journal.c

ARM 嵌入式 系统 设计与实例开发 实验教材 二源码

    /* if someone is getting the commit list, we must wait for them */
    while (atomic_read(&(jl->j_commit_flushing))) {
        sleep_on(&(jl->j_commit_wait)) ;
    }
    /* if someone is flushing this list, we must wait for them */
    while (atomic_read(&(jl->j_flushing))) {
        sleep_on(&(jl->j_flush_wait)) ;
    }
    /* was it flushed while we slept? */
    if (jl->j_len <= 0 || jl->j_trans_id != orig_trans_id) {
        return 0 ;
    }

    /* this list is now ours, we can change anything we want */
    atomic_set(&(jl->j_flushing), 1) ;

loop_start:
    cn = jl->j_realblock ;
    while(cn) {
        saved_bh = NULL ;
        /* if the blocknr == 0, this has been cleared from the hash,
        ** skip it
        */
        if (cn->blocknr == 0) {
            goto next ;
        }
        /* look for a more recent transaction that logged this
        ** buffer.  Only the most recent transaction with a buffer in
        ** it is allowed to send that buffer to disk
        */
        pjl = find_newer_jl_for_cn(cn) ;
        if (run == 0 && !pjl && cn->bh && buffer_journal_dirty(cn->bh) &&
            can_dirty(cn)) 
        {
            if (!test_bit(BH_JPrepared, &cn->bh->b_state)) {
                set_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
		submit_logged_buffer(cn->bh) ;
            } else {
                /* someone else is using this buffer.  We can't 
                ** send it to disk right now because they might
                ** be changing/logging it.
                */
                ret = 1 ;
            }
        } else if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
            clear_bit(BLOCK_NEEDS_FLUSH, &cn->state) ;
            if (!pjl && cn->bh) {
                wait_on_buffer(cn->bh) ;
            }
            /* check again, someone could have logged while we scheduled */
            pjl = find_newer_jl_for_cn(cn) ;

            /* before the JDirty_wait bit is set, the 
            ** buffer is added to the hash list.  So, if we are
            ** run in the middle of a do_journal_end, we will notice
            ** if this buffer was logged and added from the latest
            ** transaction.  In this case, we don't want to decrement
            ** b_count
            */
            if (!pjl && cn->bh && buffer_journal_dirty(cn->bh)) {
                blocknr = cn->blocknr ;
                walk_cn = cn ;
                saved_bh= cn->bh ;
                /* update all older transactions to show this block
                ** was flushed
                */
                mark_buffer_notjournal_dirty(cn->bh) ;
                while(walk_cn) {
                    if (walk_cn->bh && walk_cn->blocknr == blocknr && 
                         walk_cn->dev == cn->dev) {
                        if (walk_cn->jlist) {
                            atomic_dec(&(walk_cn->jlist->j_nonzerolen)) ;
                        }
                        walk_cn->bh = NULL ;
                    }
                    walk_cn = walk_cn->hnext ;
                }
                if (atomic_read(&saved_bh->b_count) < 1) {
                    reiserfs_warning("clm-2081: bad count on %lu\n", 
                                      saved_bh->b_blocknr) ;
                }
                brelse(saved_bh) ;
            }
        }
        /*
        ** if the more recent transaction is committed to the log,
        ** this buffer can be considered flushed.  Decrement our
        ** counters to reflect one less buffer that needs writing.
        **
        ** note, this relies on all of the above code being
        ** schedule free once pjl comes back non-null.
        */
        if (pjl && cn->bh && atomic_read(&pjl->j_commit_left) == 0) {
            atomic_dec(&cn->jlist->j_nonzerolen) ;
            cn->bh = NULL ;
        } 
next:
        cn = cn->next ;
    }
    /* the first run through the loop sends all the dirty buffers to
    ** ll_rw_block.
    ** the second run through the loop does all the accounting
    */
    if (run++ == 0) {
        goto loop_start ;
    }

    atomic_set(&(jl->j_flushing), 0) ;
    wake_up(&(jl->j_flush_wait)) ;
    return ret ;
}
/* since we never give dirty buffers to bdflush/kupdate, we have to
** flush them ourselves.  This runs through the journal lists, finds
** old metadata in need of flushing and sends it to disk.
** this does not end transactions, commit anything, or free
** cnodes.
**
** returns the highest transaction id that was flushed last time
*/
static unsigned long reiserfs_journal_kupdate(struct super_block *s) {
    struct reiserfs_journal_list *jl ;
    int i ;
    int start ;
    time_t age ;
    int ret = 0 ;

    start = SB_JOURNAL_LIST_INDEX(s) ;

