linux/fs/ext3/inode.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/ext3/inode.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  from
  10 *
  11 *  linux/fs/minix/inode.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  Goal-directed block allocation by Stephen Tweedie
  16 *      (sct@redhat.com), 1993, 1998
  17 *  Big-endian to little-endian byte-swapping/bitmaps by
  18 *        David S. Miller (davem@caip.rutgers.edu), 1995
  19 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  20 *      (jj@sunsite.ms.mff.cuni.cz)
  21 *
  22 *  Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
  23 */
  24
  25#include <linux/highuid.h>
  26#include <linux/quotaops.h>
  27#include <linux/writeback.h>
  28#include <linux/mpage.h>
  29#include <linux/namei.h>
  30#include <linux/aio.h>
  31#include "ext3.h"
  32#include "xattr.h"
  33#include "acl.h"
  34
  35static int ext3_writepage_trans_blocks(struct inode *inode);
  36static int ext3_block_truncate_page(struct inode *inode, loff_t from);
  37
  38/*
  39 * Test whether an inode is a fast symlink.
  40 */
  41static int ext3_inode_is_fast_symlink(struct inode *inode)
  42{
  43        int ea_blocks = EXT3_I(inode)->i_file_acl ?
  44                (inode->i_sb->s_blocksize >> 9) : 0;
  45
  46        return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
  47}
  48
  49/*
  50 * The ext3 forget function must perform a revoke if we are freeing data
  51 * which has been journaled.  Metadata (eg. indirect blocks) must be
  52 * revoked in all cases.
  53 *
  54 * "bh" may be NULL: a metadata block may have been freed from memory
  55 * but there may still be a record of it in the journal, and that record
  56 * still needs to be revoked.
  57 */
  58int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
  59                        struct buffer_head *bh, ext3_fsblk_t blocknr)
  60{
  61        int err;
  62
  63        might_sleep();
  64
  65        trace_ext3_forget(inode, is_metadata, blocknr);
  66        BUFFER_TRACE(bh, "enter");
  67
  68        jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
  69                  "data mode %lx\n",
  70                  bh, is_metadata, inode->i_mode,
  71                  test_opt(inode->i_sb, DATA_FLAGS));
  72
  73        /* Never use the revoke function if we are doing full data
  74         * journaling: there is no need to, and a V1 superblock won't
  75         * support it.  Otherwise, only skip the revoke on un-journaled
  76         * data blocks. */
  77
  78        if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
  79            (!is_metadata && !ext3_should_journal_data(inode))) {
  80                if (bh) {
  81                        BUFFER_TRACE(bh, "call journal_forget");
  82                        return ext3_journal_forget(handle, bh);
  83                }
  84                return 0;
  85        }
  86
  87        /*
  88         * data!=journal && (is_metadata || should_journal_data(inode))
  89         */
  90        BUFFER_TRACE(bh, "call ext3_journal_revoke");
  91        err = ext3_journal_revoke(handle, blocknr, bh);
  92        if (err)
  93                ext3_abort(inode->i_sb, __func__,
  94                           "error %d when attempting revoke", err);
  95        BUFFER_TRACE(bh, "exit");
  96        return err;
  97}
  98
  99/*
 100 * Work out how many blocks we need to proceed with the next chunk of a
 101 * truncate transaction.
 102 */
 103static unsigned long blocks_for_truncate(struct inode *inode)
 104{
 105        unsigned long needed;
 106
 107        needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
 108
 109        /* Give ourselves just enough room to cope with inodes in which
 110         * i_blocks is corrupt: we've seen disk corruptions in the past
 111         * which resulted in random data in an inode which looked enough
 112         * like a regular file for ext3 to try to delete it.  Things
 113         * will go a bit crazy if that happens, but at least we should
 114         * try not to panic the whole kernel. */
 115        if (needed < 2)
 116                needed = 2;
 117
 118        /* But we need to bound the transaction so we don't overflow the
 119         * journal. */
 120        if (needed > EXT3_MAX_TRANS_DATA)
 121                needed = EXT3_MAX_TRANS_DATA;
 122
 123        return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
 124}
 125
 126/*
 127 * Truncate transactions can be complex and absolutely huge.  So we need to
 128 * be able to restart the transaction at a conventient checkpoint to make
 129 * sure we don't overflow the journal.
 130 *
 131 * start_transaction gets us a new handle for a truncate transaction,
 132 * and extend_transaction tries to extend the existing one a bit.  If
 133 * extend fails, we need to propagate the failure up and restart the
 134 * transaction in the top-level truncate loop. --sct
 135 */
 136static handle_t *start_transaction(struct inode *inode)
 137{
 138        handle_t *result;
 139
 140        result = ext3_journal_start(inode, blocks_for_truncate(inode));
 141        if (!IS_ERR(result))
 142                return result;
 143
 144        ext3_std_error(inode->i_sb, PTR_ERR(result));
 145        return result;
 146}
 147
 148/*
 149 * Try to extend this transaction for the purposes of truncation.
 150 *
 151 * Returns 0 if we managed to create more room.  If we can't create more
 152 * room, and the transaction must be restarted we return 1.
 153 */
 154static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
 155{
 156        if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
 157                return 0;
 158        if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
 159                return 0;
 160        return 1;
 161}
 162
 163/*
 164 * Restart the transaction associated with *handle.  This does a commit,
 165 * so before we call here everything must be consistently dirtied against
 166 * this transaction.
 167 */
 168static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
 169{
 170        int ret;
 171
 172        jbd_debug(2, "restarting handle %p\n", handle);
 173        /*
 174         * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
 175         * At this moment, get_block can be called only for blocks inside
 176         * i_size since page cache has been already dropped and writes are
 177         * blocked by i_mutex. So we can safely drop the truncate_mutex.
 178         */
 179        mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 180        ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
 181        mutex_lock(&EXT3_I(inode)->truncate_mutex);
 182        return ret;
 183}
 184
 185/*
 186 * Called at inode eviction from icache
 187 */
 188void ext3_evict_inode (struct inode *inode)
 189{
 190        struct ext3_inode_info *ei = EXT3_I(inode);
 191        struct ext3_block_alloc_info *rsv;
 192        handle_t *handle;
 193        int want_delete = 0;
 194
 195        trace_ext3_evict_inode(inode);
 196        if (!inode->i_nlink && !is_bad_inode(inode)) {
 197                dquot_initialize(inode);
 198                want_delete = 1;
 199        }
 200
 201        /*
 202         * When journalling data dirty buffers are tracked only in the journal.
 203         * So although mm thinks everything is clean and ready for reaping the
 204         * inode might still have some pages to write in the running
 205         * transaction or waiting to be checkpointed. Thus calling
 206         * journal_invalidatepage() (via truncate_inode_pages()) to discard
 207         * these buffers can cause data loss. Also even if we did not discard
 208         * these buffers, we would have no way to find them after the inode
 209         * is reaped and thus user could see stale data if he tries to read
 210         * them before the transaction is checkpointed. So be careful and
 211         * force everything to disk here... We use ei->i_datasync_tid to
 212         * store the newest transaction containing inode's data.
 213         *
 214         * Note that directories do not have this problem because they don't
 215         * use page cache.
 216         *
 217         * The s_journal check handles the case when ext3_get_journal() fails
 218         * and puts the journal inode.
 219         */
 220        if (inode->i_nlink && ext3_should_journal_data(inode) &&
 221            EXT3_SB(inode->i_sb)->s_journal &&
 222            (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
 223            inode->i_ino != EXT3_JOURNAL_INO) {
 224                tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
 225                journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
 226
 227                log_start_commit(journal, commit_tid);
 228                log_wait_commit(journal, commit_tid);
 229                filemap_write_and_wait(&inode->i_data);
 230        }
 231        truncate_inode_pages(&inode->i_data, 0);
 232
 233        ext3_discard_reservation(inode);
 234        rsv = ei->i_block_alloc_info;
 235        ei->i_block_alloc_info = NULL;
 236        if (unlikely(rsv))
 237                kfree(rsv);
 238
 239        if (!want_delete)
 240                goto no_delete;
 241
 242        handle = start_transaction(inode);
 243        if (IS_ERR(handle)) {
 244                /*
 245                 * If we're going to skip the normal cleanup, we still need to
 246                 * make sure that the in-core orphan linked list is properly
 247                 * cleaned up.
 248                 */
 249                ext3_orphan_del(NULL, inode);
 250                goto no_delete;
 251        }
 252
 253        if (IS_SYNC(inode))
 254                handle->h_sync = 1;
 255        inode->i_size = 0;
 256        if (inode->i_blocks)
 257                ext3_truncate(inode);
 258        /*
 259         * Kill off the orphan record created when the inode lost the last
 260         * link.  Note that ext3_orphan_del() has to be able to cope with the
 261         * deletion of a non-existent orphan - ext3_truncate() could
 262         * have removed the record.
 263         */
 264        ext3_orphan_del(handle, inode);
 265        ei->i_dtime = get_seconds();
 266
 267        /*
 268         * One subtle ordering requirement: if anything has gone wrong
 269         * (transaction abort, IO errors, whatever), then we can still
 270         * do these next steps (the fs will already have been marked as
 271         * having errors), but we can't free the inode if the mark_dirty
 272         * fails.
 273         */
 274        if (ext3_mark_inode_dirty(handle, inode)) {
 275                /* If that failed, just dquot_drop() and be done with that */
 276                dquot_drop(inode);
 277                clear_inode(inode);
 278        } else {
 279                ext3_xattr_delete_inode(handle, inode);
 280                dquot_free_inode(inode);
 281                dquot_drop(inode);
 282                clear_inode(inode);
 283                ext3_free_inode(handle, inode);
 284        }
 285        ext3_journal_stop(handle);
 286        return;
 287no_delete:
 288        clear_inode(inode);
 289        dquot_drop(inode);
 290}
 291
 292typedef struct {
 293        __le32  *p;
 294        __le32  key;
 295        struct buffer_head *bh;
 296} Indirect;
 297
 298static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
 299{
 300        p->key = *(p->p = v);
 301        p->bh = bh;
 302}
 303
 304static int verify_chain(Indirect *from, Indirect *to)
 305{
 306        while (from <= to && from->key == *from->p)
 307                from++;
 308        return (from > to);
 309}
 310
 311/**
 312 *      ext3_block_to_path - parse the block number into array of offsets
 313 *      @inode: inode in question (we are only interested in its superblock)
 314 *      @i_block: block number to be parsed
 315 *      @offsets: array to store the offsets in
 316 *      @boundary: set this non-zero if the referred-to block is likely to be
 317 *             followed (on disk) by an indirect block.
 318 *
 319 *      To store the locations of file's data ext3 uses a data structure common
 320 *      for UNIX filesystems - tree of pointers anchored in the inode, with
 321 *      data blocks at leaves and indirect blocks in intermediate nodes.
 322 *      This function translates the block number into path in that tree -
 323 *      return value is the path length and @offsets[n] is the offset of
 324 *      pointer to (n+1)th node in the nth one. If @block is out of range
 325 *      (negative or too large) warning is printed and zero returned.
 326 *
 327 *      Note: function doesn't find node addresses, so no IO is needed. All
 328 *      we need to know is the capacity of indirect blocks (taken from the
 329 *      inode->i_sb).
 330 */
 331
 332/*
 333 * Portability note: the last comparison (check that we fit into triple
 334 * indirect block) is spelled differently, because otherwise on an
 335 * architecture with 32-bit longs and 8Kb pages we might get into trouble
 336 * if our filesystem had 8Kb blocks. We might use long long, but that would
 337 * kill us on x86. Oh, well, at least the sign propagation does not matter -
 338 * i_block would have to be negative in the very beginning, so we would not
 339 * get there at all.
 340 */
 341
 342static int ext3_block_to_path(struct inode *inode,
 343                        long i_block, int offsets[4], int *boundary)
 344{
 345        int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
 346        int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
 347        const long direct_blocks = EXT3_NDIR_BLOCKS,
 348                indirect_blocks = ptrs,
 349                double_blocks = (1 << (ptrs_bits * 2));
 350        int n = 0;
 351        int final = 0;
 352
 353        if (i_block < 0) {
 354                ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
 355        } else if (i_block < direct_blocks) {
 356                offsets[n++] = i_block;
 357                final = direct_blocks;
 358        } else if ( (i_block -= direct_blocks) < indirect_blocks) {
 359                offsets[n++] = EXT3_IND_BLOCK;
 360                offsets[n++] = i_block;
 361                final = ptrs;
 362        } else if ((i_block -= indirect_blocks) < double_blocks) {
 363                offsets[n++] = EXT3_DIND_BLOCK;
 364                offsets[n++] = i_block >> ptrs_bits;
 365                offsets[n++] = i_block & (ptrs - 1);
 366                final = ptrs;
 367        } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
 368                offsets[n++] = EXT3_TIND_BLOCK;
 369                offsets[n++] = i_block >> (ptrs_bits * 2);
 370                offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
 371                offsets[n++] = i_block & (ptrs - 1);
 372                final = ptrs;
 373        } else {
 374                ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
 375        }
 376        if (boundary)
 377                *boundary = final - 1 - (i_block & (ptrs - 1));
 378        return n;
 379}
 380
 381/**
 382 *      ext3_get_branch - read the chain of indirect blocks leading to data
 383 *      @inode: inode in question
 384 *      @depth: depth of the chain (1 - direct pointer, etc.)
 385 *      @offsets: offsets of pointers in inode/indirect blocks
 386 *      @chain: place to store the result
 387 *      @err: here we store the error value
 388 *
 389 *      Function fills the array of triples <key, p, bh> and returns %NULL
 390 *      if everything went OK or the pointer to the last filled triple
 391 *      (incomplete one) otherwise. Upon the return chain[i].key contains
 392 *      the number of (i+1)-th block in the chain (as it is stored in memory,
 393 *      i.e. little-endian 32-bit), chain[i].p contains the address of that
 394 *      number (it points into struct inode for i==0 and into the bh->b_data
 395 *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 396 *      block for i>0 and NULL for i==0. In other words, it holds the block
 397 *      numbers of the chain, addresses they were taken from (and where we can
 398 *      verify that chain did not change) and buffer_heads hosting these
 399 *      numbers.
 400 *
 401 *      Function stops when it stumbles upon zero pointer (absent block)
 402 *              (pointer to last triple returned, *@err == 0)
 403 *      or when it gets an IO error reading an indirect block
 404 *              (ditto, *@err == -EIO)
 405 *      or when it notices that chain had been changed while it was reading
 406 *              (ditto, *@err == -EAGAIN)
 407 *      or when it reads all @depth-1 indirect blocks successfully and finds
 408 *      the whole chain, all way to the data (returns %NULL, *err == 0).
 409 */
 410static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
 411                                 Indirect chain[4], int *err)
 412{
 413        struct super_block *sb = inode->i_sb;
 414        Indirect *p = chain;
 415        struct buffer_head *bh;
 416
 417        *err = 0;
 418        /* i_data is not going away, no lock needed */
 419        add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
 420        if (!p->key)
 421                goto no_block;
 422        while (--depth) {
 423                bh = sb_bread(sb, le32_to_cpu(p->key));
 424                if (!bh)
 425                        goto failure;
 426                /* Reader: pointers */
 427                if (!verify_chain(chain, p))
 428                        goto changed;
 429                add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
 430                /* Reader: end */
 431                if (!p->key)
 432                        goto no_block;
 433        }
 434        return NULL;
 435
 436changed:
 437        brelse(bh);
 438        *err = -EAGAIN;
 439        goto no_block;
 440failure:
 441        *err = -EIO;
 442no_block:
 443        return p;
 444}
 445
 446/**
 447 *      ext3_find_near - find a place for allocation with sufficient locality
 448 *      @inode: owner
 449 *      @ind: descriptor of indirect block.
 450 *
 451 *      This function returns the preferred place for block allocation.
