journal.c

ARM 嵌入式 系统 设计与实例开发 实验教材 二源码

/* buffer is in current transaction */
inline int buffer_journaled(const struct buffer_head *bh) {
  if (bh)
    return test_bit(BH_JDirty, ( struct buffer_head * ) &bh->b_state) ;
  else
    return 0 ;
}

/* disk block was taken off free list before being in a finished transation, or written to disk
** journal_new blocks can be reused immediately, for any purpose
*/ 
inline int buffer_journal_new(const struct buffer_head *bh) {
  if (bh) 
    return test_bit(BH_JNew, ( struct buffer_head * )&bh->b_state) ;
  else
    return 0 ;
}

inline int mark_buffer_journal_new(struct buffer_head *bh) {
  if (bh) {
    set_bit(BH_JNew, &bh->b_state) ;
  }
  return 0 ;
}

inline int mark_buffer_not_journaled(struct buffer_head *bh) {
  if (bh) 
    clear_bit(BH_JDirty, &bh->b_state) ;
  return 0 ;
}

/* utility function to force a BUG if it is called without the big
** kernel lock held.  caller is the string printed just before calling BUG()
*/
void reiserfs_check_lock_depth(char *caller) {
#ifdef CONFIG_SMP
  if (current->lock_depth < 0) {
    printk("%s called without kernel lock held\n", caller) ;
    show_reiserfs_locks() ;
    BUG() ;
  }
#else
  ;
#endif
}

/* return a cnode with same dev, block number and size in table, or null if not found */
static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct reiserfs_journal_cnode **table,
                                   				  kdev_t dev,long bl,int size) {
  struct reiserfs_journal_cnode *cn ;
  cn = journal_hash(table, dev, bl) ;
  while(cn) {
    if ((cn->blocknr == bl) && (cn->dev == dev))
      return cn ;
    cn = cn->hnext ;
  }
  return (struct reiserfs_journal_cnode *)0 ;
}

/* returns a cnode with same size, block number and dev as bh in the current transaction hash.  NULL if not found */
static inline struct reiserfs_journal_cnode *get_journal_hash(struct super_block *p_s_sb, struct buffer_head *bh) {
  struct reiserfs_journal_cnode *cn ;
  if (bh) {
    cn =  get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, bh->b_dev, bh->b_blocknr, bh->b_size) ;
  }
  else {
    return (struct reiserfs_journal_cnode *)0 ;
  }
  return cn ;
}

/* once upon a time, the journal would deadlock.  a lot.  Now, when
** CONFIG_REISERFS_CHECK is defined, anytime someone enters a
** transaction, it pushes itself into this ugly static list, and pops
** itself off before calling journal_end.  I made a SysRq key to dump
** the list, and tell me what the writers are when I'm deadlocked.  */

				/* are you depending on the compiler
                                   to optimize this function away
                                   everywhere it is called? It is not
                                   obvious how this works, but I
                                   suppose debugging code need not be
                                   clear.  -Hans */
static char *journal_writers[512] ;
int push_journal_writer(char *s) {
#ifdef CONFIG_REISERFS_CHECK
  int i ;
  for (i = 0 ; i < 512 ; i++) {
    if (!journal_writers[i]) {
      journal_writers[i] = s ;
      return i ;
    }
  }
  return -1 ;
#else
  return 0 ;
#endif
}
int pop_journal_writer(int index) {
#ifdef CONFIG_REISERFS_CHECK
  if (index >= 0) {
    journal_writers[index] = NULL ;
  }
#endif
  return 0 ;
}

int dump_journal_writers(void) {
  int i ;
  for (i = 0 ; i < 512 ; i++) {
    if (journal_writers[i]) {
      printk("%d: %s\n", i, journal_writers[i]) ;
    }
  }
  return 0 ;
}

/*
** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
** being overwritten by a replay after crashing.
**
** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
** sure you never write the block without logging it.
**
** next_zero_bit is a suggestion about the next block to try for find_forward.
** when bl is rejected because it is set in a journal list bitmap, we search
** for the next zero bit in the bitmap that rejected bl.  Then, we return that
** through next_zero_bit for find_forward to try.
**
** Just because we return something in next_zero_bit does not mean we won't
** reject it on the next call to reiserfs_in_journal
**
*/
int reiserfs_in_journal(struct super_block *p_s_sb, kdev_t dev, 
                        unsigned long bl, int size, int search_all, 
			unsigned long *next_zero_bit) {
  struct reiserfs_journal_cnode *cn ;
  struct reiserfs_list_bitmap *jb ;
  int i ;
  int bmap_nr = bl / (p_s_sb->s_blocksize << 3) ;
  int bit_nr = bl % (p_s_sb->s_blocksize << 3) ;
  int tmp_bit ;

