the last link), but branch->key contains the number that should
 314 *      be placed into *branch->p to fill that gap.
 315 *
 316 *      If allocation fails we free all blocks we've allocated (and forget
 317 *      their buffer_heads) and return the error value the from failed
 318 *      ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 319 *      as described above and return 0.
 320 */
 321static int ext4_alloc_branch(handle_t *handle,
 322                             struct ext4_allocation_request *ar,
 323                             int indirect_blks, ext4_lblk_t *offsets,
 324                             Indirect *branch)
 325{
 326        struct buffer_head *            bh;
 327        ext4_fsblk_t                    b, new_blocks[4];
 328        __le32                          *p;
 329        int                             i, j, err, len = 1;
 330
 331        for (i = 0; i <= indirect_blks; i++) {
 332                if (i == indirect_blks) {
 333                        new_blocks[i] = ext4_mb_new_blocks(handle, ar, &err);
 334                } else {
 335                        ar->goal = new_blocks[i] = ext4_new_meta_blocks(handle,
 336                                        ar->inode, ar->goal,
 337                                        ar->flags & EXT4_MB_DELALLOC_RESERVED,
 338                                        NULL, &err);
 339                        /* Simplify error cleanup... */
 340                        branch[i+1].bh = NULL;
 341                }
 342                if (err) {
 343                        i--;
 344                        goto failed;
 345                }
 346                branch[i].key = cpu_to_le32(new_blocks[i]);
 347                if (i == 0)
 348                        continue;
 349
 350                bh = branch[i].bh = sb_getblk(ar->inode->i_sb, new_blocks[i-1]);
 351                if (unlikely(!bh)) {
 352                        err = -ENOMEM;
 353                        goto failed;
 354                }
 355                lock_buffer(bh);
 356                BUFFER_TRACE(bh, "call get_create_access");
 357                err = ext4_journal_get_create_access(handle, bh);
 358                if (err) {
 359                        unlock_buffer(bh);
 360                        goto failed;
 361                }
 362
 363                memset(bh->b_data, 0, bh->b_size);
 364                p = branch[i].p = (__le32 *) bh->b_data + offsets[i];
 365                b = new_blocks[i];
 366
 367                if (i == indirect_blks)
 368                        len = ar->len;
 369                for (j = 0; j < len; j++)
 370                        *p++ = cpu_to_le32(b++);
 371
 372                BUFFER_TRACE(bh, "marking uptodate");
 373                set_buffer_uptodate(bh);
 374                unlock_buffer(bh);
 375
 376                BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 377                err = ext4_handle_dirty_metadata(handle, ar->inode, bh);
 378                if (err)
 379                        goto failed;
 380        }
 381        return 0;
 382failed:
 383        if (i == indirect_blks) {
 384                /* Free data blocks */
 385                ext4_free_blocks(handle, ar->inode, NULL, new_blocks[i],
 386                                 ar->len, 0);
 387                i--;
 388        }
 389        for (; i >= 0; i--) {
 390                /*
 391                 * We want to ext4_forget() only freshly allocated indirect
 392                 * blocks. Buffer for new_blocks[i] is at branch[i+1].bh
 393                 * (buffer at branch[0].bh is indirect block / inode already
 394                 * existing before ext4_alloc_branch() was called). Also
 395                 * because blocks are freshly allocated, we don't need to
 396                 * revoke them which is why we don't set
 397                 * EXT4_FREE_BLOCKS_METADATA.
 398                 */
 399                ext4_free_blocks(handle, ar->inode, branch[i+1].bh,
 400                                 new_blocks[i], 1,
 401                                 branch[i+1].bh ? EXT4_FREE_BLOCKS_FORGET : 0);
 402        }
 403        return err;
 404}
 405
 406/**
 407 * ext4_splice_branch() - splice the allocated branch onto inode.
 408 * @handle: handle for this transaction
 409 * @ar: structure describing the allocation request
 410 * @where: location of missing link
 411 * @num:   number of indirect blocks we are adding
 412 *
 413 * This function fills the missing link and does all housekeeping needed in
 414 * inode (->i_blocks, etc.). In case of success we end up with the full
 415 * chain to new block and return 0.
 416 */
 417static int ext4_splice_branch(handle_t *handle,
 418                              struct ext4_allocation_request *ar,
 419                              Indirect *where, int num)
 420{
 421        int i;
 422        int err = 0;
 423        ext4_fsblk_t current_block;
 424
 425        /*
 426         * If we're splicing into a [td]indirect block (as opposed to the
 427         * inode) then we need to get write access to the [td]indirect block
 428         * before the splice.