 452 *      It is used when heuristic for sequential allocation fails.
 453 *      Rules are:
 454 *        + if there is a block to the left of our position - allocate near it.
 455 *        + if pointer will live in indirect block - allocate near that block.
 456 *        + if pointer will live in inode - allocate in the same
 457 *          cylinder group.
 458 *
 459 * In the latter case we colour the starting block by the callers PID to
 460 * prevent it from clashing with concurrent allocations for a different inode
 461 * in the same block group.   The PID is used here so that functionally related
 462 * files will be close-by on-disk.
 463 *
 464 *      Caller must make sure that @ind is valid and will stay that way.
 465 */
 466static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
 467{
 468        struct ext3_inode_info *ei = EXT3_I(inode);
 469        __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
 470        __le32 *p;
 471        ext3_fsblk_t bg_start;
 472        ext3_grpblk_t colour;
 473
 474        /* Try to find previous block */
 475        for (p = ind->p - 1; p >= start; p--) {
 476                if (*p)
 477                        return le32_to_cpu(*p);
 478        }
 479
 480        /* No such thing, so let's try location of indirect block */
 481        if (ind->bh)
 482                return ind->bh->b_blocknr;
 483
 484        /*
 485         * It is going to be referred to from the inode itself? OK, just put it
 486         * into the same cylinder group then.
 487         */
 488        bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
 489        colour = (current->pid % 16) *
 490                        (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 491        return bg_start + colour;
 492}
 493
 494/**
 495 *      ext3_find_goal - find a preferred place for allocation.
 496 *      @inode: owner
 497 *      @block:  block we want
 498 *      @partial: pointer to the last triple within a chain
 499 *
 500 *      Normally this function find the preferred place for block allocation,
 501 *      returns it.
 502 */
 503
 504static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
 505                                   Indirect *partial)
 506{
 507        struct ext3_block_alloc_info *block_i;
 508
 509        block_i =  EXT3_I(inode)->i_block_alloc_info;
 510
 511        /*
 512         * try the heuristic for sequential allocation,
 513         * failing that at least try to get decent locality.
 514         */
 515        if (block_i && (block == block_i->last_alloc_logical_block + 1)
 516                && (block_i->last_alloc_physical_block != 0)) {
 517                return block_i->last_alloc_physical_block + 1;
 518        }
 519
 520        return ext3_find_near(inode, partial);
 521}
 522
 523/**
 524 *      ext3_blks_to_allocate - Look up the block map and count the number
 525 *      of direct blocks need to be allocated for the given branch.
 526 *
 527 *      @branch: chain of indirect blocks
 528 *      @k: number of blocks need for indirect blocks
 529 *      @blks: number of data blocks to be mapped.
 530 *      @blocks_to_boundary:  the offset in the indirect block
 531 *
 532 *      return the total number of blocks to be allocate, including the
 533 *      direct and indirect blocks.
 534 */
 535static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
 536                int blocks_to_boundary)
 537{
 538        unsigned long count = 0;
 539
 540        /*
 541         * Simple case, [t,d]Indirect block(s) has not allocated yet
 542         * then it's clear blocks on that path have not allocated
 543         */
 544        if (k > 0) {
 545                /* right now we don't handle cross boundary allocation */
 546                if (blks < blocks_to_boundary + 1)
 547                        count += blks;
 548                else
 549                        count += blocks_to_boundary + 1;
 550                return count;
 551        }
 552
 553        count++;
 554        while (count < blks && count <= blocks_to_boundary &&
 555                le32_to_cpu(*(branch[0].p + count)) == 0) {
 556                count++;
 557        }
 558        return count;
 559}
 560
 561/**
 562 *      ext3_alloc_blocks - multiple allocate blocks needed for a branch
 563 *      @handle: handle for this transaction
 564 *      @inode: owner
 565 *      @goal: preferred place for allocation
 566 *      @indirect_blks: the number of blocks need to allocate for indirect
 567 *                      blocks
 568 *      @blks:  number of blocks need to allocated for direct blocks
 569 *      @new_blocks: on return it will store the new block numbers for
 570 *      the indirect blocks(if needed) and the first direct block,
 571 *      @err: here we store the error value
 572 *
 573 *      return the number of direct blocks allocated
 574 */
 575static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
 576                        ext3_fsblk_t goal, int indirect_blks, int blks,
 577                        ext3_fsblk_t new_blocks[4], int *err)
 578{
 579        int target, i;
 580        unsigned long count = 0;
 581        int index = 0;
 582        ext3_fsblk_t current_block = 0;
 583        int ret = 0;
 584
 585        /*
 586         * Here we try to allocate the requested multiple blocks at once,
 587         * on a best-effort basis.
 588         * To build a branch, we should allocate blocks for
 589         * the indirect blocks(if not allocated yet), and at least
 590         * the first direct block of this branch.  That's the
 591         * minimum number of blocks need to allocate(required)
 592         */
 593        target = blks + indirect_blks;
 594
 595        while (1) {
 596                count = target;
 597                /* allocating blocks for indirect blocks and direct blocks */
 598                current_block = ext3_new_blocks(handle,inode,goal,&count,err);
 599                if (*err)
 600                        goto failed_out;
 601
 602                target -= count;
 603                /* allocate blocks for indirect blocks */
 604                while (index < indirect_blks && count) {
 605                        new_blocks[index++] = current_block++;
 606                        count--;
 607                }
 608
 609                if (count > 0)
 610                        break;
 611        }
 612
 613        /* save the new block number for the first direct block */
 614        new_blocks[index] = current_block;
 615
 616        /* total number of blocks allocated for direct blocks */
 617        ret = count;
 618        *err = 0;
 619        return ret;
 620failed_out:
 621        for (i = 0; i <index; i++)
 622                ext3_free_blocks(handle, inode, new_blocks[i], 1);
 623        return ret;
 624}
 625
 626/**
 627 *      ext3_alloc_branch - allocate and set up a chain of blocks.
 628 *      @handle: handle for this transaction
 629 *      @inode: owner
 630 *      @indirect_blks: number of allocated indirect blocks
 631 *      @blks: number of allocated direct blocks
 632 *      @goal: preferred place for allocation
 633 *      @offsets: offsets (in the blocks) to store the pointers to next.
 634 *      @branch: place to store the chain in.
 635 *
 636 *      This function allocates blocks, zeroes out all but the last one,
 637 *      links them into chain and (if we are synchronous) writes them to disk.
 638 *      In other words, it prepares a branch that can be spliced onto the
 639 *      inode. It stores the information about that chain in the branch[], in
 640 *      the same format as ext3_get_branch() would do. We are calling it after
 641 *      we had read the existing part of chain and partial points to the last
 642 *      triple of that (one with zero ->key). Upon the exit we have the same
 643 *      picture as after the successful ext3_get_block(), except that in one
 644 *      place chain is disconnected - *branch->p is still zero (we did not
 645 *      set the last link), but branch->key contains the number that should
 646 *      be placed into *branch->p to fill that gap.
 647 *
 648 *      If allocation fails we free all blocks we've allocated (and forget
 649 *      their buffer_heads) and return the error value the from failed
 650 *      ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 651 *      as described above and return 0.
 652 */
 653static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
 654                        int indirect_blks, int *blks, ext3_fsblk_t goal,
 655                        int *offsets, Indirect *branch)
 656{
 657        int blocksize = inode->i_sb->s_blocksize;
 658        int i, n = 0;
 659        int err = 0;
 660        struct buffer_head *bh;
 661        int num;
 662        ext3_fsblk_t new_blocks[4];
 663        ext3_fsblk_t current_block;
 664
 665        num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
 666                                *blks, new_blocks, &err);
 667        if (err)
 668                return err;
 669
 670        branch[0].key = cpu_to_le32(new_blocks[0]);
 671        /*
 672         * metadata blocks and data blocks are allocated.
 673         */
 674        for (n = 1; n <= indirect_blks;  n++) {
 675                /*
 676                 * Get buffer_head for parent block, zero it out
 677                 * and set the pointer to new one, then send
 678                 * parent to disk.
 679                 */
 680                bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
 681                if (unlikely(!bh)) {
 682                        err = -ENOMEM;
 683                        goto failed;
 684                }
 685                branch[n].bh = bh;
 686                lock_buffer(bh);
 687                BUFFER_TRACE(bh, "call get_create_access");
 688                err = ext3_journal_get_create_access(handle, bh);
 689                if (err) {
 690                        unlock_buffer(bh);
 691                        brelse(bh);
 692                        goto failed;
 693                }
 694
 695                memset(bh->b_data, 0, blocksize);
 696                branch[n].p = (__le32 *) bh->b_data + offsets[n];
 697                branch[n].key = cpu_to_le32(new_blocks[n]);
 698                *branch[n].p = branch[n].key;
 699                if ( n == indirect_blks) {
 700                        current_block = new_blocks[n];
 701                        /*
 702                         * End of chain, update the last new metablock of
 703                         * the chain to point to the new allocated
 704                         * data blocks numbers
 705                         */
 706                        for (i=1; i < num; i++)
 707                                *(branch[n].p + i) = cpu_to_le32(++current_block);
 708                }
 709                BUFFER_TRACE(bh, "marking uptodate");
 710                set_buffer_uptodate(bh);
 711                unlock_buffer(bh);
 712
 713                BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
 714                err = ext3_journal_dirty_metadata(handle, bh);
 715                if (err)
 716                        goto failed;
 717        }
 718        *blks = num;
 719        return err;
 720failed:
 721        /* Allocation failed, free what we already allocated */
 722        for (i = 1; i <= n ; i++) {
 723                BUFFER_TRACE(branch[i].bh, "call journal_forget");
 724                ext3_journal_forget(handle, branch[i].bh);
 725        }
 726        for (i = 0; i < indirect_blks; i++)
 727                ext3_free_blocks(handle, inode, new_blocks[i], 1);
 728
 729        ext3_free_blocks(handle, inode, new_blocks[i], num);
 730
 731        return err;
 732}
 733
 734/**
 735 * ext3_splice_branch - splice the allocated branch onto inode.
 736 * @handle: handle for this transaction
 737 * @inode: owner
 738 * @block: (logical) number of block we are adding
 739 * @where: location of missing link
 740 * @num:   number of indirect blocks we are adding
 741 * @blks:  number of direct blocks we are adding
 742 *
 743 * This function fills the missing link and does all housekeeping needed in
 744 * inode (->i_blocks, etc.). In case of success we end up with the full
 745 * chain to new block and return 0.
 746 */
 747static int ext3_splice_branch(handle_t *handle, struct inode *inode,
 748                        long block, Indirect *where, int num, int blks)
 749{
 750        int i;
 751        int err = 0;
 752        struct ext3_block_alloc_info *block_i;
 753        ext3_fsblk_t current_block;
 754        struct ext3_inode_info *ei = EXT3_I(inode);
 755        struct timespec now;
 756
 757        block_i = ei->i_block_alloc_info;
 758        /*
 759         * If we're splicing into a [td]indirect block (as opposed to the
 760         * inode) then we need to get write access to the [td]indirect block
 761         * before the splice.
 762         */
 763        if (where->bh) {
 764                BUFFER_TRACE(where->bh, "get_write_access");
 765                err = ext3_journal_get_write_access(handle, where->bh);
 766                if (err)
 767                        goto err_out;
 768        }
 769        /* That's it */
 770
 771        *where->p = where->key;
 772
 773        /*
 774         * Update the host buffer_head or inode to point to more just allocated
 775         * direct blocks blocks
 776         */
 777        if (num == 0 && blks > 1) {
 778                current_block = le32_to_cpu(where->key) + 1;
 779                for (i = 1; i < blks; i++)
 780                        *(where->p + i ) = cpu_to_le32(current_block++);
 781        }
 782
 783        /*
 784         * update the most recently allocated logical & physical block
 785         * in i_block_alloc_info, to assist find the proper goal block for next
 786         * allocation
 787         */
 788        if (block_i) {
 789                block_i->last_alloc_logical_block = block + blks - 1;
 790                block_i->last_alloc_physical_block =
 791                                le32_to_cpu(where[num].key) + blks - 1;
 792        }
 793
 794        /* We are done with atomic stuff, now do the rest of housekeeping */
 795        now = CURRENT_TIME_SEC;
 796        if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
 797                inode->i_ctime = now;
 798                ext3_mark_inode_dirty(handle, inode);
 799        }
 800        /* ext3_mark_inode_dirty already updated i_sync_tid */
 801        atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
 802
 803        /* had we spliced it onto indirect block? */
 804        if (where->bh) {
 805                /*
 806                 * If we spliced it onto an indirect block, we haven't
 807                 * altered the inode.  Note however that if it is being spliced
 808                 * onto an indirect block at the very end of the file (the
 809                 * file is growing) then we *will* alter the inode to reflect
 810                 * the new i_size.  But that is not done here - it is done in
 811                 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
 812                 */
 813                jbd_debug(5, "splicing indirect only\n");
 814                BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
 815                err = ext3_journal_dirty_metadata(handle, where->bh);
 816                if (err)
 817                        goto err_out;
 818        } else {
 819                /*
 820                 * OK, we spliced it into the inode itself on a direct block.
 821                 * Inode was dirtied above.
 822                 */
 823                jbd_debug(5, "splicing direct\n");
 824        }
 825        return err;
 826
 827err_out:
 828        for (i = 1; i <= num; i++) {
 829                BUFFER_TRACE(where[i].bh, "call journal_forget");
 830                ext3_journal_forget(handle, where[i].bh);
 831                ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
 832        }
 833        ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
 834
 835        return err;
 836}
 837
 838/*
 839 * Allocation strategy is simple: if we have to allocate something, we will
 840 * have to go the whole way to leaf. So let's do it before attaching anything
 841 * to tree, set linkage between the newborn blocks, write them if sync is
 842 * required, recheck the path, free and repeat if check fails, otherwise
 843 * set the last missing link (that will protect us from any truncate-generated
 844 * removals - all blocks on the path are immune now) and possibly force the
 845 * write on the parent block.
 846 * That has a nice additional property: no special recovery from the failed
 847 * allocations is needed - we simply release blocks and do not touch anything
 848 * reachable from inode.
 849 *
 850 * `handle' can be NULL if create == 0.
 851 *
 852 * The BKL may not be held on entry here.  Be sure to take it early.
 853 * return > 0, # of blocks mapped or allocated.
 854 * return = 0, if plain lookup failed.
 855 * return < 0, error case.
 856 */
 857int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
 858                sector_t iblock, unsigned long maxblocks,
 859                struct buffer_head *bh_result,
 860                int create)
 861{
 862        int err = -EIO;
 863        int offsets[4];
 864        Indirect chain[4];
 865        Indirect *partial;
 866        ext3_fsblk_t goal;
 867        int indirect_blks;
 868        int blocks_to_boundary = 0;
 869        int depth;
 870        struct ext3_inode_info *ei = EXT3_I(inode);
 871        int count = 0;
 872        ext3_fsblk_t first_block = 0;
 873
 874
 875        trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
 876        J_ASSERT(handle != NULL || create == 0);
 877        depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
 878
 879        if (depth == 0)
 880                goto out;
 881
 882        partial = ext3_get_branch(inode, depth, offsets, chain, &err);
 883
 884        /* Simplest case - block found, no allocation needed */
 885        if (!partial) {
 886                first_block = le32_to_cpu(chain[depth - 1].key);
 887                clear_buffer_new(bh_result);
 888                count++;
 889                /*map more blocks*/
 890                while (count < maxblocks && count <= blocks_to_boundary) {
 891                        ext3_fsblk_t blk;
 892
 893                        if (!verify_chain(chain, chain + depth - 1)) {
 894                                /*
 895                                 * Indirect block might be removed by
 896                                 * truncate while we were reading it.
 897                                 * Handling of that case: forget what we've
 898                                 * got now. Flag the err as EAGAIN, so it
 899                                 * will reread.