    /* safety check to prevent flush attempts during a mount */
    if (start < 0) {
        return 0 ;
    }
    i = (start + 1) % JOURNAL_LIST_COUNT ;
    while(i != start) {
        jl = SB_JOURNAL_LIST(s) + i  ;
        age = CURRENT_TIME - jl->j_timestamp ;
        if (jl->j_len > 0 && // age >= (JOURNAL_MAX_COMMIT_AGE * 2) && 
            atomic_read(&(jl->j_nonzerolen)) > 0 &&
            atomic_read(&(jl->j_commit_left)) == 0) {

            if (jl->j_trans_id == SB_JOURNAL(s)->j_trans_id) {
                break ;
            }
            /* if ret was already 1, we want to preserve that */
            ret |= kupdate_one_transaction(s, jl) ;
        } 
        if (atomic_read(&(jl->j_nonzerolen)) > 0) {
            ret |= 1 ;
        }
        i = (i + 1) % JOURNAL_LIST_COUNT ;
    }
    return ret ;
}

/*
** removes any nodes in table with name block and dev as bh.
** only touchs the hnext and hprev pointers.
*/
void remove_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_list *jl,struct buffer_head *bh,
                         int remove_freed){
  struct reiserfs_journal_cnode *cur ;
  struct reiserfs_journal_cnode **head ;

  if (!bh)
    return ;

  head= &(journal_hash(table, bh->b_dev, bh->b_blocknr)) ;
  if (!head) {
    return ;
  }
  cur = *head ;
  while(cur) {
    if (cur->blocknr == bh->b_blocknr && cur->dev == bh->b_dev && (jl == NULL || jl == cur->jlist) && 
        (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
      if (cur->hnext) {
        cur->hnext->hprev = cur->hprev ;
      }
      if (cur->hprev) {
	cur->hprev->hnext = cur->hnext ;
      } else {
	*head = cur->hnext ;
      }
      cur->blocknr = 0 ;
      cur->dev = 0 ;
      cur->state = 0 ;
      if (cur->bh && cur->jlist) /* anybody who clears the cur->bh will also dec the nonzerolen */
	atomic_dec(&(cur->jlist->j_nonzerolen)) ;
      cur->bh = NULL ;
      cur->jlist = NULL ;
    } 
    cur = cur->hnext ;
  }
}

static void free_journal_ram(struct super_block *p_s_sb) {
  vfree(SB_JOURNAL(p_s_sb)->j_cnode_free_orig) ;
  free_list_bitmaps(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_bitmap) ;
  free_bitmap_nodes(p_s_sb) ; /* must be after free_list_bitmaps */
  if (SB_JOURNAL(p_s_sb)->j_header_bh) {
    brelse(SB_JOURNAL(p_s_sb)->j_header_bh) ;
  }
  vfree(SB_JOURNAL(p_s_sb)) ;
}

/*
** call on unmount.  Only set error to 1 if you haven't made your way out
** of read_super() yet.  Any other caller must keep error at 0.
*/
static int do_journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb, int error) {
  struct reiserfs_transaction_handle myth ;

  /* we only want to flush out transactions if we were called with error == 0
  */
  if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
    /* end the current trans */
    do_journal_end(th, p_s_sb,10, FLUSH_ALL) ;

    /* make sure something gets logged to force our way into the flush code */
    journal_join(&myth, p_s_sb, 1) ;
    reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb), 1) ;
    journal_mark_dirty(&myth, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb)) ;
    do_journal_end(&myth, p_s_sb,1, FLUSH_ALL) ;
  }

  /* we decrement before we wake up, because the commit thread dies off
  ** when it has been woken up and the count is <= 0
  */
  reiserfs_mounted_fs_count-- ;
  wake_up(&reiserfs_commit_thread_wait) ;
  sleep_on(&reiserfs_commit_thread_done) ;

  free_journal_ram(p_s_sb) ;

  return 0 ;
}

/*
** call on unmount.  flush all journal trans, release all alloc'd ram
*/
int journal_release(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
  return do_journal_release(th, p_s_sb, 0) ;
}
/*
** only call from an error condition inside reiserfs_read_super!
*/
int journal_release_error(struct reiserfs_transaction_handle *th, struct super_block *p_s_sb) {
  return do_journal_release(th, p_s_sb, 1) ;
}

/* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
static int journal_compare_desc_commit(struct super_block *p_s_sb, struct reiserfs_journal_desc *desc, 
			               struct reiserfs_journal_commit *commit) {
  if (le32_to_cpu(commit->j_trans_id) != le32_to_cpu(desc->j_trans_id) || 
      le32_to_cpu(commit->j_len) != le32_to_cpu(desc->j_len) || 
      le32_to_cpu(commit->j_len) > JOURNAL_TRANS_MAX || 
      le32_to_cpu(commit->j_len) <= 0 
  ) {
    return 1 ;
  }
  return 0 ;
}
/* returns 0 if it did not find a description block  
** returns -1 if it found a corrupt commit block
** returns 1 if both desc and commit were valid 
*/
static int journal_transaction_is_valid(struct super_block *p_s_sb, struct buffer_head *d_bh, unsigned long *oldest_invalid_trans_id, unsigned long *newest_mount_id) {
  struct reiserfs_journal_desc *desc ;
  struct reiserfs_journal_commit *commit ;
  struct buffer_head *c_bh ;
  unsigned long offset ;

  if (!d_bh)
      return 0 ;

  desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
  if (le32_to_cpu(desc->j_len) > 0 && !memcmp(desc->j_magic, JOURNAL_DESC_MAGIC, 8)) {
    if (oldest_invalid_trans_id && *oldest_invalid_trans_id && le32_to_cpu(desc->j_trans_id) > *oldest_invalid_trans_id) {
      reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-986: transaction "
	              "is valid returning because trans_id %d is greater than "
		      "oldest_invalid %lu\n", le32_to_cpu(desc->j_trans_id), 
		       *oldest_invalid_trans_id);
      return 0 ;
    }
    if (newest_mount_id && *newest_mount_id > le32_to_cpu(desc->j_mount_id)) {
      reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1087: transaction "
                     "is valid returning because mount_id %d is less than "
		     "newest_mount_id %lu\n", desc->j_mount_id, 
		     *newest_mount_id) ;
      return -1 ;
    }
    offset = d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb) ;

    /* ok, we have a journal description block, lets see if the transaction was valid */
    c_bh = sb_bread(p_s_sb, reiserfs_get_journal_block(p_s_sb) + ((offset + le32_to_cpu(desc->j_len) + 1) % JOURNAL_BLOCK_COUNT)) ;
    if (!c_bh)
      return 0 ;
    commit = (struct reiserfs_journal_commit *)c_bh->b_data ;
    if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
      reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, 
                     "journal_transaction_is_valid, commit offset %ld had bad "
		     "time %d or length %d\n", 
		     c_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb),
		     le32_to_cpu(commit->j_trans_id), 
		     le32_to_cpu(commit->j_len));
      brelse(c_bh) ;
      if (oldest_invalid_trans_id)
        *oldest_invalid_trans_id = le32_to_cpu(desc->j_trans_id) ;
	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1004: "
	               "transaction_is_valid setting oldest invalid trans_id "
		       "to %d\n", le32_to_cpu(desc->j_trans_id)) ;
      return -1; 
    }
    brelse(c_bh) ;
    reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1006: found valid "
                   "transaction start offset %lu, len %d id %d\n", 
		   d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb), 
		   le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_trans_id)) ;
    return 1 ;
  } else {
    return 0 ;
  }
}

static void brelse_array(struct buffer_head **heads, int num) {
  int i ;
  for (i = 0 ; i < num ; i++) {
    brelse(heads[i]) ;
  }
}

/*
** given the start, and values for the oldest acceptable transactions,
** this either reads in a replays a transaction, or returns because the transaction
** is invalid, or too old.
*/
static int journal_read_transaction(struct super_block *p_s_sb, unsigned long cur_dblock, unsigned long oldest_start, 
				    unsigned long oldest_trans_id, unsigned long newest_mount_id) {
  struct reiserfs_journal_desc *desc ;
  struct reiserfs_journal_commit *commit ;
  unsigned long trans_id = 0 ;
  struct buffer_head *c_bh ;
  struct buffer_head *d_bh ;
  struct buffer_head **log_blocks = NULL ;
  struct buffer_head **real_blocks = NULL ;
  unsigned long trans_offset ;
  int i;

  d_bh = sb_bread(p_s_sb, cur_dblock) ;
  if (!d_bh)
    return 1 ;
  desc = (struct reiserfs_journal_desc *)d_bh->b_data ;
  trans_offset = d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb) ;
  reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
                 "journal_read_transaction, offset %lu, len %d mount_id %d\n", 
		 d_bh->b_blocknr - reiserfs_get_journal_block(p_s_sb), 
		 le32_to_cpu(desc->j_len), le32_to_cpu(desc->j_mount_id)) ;
  if (le32_to_cpu(desc->j_trans_id) < oldest_trans_id) {
    reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
                   "journal_read_trans skipping because %lu is too old\n", 
		   cur_dblock - reiserfs_get_journal_block(p_s_sb)) ;
    brelse(d_bh) ;
    return 1 ;
  }
  if (le32_to_cpu(desc->j_mount_id) != newest_mount_id) {
    reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
                   "journal_read_trans skipping because %d is != "
		   "newest_mount_id %lu\n", le32_to_cpu(desc->j_mount_id), 
		    newest_mount_id) ;
    brelse(d_bh) ;
    return 1 ;