 429         */
 430        if (where->bh) {
 431                BUFFER_TRACE(where->bh, "get_write_access");
 432                err = ext4_journal_get_write_access(handle, where->bh);
 433                if (err)
 434                        goto err_out;
 435        }
 436        /* That's it */
 437
 438        *where->p = where->key;
 439
 440        /*
 441         * Update the host buffer_head or inode to point to more just allocated
 442         * direct blocks blocks
 443         */
 444        if (num == 0 && ar->len > 1) {
 445                current_block = le32_to_cpu(where->key) + 1;
 446                for (i = 1; i < ar->len; i++)
 447                        *(where->p + i) = cpu_to_le32(current_block++);
 448        }
 449
 450        /* We are done with atomic stuff, now do the rest of housekeeping */
 451        /* had we spliced it onto indirect block? */
 452        if (where->bh) {
 453                /*
 454                 * If we spliced it onto an indirect block, we haven't
 455                 * altered the inode.  Note however that if it is being spliced
 456                 * onto an indirect block at the very end of the file (the
 457                 * file is growing) then we *will* alter the inode to reflect
 458                 * the new i_size.  But that is not done here - it is done in
 459                 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
 460                 */
 461                jbd_debug(5, "splicing indirect only\n");
 462                BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
 463                err = ext4_handle_dirty_metadata(handle, ar->inode, where->bh);
 464                if (err)
 465                        goto err_out;
 466        } else {
 467                /*
 468                 * OK, we spliced it into the inode itself on a direct block.
 469                 */
 470                err = ext4_mark_inode_dirty(handle, ar->inode);
 471                if (unlikely(err))
 472                        goto err_out;
 473                jbd_debug(5, "splicing direct\n");
 474        }
 475        return err;
 476
 477err_out:
 478        for (i = 1; i <= num; i++) {
 479                /*
 480                 * branch[i].bh is newly allocated, so there is no
 481                 * need to revoke the block, which is why we don't
 482                 * need to set EXT4_FREE_BLOCKS_METADATA.
 483                 */
 484                ext4_free_blocks(handle, ar->inode, where[i].bh, 0, 1,
 485                                 EXT4_FREE_BLOCKS_FORGET);
 486        }
 487        ext4_free_blocks(handle, ar->inode, NULL, le32_to_cpu(where[num].key),
 488                         ar->len, 0);
 489
 490        return err;
 491}
 492
 493/*
 494 * The ext4_ind_map_blocks() function handles non-extents inodes
 495 * (i.e., using the traditional indirect/double-indirect i_blocks
 496 * scheme) for ext4_map_blocks().
 497 *
 498 * Allocation strategy is simple: if we have to allocate something, we will
 499 * have to go the whole way to leaf. So let's do it before attaching anything
 500 * to tree, set linkage between the newborn blocks, write them if sync is
 501 * required, recheck the path, free and repeat if check fails, otherwise
 502 * set the last missing link (that will protect us from any truncate-generated
 503 * removals - all blocks on the path are immune now) and possibly force the
 504 * write on the parent block.
 505 * That has a nice additional property: no special recovery from the failed
 506 * allocations is needed - we simply release blocks and do not touch anything
 507 * reachable from inode.
 508 *
 509 * `handle' can be NULL if create == 0.
 510 *
 511 * return > 0, # of blocks mapped or allocated.
 512 * return = 0, if plain lookup failed.
 513 * return < 0, error case.
 514 *
 515 * The ext4_ind_get_blocks() function should be called with
 516 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
 517 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
 518 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
 519 * blocks.
 520 */
 521int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
 522                        struct ext4_map_blocks *map,
 523                        int flags)
 524{
 525        struct ext4_allocation_request ar;
 526        int err = -EIO;
 527        ext4_lblk_t offsets[4];
 528        Indirect chain[4];
 529        Indirect *partial;
 530        int indirect_blks;
 531        int blocks_to_boundary = 0;
 532        int depth;
 533        int count = 0;
 534        ext4_fsblk_t first_block = 0;
 535
 536        trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
 537        ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
 538        ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
 539        depth = ext4_block_to_path(inode, map->m_lblk, offsets,
 540                                   &blocks_to_boundary);
 541
 542        if (depth == 0)
 543                goto out;
 544
 545        partial = ext4_get_branch(inode, depth, offsets, chain, &err);
 546
 547        /* Simplest case - block found, no allocation needed */
 548        if (!partial) {
 549                first_block = le32_to_cpu(chain[depth - 1].key);
 550                count++;
 551                /*map more blocks*/
 552                while (count < map->m_len && count <= blocks_to_boundary) {
 553                        ext4_fsblk_t blk;
 554
 555                        blk = le32_to_cpu(*(chain[depth-1].p + count));
 556
 557                        if (blk == first_block + count)
 558                                count++;
 559                        else
 560                                break;
 561                }
 562                goto got_it;
 563        }
 564
 565        /* Next simple case - plain lookup failed */
 566        if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
 567                unsigned epb = inode->i_sb->s_blocksize / sizeof(u32);
 568                int i;
 569
 570                /*
 571                 * Count number blocks in a subtree under 'partial'. At each
 572                 * level we count number of complete empty subtrees beyond
 573                 * current offset and then descend into the subtree only
 574                 * partially beyond current offset.
 575                 */
 576                count = 0;
 577                for (i = partial - chain + 1; i < depth; i++)
 578                        count = count * epb + (epb - offsets[i] - 1);
 579                count++;
 580                /* Fill in size of a hole we found */
 581                map->m_pblk = 0;
 582                map->m_len = min_t(unsigned int, map->m_len, count);
 583                goto cleanup;
 584        }
 585
 586        /* Failed read of indirect block */
 587        if (err == -EIO)
 588                goto cleanup;
 589
 590        /*
 591         * Okay, we need to do block allocation.