 900                                 */
 901                                err = -EAGAIN;
 902                                count = 0;
 903                                break;
 904                        }
 905                        blk = le32_to_cpu(*(chain[depth-1].p + count));
 906
 907                        if (blk == first_block + count)
 908                                count++;
 909                        else
 910                                break;
 911                }
 912                if (err != -EAGAIN)
 913                        goto got_it;
 914        }
 915
 916        /* Next simple case - plain lookup or failed read of indirect block */
 917        if (!create || err == -EIO)
 918                goto cleanup;
 919
 920        /*
 921         * Block out ext3_truncate while we alter the tree
 922         */
 923        mutex_lock(&ei->truncate_mutex);
 924
 925        /*
 926         * If the indirect block is missing while we are reading
 927         * the chain(ext3_get_branch() returns -EAGAIN err), or
 928         * if the chain has been changed after we grab the semaphore,
 929         * (either because another process truncated this branch, or
 930         * another get_block allocated this branch) re-grab the chain to see if
 931         * the request block has been allocated or not.
 932         *
 933         * Since we already block the truncate/other get_block
 934         * at this point, we will have the current copy of the chain when we
 935         * splice the branch into the tree.
 936         */
 937        if (err == -EAGAIN || !verify_chain(chain, partial)) {
 938                while (partial > chain) {
 939                        brelse(partial->bh);
 940                        partial--;
 941                }
 942                partial = ext3_get_branch(inode, depth, offsets, chain, &err);
 943                if (!partial) {
 944                        count++;
 945                        mutex_unlock(&ei->truncate_mutex);
 946                        if (err)
 947                                goto cleanup;
 948                        clear_buffer_new(bh_result);
 949                        goto got_it;
 950                }
 951        }
 952
 953        /*
 954         * Okay, we need to do block allocation.  Lazily initialize the block
 955         * allocation info here if necessary
 956        */
 957        if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
 958                ext3_init_block_alloc_info(inode);
 959
 960        goal = ext3_find_goal(inode, iblock, partial);
 961
 962        /* the number of blocks need to allocate for [d,t]indirect blocks */
 963        indirect_blks = (chain + depth) - partial - 1;
 964
 965        /*
 966         * Next look up the indirect map to count the totoal number of
 967         * direct blocks to allocate for this branch.
 968         */
 969        count = ext3_blks_to_allocate(partial, indirect_blks,
 970                                        maxblocks, blocks_to_boundary);
 971        err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
 972                                offsets + (partial - chain), partial);
 973
 974        /*
 975         * The ext3_splice_branch call will free and forget any buffers
 976         * on the new chain if there is a failure, but that risks using
 977         * up transaction credits, especially for bitmaps where the
 978         * credits cannot be returned.  Can we handle this somehow?  We
 979         * may need to return -EAGAIN upwards in the worst case.  --sct
 980         */
 981        if (!err)
 982                err = ext3_splice_branch(handle, inode, iblock,
 983                                        partial, indirect_blks, count);
 984        mutex_unlock(&ei->truncate_mutex);
 985        if (err)
 986                goto cleanup;
 987
 988        set_buffer_new(bh_result);
 989got_it:
 990        map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
 991        if (count > blocks_to_boundary)
 992                set_buffer_boundary(bh_result);
 993        err = count;
 994        /* Clean up and exit */
 995        partial = chain + depth - 1;    /* the whole chain */
 996cleanup:
 997        while (partial > chain) {
 998                BUFFER_TRACE(partial->bh, "call brelse");
 999                brelse(partial->bh);
1000                partial--;
1001        }
1002        BUFFER_TRACE(bh_result, "returned");
1003out:
1004        trace_ext3_get_blocks_exit(inode, iblock,
1005                                   depth ? le32_to_cpu(chain[depth-1].key) : 0,
1006                                   count, err);
1007        return err;
1008}
1009
1010/* Maximum number of blocks we map for direct IO at once. */
1011#define DIO_MAX_BLOCKS 4096
1012/*
1013 * Number of credits we need for writing DIO_MAX_BLOCKS:
1014 * We need sb + group descriptor + bitmap + inode -> 4
1015 * For B blocks with A block pointers per block we need:
1016 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1017 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1018 */
1019#define DIO_CREDITS 25
1020
1021static int ext3_get_block(struct inode *inode, sector_t iblock,
1022                        struct buffer_head *bh_result, int create)
1023{
1024        handle_t *handle = ext3_journal_current_handle();
1025        int ret = 0, started = 0;
1026        unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1027
1028        if (create && !handle) {        /* Direct IO write... */
1029                if (max_blocks > DIO_MAX_BLOCKS)
1030                        max_blocks = DIO_MAX_BLOCKS;
1031                handle = ext3_journal_start(inode, DIO_CREDITS +
1032                                EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
1033                if (IS_ERR(handle)) {
1034                        ret = PTR_ERR(handle);
1035                        goto out;
1036                }
1037                started = 1;
1038        }
1039
1040        ret = ext3_get_blocks_handle(handle, inode, iblock,
1041                                        max_blocks, bh_result, create);
1042        if (ret > 0) {
1043                bh_result->b_size = (ret << inode->i_blkbits);
1044                ret = 0;
1045        }
1046        if (started)
1047                ext3_journal_stop(handle);
1048out:
1049        return ret;
1050}
1051
1052int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1053                u64 start, u64 len)
1054{
1055        return generic_block_fiemap(inode, fieinfo, start, len,
1056                                    ext3_get_block);
1057}
1058
1059/*
1060 * `handle' can be NULL if create is zero
1061 */
1062struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1063                                long block, int create, int *errp)
1064{
1065        struct buffer_head dummy;
1066        int fatal = 0, err;
1067
1068        J_ASSERT(handle != NULL || create == 0);
1069
1070        dummy.b_state = 0;
1071        dummy.b_blocknr = -1000;
1072        buffer_trace_init(&dummy.b_history);
1073        err = ext3_get_blocks_handle(handle, inode, block, 1,
1074                                        &dummy, create);
1075        /*
1076         * ext3_get_blocks_handle() returns number of blocks
1077         * mapped. 0 in case of a HOLE.
1078         */
1079        if (err > 0) {
1080                WARN_ON(err > 1);
1081                err = 0;
1082        }
1083        *errp = err;
1084        if (!err && buffer_mapped(&dummy)) {
1085                struct buffer_head *bh;
1086                bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1087                if (unlikely(!bh)) {
1088                        *errp = -ENOMEM;
1089                        goto err;
1090                }
1091                if (buffer_new(&dummy)) {
1092                        J_ASSERT(create != 0);
1093                        J_ASSERT(handle != NULL);
1094
1095                        /*
1096                         * Now that we do not always journal data, we should
1097                         * keep in mind whether this should always journal the
1098                         * new buffer as metadata.  For now, regular file
1099                         * writes use ext3_get_block instead, so it's not a
1100                         * problem.
1101                         */
1102                        lock_buffer(bh);
1103                        BUFFER_TRACE(bh, "call get_create_access");
1104                        fatal = ext3_journal_get_create_access(handle, bh);
1105                        if (!fatal && !buffer_uptodate(bh)) {
1106                                memset(bh->b_data,0,inode->i_sb->s_blocksize);
1107                                set_buffer_uptodate(bh);
1108                        }
1109                        unlock_buffer(bh);
1110                        BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1111                        err = ext3_journal_dirty_metadata(handle, bh);
1112                        if (!fatal)
1113                                fatal = err;
1114                } else {
1115                        BUFFER_TRACE(bh, "not a new buffer");
1116                }
1117                if (fatal) {
1118                        *errp = fatal;
1119                        brelse(bh);
1120                        bh = NULL;
1121                }
1122                return bh;
1123        }
1124err:
1125        return NULL;
1126}
1127
1128struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1129                               int block, int create, int *err)
1130{
1131        struct buffer_head * bh;
1132
1133        bh = ext3_getblk(handle, inode, block, create, err);
1134        if (!bh)
1135                return bh;
1136        if (bh_uptodate_or_lock(bh))
1137                return bh;
1138        get_bh(bh);
1139        bh->b_end_io = end_buffer_read_sync;
1140        submit_bh(READ | REQ_META | REQ_PRIO, bh);
1141        wait_on_buffer(bh);
1142        if (buffer_uptodate(bh))
1143                return bh;
1144        put_bh(bh);
1145        *err = -EIO;
1146        return NULL;
1147}
1148
1149static int walk_page_buffers(   handle_t *handle,
1150                                struct buffer_head *head,
1151                                unsigned from,
1152                                unsigned to,
1153                                int *partial,
1154                                int (*fn)(      handle_t *handle,
1155                                                struct buffer_head *bh))
1156{
1157        struct buffer_head *bh;
1158        unsigned block_start, block_end;
1159        unsigned blocksize = head->b_size;
1160        int err, ret = 0;
1161        struct buffer_head *next;
1162
1163        for (   bh = head, block_start = 0;
1164                ret == 0 && (bh != head || !block_start);
1165                block_start = block_end, bh = next)
1166        {
1167                next = bh->b_this_page;
1168                block_end = block_start + blocksize;
1169                if (block_end <= from || block_start >= to) {
1170                        if (partial && !buffer_uptodate(bh))
1171                                *partial = 1;
1172                        continue;
1173                }
1174                err = (*fn)(handle, bh);
1175                if (!ret)
1176                        ret = err;
1177        }
1178        return ret;
1179}
1180
1181/*
1182 * To preserve ordering, it is essential that the hole instantiation and
1183 * the data write be encapsulated in a single transaction.  We cannot
1184 * close off a transaction and start a new one between the ext3_get_block()
1185 * and the commit_write().  So doing the journal_start at the start of
1186 * prepare_write() is the right place.
1187 *
1188 * Also, this function can nest inside ext3_writepage() ->
1189 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1190 * has generated enough buffer credits to do the whole page.  So we won't
1191 * block on the journal in that case, which is good, because the caller may
1192 * be PF_MEMALLOC.
1193 *
1194 * By accident, ext3 can be reentered when a transaction is open via
1195 * quota file writes.  If we were to commit the transaction while thus
1196 * reentered, there can be a deadlock - we would be holding a quota
1197 * lock, and the commit would never complete if another thread had a
1198 * transaction open and was blocking on the quota lock - a ranking
1199 * violation.
1200 *
1201 * So what we do is to rely on the fact that journal_stop/journal_start
1202 * will _not_ run commit under these circumstances because handle->h_ref
1203 * is elevated.  We'll still have enough credits for the tiny quotafile
1204 * write.
1205 */
1206static int do_journal_get_write_access(handle_t *handle,
1207                                        struct buffer_head *bh)
1208{
1209        int dirty = buffer_dirty(bh);
1210        int ret;
1211
1212        if (!buffer_mapped(bh) || buffer_freed(bh))
1213                return 0;
1214        /*
1215         * __block_prepare_write() could have dirtied some buffers. Clean
1216         * the dirty bit as jbd2_journal_get_write_access() could complain
1217         * otherwise about fs integrity issues. Setting of the dirty bit
1218         * by __block_prepare_write() isn't a real problem here as we clear
1219         * the bit before releasing a page lock and thus writeback cannot
1220         * ever write the buffer.
1221         */
1222        if (dirty)
1223                clear_buffer_dirty(bh);
1224        ret = ext3_journal_get_write_access(handle, bh);
1225        if (!ret && dirty)
1226                ret = ext3_journal_dirty_metadata(handle, bh);
1227        return ret;
1228}
1229
1230/*
1231 * Truncate blocks that were not used by write. We have to truncate the
1232 * pagecache as well so that corresponding buffers get properly unmapped.
1233 */
1234static void ext3_truncate_failed_write(struct inode *inode)
1235{
1236        truncate_inode_pages(inode->i_mapping, inode->i_size);
1237        ext3_truncate(inode);
1238}
1239
1240/*
1241 * Truncate blocks that were not used by direct IO write. We have to zero out
1242 * the last file block as well because direct IO might have written to it.
1243 */
1244static void ext3_truncate_failed_direct_write(struct inode *inode)
1245{
1246        ext3_block_truncate_page(inode, inode->i_size);
1247        ext3_truncate(inode);
1248}
1249
1250static int ext3_write_begin(struct file *file, struct address_space *mapping,
1251                                loff_t pos, unsigned len, unsigned flags,
1252                                struct page **pagep, void **fsdata)
1253{
1254        struct inode *inode = mapping->host;
1255        int ret;
1256        handle_t *handle;
1257        int retries = 0;
1258        struct page *page;
1259        pgoff_t index;
1260        unsigned from, to;
1261        /* Reserve one block more for addition to orphan list in case
1262         * we allocate blocks but write fails for some reason */
1263        int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
1264
1265        trace_ext3_write_begin(inode, pos, len, flags);
1266
1267        index = pos >> PAGE_CACHE_SHIFT;
1268        from = pos & (PAGE_CACHE_SIZE - 1);
1269        to = from + len;
1270
1271retry:
1272        page = grab_cache_page_write_begin(mapping, index, flags);
1273        if (!page)
1274                return -ENOMEM;
1275        *pagep = page;
1276
1277        handle = ext3_journal_start(inode, needed_blocks);
1278        if (IS_ERR(handle)) {
1279                unlock_page(page);
1280                page_cache_release(page);
1281                ret = PTR_ERR(handle);
1282                goto out;
1283        }
1284        ret = __block_write_begin(page, pos, len, ext3_get_block);
1285        if (ret)
1286                goto write_begin_failed;
1287
1288        if (ext3_should_journal_data(inode)) {
1289                ret = walk_page_buffers(handle, page_buffers(page),
1290                                from, to, NULL, do_journal_get_write_access);
1291        }
1292write_begin_failed:
1293        if (ret) {
1294                /*
1295                 * block_write_begin may have instantiated a few blocks
1296                 * outside i_size.  Trim these off again. Don't need
1297                 * i_size_read because we hold i_mutex.
1298                 *
1299                 * Add inode to orphan list in case we crash before truncate
1300                 * finishes. Do this only if ext3_can_truncate() agrees so
1301                 * that orphan processing code is happy.
1302                 */
1303                if (pos + len > inode->i_size && ext3_can_truncate(inode))
1304                        ext3_orphan_add(handle, inode);
1305                ext3_journal_stop(handle);
1306                unlock_page(page);
1307                page_cache_release(page);
1308                if (pos + len > inode->i_size)
1309                        ext3_truncate_failed_write(inode);
1310        }
1311        if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1312                goto retry;
1313out:
1314        return ret;
1315}
1316
1317
1318int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1319{
1320        int err = journal_dirty_data(handle, bh);
1321        if (err)
1322                ext3_journal_abort_handle(__func__, __func__,
1323                                                bh, handle, err);
1324        return err;
1325}
1326
1327/* For ordered writepage and write_end functions */
1328static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1329{
1330        /*
1331         * Write could have mapped the buffer but it didn't copy the data in
1332         * yet. So avoid filing such buffer into a transaction.
1333         */
1334        if (buffer_mapped(bh) && buffer_uptodate(bh))
1335                return ext3_journal_dirty_data(handle, bh);
1336        return 0;
1337}
1338
1339/* For write_end() in data=journal mode */
1340static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1341{
1342        if (!buffer_mapped(bh) || buffer_freed(bh))
1343                return 0;
1344        set_buffer_uptodate(bh);
1345        return ext3_journal_dirty_metadata(handle, bh);
1346}
1347
1348/*
1349 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1350 * for the whole page but later we failed to copy the data in. Update inode
1351 * size according to what we managed to copy. The rest is going to be
1352 * truncated in write_end function.
1353 */
1354static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
1355{
1356        /* What matters to us is i_disksize. We don't write i_size anywhere */
1357        if (pos + copied > inode->i_size)
1358                i_size_write(inode, pos + copied);
1359        if (pos + copied > EXT3_I(inode)->i_disksize) {
1360                EXT3_I(inode)->i_disksize = pos + copied;
1361                mark_inode_dirty(inode);
1362        }
1363}
1364
1365/*
1366 * We need to pick up the new inode size which generic_commit_write gave us
1367 * `file' can be NULL - eg, when called from page_symlink().