  *next_zero_bit = 0 ; /* always start this at zero. */

  /* we aren't logging all blocks are safe for reuse */
  if (reiserfs_dont_log(p_s_sb)) {
    return 0 ;
  }

  PROC_INFO_INC( p_s_sb, journal.in_journal );
  /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
  ** if we crash before the transaction that freed it commits,  this transaction won't
  ** have committed either, and the block will never be written
  */
  if (search_all) {
    for (i = 0 ; i < JOURNAL_NUM_BITMAPS ; i++) {
      PROC_INFO_INC( p_s_sb, journal.in_journal_bitmap );
      jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
      if (jb->journal_list && jb->bitmaps[bmap_nr] &&
          test_bit(bit_nr, jb->bitmaps[bmap_nr]->data)) {
	tmp_bit = find_next_zero_bit((unsigned long *)
	                             (jb->bitmaps[bmap_nr]->data),
	                             p_s_sb->s_blocksize << 3, bit_nr+1) ; 
	*next_zero_bit = bmap_nr * (p_s_sb->s_blocksize << 3) + tmp_bit ;
	return 1 ;
      }
    }
  }

  /* is it in any old transactions? */
  if (search_all && (cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_list_hash_table, dev,bl,size))) {
    return 1; 
  }

  /* is it in the current transaction.  This should never happen */
  if ((cn = get_journal_hash_dev(SB_JOURNAL(p_s_sb)->j_hash_table, dev,bl,size))) {
    return 1; 
  }

  PROC_INFO_INC( p_s_sb, journal.in_journal_reusable );
  /* safe for reuse */
  return 0 ;
}

/* insert cn into table
*/
inline void insert_journal_hash(struct reiserfs_journal_cnode **table, struct reiserfs_journal_cnode *cn) {
  struct reiserfs_journal_cnode *cn_orig ;

  cn_orig = journal_hash(table, cn->dev, cn->blocknr) ;
  cn->hnext = cn_orig ;
  cn->hprev = NULL ;
  if (cn_orig) {
    cn_orig->hprev = cn ;
  }
  journal_hash(table, cn->dev, cn->blocknr) =  cn ;
}

/* lock the current transaction */
inline static void lock_journal(struct super_block *p_s_sb) {
  PROC_INFO_INC( p_s_sb, journal.lock_journal );
  while(atomic_read(&(SB_JOURNAL(p_s_sb)->j_wlock)) > 0) {
    PROC_INFO_INC( p_s_sb, journal.lock_journal_wait );
    sleep_on(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
  }
  atomic_set(&(SB_JOURNAL(p_s_sb)->j_wlock), 1) ;
}

/* unlock the current transaction */
inline static void unlock_journal(struct super_block *p_s_sb) {
  atomic_dec(&(SB_JOURNAL(p_s_sb)->j_wlock)) ;
  wake_up(&(SB_JOURNAL(p_s_sb)->j_wait)) ;
}

/*
** this used to be much more involved, and I'm keeping it just in case things get ugly again.
** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
** transaction.
*/
static void cleanup_freed_for_journal_list(struct super_block *p_s_sb, struct reiserfs_journal_list *jl) {

  struct reiserfs_list_bitmap *jb = jl->j_list_bitmap ;
  if (jb) {
    cleanup_bitmap_list(p_s_sb, jb) ;
  }
  jl->j_list_bitmap->journal_list = NULL ;
  jl->j_list_bitmap = NULL ;
}