 592        */
 593        if (ext4_has_feature_bigalloc(inode->i_sb)) {
 594                EXT4_ERROR_INODE(inode, "Can't allocate blocks for "
 595                                 "non-extent mapped inodes with bigalloc");
 596                err = -EFSCORRUPTED;
 597                goto out;
 598        }
 599
 600        /* Set up for the direct block allocation */
 601        memset(&ar, 0, sizeof(ar));
 602        ar.inode = inode;
 603        ar.logical = map->m_lblk;
 604        if (S_ISREG(inode->i_mode))
 605                ar.flags = EXT4_MB_HINT_DATA;
 606        if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 607                ar.flags |= EXT4_MB_DELALLOC_RESERVED;
 608        if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
 609                ar.flags |= EXT4_MB_USE_RESERVED;
 610
 611        ar.goal = ext4_find_goal(inode, map->m_lblk, partial);
 612
 613        /* the number of blocks need to allocate for [d,t]indirect blocks */
 614        indirect_blks = (chain + depth) - partial - 1;
 615
 616        /*
 617         * Next look up the indirect map to count the totoal number of
 618         * direct blocks to allocate for this branch.
 619         */
 620        ar.len = ext4_blks_to_allocate(partial, indirect_blks,
 621                                       map->m_len, blocks_to_boundary);
 622
 623        /*
 624         * Block out ext4_truncate while we alter the tree
 625         */
 626        err = ext4_alloc_branch(handle, &ar, indirect_blks,
 627                                offsets + (partial - chain), partial);
 628
 629        /*
 630         * The ext4_splice_branch call will free and forget any buffers
 631         * on the new chain if there is a failure, but that risks using
 632         * up transaction credits, especially for bitmaps where the
 633         * credits cannot be returned.  Can we handle this somehow?  We
 634         * may need to return -EAGAIN upwards in the worst case.  --sct
 635         */
 636        if (!err)
 637                err = ext4_splice_branch(handle, &ar, partial, indirect_blks);
 638        if (err)
 639                goto cleanup;
 640
 641        map->m_flags |= EXT4_MAP_NEW;
 642
 643        ext4_update_inode_fsync_trans(handle, inode, 1);
 644        count = ar.len;
 645got_it:
 646        map->m_flags |= EXT4_MAP_MAPPED;
 647        map->m_pblk = le32_to_cpu(chain[depth-1].key);
 648        map->m_len = count;
 649        if (count > blocks_to_boundary)
 650                map->m_flags |= EXT4_MAP_BOUNDARY;
 651        err = count;
 652        /* Clean up and exit */
 653        partial = chain + depth - 1;    /* the whole chain */
 654cleanup:
 655        while (partial > chain) {
 656                BUFFER_TRACE(partial->bh, "call brelse");
 657                brelse(partial->bh);
 658                partial--;
 659        }
 660out:
 661        trace_ext4_ind_map_blocks_exit(inode, flags, map, err);
 662        return err;
 663}
 664
 665/*
 666 * Calculate number of indirect blocks touched by mapping @nrblocks logically
 667 * contiguous blocks
 668 */
 669int ext4_ind_trans_blocks(struct inode *inode, int nrblocks)
 670{
 671        /*
 672         * With N contiguous data blocks, we need at most
 673         * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
 674         * 2 dindirect blocks, and 1 tindirect block
 675         */
 676        return DIV_ROUND_UP(nrblocks, EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
 677}
 678
 679static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
 680                                     struct buffer_head *bh, int *dropped)
 681{
 682        int err;
 683
 684        if (bh) {
 685                BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 686                err = ext4_handle_dirty_metadata(handle, inode, bh);
 687                if (unlikely(err))
 688                        return err;
 689        }
 690        err = ext4_mark_inode_dirty(handle, inode);
 691        if (unlikely(err))
 692                return err;
 693        /*
 694         * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 695         * moment, get_block can be called only for blocks inside i_size since
 696         * page cache has been already dropped and writes are blocked by
 697         * i_mutex. So we can safely drop the i_data_sem here.
 698         */
 699        BUG_ON(EXT4_JOURNAL(inode) == NULL);
 700        ext4_discard_preallocations(inode, 0);
 701        up_write(&EXT4_I(inode)->i_data_sem);
 702        *dropped = 1;
 703        return 0;
 704}
 705
 706/*
 707 * Truncate transactions can be complex and absolutely huge.  So we need to
 708 * be able to restart the transaction at a convenient checkpoint to make
 709 * sure we don't overflow the journal.
 710 *
 711 * Try to extend this transaction for the purposes of truncation.  If
 712 * extend fails, we restart transaction.
 713 */
 714static int ext4_ind_truncate_ensure_credits(handle_t *handle,
 715                                            struct inode *inode,
 716                                            struct buffer_head *bh,
 717                                            int revoke_creds)
 718{
 719        int ret;
 720        int dropped = 0;
 721
 722        ret = ext4_journal_ensure_credits_fn(handle, EXT4_RESERVE_TRANS_BLOCKS,
 723                        ext4_blocks_for_truncate(inode), revoke_creds,
 724                        ext4_ind_trunc_restart_fn(handle, inode, bh, &dropped));
 725        if (dropped)
 726                down_write(&EXT4_I(inode)->i_data_sem);
 727        if (ret <= 0)
 728                return ret;
 729        if (bh) {
 730                BUFFER_TRACE(bh, "retaking write access");
 731                ret = ext4_journal_get_write_access(handle, bh);
 732                if (unlikely(ret))
 733                        return ret;
 734        }
 735        return 0;
 736}
 737
 738/*
 739 * Probably it should be a library function... search for first non-zero word
 740 * or memcmp with zero_page, whatever is better for particular architecture.