1368 *
1369 * ext3 never places buffers on inode->i_mapping->private_list.  metadata
1370 * buffers are managed internally.
1371 */
1372static int ext3_ordered_write_end(struct file *file,
1373                                struct address_space *mapping,
1374                                loff_t pos, unsigned len, unsigned copied,
1375                                struct page *page, void *fsdata)
1376{
1377        handle_t *handle = ext3_journal_current_handle();
1378        struct inode *inode = file->f_mapping->host;
1379        unsigned from, to;
1380        int ret = 0, ret2;
1381
1382        trace_ext3_ordered_write_end(inode, pos, len, copied);
1383        copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1384
1385        from = pos & (PAGE_CACHE_SIZE - 1);
1386        to = from + copied;
1387        ret = walk_page_buffers(handle, page_buffers(page),
1388                from, to, NULL, journal_dirty_data_fn);
1389
1390        if (ret == 0)
1391                update_file_sizes(inode, pos, copied);
1392        /*
1393         * There may be allocated blocks outside of i_size because
1394         * we failed to copy some data. Prepare for truncate.
1395         */
1396        if (pos + len > inode->i_size && ext3_can_truncate(inode))
1397                ext3_orphan_add(handle, inode);
1398        ret2 = ext3_journal_stop(handle);
1399        if (!ret)
1400                ret = ret2;
1401        unlock_page(page);
1402        page_cache_release(page);
1403
1404        if (pos + len > inode->i_size)
1405                ext3_truncate_failed_write(inode);
1406        return ret ? ret : copied;
1407}
1408
1409static int ext3_writeback_write_end(struct file *file,
1410                                struct address_space *mapping,
1411                                loff_t pos, unsigned len, unsigned copied,
1412                                struct page *page, void *fsdata)
1413{
1414        handle_t *handle = ext3_journal_current_handle();
1415        struct inode *inode = file->f_mapping->host;
1416        int ret;
1417
1418        trace_ext3_writeback_write_end(inode, pos, len, copied);
1419        copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1420        update_file_sizes(inode, pos, copied);
1421        /*
1422         * There may be allocated blocks outside of i_size because
1423         * we failed to copy some data. Prepare for truncate.
1424         */
1425        if (pos + len > inode->i_size && ext3_can_truncate(inode))
1426                ext3_orphan_add(handle, inode);
1427        ret = ext3_journal_stop(handle);
1428        unlock_page(page);
1429        page_cache_release(page);
1430
1431        if (pos + len > inode->i_size)
1432                ext3_truncate_failed_write(inode);
1433        return ret ? ret : copied;
1434}
1435
1436static int ext3_journalled_write_end(struct file *file,
1437                                struct address_space *mapping,
1438                                loff_t pos, unsigned len, unsigned copied,
1439                                struct page *page, void *fsdata)
1440{
1441        handle_t *handle = ext3_journal_current_handle();
1442        struct inode *inode = mapping->host;
1443        struct ext3_inode_info *ei = EXT3_I(inode);
1444        int ret = 0, ret2;
1445        int partial = 0;
1446        unsigned from, to;
1447
1448        trace_ext3_journalled_write_end(inode, pos, len, copied);
1449        from = pos & (PAGE_CACHE_SIZE - 1);
1450        to = from + len;
1451
1452        if (copied < len) {
1453                if (!PageUptodate(page))
1454                        copied = 0;
1455                page_zero_new_buffers(page, from + copied, to);
1456                to = from + copied;
1457        }
1458
1459        ret = walk_page_buffers(handle, page_buffers(page), from,
1460                                to, &partial, write_end_fn);
1461        if (!partial)
1462                SetPageUptodate(page);
1463
1464        if (pos + copied > inode->i_size)
1465                i_size_write(inode, pos + copied);
1466        /*
1467         * There may be allocated blocks outside of i_size because
1468         * we failed to copy some data. Prepare for truncate.
1469         */
1470        if (pos + len > inode->i_size && ext3_can_truncate(inode))
1471                ext3_orphan_add(handle, inode);
1472        ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1473        atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1474        if (inode->i_size > ei->i_disksize) {
1475                ei->i_disksize = inode->i_size;
1476                ret2 = ext3_mark_inode_dirty(handle, inode);
1477                if (!ret)
1478                        ret = ret2;
1479        }
1480
1481        ret2 = ext3_journal_stop(handle);
1482        if (!ret)
1483                ret = ret2;
1484        unlock_page(page);
1485        page_cache_release(page);
1486
1487        if (pos + len > inode->i_size)
1488                ext3_truncate_failed_write(inode);
1489        return ret ? ret : copied;
1490}
1491
1492/*
1493 * bmap() is special.  It gets used by applications such as lilo and by
1494 * the swapper to find the on-disk block of a specific piece of data.
1495 *
1496 * Naturally, this is dangerous if the block concerned is still in the
1497 * journal.  If somebody makes a swapfile on an ext3 data-journaling
1498 * filesystem and enables swap, then they may get a nasty shock when the
1499 * data getting swapped to that swapfile suddenly gets overwritten by
1500 * the original zero's written out previously to the journal and
1501 * awaiting writeback in the kernel's buffer cache.
1502 *
1503 * So, if we see any bmap calls here on a modified, data-journaled file,
1504 * take extra steps to flush any blocks which might be in the cache.
1505 */
1506static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1507{
1508        struct inode *inode = mapping->host;
1509        journal_t *journal;
1510        int err;
1511
1512        if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
1513                /*
1514                 * This is a REALLY heavyweight approach, but the use of
1515                 * bmap on dirty files is expected to be extremely rare:
1516                 * only if we run lilo or swapon on a freshly made file
1517                 * do we expect this to happen.
1518                 *
1519                 * (bmap requires CAP_SYS_RAWIO so this does not
1520                 * represent an unprivileged user DOS attack --- we'd be
1521                 * in trouble if mortal users could trigger this path at
1522                 * will.)
1523                 *
1524                 * NB. EXT3_STATE_JDATA is not set on files other than
1525                 * regular files.  If somebody wants to bmap a directory
1526                 * or symlink and gets confused because the buffer
1527                 * hasn't yet been flushed to disk, they deserve
1528                 * everything they get.
1529                 */
1530
1531                ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
1532                journal = EXT3_JOURNAL(inode);
1533                journal_lock_updates(journal);
1534                err = journal_flush(journal);
1535                journal_unlock_updates(journal);
1536
1537                if (err)
1538                        return 0;
1539        }
1540
1541        return generic_block_bmap(mapping,block,ext3_get_block);
1542}
1543
1544static int bget_one(handle_t *handle, struct buffer_head *bh)
1545{
1546        get_bh(bh);
1547        return 0;
1548}
1549
1550static int bput_one(handle_t *handle, struct buffer_head *bh)
1551{
1552        put_bh(bh);
1553        return 0;
1554}
1555
1556static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1557{
1558        return !buffer_mapped(bh);
1559}
1560
1561/*
1562 * Note that we always start a transaction even if we're not journalling
1563 * data.  This is to preserve ordering: any hole instantiation within
1564 * __block_write_full_page -> ext3_get_block() should be journalled
1565 * along with the data so we don't crash and then get metadata which
1566 * refers to old data.
1567 *
1568 * In all journalling modes block_write_full_page() will start the I/O.
1569 *
1570 * Problem:
1571 *
1572 *      ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1573 *              ext3_writepage()
1574 *
1575 * Similar for:
1576 *
1577 *      ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1578 *
1579 * Same applies to ext3_get_block().  We will deadlock on various things like
1580 * lock_journal and i_truncate_mutex.
1581 *
1582 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1583 * allocations fail.
1584 *
1585 * 16May01: If we're reentered then journal_current_handle() will be
1586 *          non-zero. We simply *return*.
1587 *
1588 * 1 July 2001: @@@ FIXME:
1589 *   In journalled data mode, a data buffer may be metadata against the
1590 *   current transaction.  But the same file is part of a shared mapping
1591 *   and someone does a writepage() on it.
1592 *
1593 *   We will move the buffer onto the async_data list, but *after* it has
1594 *   been dirtied. So there's a small window where we have dirty data on
1595 *   BJ_Metadata.
1596 *
1597 *   Note that this only applies to the last partial page in the file.  The
1598 *   bit which block_write_full_page() uses prepare/commit for.  (That's
1599 *   broken code anyway: it's wrong for msync()).
1600 *
1601 *   It's a rare case: affects the final partial page, for journalled data
1602 *   where the file is subject to bith write() and writepage() in the same
1603 *   transction.  To fix it we'll need a custom block_write_full_page().
1604 *   We'll probably need that anyway for journalling writepage() output.
1605 *
1606 * We don't honour synchronous mounts for writepage().  That would be
1607 * disastrous.  Any write() or metadata operation will sync the fs for
1608 * us.
1609 *
1610 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1611 * we don't need to open a transaction here.
1612 */
1613static int ext3_ordered_writepage(struct page *page,
1614                                struct writeback_control *wbc)
1615{
1616        struct inode *inode = page->mapping->host;
1617        struct buffer_head *page_bufs;
1618        handle_t *handle = NULL;
1619        int ret = 0;
1620        int err;
1621
1622        J_ASSERT(PageLocked(page));
1623        /*
1624         * We don't want to warn for emergency remount. The condition is
1625         * ordered to avoid dereferencing inode->i_sb in non-error case to
1626         * avoid slow-downs.
1627         */
1628        WARN_ON_ONCE(IS_RDONLY(inode) &&
1629                     !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1630
1631        /*
1632         * We give up here if we're reentered, because it might be for a
1633         * different filesystem.
1634         */
1635        if (ext3_journal_current_handle())
1636                goto out_fail;
1637
1638        trace_ext3_ordered_writepage(page);
1639        if (!page_has_buffers(page)) {
1640                create_empty_buffers(page, inode->i_sb->s_blocksize,
1641                                (1 << BH_Dirty)|(1 << BH_Uptodate));
1642                page_bufs = page_buffers(page);
1643        } else {
1644                page_bufs = page_buffers(page);
1645                if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1646                                       NULL, buffer_unmapped)) {
1647                        /* Provide NULL get_block() to catch bugs if buffers
1648                         * weren't really mapped */
1649                        return block_write_full_page(page, NULL, wbc);
1650                }
1651        }
1652        handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1653
1654        if (IS_ERR(handle)) {
1655                ret = PTR_ERR(handle);
1656                goto out_fail;
1657        }
1658
1659        walk_page_buffers(handle, page_bufs, 0,
1660                        PAGE_CACHE_SIZE, NULL, bget_one);
1661
1662        ret = block_write_full_page(page, ext3_get_block, wbc);
1663
1664        /*
1665         * The page can become unlocked at any point now, and
1666         * truncate can then come in and change things.  So we
1667         * can't touch *page from now on.  But *page_bufs is
1668         * safe due to elevated refcount.
1669         */
1670
1671        /*
1672         * And attach them to the current transaction.  But only if
1673         * block_write_full_page() succeeded.  Otherwise they are unmapped,
1674         * and generally junk.
1675         */
1676        if (ret == 0) {
1677                err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1678                                        NULL, journal_dirty_data_fn);
1679                if (!ret)
1680                        ret = err;
1681        }
1682        walk_page_buffers(handle, page_bufs, 0,
1683                        PAGE_CACHE_SIZE, NULL, bput_one);
1684        err = ext3_journal_stop(handle);
1685        if (!ret)
1686                ret = err;
1687        return ret;
1688
1689out_fail:
1690        redirty_page_for_writepage(wbc, page);
1691        unlock_page(page);
1692        return ret;
1693}
1694
1695static int ext3_writeback_writepage(struct page *page,
1696                                struct writeback_control *wbc)
1697{
1698        struct inode *inode = page->mapping->host;
1699        handle_t *handle = NULL;
1700        int ret = 0;
1701        int err;
1702
1703        J_ASSERT(PageLocked(page));
1704        /*
1705         * We don't want to warn for emergency remount. The condition is
1706         * ordered to avoid dereferencing inode->i_sb in non-error case to
1707         * avoid slow-downs.
1708         */
1709        WARN_ON_ONCE(IS_RDONLY(inode) &&
1710                     !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1711
1712        if (ext3_journal_current_handle())
1713                goto out_fail;
1714
1715        trace_ext3_writeback_writepage(page);
1716        if (page_has_buffers(page)) {
1717                if (!walk_page_buffers(NULL, page_buffers(page), 0,
1718                                      PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1719                        /* Provide NULL get_block() to catch bugs if buffers
1720                         * weren't really mapped */
1721                        return block_write_full_page(page, NULL, wbc);
1722                }
1723        }
1724
1725        handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1726        if (IS_ERR(handle)) {
1727                ret = PTR_ERR(handle);
1728                goto out_fail;
1729        }
1730
1731        ret = block_write_full_page(page, ext3_get_block, wbc);
1732
1733        err = ext3_journal_stop(handle);
1734        if (!ret)
1735                ret = err;
1736        return ret;
1737
1738out_fail:
1739        redirty_page_for_writepage(wbc, page);
1740        unlock_page(page);
1741        return ret;
1742}
1743
1744static int ext3_journalled_writepage(struct page *page,
1745                                struct writeback_control *wbc)
1746{
1747        struct inode *inode = page->mapping->host;
1748        handle_t *handle = NULL;
1749        int ret = 0;
1750        int err;
1751
1752        J_ASSERT(PageLocked(page));
1753        /*
1754         * We don't want to warn for emergency remount. The condition is
1755         * ordered to avoid dereferencing inode->i_sb in non-error case to
1756         * avoid slow-downs.
1757         */
1758        WARN_ON_ONCE(IS_RDONLY(inode) &&
1759                     !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1760
1761        if (ext3_journal_current_handle())
1762                goto no_write;
1763
1764        trace_ext3_journalled_writepage(page);
1765        handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1766        if (IS_ERR(handle)) {
1767                ret = PTR_ERR(handle);
1768                goto no_write;
1769        }
1770
1771        if (!page_has_buffers(page) || PageChecked(page)) {
1772                /*
1773                 * It's mmapped pagecache.  Add buffers and journal it.  There
1774                 * doesn't seem much point in redirtying the page here.
1775                 */
1776                ClearPageChecked(page);
1777                ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1778                                          ext3_get_block);
1779                if (ret != 0) {
1780                        ext3_journal_stop(handle);
1781                        goto out_unlock;
1782                }
1783                ret = walk_page_buffers(handle, page_buffers(page), 0,
1784                        PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1785
1786                err = walk_page_buffers(handle, page_buffers(page), 0,
1787                                PAGE_CACHE_SIZE, NULL, write_end_fn);
1788                if (ret == 0)
1789                        ret = err;
1790                ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1791                atomic_set(&EXT3_I(inode)->i_datasync_tid,
1792                           handle->h_transaction->t_tid);
1793                unlock_page(page);
1794        } else {
1795                /*
1796                 * It may be a page full of checkpoint-mode buffers.  We don't
1797                 * really know unless we go poke around in the buffer_heads.
1798                 * But block_write_full_page will do the right thing.