 741 * Linus?
 742 */
 743static inline int all_zeroes(__le32 *p, __le32 *q)
 744{
 745        while (p < q)
 746                if (*p++)
 747                        return 0;
 748        return 1;
 749}
 750
 751/**
 752 *      ext4_find_shared - find the indirect blocks for partial truncation.
 753 *      @inode:   inode in question
 754 *      @depth:   depth of the affected branch
 755 *      @offsets: offsets of pointers in that branch (see ext4_block_to_path)
 756 *      @chain:   place to store the pointers to partial indirect blocks
 757 *      @top:     place to the (detached) top of branch
 758 *
 759 *      This is a helper function used by ext4_truncate().
 760 *
 761 *      When we do truncate() we may have to clean the ends of several
 762 *      indirect blocks but leave the blocks themselves alive. Block is
 763 *      partially truncated if some data below the new i_size is referred
 764 *      from it (and it is on the path to the first completely truncated
 765 *      data block, indeed).  We have to free the top of that path along
 766 *      with everything to the right of the path. Since no allocation
 767 *      past the truncation point is possible until ext4_truncate()
 768 *      finishes, we may safely do the latter, but top of branch may
 769 *      require special attention - pageout below the truncation point
 770 *      might try to populate it.
 771 *
 772 *      We atomically detach the top of branch from the tree, store the
 773 *      block number of its root in *@top, pointers to buffer_heads of
 774 *      partially truncated blocks - in @chain[].bh and pointers to
 775 *      their last elements that should not be removed - in
 776 *      @chain[].p. Return value is the pointer to last filled element
 777 *      of @chain.
 778 *
 779 *      The work left to caller to do the actual freeing of subtrees:
 780 *              a) free the subtree starting from *@top
 781 *              b) free the subtrees whose roots are stored in
 782 *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
 783 *              c) free the subtrees growing from the inode past the @chain[0].
 784 *                      (no partially truncated stuff there).  */
 785
 786static Indirect *ext4_find_shared(struct inode *inode, int depth,
 787                                  ext4_lblk_t offsets[4], Indirect chain[4],
 788                                  __le32 *top)
 789{
 790        Indirect *partial, *p;
 791        int k, err;
 792
 793        *top = 0;
 794        /* Make k index the deepest non-null offset + 1 */
 795        for (k = depth; k > 1 && !offsets[k-1]; k--)
 796                ;
 797        partial = ext4_get_branch(inode, k, offsets, chain, &err);
 798        /* Writer: pointers */
 799        if (!partial)
 800                partial = chain + k-1;
 801        /*
 802         * If the branch acquired continuation since we've looked at it -
 803         * fine, it should all survive and (new) top doesn't belong to us.
 804         */
 805        if (!partial->key && *partial->p)
 806                /* Writer: end */
 807                goto no_top;
 808        for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
 809                ;
 810        /*
 811         * OK, we've found the last block that must survive. The rest of our
 812         * branch should be detached before unlocking. However, if that rest
 813         * of branch is all ours and does not grow immediately from the inode
 814         * it's easier to cheat and just decrement partial->p.
 815         */
 816        if (p == chain + k - 1 && p > chain) {
 817                p->p--;
 818        } else {
 819                *top = *p->p;
 820                /* Nope, don't do this in ext4.  Must leave the tree intact */
 821#if 0
 822                *p->p = 0;
 823#endif
 824        }
 825        /* Writer: end */
 826
 827        while (partial > p) {
 828                brelse(partial->bh);
 829                partial--;
 830        }
 831no_top:
 832        return partial;
 833}
 834
 835/*
 836 * Zero a number of block pointers in either an inode or an indirect block.
 837 * If we restart the transaction we must again get write access to the
 838 * indirect block for further modification.
 839 *
 840 * We release `count' blocks on disk, but (last - first) may be greater
 841 * than `count' because there can be holes in there.
 842 *
 843 * Return 0 on success, 1 on invalid block range
 844 * and < 0 on fatal error.
 845 */
 846static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
 847                             struct buffer_head *bh,
 848                             ext4_fsblk_t block_to_free,
 849                             unsigned long count, __le32 *first,
 850                             __le32 *last)
 851{
 852        __le32 *p;
 853        int     flags = EXT4_FREE_BLOCKS_VALIDATED;
 854        int     err;
 855
 856        if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
 857            ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
 858                flags |= EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_METADATA;
 859        else if (ext4_should_journal_data(inode))
 860                flags |= EXT4_FREE_BLOCKS_FORGET;
 861
 862        if (!ext4_inode_block_valid(inode, block_to_free, count)) {
 863                EXT4_ERROR_INODE(inode, "attempt to clear invalid "
 864                                 "blocks %llu len %lu",
 865                                 (unsigned long long) block_to_free, count);
 866                return 1;
 867        }
 868
 869        err = ext4_ind_truncate_ensure_credits(handle, inode, bh,
 870                                ext4_free_data_revoke_credits(inode, count));
 871        if (err < 0)
 872                goto out_err;
 873
 874        for (p = first; p < last; p++)
 875                *p = 0;
 876
 877        ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
 878        return 0;
 879out_err:
 880        ext4_std_error(inode->i_sb, err);
 881        return err;
 882}
 883
 884/**
 885 * ext4_free_data - free a list of data blocks
 886 * @handle:     handle for this transaction
 887 * @inode:      inode we are dealing with
 888 * @this_bh:    indirect buffer_head which contains *@first and *@last
 889 * @first:      array of block numbers
 890 * @last:       points immediately past the end of array
 891 *
 892 * We are freeing all blocks referred from that array (numbers are stored as
 893 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
 894 *
 895 * We accumulate contiguous runs of blocks to free.  Conveniently, if these
 896 * blocks are contiguous then releasing them at one time will only affect one
 897 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
 898 * actually use a lot of journal space.