1799                 */
1800                ret = block_write_full_page(page, ext3_get_block, wbc);
1801        }
1802        err = ext3_journal_stop(handle);
1803        if (!ret)
1804                ret = err;
1805out:
1806        return ret;
1807
1808no_write:
1809        redirty_page_for_writepage(wbc, page);
1810out_unlock:
1811        unlock_page(page);
1812        goto out;
1813}
1814
1815static int ext3_readpage(struct file *file, struct page *page)
1816{
1817        trace_ext3_readpage(page);
1818        return mpage_readpage(page, ext3_get_block);
1819}
1820
1821static int
1822ext3_readpages(struct file *file, struct address_space *mapping,
1823                struct list_head *pages, unsigned nr_pages)
1824{
1825        return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1826}
1827
1828static void ext3_invalidatepage(struct page *page, unsigned int offset,
1829                                unsigned int length)
1830{
1831        journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1832
1833        trace_ext3_invalidatepage(page, offset, length);
1834
1835        /*
1836         * If it's a full truncate we just forget about the pending dirtying
1837         */
1838        if (offset == 0 && length == PAGE_CACHE_SIZE)
1839                ClearPageChecked(page);
1840
1841        journal_invalidatepage(journal, page, offset, length);
1842}
1843
1844static int ext3_releasepage(struct page *page, gfp_t wait)
1845{
1846        journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1847
1848        trace_ext3_releasepage(page);
1849        WARN_ON(PageChecked(page));
1850        if (!page_has_buffers(page))
1851                return 0;
1852        return journal_try_to_free_buffers(journal, page, wait);
1853}
1854
1855/*
1856 * If the O_DIRECT write will extend the file then add this inode to the
1857 * orphan list.  So recovery will truncate it back to the original size
1858 * if the machine crashes during the write.
1859 *
1860 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1861 * crashes then stale disk data _may_ be exposed inside the file. But current
1862 * VFS code falls back into buffered path in that case so we are safe.
1863 */
1864static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1865                        const struct iovec *iov, loff_t offset,
1866                        unsigned long nr_segs)
1867{
1868        struct file *file = iocb->ki_filp;
1869        struct inode *inode = file->f_mapping->host;
1870        struct ext3_inode_info *ei = EXT3_I(inode);
1871        handle_t *handle;
1872        ssize_t ret;
1873        int orphan = 0;
1874        size_t count = iov_length(iov, nr_segs);
1875        int retries = 0;
1876
1877        trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1878
1879        if (rw == WRITE) {
1880                loff_t final_size = offset + count;
1881
1882                if (final_size > inode->i_size) {
1883                        /* Credits for sb + inode write */
1884                        handle = ext3_journal_start(inode, 2);
1885                        if (IS_ERR(handle)) {
1886                                ret = PTR_ERR(handle);
1887                                goto out;
1888                        }
1889                        ret = ext3_orphan_add(handle, inode);
1890                        if (ret) {
1891                                ext3_journal_stop(handle);
1892                                goto out;
1893                        }
1894                        orphan = 1;
1895                        ei->i_disksize = inode->i_size;
1896                        ext3_journal_stop(handle);
1897                }
1898        }
1899
1900retry:
1901        ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1902                                 ext3_get_block);
1903        /*
1904         * In case of error extending write may have instantiated a few
1905         * blocks outside i_size. Trim these off again.
1906         */
1907        if (unlikely((rw & WRITE) && ret < 0)) {
1908                loff_t isize = i_size_read(inode);
1909                loff_t end = offset + iov_length(iov, nr_segs);
1910
1911                if (end > isize)
1912                        ext3_truncate_failed_direct_write(inode);
1913        }
1914        if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1915                goto retry;
1916
1917        if (orphan) {
1918                int err;
1919
1920                /* Credits for sb + inode write */
1921                handle = ext3_journal_start(inode, 2);
1922                if (IS_ERR(handle)) {
1923                        /* This is really bad luck. We've written the data
1924                         * but cannot extend i_size. Truncate allocated blocks
1925                         * and pretend the write failed... */
1926                        ext3_truncate_failed_direct_write(inode);
1927                        ret = PTR_ERR(handle);
1928                        goto out;
1929                }
1930                if (inode->i_nlink)
1931                        ext3_orphan_del(handle, inode);
1932                if (ret > 0) {
1933                        loff_t end = offset + ret;
1934                        if (end > inode->i_size) {
1935                                ei->i_disksize = end;
1936                                i_size_write(inode, end);
1937                                /*
1938                                 * We're going to return a positive `ret'
1939                                 * here due to non-zero-length I/O, so there's
1940                                 * no way of reporting error returns from
1941                                 * ext3_mark_inode_dirty() to userspace.  So
1942                                 * ignore it.
1943                                 */
1944                                ext3_mark_inode_dirty(handle, inode);
1945                        }
1946                }
1947                err = ext3_journal_stop(handle);
1948                if (ret == 0)
1949                        ret = err;
1950        }
1951out:
1952        trace_ext3_direct_IO_exit(inode, offset,
1953                                iov_length(iov, nr_segs), rw, ret);
1954        return ret;
1955}
1956
1957/*
1958 * Pages can be marked dirty completely asynchronously from ext3's journalling
1959 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
1960 * much here because ->set_page_dirty is called under VFS locks.  The page is
1961 * not necessarily locked.
1962 *
1963 * We cannot just dirty the page and leave attached buffers clean, because the
1964 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
1965 * or jbddirty because all the journalling code will explode.
1966 *
1967 * So what we do is to mark the page "pending dirty" and next time writepage
1968 * is called, propagate that into the buffers appropriately.
1969 */
1970static int ext3_journalled_set_page_dirty(struct page *page)
1971{
1972        SetPageChecked(page);
1973        return __set_page_dirty_nobuffers(page);
1974}
1975
1976static const struct address_space_operations ext3_ordered_aops = {
1977        .readpage               = ext3_readpage,
1978        .readpages              = ext3_readpages,
1979        .writepage              = ext3_ordered_writepage,
1980        .write_begin            = ext3_write_begin,
1981        .write_end              = ext3_ordered_write_end,
1982        .bmap                   = ext3_bmap,
1983        .invalidatepage         = ext3_invalidatepage,
1984        .releasepage            = ext3_releasepage,
1985        .direct_IO              = ext3_direct_IO,
1986        .migratepage            = buffer_migrate_page,
1987        .is_partially_uptodate  = block_is_partially_uptodate,
1988        .is_dirty_writeback     = buffer_check_dirty_writeback,
1989        .error_remove_page      = generic_error_remove_page,
1990};
1991
1992static const struct address_space_operations ext3_writeback_aops = {
1993        .readpage               = ext3_readpage,
1994        .readpages              = ext3_readpages,
1995        .writepage              = ext3_writeback_writepage,
1996        .write_begin            = ext3_write_begin,
1997        .write_end              = ext3_writeback_write_end,
1998        .bmap                   = ext3_bmap,
1999        .invalidatepage         = ext3_invalidatepage,
2000        .releasepage            = ext3_releasepage,
2001        .direct_IO              = ext3_direct_IO,
2002        .migratepage            = buffer_migrate_page,
2003        .is_partially_uptodate  = block_is_partially_uptodate,
2004        .error_remove_page      = generic_error_remove_page,
2005};
2006
2007static const struct address_space_operations ext3_journalled_aops = {
2008        .readpage               = ext3_readpage,
2009        .readpages              = ext3_readpages,
2010        .writepage              = ext3_journalled_writepage,
2011        .write_begin            = ext3_write_begin,
2012        .write_end              = ext3_journalled_write_end,
2013        .set_page_dirty         = ext3_journalled_set_page_dirty,
2014        .bmap                   = ext3_bmap,
2015        .invalidatepage         = ext3_invalidatepage,
2016        .releasepage            = ext3_releasepage,
2017        .is_partially_uptodate  = block_is_partially_uptodate,
2018        .error_remove_page      = generic_error_remove_page,
2019};
2020
2021void ext3_set_aops(struct inode *inode)
2022{
2023        if (ext3_should_order_data(inode))
2024                inode->i_mapping->a_ops = &ext3_ordered_aops;
2025        else if (ext3_should_writeback_data(inode))
2026                inode->i_mapping->a_ops = &ext3_writeback_aops;
2027        else
2028                inode->i_mapping->a_ops = &ext3_journalled_aops;
2029}
2030
2031/*
2032 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2033 * up to the end of the block which corresponds to `from'.
2034 * This required during truncate. We need to physically zero the tail end
2035 * of that block so it doesn't yield old data if the file is later grown.
2036 */
2037static int ext3_block_truncate_page(struct inode *inode, loff_t from)
2038{
2039        ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2040        unsigned offset = from & (PAGE_CACHE_SIZE - 1);
2041        unsigned blocksize, iblock, length, pos;
2042        struct page *page;
2043        handle_t *handle = NULL;
2044        struct buffer_head *bh;
2045        int err = 0;
2046
2047        /* Truncated on block boundary - nothing to do */
2048        blocksize = inode->i_sb->s_blocksize;
2049        if ((from & (blocksize - 1)) == 0)
2050                return 0;
2051
2052        page = grab_cache_page(inode->i_mapping, index);
2053        if (!page)
2054                return -ENOMEM;
2055        length = blocksize - (offset & (blocksize - 1));
2056        iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2057
2058        if (!page_has_buffers(page))
2059                create_empty_buffers(page, blocksize, 0);
2060
2061        /* Find the buffer that contains "offset" */
2062        bh = page_buffers(page);
2063        pos = blocksize;
2064        while (offset >= pos) {
2065                bh = bh->b_this_page;
2066                iblock++;
2067                pos += blocksize;
2068        }
2069
2070        err = 0;
2071        if (buffer_freed(bh)) {
2072                BUFFER_TRACE(bh, "freed: skip");
2073                goto unlock;
2074        }
2075
2076        if (!buffer_mapped(bh)) {
2077                BUFFER_TRACE(bh, "unmapped");
2078                ext3_get_block(inode, iblock, bh, 0);
2079                /* unmapped? It's a hole - nothing to do */
2080                if (!buffer_mapped(bh)) {
2081                        BUFFER_TRACE(bh, "still unmapped");
2082                        goto unlock;
2083                }
2084        }
2085
2086        /* Ok, it's mapped. Make sure it's up-to-date */
2087        if (PageUptodate(page))
2088                set_buffer_uptodate(bh);
2089
2090        if (!bh_uptodate_or_lock(bh)) {
2091                err = bh_submit_read(bh);
2092                /* Uhhuh. Read error. Complain and punt. */
2093                if (err)
2094                        goto unlock;
2095        }
2096
2097        /* data=writeback mode doesn't need transaction to zero-out data */
2098        if (!ext3_should_writeback_data(inode)) {
2099                /* We journal at most one block */
2100                handle = ext3_journal_start(inode, 1);
2101                if (IS_ERR(handle)) {
2102                        clear_highpage(page);
2103                        flush_dcache_page(page);
2104                        err = PTR_ERR(handle);
2105                        goto unlock;
2106                }
2107        }
2108
2109        if (ext3_should_journal_data(inode)) {
2110                BUFFER_TRACE(bh, "get write access");
2111                err = ext3_journal_get_write_access(handle, bh);
2112                if (err)
2113                        goto stop;
2114        }
2115
2116        zero_user(page, offset, length);
2117        BUFFER_TRACE(bh, "zeroed end of block");
2118
2119        err = 0;
2120        if (ext3_should_journal_data(inode)) {
2121                err = ext3_journal_dirty_metadata(handle, bh);
2122        } else {
2123                if (ext3_should_order_data(inode))
2124                        err = ext3_journal_dirty_data(handle, bh);
2125                mark_buffer_dirty(bh);
2126        }
2127stop:
2128        if (handle)
2129                ext3_journal_stop(handle);
2130
2131unlock:
2132        unlock_page(page);
2133        page_cache_release(page);
2134        return err;
2135}
2136
2137/*
2138 * Probably it should be a library function... search for first non-zero word
2139 * or memcmp with zero_page, whatever is better for particular architecture.
2140 * Linus?
2141 */
2142static inline int all_zeroes(__le32 *p, __le32 *q)
2143{
2144        while (p < q)
2145                if (*p++)
2146                        return 0;
2147        return 1;
2148}
2149
2150/**
2151 *      ext3_find_shared - find the indirect blocks for partial truncation.
2152 *      @inode:   inode in question
2153 *      @depth:   depth of the affected branch
2154 *      @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2155 *      @chain:   place to store the pointers to partial indirect blocks
2156 *      @top:     place to the (detached) top of branch
2157 *
2158 *      This is a helper function used by ext3_truncate().
2159 *
2160 *      When we do truncate() we may have to clean the ends of several
2161 *      indirect blocks but leave the blocks themselves alive. Block is
2162 *      partially truncated if some data below the new i_size is referred
2163 *      from it (and it is on the path to the first completely truncated
2164 *      data block, indeed).  We have to free the top of that path along
2165 *      with everything to the right of the path. Since no allocation
2166 *      past the truncation point is possible until ext3_truncate()
2167 *      finishes, we may safely do the latter, but top of branch may
2168 *      require special attention - pageout below the truncation point
2169 *      might try to populate it.
2170 *
2171 *      We atomically detach the top of branch from the tree, store the
2172 *      block number of its root in *@top, pointers to buffer_heads of
2173 *      partially truncated blocks - in @chain[].bh and pointers to
2174 *      their last elements that should not be removed - in
2175 *      @chain[].p. Return value is the pointer to last filled element
2176 *      of @chain.
2177 *
2178 *      The work left to caller to do the actual freeing of subtrees:
2179 *              a) free the subtree starting from *@top
2180 *              b) free the subtrees whose roots are stored in
2181 *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2182 *              c) free the subtrees growing from the inode past the @chain[0].
2183 *                      (no partially truncated stuff there).  */
2184
2185static Indirect *ext3_find_shared(struct inode *inode, int depth,
2186                        int offsets[4], Indirect chain[4], __le32 *top)
2187{
2188        Indirect *partial, *p;
2189        int k, err;
2190
2191        *top = 0;
2192        /* Make k index the deepest non-null offset + 1 */
2193        for (k = depth; k > 1 && !offsets[k-1]; k--)
2194                ;
2195        partial = ext3_get_branch(inode, k, offsets, chain, &err);
2196        /* Writer: pointers */
2197        if (!partial)
2198                partial = chain + k-1;
2199        /*
2200         * If the branch acquired continuation since we've looked at it -
2201         * fine, it should all survive and (new) top doesn't belong to us.
2202         */
2203        if (!partial->key && *partial->p)
2204                /* Writer: end */
2205                goto no_top;
2206        for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2207                ;
2208        /*
2209         * OK, we've found the last block that must survive. The rest of our
2210         * branch should be detached before unlocking. However, if that rest
2211         * of branch is all ours and does not grow immediately from the inode
2212         * it's easier to cheat and just decrement partial->p.
2213         */
2214        if (p == chain + k - 1 && p > chain) {
2215                p->p--;
2216        } else {
2217                *top = *p->p;
2218                /* Nope, don't do this in ext3.  Must leave the tree intact */
2219#if 0
2220                *p->p = 0;
2221#endif
2222        }
2223        /* Writer: end */
2224
2225        while(partial > p) {
2226                brelse(partial->bh);
2227                partial--;
2228        }
2229no_top:
2230        return partial;
2231}
2232
2233/*
2234 * Zero a number of block pointers in either an inode or an indirect block.
2235 * If we restart the transaction we must again get write access to the
2236 * indirect block for further modification.
2237 *
2238 * We release `count' blocks on disk, but (last - first) may be greater
2239 * than `count' because there can be holes in there.
2240 */
2241static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
2242                struct buffer_head *bh, ext3_fsblk_t block_to_free,
2243                unsigned long count, __le32 *first, __le32 *last)
2244{
2245        __le32 *p;
2246        if (try_to_extend_transaction(handle, inode)) {
2247                if (bh) {
2248                        BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2249                        if (ext3_journal_dirty_metadata(handle, bh))
2250                                return;
2251                }
2252                ext3_mark_inode_dirty(handle, inode);
2253                truncate_restart_transaction(handle, inode);
2254                if (bh) {
2255                        BUFFER_TRACE(bh, "retaking write access");
2256                        if (ext3_journal_get_write_access(handle, bh))
2257                                return;
2258                }
2259        }
2260
2261        /*
2262         * Any buffers which are on the journal will be in memory. We find
2263         * them on the hash table so journal_revoke() will run journal_forget()
2264         * on them.  We've already detached each block from the file, so
2265         * bforget() in journal_forget() should be safe.