 899 *
 900 * @this_bh will be %NULL if @first and @last point into the inode's direct
 901 * block pointers.
 902 */
 903static void ext4_free_data(handle_t *handle, struct inode *inode,
 904                           struct buffer_head *this_bh,
 905                           __le32 *first, __le32 *last)
 906{
 907        ext4_fsblk_t block_to_free = 0;    /* Starting block # of a run */
 908        unsigned long count = 0;            /* Number of blocks in the run */
 909        __le32 *block_to_free_p = NULL;     /* Pointer into inode/ind
 910                                               corresponding to
 911                                               block_to_free */
 912        ext4_fsblk_t nr;                    /* Current block # */
 913        __le32 *p;                          /* Pointer into inode/ind
 914                                               for current block */
 915        int err = 0;
 916
 917        if (this_bh) {                          /* For indirect block */
 918                BUFFER_TRACE(this_bh, "get_write_access");
 919                err = ext4_journal_get_write_access(handle, this_bh);
 920                /* Important: if we can't update the indirect pointers
 921                 * to the blocks, we can't free them. */
 922                if (err)
 923                        return;
 924        }
 925
 926        for (p = first; p < last; p++) {
 927                nr = le32_to_cpu(*p);
 928                if (nr) {
 929                        /* accumulate blocks to free if they're contiguous */
 930                        if (count == 0) {
 931                                block_to_free = nr;
 932                                block_to_free_p = p;
 933                                count = 1;
 934                        } else if (nr == block_to_free + count) {
 935                                count++;
 936                        } else {
 937                                err = ext4_clear_blocks(handle, inode, this_bh,
 938                                                        block_to_free, count,
 939                                                        block_to_free_p, p);
 940                                if (err)
 941                                        break;
 942                                block_to_free = nr;
 943                                block_to_free_p = p;
 944                                count = 1;
 945                        }
 946                }
 947        }
 948
 949        if (!err && count > 0)
 950                err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
 951                                        count, block_to_free_p, p);
 952        if (err < 0)
 953                /* fatal error */
 954                return;
 955
 956        if (this_bh) {
 957                BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
 958
 959                /*
 960                 * The buffer head should have an attached journal head at this
 961                 * point. However, if the data is corrupted and an indirect
 962                 * block pointed to itself, it would have been detached when
 963                 * the block was cleared. Check for this instead of OOPSing.
 964                 */
 965                if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
 966                        ext4_handle_dirty_metadata(handle, inode, this_bh);
 967                else
 968                        EXT4_ERROR_INODE(inode,
 969                                         "circular indirect block detected at "
 970                                         "block %llu",
 971                                (unsigned long long) this_bh->b_blocknr);
 972        }
 973}
 974
 975/**
 976 *      ext4_free_branches - free an array of branches
 977 *      @handle: JBD handle for this transaction
 978 *      @inode: inode we are dealing with
 979 *      @parent_bh: the buffer_head which contains *@first and *@last
 980 *      @first: array of block numbers
 981 *      @last:  pointer immediately past the end of array
 982 *      @depth: depth of the branches to free
 983 *
 984 *      We are freeing all blocks referred from these branches (numbers are
 985 *      stored as little-endian 32-bit) and updating @inode->i_blocks
 986 *      appropriately.
 987 */
 988static void ext4_free_branches(handle_t *handle, struct inode *inode,
 989                               struct buffer_head *parent_bh,
 990                               __le32 *first, __le32 *last, int depth)
 991{
 992        ext4_fsblk_t nr;
 993        __le32 *p;
 994
 995        if (ext4_handle_is_aborted(handle))
 996                return;
 997
 998        if (depth--) {
 999                struct buffer_head *bh;
1000                int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1001                p = last;
1002                while (--p >= first) {
1003                        nr = le32_to_cpu(*p);
1004                        if (!nr)
1005                                continue;               /* A hole */
1006
1007                        if (!ext4_inode_block_valid(inode, nr, 1)) {
1008                                EXT4_ERROR_INODE(inode,
1009                                                 "invalid indirect mapped "
1010                                                 "block %lu (level %d)",
1011                                                 (unsigned long) nr, depth);
1012                                break;
1013                        }
1014
1015                        /* Go read the buffer for the next level down */
1016                        bh = ext4_sb_bread(inode->i_sb, nr, 0);
1017
1018                        /*
1019                         * A read failure? Report error and clear slot
1020                         * (should be rare).