2266         *
2267         * AKPM: turn on bforget in journal_forget()!!!
2268         */
2269        for (p = first; p < last; p++) {
2270                u32 nr = le32_to_cpu(*p);
2271                if (nr) {
2272                        struct buffer_head *bh;
2273
2274                        *p = 0;
2275                        bh = sb_find_get_block(inode->i_sb, nr);
2276                        ext3_forget(handle, 0, inode, bh, nr);
2277                }
2278        }
2279
2280        ext3_free_blocks(handle, inode, block_to_free, count);
2281}
2282
2283/**
2284 * ext3_free_data - free a list of data blocks
2285 * @handle:     handle for this transaction
2286 * @inode:      inode we are dealing with
2287 * @this_bh:    indirect buffer_head which contains *@first and *@last
2288 * @first:      array of block numbers
2289 * @last:       points immediately past the end of array
2290 *
2291 * We are freeing all blocks referred from that array (numbers are stored as
2292 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2293 *
2294 * We accumulate contiguous runs of blocks to free.  Conveniently, if these
2295 * blocks are contiguous then releasing them at one time will only affect one
2296 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2297 * actually use a lot of journal space.
2298 *
2299 * @this_bh will be %NULL if @first and @last point into the inode's direct
2300 * block pointers.
2301 */
2302static void ext3_free_data(handle_t *handle, struct inode *inode,
2303                           struct buffer_head *this_bh,
2304                           __le32 *first, __le32 *last)
2305{
2306        ext3_fsblk_t block_to_free = 0;    /* Starting block # of a run */
2307        unsigned long count = 0;            /* Number of blocks in the run */
2308        __le32 *block_to_free_p = NULL;     /* Pointer into inode/ind
2309                                               corresponding to
2310                                               block_to_free */
2311        ext3_fsblk_t nr;                    /* Current block # */
2312        __le32 *p;                          /* Pointer into inode/ind
2313                                               for current block */
2314        int err;
2315
2316        if (this_bh) {                          /* For indirect block */
2317                BUFFER_TRACE(this_bh, "get_write_access");
2318                err = ext3_journal_get_write_access(handle, this_bh);
2319                /* Important: if we can't update the indirect pointers
2320                 * to the blocks, we can't free them. */
2321                if (err)
2322                        return;
2323        }
2324
2325        for (p = first; p < last; p++) {
2326                nr = le32_to_cpu(*p);
2327                if (nr) {
2328                        /* accumulate blocks to free if they're contiguous */
2329                        if (count == 0) {
2330                                block_to_free = nr;
2331                                block_to_free_p = p;
2332                                count = 1;
2333                        } else if (nr == block_to_free + count) {
2334                                count++;
2335                        } else {
2336                                ext3_clear_blocks(handle, inode, this_bh,
2337                                                  block_to_free,
2338                                                  count, block_to_free_p, p);
2339                                block_to_free = nr;
2340                                block_to_free_p = p;
2341                                count = 1;
2342                        }
2343                }
2344        }
2345
2346        if (count > 0)
2347                ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2348                                  count, block_to_free_p, p);
2349
2350        if (this_bh) {
2351                BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2352
2353                /*
2354                 * The buffer head should have an attached journal head at this
2355                 * point. However, if the data is corrupted and an indirect
2356                 * block pointed to itself, it would have been detached when
2357                 * the block was cleared. Check for this instead of OOPSing.
2358                 */
2359                if (bh2jh(this_bh))
2360                        ext3_journal_dirty_metadata(handle, this_bh);
2361                else
2362                        ext3_error(inode->i_sb, "ext3_free_data",
2363                                   "circular indirect block detected, "
2364                                   "inode=%lu, block=%llu",
2365                                   inode->i_ino,
2366                                   (unsigned long long)this_bh->b_blocknr);
2367        }
2368}
2369
2370/**
2371 *      ext3_free_branches - free an array of branches
2372 *      @handle: JBD handle for this transaction
2373 *      @inode: inode we are dealing with
2374 *      @parent_bh: the buffer_head which contains *@first and *@last
2375 *      @first: array of block numbers
2376 *      @last:  pointer immediately past the end of array
2377 *      @depth: depth of the branches to free
2378 *
2379 *      We are freeing all blocks referred from these branches (numbers are
2380 *      stored as little-endian 32-bit) and updating @inode->i_blocks
2381 *      appropriately.
2382 */
2383static void ext3_free_branches(handle_t *handle, struct inode *inode,
2384                               struct buffer_head *parent_bh,
2385                               __le32 *first, __le32 *last, int depth)
2386{
2387        ext3_fsblk_t nr;
2388        __le32 *p;
2389
2390        if (is_handle_aborted(handle))
2391                return;
2392
2393        if (depth--) {
2394                struct buffer_head *bh;
2395                int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2396                p = last;
2397                while (--p >= first) {
2398                        nr = le32_to_cpu(*p);
2399                        if (!nr)
2400                                continue;               /* A hole */
2401
2402                        /* Go read the buffer for the next level down */
2403                        bh = sb_bread(inode->i_sb, nr);
2404
2405                        /*
2406                         * A read failure? Report error and clear slot
2407                         * (should be rare).
2408                         */
2409                        if (!bh) {
2410                                ext3_error(inode->i_sb, "ext3_free_branches",
2411                                           "Read failure, inode=%lu, block="E3FSBLK,
2412                                           inode->i_ino, nr);
2413                                continue;
2414                        }
2415
2416                        /* This zaps the entire block.  Bottom up. */
2417                        BUFFER_TRACE(bh, "free child branches");
2418                        ext3_free_branches(handle, inode, bh,
2419                                           (__le32*)bh->b_data,
2420                                           (__le32*)bh->b_data + addr_per_block,
2421                                           depth);
2422
2423                        /*
2424                         * Everything below this this pointer has been
2425                         * released.  Now let this top-of-subtree go.
2426                         *
2427                         * We want the freeing of this indirect block to be
2428                         * atomic in the journal with the updating of the
2429                         * bitmap block which owns it.  So make some room in
2430                         * the journal.
2431                         *
2432                         * We zero the parent pointer *after* freeing its
2433                         * pointee in the bitmaps, so if extend_transaction()
2434                         * for some reason fails to put the bitmap changes and
2435                         * the release into the same transaction, recovery
2436                         * will merely complain about releasing a free block,
2437                         * rather than leaking blocks.
2438                         */
2439                        if (is_handle_aborted(handle))
2440                                return;
2441                        if (try_to_extend_transaction(handle, inode)) {
2442                                ext3_mark_inode_dirty(handle, inode);
2443                                truncate_restart_transaction(handle, inode);
2444                        }
2445
2446                        /*
2447                         * We've probably journalled the indirect block several
2448                         * times during the truncate.  But it's no longer
2449                         * needed and we now drop it from the transaction via
2450                         * journal_revoke().
2451                         *
2452                         * That's easy if it's exclusively part of this
2453                         * transaction.  But if it's part of the committing
2454                         * transaction then journal_forget() will simply
2455                         * brelse() it.  That means that if the underlying
2456                         * block is reallocated in ext3_get_block(),
2457                         * unmap_underlying_metadata() will find this block
2458                         * and will try to get rid of it.  damn, damn. Thus
2459                         * we don't allow a block to be reallocated until
2460                         * a transaction freeing it has fully committed.
2461                         *
2462                         * We also have to make sure journal replay after a
2463                         * crash does not overwrite non-journaled data blocks
2464                         * with old metadata when the block got reallocated for
2465                         * data.  Thus we have to store a revoke record for a
2466                         * block in the same transaction in which we free the
2467                         * block.
2468                         */
2469                        ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2470
2471                        ext3_free_blocks(handle, inode, nr, 1);
2472
2473                        if (parent_bh) {
2474                                /*
2475                                 * The block which we have just freed is
2476                                 * pointed to by an indirect block: journal it
2477                                 */
2478                                BUFFER_TRACE(parent_bh, "get_write_access");
2479                                if (!ext3_journal_get_write_access(handle,
2480                                                                   parent_bh)){
2481                                        *p = 0;
2482                                        BUFFER_TRACE(parent_bh,
2483                                        "call ext3_journal_dirty_metadata");
2484                                        ext3_journal_dirty_metadata(handle,
2485                                                                    parent_bh);
2486                                }
2487                        }
2488                }
2489        } else {
2490                /* We have reached the bottom of the tree. */
2491                BUFFER_TRACE(parent_bh, "free data blocks");
2492                ext3_free_data(handle, inode, parent_bh, first, last);
2493        }
2494}
2495
2496int ext3_can_truncate(struct inode *inode)
2497{
2498        if (S_ISREG(inode->i_mode))
2499                return 1;
2500        if (S_ISDIR(inode->i_mode))
2501                return 1;
2502        if (S_ISLNK(inode->i_mode))
2503                return !ext3_inode_is_fast_symlink(inode);
2504        return 0;
2505}
2506
2507/*
2508 * ext3_truncate()
2509 *
2510 * We block out ext3_get_block() block instantiations across the entire
2511 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2512 * simultaneously on behalf of the same inode.
2513 *
2514 * As we work through the truncate and commit bits of it to the journal there
2515 * is one core, guiding principle: the file's tree must always be consistent on
2516 * disk.  We must be able to restart the truncate after a crash.
2517 *
2518 * The file's tree may be transiently inconsistent in memory (although it
2519 * probably isn't), but whenever we close off and commit a journal transaction,
2520 * the contents of (the filesystem + the journal) must be consistent and
2521 * restartable.  It's pretty simple, really: bottom up, right to left (although
2522 * left-to-right works OK too).
2523 *
2524 * Note that at recovery time, journal replay occurs *before* the restart of
2525 * truncate against the orphan inode list.
2526 *
2527 * The committed inode has the new, desired i_size (which is the same as
2528 * i_disksize in this case).  After a crash, ext3_orphan_cleanup() will see
2529 * that this inode's truncate did not complete and it will again call
2530 * ext3_truncate() to have another go.  So there will be instantiated blocks
2531 * to the right of the truncation point in a crashed ext3 filesystem.  But
2532 * that's fine - as long as they are linked from the inode, the post-crash
2533 * ext3_truncate() run will find them and release them.
2534 */
2535void ext3_truncate(struct inode *inode)
2536{
2537        handle_t *handle;
2538        struct ext3_inode_info *ei = EXT3_I(inode);
2539        __le32 *i_data = ei->i_data;
2540        int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2541        int offsets[4];
2542        Indirect chain[4];
2543        Indirect *partial;
2544        __le32 nr = 0;
2545        int n;
2546        long last_block;
2547        unsigned blocksize = inode->i_sb->s_blocksize;
2548
2549        trace_ext3_truncate_enter(inode);
2550
2551        if (!ext3_can_truncate(inode))
2552                goto out_notrans;
2553
2554        if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2555                ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
2556
2557        handle = start_transaction(inode);
2558        if (IS_ERR(handle))
2559                goto out_notrans;
2560
2561        last_block = (inode->i_size + blocksize-1)
2562                                        >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2563        n = ext3_block_to_path(inode, last_block, offsets, NULL);
2564        if (n == 0)
2565                goto out_stop;  /* error */
2566
2567        /*
2568         * OK.  This truncate is going to happen.  We add the inode to the
2569         * orphan list, so that if this truncate spans multiple transactions,
2570         * and we crash, we will resume the truncate when the filesystem
2571         * recovers.  It also marks the inode dirty, to catch the new size.
2572         *
2573         * Implication: the file must always be in a sane, consistent
2574         * truncatable state while each transaction commits.
2575         */
2576        if (ext3_orphan_add(handle, inode))
2577                goto out_stop;
2578
2579        /*
2580         * The orphan list entry will now protect us from any crash which
2581         * occurs before the truncate completes, so it is now safe to propagate
2582         * the new, shorter inode size (held for now in i_size) into the
2583         * on-disk inode. We do this via i_disksize, which is the value which
2584         * ext3 *really* writes onto the disk inode.
2585         */
2586        ei->i_disksize = inode->i_size;
2587
2588        /*
2589         * From here we block out all ext3_get_block() callers who want to
2590         * modify the block allocation tree.
2591         */
2592        mutex_lock(&ei->truncate_mutex);
2593
2594        if (n == 1) {           /* direct blocks */
2595                ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2596                               i_data + EXT3_NDIR_BLOCKS);
2597                goto do_indirects;
2598        }
2599
2600        partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2601        /* Kill the top of shared branch (not detached) */
2602        if (nr) {
2603                if (partial == chain) {
2604                        /* Shared branch grows from the inode */
2605                        ext3_free_branches(handle, inode, NULL,
2606                                           &nr, &nr+1, (chain+n-1) - partial);
2607                        *partial->p = 0;
2608                        /*
2609                         * We mark the inode dirty prior to restart,
2610                         * and prior to stop.  No need for it here.
2611                         */
2612                } else {
2613                        /* Shared branch grows from an indirect block */
2614                        ext3_free_branches(handle, inode, partial->bh,
2615                                        partial->p,
2616                                        partial->p+1, (chain+n-1) - partial);
2617                }
2618        }
2619        /* Clear the ends of indirect blocks on the shared branch */
2620        while (partial > chain) {
2621                ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2622                                   (__le32*)partial->bh->b_data+addr_per_block,
2623                                   (chain+n-1) - partial);
2624                BUFFER_TRACE(partial->bh, "call brelse");
2625                brelse (partial->bh);
2626                partial--;
2627        }
2628do_indirects:
2629        /* Kill the remaining (whole) subtrees */
2630        switch (offsets[0]) {
2631        default:
2632                nr = i_data[EXT3_IND_BLOCK];
2633                if (nr) {
2634                        ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2635                        i_data[EXT3_IND_BLOCK] = 0;
2636                }
2637        case EXT3_IND_BLOCK:
2638                nr = i_data[EXT3_DIND_BLOCK];
2639                if (nr) {
2640                        ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2641                        i_data[EXT3_DIND_BLOCK] = 0;
2642                }
2643        case EXT3_DIND_BLOCK:
2644                nr = i_data[EXT3_TIND_BLOCK];
2645                if (nr) {
2646                        ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2647                        i_data[EXT3_TIND_BLOCK] = 0;
2648                }
2649        case EXT3_TIND_BLOCK:
2650                ;
2651        }
2652
2653        ext3_discard_reservation(inode);
2654
2655        mutex_unlock(&ei->truncate_mutex);
2656        inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2657        ext3_mark_inode_dirty(handle, inode);
2658
2659        /*
2660         * In a multi-transaction truncate, we only make the final transaction
2661         * synchronous
2662         */
2663        if (IS_SYNC(inode))
2664                handle->h_sync = 1;
2665out_stop:
2666        /*
2667         * If this was a simple ftruncate(), and the file will remain alive
2668         * then we need to clear up the orphan record which we created above.
2669         * However, if this was a real unlink then we were called by
2670         * ext3_evict_inode(), and we allow that function to clean up the
2671         * orphan info for us.
2672         */
2673        if (inode->i_nlink)
2674                ext3_orphan_del(handle, inode);
2675
2676        ext3_journal_stop(handle);
2677        trace_ext3_truncate_exit(inode);
2678        return;
2679out_notrans:
2680        /*
2681         * Delete the inode from orphan list so that it doesn't stay there
2682         * forever and trigger assertion on umount.
2683         */
2684        if (inode->i_nlink)
2685                ext3_orphan_del(NULL, inode);
2686        trace_ext3_truncate_exit(inode);
2687}
2688
2689static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2690                unsigned long ino, struct ext3_iloc *iloc)
2691{
2692        unsigned long block_group;
2693        unsigned long offset;
2694        ext3_fsblk_t block;
2695        struct ext3_group_desc *gdp;
2696
2697        if (!ext3_valid_inum(sb, ino)) {
2698                /*
2699                 * This error is already checked for in namei.c unless we are
2700                 * looking at an NFS filehandle, in which case no error
2701                 * report is needed
2702                 */
2703                return 0;
2704        }
2705
2706        block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2707        gdp = ext3_get_group_desc(sb, block_group, NULL);
2708        if (!gdp)
2709                return 0;
2710        /*
2711         * Figure out the offset within the block group inode table
2712         */
2713        offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2714                EXT3_INODE_SIZE(sb);
2715        block = le32_to_cpu(gdp->bg_inode_table) +
2716                (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2717
2718        iloc->block_group = block_group;
2719        iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2720        return block;
2721}
2722
2723/*
2724 * ext3_get_inode_loc returns with an extra refcount against the inode's
2725 * underlying buffer_head on success. If 'in_mem' is true, we have all
2726 * data in memory that is needed to recreate the on-disk version of this
2727 * inode.
2728 */
2729static int __ext3_get_inode_loc(struct inode *inode,
2730                                struct ext3_iloc *iloc, int in_mem)
2731{
2732        ext3_fsblk_t block;
2733        struct buffer_head *bh;
2734
2735        block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2736        if (!block)
2737                return -EIO;
2738
2739        bh = sb_getblk(inode->i_sb, block);
2740        if (unlikely(!bh)) {
2741                ext3_error (inode->i_sb, "ext3_get_inode_loc",
2742                                "unable to read inode block - "
2743                                "inode=%lu, block="E3FSBLK,
2744                                 inode->i_ino, block);
2745                return -ENOMEM;
2746        }
2747        if (!buffer_uptodate(bh)) {
2748                lock_buffer(bh);
2749
2750                /*
2751                 * If the buffer has the write error flag, we have failed
2752                 * to write out another inode in the same block.  In this
2753                 * case, we don't have to read the block because we may
2754                 * read the old inode data successfully.
2755                 */
2756                if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2757                        set_buffer_uptodate(bh);
2758
2759                if (buffer_uptodate(bh)) {
2760                        /* someone brought it uptodate while we waited */
2761                        unlock_buffer(bh);
2762                        goto has_buffer;
2763                }
2764
2765                /*
2766                 * If we have all information of the inode in memory and this
2767                 * is the only valid inode in the block, we need not read the
2768                 * block.
2769                 */
2770                if (in_mem) {
2771                        struct buffer_head *bitmap_bh;
2772                        struct ext3_group_desc *desc;
2773                        int inodes_per_buffer;
2774                        int inode_offset, i;
2775                        int block_group;
2776                        int start;
2777
2778                        block_group = (inode->i_ino - 1) /
2779                                        EXT3_INODES_PER_GROUP(inode->i_sb);
2780                        inodes_per_buffer = bh->b_size /
2781                                EXT3_INODE_SIZE(inode->i_sb);
2782                        inode_offset = ((inode->i_ino - 1) %
2783                                        EXT3_INODES_PER_GROUP(inode->i_sb));
2784                        start = inode_offset & ~(inodes_per_buffer - 1);
2785
2786                        /* Is the inode bitmap in cache? */
2787                        desc = ext3_get_group_desc(inode->i_sb,
2788                                                block_group, NULL);
2789                        if (!desc)
2790                                goto make_io;
2791
2792                        bitmap_bh = sb_getblk(inode->i_sb,
2793                                        le32_to_cpu(desc->bg_inode_bitmap));
2794                        if (unlikely(!bitmap_bh))
2795                                goto make_io;
2796
2797                        /*
2798                         * If the inode bitmap isn't in cache then the
2799                         * optimisation may end up performing two reads instead
2800                         * of one, so skip it.
2801                         */
2802                        if (!buffer_uptodate(bitmap_bh)) {
2803                                brelse(bitmap_bh);
2804                                goto make_io;
2805                        }
2806                        for (i = start; i < start + inodes_per_buffer; i++) {
2807                                if (i == inode_offset)
2808                                        continue;
2809                                if (ext3_test_bit(i, bitmap_bh->b_data))
2810                                        break;
2811                        }
2812                        brelse(bitmap_bh);
2813                        if (i == start + inodes_per_buffer) {
2814                                /* all other inodes are free, so skip I/O */
2815                                memset(bh->b_data, 0, bh->b_size);
2816                                set_buffer_uptodate(bh);
2817                                unlock_buffer(bh);
2818                                goto has_buffer;
2819                        }
2820                }
2821
2822make_io:
2823                /*
2824                 * There are other valid inodes in the buffer, this inode
2825                 * has in-inode xattrs, or we don't have this inode in memory.
2826                 * Read the block from disk.
2827                 */
2828                trace_ext3_load_inode(inode);
2829                get_bh(bh);
2830                bh->b_end_io = end_buffer_read_sync;
2831                submit_bh(READ | REQ_META | REQ_PRIO, bh);
2832                wait_on_buffer(bh);
2833                if (!buffer_uptodate(bh)) {
2834                        ext3_error(inode->i_sb, "ext3_get_inode_loc",
2835                                        "unable to read inode block - "
2836                                        "inode=%lu, block="E3FSBLK,
2837                                        inode->i_ino, block);
2838                        brelse(bh);
2839                        return -EIO;
2840                }
2841        }
2842has_buffer:
2843        iloc->bh = bh;
2844        return 0;
2845}
2846
2847int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2848{
2849        /* We have all inode data except xattrs in memory here. */
2850        return __ext3_get_inode_loc(inode, iloc,
2851                !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
2852}
2853
2854void ext3_set_inode_flags(struct inode *inode)
2855{
2856        unsigned int flags = EXT3_I(inode)->i_flags;
2857
2858        inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2859        if (flags & EXT3_SYNC_FL)
2860                inode->i_flags |= S_SYNC;
2861        if (flags & EXT3_APPEND_FL)
2862                inode->i_flags |= S_APPEND;
2863        if (flags & EXT3_IMMUTABLE_FL)
2864                inode->i_flags |= S_IMMUTABLE;
2865        if (flags & EXT3_NOATIME_FL)
2866                inode->i_flags |= S_NOATIME;
2867        if (flags & EXT3_DIRSYNC_FL)
2868                inode->i_flags |= S_DIRSYNC;
2869}
2870
2871/* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2872void ext3_get_inode_flags(struct ext3_inode_info *ei)
2873{
2874        unsigned int flags = ei->vfs_inode.i_flags;
2875
2876        ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2877                        EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2878        if (flags & S_SYNC)
2879                ei->i_flags |= EXT3_SYNC_FL;
2880        if (flags & S_APPEND)
2881                ei->i_flags |= EXT3_APPEND_FL;
2882        if (flags & S_IMMUTABLE)
2883                ei->i_flags |= EXT3_IMMUTABLE_FL;
2884        if (flags & S_NOATIME)
2885                ei->i_flags |= EXT3_NOATIME_FL;
2886        if (flags & S_DIRSYNC)
2887                ei->i_flags |= EXT3_DIRSYNC_FL;
2888}
2889
2890struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
2891{
2892        struct ext3_iloc iloc;
2893        struct ext3_inode *raw_inode;
2894        struct ext3_inode_info *ei;
2895        struct buffer_head *bh;
2896        struct inode *inode;
2897        journal_t *journal = EXT3_SB(sb)->s_journal;
2898        transaction_t *transaction;
2899        long ret;
2900        int block;
2901        uid_t i_uid;
2902        gid_t i_gid;
2903
2904        inode = iget_locked(sb, ino);
2905        if (!inode)
2906                return ERR_PTR(-ENOMEM);
2907        if (!(inode->i_state & I_NEW))
2908                return inode;
2909
2910        ei = EXT3_I(inode);
2911        ei->i_block_alloc_info = NULL;
2912
2913        ret = __ext3_get_inode_loc(inode, &iloc, 0);
2914        if (ret < 0)
2915                goto bad_inode;
2916        bh = iloc.bh;
2917        raw_inode = ext3_raw_inode(&iloc);
2918        inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2919        i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2920        i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2921        if(!(test_opt (inode->i_sb, NO_UID32))) {
2922                i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2923                i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2924        }
2925        i_uid_write(inode, i_uid);
2926        i_gid_write(inode, i_gid);
2927        set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
2928        inode->i_size = le32_to_cpu(raw_inode->i_size);
2929        inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2930        inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2931        inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2932        inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2933
2934        ei->i_state_flags = 0;
2935        ei->i_dir_start_lookup = 0;
2936        ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2937        /* We now have enough fields to check if the inode was active or not.
2938         * This is needed because nfsd might try to access dead inodes
2939         * the test is that same one that e2fsck uses
2940         * NeilBrown 1999oct15
2941         */
2942        if (inode->i_nlink == 0) {
2943                if (inode->i_mode == 0 ||
2944                    !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2945                        /* this inode is deleted */
2946                        brelse (bh);
2947                        ret = -ESTALE;
2948                        goto bad_inode;
2949                }
2950                /* The only unlinked inodes we let through here have
2951                 * valid i_mode and are being read by the orphan
2952                 * recovery code: that's fine, we're about to complete
2953                 * the process of deleting those. */
2954        }
2955        inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2956        ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2957#ifdef EXT3_FRAGMENTS
2958        ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2959        ei->i_frag_no = raw_inode->i_frag;
2960        ei->i_frag_size = raw_inode->i_fsize;
2961#endif
2962        ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2963        if (!S_ISREG(inode->i_mode)) {
2964                ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2965        } else {
2966                inode->i_size |=
2967                        ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2968        }
2969        ei->i_disksize = inode->i_size;
2970        inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2971        ei->i_block_group = iloc.block_group;
2972        /*
2973         * NOTE! The in-memory inode i_data array is in little-endian order
2974         * even on big-endian machines: we do NOT byteswap the block numbers!
2975         */
2976        for (block = 0; block < EXT3_N_BLOCKS; block++)
2977                ei->i_data[block] = raw_inode->i_block[block];
2978        INIT_LIST_HEAD(&ei->i_orphan);
2979
2980        /*
2981         * Set transaction id's of transactions that have to be committed
2982         * to finish f[data]sync. We set them to currently running transaction
2983         * as we cannot be sure that the inode or some of its metadata isn't
2984         * part of the transaction - the inode could have been reclaimed and
2985         * now it is reread from disk.
2986         */
2987        if (journal) {
2988                tid_t tid;
2989
2990                spin_lock(&journal->j_state_lock);
2991                if (journal->j_running_transaction)
2992                        transaction = journal->j_running_transaction;
2993                else
2994                        transaction = journal->j_committing_transaction;
2995                if (transaction)
2996                        tid = transaction->t_tid;
2997                else
2998                        tid = journal->j_commit_sequence;
2999                spin_unlock(&journal->j_state_lock);
3000                atomic_set(&ei->i_sync_tid, tid);
3001                atomic_set(&ei->i_datasync_tid, tid);
3002        }
3003
3004        if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
3005            EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
3006                /*
3007                 * When mke2fs creates big inodes it does not zero out
3008                 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3009                 * so ignore those first few inodes.
3010                 */
3011                ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3012                if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3013                    EXT3_INODE_SIZE(inode->i_sb)) {
3014                        brelse (bh);
3015                        ret = -EIO;
3016                        goto bad_inode;
3017                }
3018                if (ei->i_extra_isize == 0) {
3019                        /* The extra space is currently unused. Use it. */
3020                        ei->i_extra_isize = sizeof(struct ext3_inode) -
3021                                            EXT3_GOOD_OLD_INODE_SIZE;
3022                } else {
3023                        __le32 *magic = (void *)raw_inode +
3024                                        EXT3_GOOD_OLD_INODE_SIZE +
3025                                        ei->i_extra_isize;
3026                        if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
3027                                 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
3028                }
3029        } else
3030                ei->i_extra_isize = 0;
3031
3032        if (S_ISREG(inode->i_mode)) {
3033                inode->i_op = &ext3_file_inode_operations;
3034                inode->i_fop = &ext3_file_operations;
3035                ext3_set_aops(inode);
3036        } else if (S_ISDIR(inode->i_mode)) {
3037                inode->i_op = &ext3_dir_inode_operations;
3038                inode->i_fop = &ext3_dir_operations;
3039        } else if (S_ISLNK(inode->i_mode)) {
3040                if (ext3_inode_is_fast_symlink(inode)) {
3041                        inode->i_op = &ext3_fast_symlink_inode_operations;
3042                        nd_terminate_link(ei->i_data, inode->i_size,
3043                                sizeof(ei->i_data) - 1);
3044                } else {
3045                        inode->i_op = &ext3_symlink_inode_operations;
3046                        ext3_set_aops(inode);
3047                }
3048        } else {
3049                inode->i_op = &ext3_special_inode_operations;
3050                if (raw_inode->i_block[0])
3051                        init_special_inode(inode, inode->i_mode,
3052                           old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3053                else
3054                        init_special_inode(inode, inode->i_mode,
3055                           new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3056        }
3057        brelse (iloc.bh);
3058        ext3_set_inode_flags(inode);
3059        unlock_new_inode(inode);
3060        return inode;
3061
3062bad_inode:
3063        iget_failed(inode);
3064        return ERR_PTR(ret);
3065}
3066
3067/*
3068 * Post the struct inode info into an on-disk inode location in the
3069 * buffer-cache.  This gobbles the caller's reference to the
3070 * buffer_head in the inode location struct.
3071 *
3072 * The caller must have write access to iloc->bh.
3073 */
3074static int ext3_do_update_inode(handle_t *handle,
3075                                struct inode *inode,
3076                                struct ext3_iloc *iloc)
3077{
3078        struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3079        struct ext3_inode_info *ei = EXT3_I(inode);
3080        struct buffer_head *bh = iloc->bh;
3081        int err = 0, rc, block;
3082        int need_datasync = 0;
3083        __le32 disksize;
3084        uid_t i_uid;
3085        gid_t i_gid;
3086
3087again:
3088        /* we can't allow multiple procs in here at once, its a bit racey */
3089        lock_buffer(bh);
3090
3091        /* For fields not not tracking in the in-memory inode,
3092         * initialise them to zero for new inodes. */
3093        if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
3094                memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3095
3096        ext3_get_inode_flags(ei);
3097        raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3098        i_uid = i_uid_read(inode);
3099        i_gid = i_gid_read(inode);
3100        if(!(test_opt(inode->i_sb, NO_UID32))) {
3101                raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3102                raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3103/*
3104 * Fix up interoperability with old kernels. Otherwise, old inodes get
3105 * re-used with the upper 16 bits of the uid/gid intact
3106 */
3107                if(!ei->i_dtime) {
3108                        raw_inode->i_uid_high =
3109                                cpu_to_le16(high_16_bits(i_uid));
3110                        raw_inode->i_gid_high =
3111                                cpu_to_le16(high_16_bits(i_gid));
3112                } else {
3113                        raw_inode->i_uid_high = 0;
3114                        raw_inode->i_gid_high = 0;
3115                }
3116        } else {
3117                raw_inode->i_uid_low =
3118                        cpu_to_le16(fs_high2lowuid(i_uid));
3119                raw_inode->i_gid_low =
3120                        cpu_to_le16(fs_high2lowgid(i_gid));
3121                raw_inode->i_uid_high = 0;
3122                raw_inode->i_gid_high = 0;
3123        }
3124        raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3125        disksize = cpu_to_le32(ei->i_disksize);
3126        if (disksize != raw_inode->i_size) {
3127                need_datasync = 1;
3128                raw_inode->i_size = disksize;
3129        }
3130        raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3131        raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3132        raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3133        raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3134        raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3135        raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3136#ifdef EXT3_FRAGMENTS
3137        raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3138        raw_inode->i_frag = ei->i_frag_no;
3139        raw_inode->i_fsize = ei->i_frag_size;
3140#endif
3141        raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3142        if (!S_ISREG(inode->i_mode)) {
3143                raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3144        } else {
3145                disksize = cpu_to_le32(ei->i_disksize >> 32);
3146                if (disksize != raw_inode->i_size_high) {
3147                        raw_inode->i_size_high = disksize;
3148                        need_datasync = 1;
3149                }
3150                if (ei->i_disksize > 0x7fffffffULL) {
3151                        struct super_block *sb = inode->i_sb;
3152                        if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3153                                        EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3154                            EXT3_SB(sb)->s_es->s_rev_level ==
3155                                        cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3156                               /* If this is the first large file
3157                                * created, add a flag to the superblock.