1021                         */
1022                        if (IS_ERR(bh)) {
1023                                ext4_error_inode_block(inode, nr, -PTR_ERR(bh),
1024                                                       "Read failure");
1025                                continue;
1026                        }
1027
1028                        /* This zaps the entire block.  Bottom up. */
1029                        BUFFER_TRACE(bh, "free child branches");
1030                        ext4_free_branches(handle, inode, bh,
1031                                        (__le32 *) bh->b_data,
1032                                        (__le32 *) bh->b_data + addr_per_block,
1033                                        depth);
1034                        brelse(bh);
1035
1036                        /*
1037                         * Everything below this pointer has been
1038                         * released.  Now let this top-of-subtree go.
1039                         *
1040                         * We want the freeing of this indirect block to be
1041                         * atomic in the journal with the updating of the
1042                         * bitmap block which owns it.  So make some room in
1043                         * the journal.
1044                         *
1045                         * We zero the parent pointer *after* freeing its
1046                         * pointee in the bitmaps, so if extend_transaction()
1047                         * for some reason fails to put the bitmap changes and
1048                         * the release into the same transaction, recovery
1049                         * will merely complain about releasing a free block,
1050                         * rather than leaking blocks.
1051                         */
1052                        if (ext4_handle_is_aborted(handle))
1053                                return;
1054                        if (ext4_ind_truncate_ensure_credits(handle, inode,
1055                                        NULL,
1056                                        ext4_free_metadata_revoke_credits(
1057                                                        inode->i_sb, 1)) < 0)
1058                                return;
1059
1060                        /*
1061                         * The forget flag here is critical because if
1062                         * we are journaling (and not doing data
1063                         * journaling), we have to make sure a revoke
1064                         * record is written to prevent the journal
1065                         * replay from overwriting the (former)
1066                         * indirect block if it gets reallocated as a
1067                         * data block.  This must happen in the same
1068                         * transaction where the data blocks are
1069                         * actually freed.
1070                         */
1071                        ext4_free_blocks(handle, inode, NULL, nr, 1,
1072                                         EXT4_FREE_BLOCKS_METADATA|
1073                                         EXT4_FREE_BLOCKS_FORGET);
1074
1075                        if (parent_bh) {
1076                                /*
1077                                 * The block which we have just freed is
1078                                 * pointed to by an indirect block: journal it
1079                                 */
1080                                BUFFER_TRACE(parent_bh, "get_write_access");
1081                                if (!ext4_journal_get_write_access(handle,
1082                                                                   parent_bh)){
1083                                        *p = 0;
1084                                        BUFFER_TRACE(parent_bh,
1085                                        "call ext4_handle_dirty_metadata");
1086                                        ext4_handle_dirty_metadata(handle,
1087                                                                   inode,
1088                                                                   parent_bh);
1089                                }
1090                        }
1091                }
1092        } else {
1093                /* We have reached the bottom of the tree. */
1094                BUFFER_TRACE(parent_bh, "free data blocks");
1095                ext4_free_data(handle, inode, parent_bh, first, last);
1096        }
1097}
1098
1099void ext4_ind_truncate(handle_t *handle, struct inode *inode)
1100{
1101        struct ext4_inode_info *ei = EXT4_I(inode);
1102        __le32 *i_data = ei->i_data;
1103        int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1104        ext4_lblk_t offsets[4];
1105        Indirect chain[4];
1106        Indirect *partial;
1107        __le32 nr = 0;
1108        int n = 0;
1109        ext4_lblk_t last_block, max_block;
1110        unsigned blocksize = inode->i_sb->s_blocksize;
1111
1112        last_block = (inode->i_size + blocksize-1)
1113                                        >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1114        max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1115                                        >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1116
1117        if (last_block != max_block) {
1118                n = ext4_block_to_path(inode, last_block, offsets, NULL);
1119                if (n == 0)
1120                        return;
1121        }
1122
1123        ext4_es_remove_extent(inode, last_block, EXT_MAX_BLOCKS - last_block);
1124
1125        /*
1126         * The orphan list entry will now protect us from any crash which
1127         * occurs before the truncate completes, so it is now safe to propagate
1128         * the new, shorter inode size (held for now in i_size) into the
1129         * on-disk inode. We do this via i_disksize, which is the value which
1130         * ext4 *really* writes onto the disk inode.
1131         */
1132        ei->i_disksize = inode->i_size;
1133
1134        if (last_block == max_block) {
1135                /*
1136                 * It is unnecessary to free any data blocks if last_block is
1137                 * equal to the indirect block limit.
1138                 */
1139                return;
1140        } else if (n == 1) {            /* direct blocks */
1141                ext4_free_data(handle, inode, NULL, i_data+offsets[0],
1142                               i_data + EXT4_NDIR_BLOCKS);
1143                goto do_indirects;
1144        }
1145
1146        partial = ext4_find_shared(inode, n, offsets, chain, &nr);
1147        /* Kill the top of shared branch (not detached) */
1148        if (nr) {
1149                if (partial == chain) {
1150                        /* Shared branch grows from the inode */
1151                        ext4_free_branches(handle, inode, NULL,
1152                                           &nr, &nr+1, (chain+n-1) - partial);
1153                        *partial->p = 0;
1154                        /*
1155                         * We mark the inode dirty prior to restart,
1156                         * and prior to stop.  No need for it here.
1157                         */
1158                } else {
1159                        /* Shared branch grows from an indirect block */
1160                        BUFFER_TRACE(partial->bh, "get_write_access");
1161                        ext4_free_branches(handle, inode, partial->bh,
1162                                        partial->p,
1163                                        partial->p+1, (chain+n-1) - partial);
1164                }
1165        }
1166        /* Clear the ends of indirect blocks on the shared branch */
1167        while (partial > chain) {
1168                ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
1169                                   (__le32*)partial->bh->b_data+addr_per_block,
1170                                   (chain+n-1) - partial);
1171                BUFFER_TRACE(partial->bh, "call brelse");
1172                brelse(partial->bh);
1173                partial--;
1174        }
1175do_indirects:
1176        /* Kill the remaining (whole) subtrees */
1177        switch (offsets[0]) {
1178        default:
1179                nr = i_data[EXT4_IND_BLOCK];
1180                if (nr) {
1181                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1182                        i_data[EXT4_IND_BLOCK] = 0;
1183                }
1184                fallthrough;
1185        case EXT4_IND_BLOCK:
1186                nr = i_data[EXT4_DIND_BLOCK];
1187                if (nr) {
1188                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1189                        i_data[EXT4_DIND_BLOCK] = 0;
1190                }
1191                fallthrough;
1192        case EXT4_DIND_BLOCK:
1193                nr = i_data[EXT4_TIND_BLOCK];
1194                if (nr) {
1195                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1196                        i_data[EXT4_TIND_BLOCK] = 0;
1197                }
1198                fallthrough;
1199        case EXT4_TIND_BLOCK:
1200                ;
1201        }
1202}
1203
1204/**
1205 *      ext4_ind_remove_space - remove space from the range
1206 *      @handle: JBD handle for this transaction
1207 *      @inode: inode we are dealing with
1208 *      @start: First block to remove
1209 *      @end:   One block after the last block to remove (exclusive)
1210 *
1211 *      Free the blocks in the defined range (end is exclusive endpoint of
1212 *      range). This is used by ext4_punch_hole().
1213 */
1214int ext4_ind_remove_space(handle_t *handle, struct inode *inode,
1215                          ext4_lblk_t start, ext4_lblk_t end)
1216{
1217        struct ext4_inode_info *ei = EXT4_I(inode);
1218        __le32 *i_data = ei->i_data;
1219        int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1220        ext4_lblk_t offsets[4], offsets2[4];
1221        Indirect chain[4], chain2[4];
1222        Indirect *partial, *partial2;
1223        Indirect *p = NULL, *p2 = NULL;
1224        ext4_lblk_t max_block;
1225        __le32 nr = 0, nr2 = 0;
1226        int n = 0, n2 = 0;
1227        unsigned blocksize = inode->i_sb->s_blocksize;
1228
1229        max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
1230                                        >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
1231        if (end >= max_block)
1232                end = max_block;
1233        if ((start >= end) || (start > max_block))
1234                return 0;
1235
1236        n = ext4_block_to_path(inode, start, offsets, NULL);
1237        n2 = ext4_block_to_path(inode, end, offsets2, NULL);
1238
1239        BUG_ON(n > n2);
1240
1241        if ((n == 1) && (n == n2)) {
1242                /* We're punching only within direct block range */
1243                ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1244                               i_data + offsets2[0]);
1245                return 0;
1246        } else if (n2 > n) {
1247                /*
1248                 * Start and end are on a different levels so we're going to
1249                 * free partial block at start, and partial block at end of
1250                 * the range. If there are some levels in between then
1251                 * do_indirects label will take care of that.
1252                 */
1253
1254                if (n == 1) {
1255                        /*
1256                         * Start is at the direct block level, free
1257                         * everything to the end of the level.
1258                         */
1259                        ext4_free_data(handle, inode, NULL, i_data + offsets[0],
1260                                       i_data + EXT4_NDIR_BLOCKS);
1261                        goto end_range;
1262                }
1263
1264
1265                partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1266                if (nr) {
1267                        if (partial == chain) {
1268                                /* Shared branch grows from the inode */
1269                                ext4_free_branches(handle, inode, NULL,
1270                                           &nr, &nr+1, (chain+n-1) - partial);
1271                                *partial->p = 0;
1272                        } else {
1273                                /* Shared branch grows from an indirect block */
1274                                BUFFER_TRACE(partial->bh, "get_write_access");
1275                                ext4_free_branches(handle, inode, partial->bh,
1276                                        partial->p,
1277                                        partial->p+1, (chain+n-1) - partial);
1278                        }
1279                }
1280
1281                /*
1282                 * Clear the ends of indirect blocks on the shared branch
1283                 * at the start of the range
1284                 */
1285                while (partial > chain) {
1286                        ext4_free_branches(handle, inode, partial->bh,
1287                                partial->p + 1,
1288                                (__le32 *)partial->bh->b_data+addr_per_block,
1289                                (chain+n-1) - partial);
1290                        partial--;
1291                }
1292
1293end_range:
1294                partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1295                if (nr2) {
1296                        if (partial2 == chain2) {
1297                                /*
1298                                 * Remember, end is exclusive so here we're at
1299                                 * the start of the next level we're not going
1300                                 * to free. Everything was covered by the start
1301                                 * of the range.