3158                                */
3159                                unlock_buffer(bh);
3160                                err = ext3_journal_get_write_access(handle,
3161                                                EXT3_SB(sb)->s_sbh);
3162                                if (err)
3163                                        goto out_brelse;
3164
3165                                ext3_update_dynamic_rev(sb);
3166                                EXT3_SET_RO_COMPAT_FEATURE(sb,
3167                                        EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
3168                                handle->h_sync = 1;
3169                                err = ext3_journal_dirty_metadata(handle,
3170                                                EXT3_SB(sb)->s_sbh);
3171                                /* get our lock and start over */
3172                                goto again;
3173                        }
3174                }
3175        }
3176        raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3177        if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3178                if (old_valid_dev(inode->i_rdev)) {
3179                        raw_inode->i_block[0] =
3180                                cpu_to_le32(old_encode_dev(inode->i_rdev));
3181                        raw_inode->i_block[1] = 0;
3182                } else {
3183                        raw_inode->i_block[0] = 0;
3184                        raw_inode->i_block[1] =
3185                                cpu_to_le32(new_encode_dev(inode->i_rdev));
3186                        raw_inode->i_block[2] = 0;
3187                }
3188        } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3189                raw_inode->i_block[block] = ei->i_data[block];
3190
3191        if (ei->i_extra_isize)
3192                raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3193
3194        BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
3195        unlock_buffer(bh);
3196        rc = ext3_journal_dirty_metadata(handle, bh);
3197        if (!err)
3198                err = rc;
3199        ext3_clear_inode_state(inode, EXT3_STATE_NEW);
3200
3201        atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
3202        if (need_datasync)
3203                atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
3204out_brelse:
3205        brelse (bh);
3206        ext3_std_error(inode->i_sb, err);
3207        return err;
3208}
3209
3210/*
3211 * ext3_write_inode()
3212 *
3213 * We are called from a few places:
3214 *
3215 * - Within generic_file_write() for O_SYNC files.
3216 *   Here, there will be no transaction running. We wait for any running
3217 *   transaction to commit.
3218 *
3219 * - Within sys_sync(), kupdate and such.
3220 *   We wait on commit, if tol to.
3221 *
3222 * - Within prune_icache() (PF_MEMALLOC == true)
3223 *   Here we simply return.  We can't afford to block kswapd on the
3224 *   journal commit.
3225 *
3226 * In all cases it is actually safe for us to return without doing anything,
3227 * because the inode has been copied into a raw inode buffer in
3228 * ext3_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
3229 * knfsd.
3230 *
3231 * Note that we are absolutely dependent upon all inode dirtiers doing the
3232 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3233 * which we are interested.
3234 *
3235 * It would be a bug for them to not do this.  The code:
3236 *
3237 *      mark_inode_dirty(inode)
3238 *      stuff();
3239 *      inode->i_size = expr;
3240 *
3241 * is in error because a kswapd-driven write_inode() could occur while
3242 * `stuff()' is running, and the new i_size will be lost.  Plus the inode
3243 * will no longer be on the superblock's dirty inode list.
3244 */
3245int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
3246{
3247        if (current->flags & PF_MEMALLOC)
3248                return 0;
3249
3250        if (ext3_journal_current_handle()) {
3251                jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3252                dump_stack();
3253                return -EIO;
3254        }
3255
3256        if (wbc->sync_mode != WB_SYNC_ALL)
3257                return 0;
3258
3259        return ext3_force_commit(inode->i_sb);
3260}
3261
3262/*
3263 * ext3_setattr()
3264 *
3265 * Called from notify_change.
3266 *
3267 * We want to trap VFS attempts to truncate the file as soon as
3268 * possible.  In particular, we want to make sure that when the VFS
3269 * shrinks i_size, we put the inode on the orphan list and modify
3270 * i_disksize immediately, so that during the subsequent flushing of
3271 * dirty pages and freeing of disk blocks, we can guarantee that any
3272 * commit will leave the blocks being flushed in an unused state on
3273 * disk.  (On recovery, the inode will get truncated and the blocks will
3274 * be freed, so we have a strong guarantee that no future commit will
3275 * leave these blocks visible to the user.)
3276 *
3277 * Called with inode->sem down.
3278 */
3279int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3280{
3281        struct inode *inode = dentry->d_inode;
3282        int error, rc = 0;
3283        const unsigned int ia_valid = attr->ia_valid;
3284
3285        error = inode_change_ok(inode, attr);
3286        if (error)
3287                return error;
3288
3289        if (is_quota_modification(inode, attr))
3290                dquot_initialize(inode);
3291        if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3292            (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3293                handle_t *handle;
3294
3295                /* (user+group)*(old+new) structure, inode write (sb,
3296                 * inode block, ? - but truncate inode update has it) */
3297                handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3298                                        EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
3299                if (IS_ERR(handle)) {
3300                        error = PTR_ERR(handle);
3301                        goto err_out;
3302                }
3303                error = dquot_transfer(inode, attr);
3304                if (error) {
3305                        ext3_journal_stop(handle);
3306                        return error;
3307                }
3308                /* Update corresponding info in inode so that everything is in
3309                 * one transaction */
3310                if (attr->ia_valid & ATTR_UID)
3311                        inode->i_uid = attr->ia_uid;
3312                if (attr->ia_valid & ATTR_GID)
3313                        inode->i_gid = attr->ia_gid;
3314                error = ext3_mark_inode_dirty(handle, inode);
3315                ext3_journal_stop(handle);
3316        }
3317
3318        if (attr->ia_valid & ATTR_SIZE)
3319                inode_dio_wait(inode);
3320
3321        if (S_ISREG(inode->i_mode) &&
3322            attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3323                handle_t *handle;
3324
3325                handle = ext3_journal_start(inode, 3);
3326                if (IS_ERR(handle)) {
3327                        error = PTR_ERR(handle);
3328                        goto err_out;
3329                }
3330
3331                error = ext3_orphan_add(handle, inode);
3332                if (error) {
3333                        ext3_journal_stop(handle);
3334                        goto err_out;
3335                }
3336                EXT3_I(inode)->i_disksize = attr->ia_size;
3337                error = ext3_mark_inode_dirty(handle, inode);
3338                ext3_journal_stop(handle);
3339                if (error) {
3340                        /* Some hard fs error must have happened. Bail out. */
3341                        ext3_orphan_del(NULL, inode);
3342                        goto err_out;
3343                }
3344                rc = ext3_block_truncate_page(inode, attr->ia_size);
3345                if (rc) {
3346                        /* Cleanup orphan list and exit */
3347                        handle = ext3_journal_start(inode, 3);
3348                        if (IS_ERR(handle)) {
3349                                ext3_orphan_del(NULL, inode);
3350                                goto err_out;
3351                        }
3352                        ext3_orphan_del(handle, inode);
3353                        ext3_journal_stop(handle);
3354                        goto err_out;
3355                }
3356        }
3357
3358        if ((attr->ia_valid & ATTR_SIZE) &&
3359            attr->ia_size != i_size_read(inode)) {
3360                truncate_setsize(inode, attr->ia_size);
3361                ext3_truncate(inode);
3362        }
3363
3364        setattr_copy(inode, attr);
3365        mark_inode_dirty(inode);
3366
3367        if (ia_valid & ATTR_MODE)
3368                rc = ext3_acl_chmod(inode);
3369
3370err_out:
3371        ext3_std_error(inode->i_sb, error);
3372        if (!error)
3373                error = rc;
3374        return error;
3375}
3376
3377
3378/*
3379 * How many blocks doth make a writepage()?
3380 *
3381 * With N blocks per page, it may be:
3382 * N data blocks
3383 * 2 indirect block
3384 * 2 dindirect
3385 * 1 tindirect
3386 * N+5 bitmap blocks (from the above)
3387 * N+5 group descriptor summary blocks
3388 * 1 inode block
3389 * 1 superblock.
3390 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3391 *
3392 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3393 *
3394 * With ordered or writeback data it's the same, less the N data blocks.
3395 *
3396 * If the inode's direct blocks can hold an integral number of pages then a
3397 * page cannot straddle two indirect blocks, and we can only touch one indirect
3398 * and dindirect block, and the "5" above becomes "3".
3399 *
3400 * This still overestimates under most circumstances.  If we were to pass the
3401 * start and end offsets in here as well we could do block_to_path() on each
3402 * block and work out the exact number of indirects which are touched.  Pah.
3403 */
3404
3405static int ext3_writepage_trans_blocks(struct inode *inode)
3406{
3407        int bpp = ext3_journal_blocks_per_page(inode);
3408        int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3409        int ret;
3410
3411        if (ext3_should_journal_data(inode))
3412                ret = 3 * (bpp + indirects) + 2;
3413        else
3414                ret = 2 * (bpp + indirects) + indirects + 2;
3415
3416#ifdef CONFIG_QUOTA
3417        /* We know that structure was already allocated during dquot_initialize so
3418         * we will be updating only the data blocks + inodes */
3419        ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
3420#endif
3421
3422        return ret;
3423}
3424
3425/*
3426 * The caller must have previously called ext3_reserve_inode_write().
3427 * Give this, we know that the caller already has write access to iloc->bh.
3428 */
3429int ext3_mark_iloc_dirty(handle_t *handle,
3430                struct inode *inode, struct ext3_iloc *iloc)
3431{
3432        int err = 0;
3433
3434        /* the do_update_inode consumes one bh->b_count */
3435        get_bh(iloc->bh);
3436
3437        /* ext3_do_update_inode() does journal_dirty_metadata */
3438        err = ext3_do_update_inode(handle, inode, iloc);
3439        put_bh(iloc->bh);
3440        return err;
3441}
3442
3443/*
3444 * On success, We end up with an outstanding reference count against
3445 * iloc->bh.  This _must_ be cleaned up later.
3446 */
3447
3448int
3449ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3450                         struct ext3_iloc *iloc)
3451{
3452        int err = 0;
3453        if (handle) {
3454                err = ext3_get_inode_loc(inode, iloc);
3455                if (!err) {
3456                        BUFFER_TRACE(iloc->bh, "get_write_access");
3457                        err = ext3_journal_get_write_access(handle, iloc->bh);
3458                        if (err) {
3459                                brelse(iloc->bh);
3460                                iloc->bh = NULL;
3461                        }
3462                }
3463        }
3464        ext3_std_error(inode->i_sb, err);
3465        return err;
3466}
3467
3468/*
3469 * What we do here is to mark the in-core inode as clean with respect to inode
3470 * dirtiness (it may still be data-dirty).
3471 * This means that the in-core inode may be reaped by prune_icache
3472 * without having to perform any I/O.  This is a very good thing,
3473 * because *any* task may call prune_icache - even ones which
3474 * have a transaction open against a different journal.
3475 *
3476 * Is this cheating?  Not really.  Sure, we haven't written the
3477 * inode out, but prune_icache isn't a user-visible syncing function.
3478 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3479 * we start and wait on commits.
3480 */
3481int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3482{
3483        struct ext3_iloc iloc;
3484        int err;
3485
3486        might_sleep();
3487        trace_ext3_mark_inode_dirty(inode, _RET_IP_);
3488        err = ext3_reserve_inode_write(handle, inode, &iloc);
3489        if (!err)
3490                err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3491        return err;
3492}
3493
3494/*
3495 * ext3_dirty_inode() is called from __mark_inode_dirty()
3496 *
3497 * We're really interested in the case where a file is being extended.
3498 * i_size has been changed by generic_commit_write() and we thus need
3499 * to include the updated inode in the current transaction.
3500 *
3501 * Also, dquot_alloc_space() will always dirty the inode when blocks
3502 * are allocated to the file.
3503 *
3504 * If the inode is marked synchronous, we don't honour that here - doing
3505 * so would cause a commit on atime updates, which we don't bother doing.
3506 * We handle synchronous inodes at the highest possible level.
3507 */
3508void ext3_dirty_inode(struct inode *inode, int flags)
3509{
3510        handle_t *current_handle = ext3_journal_current_handle();
3511        handle_t *handle;
3512
3513        handle = ext3_journal_start(inode, 2);
3514        if (IS_ERR(handle))
3515                goto out;
3516        if (current_handle &&
3517                current_handle->h_transaction != handle->h_transaction) {
3518                /* This task has a transaction open against a different fs */
3519                printk(KERN_EMERG "%s: transactions do not match!\n",
3520                       __func__);
3521        } else {
3522                jbd_debug(5, "marking dirty.  outer handle=%p\n",
3523                                current_handle);
3524                ext3_mark_inode_dirty(handle, inode);
3525        }
3526        ext3_journal_stop(handle);
3527out:
3528        return;
3529}
3530
3531#if 0
3532/*
3533 * Bind an inode's backing buffer_head into this transaction, to prevent
3534 * it from being flushed to disk early.  Unlike
3535 * ext3_reserve_inode_write, this leaves behind no bh reference and
3536 * returns no iloc structure, so the caller needs to repeat the iloc
3537 * lookup to mark the inode dirty later.
3538 */
3539static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3540{
3541        struct ext3_iloc iloc;
3542
3543        int err = 0;
3544        if (handle) {
3545                err = ext3_get_inode_loc(inode, &iloc);
3546                if (!err) {
3547                        BUFFER_TRACE(iloc.bh, "get_write_access");
3548                        err = journal_get_write_access(handle, iloc.bh);
3549                        if (!err)
3550                                err = ext3_journal_dirty_metadata(handle,
3551                                                                  iloc.bh);
3552                        brelse(iloc.bh);
3553                }
3554        }
3555        ext3_std_error(inode->i_sb, err);
3556        return err;
3557}
3558#endif
3559
3560int ext3_change_inode_journal_flag(struct inode *inode, int val)
3561{
3562        journal_t *journal;
3563        handle_t *handle;
3564        int err;
3565
3566        /*
3567         * We have to be very careful here: changing a data block's
3568         * journaling status dynamically is dangerous.  If we write a
3569         * data block to the journal, change the status and then delete
3570         * that block, we risk forgetting to revoke the old log record
3571         * from the journal and so a subsequent replay can corrupt data.
3572         * So, first we make sure that the journal is empty and that
3573         * nobody is changing anything.
3574         */
3575
3576        journal = EXT3_JOURNAL(inode);
3577        if (is_journal_aborted(journal))
3578                return -EROFS;
3579
3580        journal_lock_updates(journal);
3581        journal_flush(journal);
3582
3583        /*
3584         * OK, there are no updates running now, and all cached data is
3585         * synced to disk.  We are now in a completely consistent state
3586         * which doesn't have anything in the journal, and we know that
3587         * no filesystem updates are running, so it is safe to modify
3588         * the inode's in-core data-journaling state flag now.
3589         */
3590
3591        if (val)
3592                EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3593        else
3594                EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3595        ext3_set_aops(inode);
3596
3597        journal_unlock_updates(journal);
3598
3599        /* Finally we can mark the inode as dirty. */
3600
3601        handle = ext3_journal_start(inode, 1);
3602        if (IS_ERR(handle))
3603                return PTR_ERR(handle);
3604
3605        err = ext3_mark_inode_dirty(handle, inode);
3606        handle->h_sync = 1;
3607        ext3_journal_stop(handle);
3608        ext3_std_error(inode->i_sb, err);
3609
3610        return err;
3611}
3612
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.