1302                                 */
1303                                goto do_indirects;
1304                        }
1305                } else {
1306                        /*
1307                         * ext4_find_shared returns Indirect structure which
1308                         * points to the last element which should not be
1309                         * removed by truncate. But this is end of the range
1310                         * in punch_hole so we need to point to the next element
1311                         */
1312                        partial2->p++;
1313                }
1314
1315                /*
1316                 * Clear the ends of indirect blocks on the shared branch
1317                 * at the end of the range
1318                 */
1319                while (partial2 > chain2) {
1320                        ext4_free_branches(handle, inode, partial2->bh,
1321                                           (__le32 *)partial2->bh->b_data,
1322                                           partial2->p,
1323                                           (chain2+n2-1) - partial2);
1324                        partial2--;
1325                }
1326                goto do_indirects;
1327        }
1328
1329        /* Punch happened within the same level (n == n2) */
1330        partial = p = ext4_find_shared(inode, n, offsets, chain, &nr);
1331        partial2 = p2 = ext4_find_shared(inode, n2, offsets2, chain2, &nr2);
1332
1333        /* Free top, but only if partial2 isn't its subtree. */
1334        if (nr) {
1335                int level = min(partial - chain, partial2 - chain2);
1336                int i;
1337                int subtree = 1;
1338
1339                for (i = 0; i <= level; i++) {
1340                        if (offsets[i] != offsets2[i]) {
1341                                subtree = 0;
1342                                break;
1343                        }
1344                }
1345
1346                if (!subtree) {
1347                        if (partial == chain) {
1348                                /* Shared branch grows from the inode */
1349                                ext4_free_branches(handle, inode, NULL,
1350                                                   &nr, &nr+1,
1351                                                   (chain+n-1) - partial);
1352                                *partial->p = 0;
1353                        } else {
1354                                /* Shared branch grows from an indirect block */
1355                                BUFFER_TRACE(partial->bh, "get_write_access");
1356                                ext4_free_branches(handle, inode, partial->bh,
1357                                                   partial->p,
1358                                                   partial->p+1,
1359                                                   (chain+n-1) - partial);
1360                        }
1361                }
1362        }
1363
1364        if (!nr2) {
1365                /*
1366                 * ext4_find_shared returns Indirect structure which
1367                 * points to the last element which should not be
1368                 * removed by truncate. But this is end of the range
1369                 * in punch_hole so we need to point to the next element
1370                 */
1371                partial2->p++;
1372        }
1373
1374        while (partial > chain || partial2 > chain2) {
1375                int depth = (chain+n-1) - partial;
1376                int depth2 = (chain2+n2-1) - partial2;
1377
1378                if (partial > chain && partial2 > chain2 &&
1379                    partial->bh->b_blocknr == partial2->bh->b_blocknr) {
1380                        /*
1381                         * We've converged on the same block. Clear the range,
1382                         * then we're done.
1383                         */
1384                        ext4_free_branches(handle, inode, partial->bh,
1385                                           partial->p + 1,
1386                                           partial2->p,
1387                                           (chain+n-1) - partial);
1388                        goto cleanup;
1389                }
1390
1391                /*
1392                 * The start and end partial branches may not be at the same
1393                 * level even though the punch happened within one level. So, we
1394                 * give them a chance to arrive at the same level, then walk
1395                 * them in step with each other until we converge on the same
1396                 * block.
1397                 */
1398                if (partial > chain && depth <= depth2) {
1399                        ext4_free_branches(handle, inode, partial->bh,
1400                                           partial->p + 1,
1401                                           (__le32 *)partial->bh->b_data+addr_per_block,
1402                                           (chain+n-1) - partial);
1403                        partial--;
1404                }
1405                if (partial2 > chain2 && depth2 <= depth) {
1406                        ext4_free_branches(handle, inode, partial2->bh,
1407                                           (__le32 *)partial2->bh->b_data,
1408                                           partial2->p,
1409                                           (chain2+n2-1) - partial2);
1410                        partial2--;
1411                }
1412        }
1413
1414cleanup:
1415        while (p && p > chain) {
1416                BUFFER_TRACE(p->bh, "call brelse");
1417                brelse(p->bh);
1418                p--;
1419        }
1420        while (p2 && p2 > chain2) {
1421                BUFFER_TRACE(p2->bh, "call brelse");
1422                brelse(p2->bh);
1423                p2--;
1424        }
1425        return 0;
1426
1427do_indirects:
1428        /* Kill the remaining (whole) subtrees */
1429        switch (offsets[0]) {
1430        default:
1431                if (++n >= n2)
1432                        break;
1433                nr = i_data[EXT4_IND_BLOCK];
1434                if (nr) {
1435                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
1436                        i_data[EXT4_IND_BLOCK] = 0;
1437                }
1438                fallthrough;
1439        case EXT4_IND_BLOCK:
1440                if (++n >= n2)
1441                        break;
1442                nr = i_data[EXT4_DIND_BLOCK];
1443                if (nr) {
1444                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
1445                        i_data[EXT4_DIND_BLOCK] = 0;
1446                }
1447                fallthrough;
1448        case EXT4_DIND_BLOCK:
1449                if (++n >= n2)
1450                        break;
1451                nr = i_data[EXT4_TIND_BLOCK];
1452                if (nr) {
1453                        ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
1454                        i_data[EXT4_TIND_BLOCK] = 0;
1455                }
1456                fallthrough;
1457        case EXT4_TIND_BLOCK:
1458                ;
1459        }
1460        goto cleanup;
1461}
1462