/*
** if this journal list still has commit blocks unflushed, send them to disk.
**
** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
** Before the commit block can by written, every other log block must be safely on disk
**
*/
static int flush_commit_list(struct super_block *s, struct reiserfs_journal_list *jl, int flushall) {
  int i, count ;
  int index = 0 ;
  int bn ;
  int retry_count = 0 ;
  int orig_commit_left = 0 ;
  struct buffer_head *tbh = NULL ;
  struct reiserfs_journal_list *other_jl ;

  reiserfs_check_lock_depth("flush_commit_list") ;

  if (atomic_read(&jl->j_older_commits_done)) {
    return 0 ;
  }

  /* before we can put our commit blocks on disk, we have to make sure everyone older than
  ** us is on disk too
  */
  if (jl->j_len <= 0) {
    return 0 ;
  }
  if (flushall) {
    /* we _must_ make sure the transactions are committed in order.  Start with the
    ** index after this one, wrap all the way around 
    */
    index = (jl - SB_JOURNAL_LIST(s)) + 1 ;
    for (i = 0 ; i < JOURNAL_LIST_COUNT ; i++) {
      other_jl = SB_JOURNAL_LIST(s) + ( (index + i) % JOURNAL_LIST_COUNT) ;
      if (other_jl && other_jl != jl && other_jl->j_len > 0 && other_jl->j_trans_id > 0 && 
          other_jl->j_trans_id <= jl->j_trans_id && (atomic_read(&(jl->j_older_commits_done)) == 0)) {
        flush_commit_list(s, other_jl, 0) ;
      }
    }
  }

  count = 0 ;
  /* don't flush the commit list for the current transactoin */
  if (jl == ((SB_JOURNAL_LIST(s) + SB_JOURNAL_LIST_INDEX(s)))) {
    return 0 ;
  }

  /* make sure nobody is trying to flush this one at the same time */
  if (atomic_read(&(jl->j_commit_flushing))) {
    sleep_on(&(jl->j_commit_wait)) ;
    if (flushall) {
      atomic_set(&(jl->j_older_commits_done), 1) ;
    }
    return 0 ;
  }
  
  /* this commit is done, exit */
  if (atomic_read(&(jl->j_commit_left)) <= 0) {
    if (flushall) {
      atomic_set(&(jl->j_older_commits_done), 1) ;
    }
    return 0 ;
  }
  /* keeps others from flushing while we are flushing */
  atomic_set(&(jl->j_commit_flushing), 1) ; 


  if (jl->j_len > JOURNAL_TRANS_MAX) {
    reiserfs_panic(s, "journal-512: flush_commit_list: length is %lu, list number %d\n", jl->j_len, jl - SB_JOURNAL_LIST(s)) ;
    return 0 ;
  }

  orig_commit_left = atomic_read(&(jl->j_commit_left)) ; 

  /* start by checking all the commit blocks in this transaction.  
  ** Add anyone not on disk into tbh.  Stop checking once commit_left <= 1, because that means we
  ** only have the commit block left 
  */
retry:
  count = 0 ;
  for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 && i < (jl->j_len + 1) ; i++) {  /* everything but commit_bh */
    bn = reiserfs_get_journal_block(s) + (jl->j_start+i) % JOURNAL_BLOCK_COUNT;
    tbh = sb_get_hash_table(s, bn) ;

/* kill this sanity check */
if (count > (orig_commit_left + 2)) {
reiserfs_panic(s, "journal-539: flush_commit_list: BAD count(%d) > orig_commit_left(%d)!\n", count, orig_commit_left) ;
}
    if (tbh) {
      if (buffer_locked(tbh)) { /* wait on it, redo it just to make sure */
	wait_on_buffer(tbh) ;
	if (!buffer_uptodate(tbh)) {
	  reiserfs_panic(s, "journal-584, buffer write failed\n") ;
	}
      } 
      if (buffer_dirty(tbh)) {
	printk("journal-569: flush_commit_list, block already dirty!\n") ;
      } else {				
	mark_buffer_dirty(tbh) ;
      }
      ll_rw_block(WRITE, 1, &tbh) ;
      count++ ;
      put_bh(tbh) ; /* once for our get_hash */
    } 
  }

  /* wait on everyone in tbh before writing commit block*/
  if (count > 0) {
    for (i = 0 ; atomic_read(&(jl->j_commit_left)) > 1 && 
                 i < (jl->j_len + 1) ; i++) {  /* everything but commit_bh */
      bn = reiserfs_get_journal_block(s) + (jl->j_start + i) % JOURNAL_BLOCK_COUNT  ;
      tbh = sb_get_hash_table(s, bn) ;

      wait_on_buffer(tbh) ;
      if (!buffer_uptodate(tbh)) {
	reiserfs_panic(s, "journal-601, buffer write failed\n") ;
      }
      put_bh(tbh) ; /* once for our get_hash */
      bforget(tbh) ;    /* once due to original getblk in do_journal_end */
      atomic_dec(&(jl->j_commit_left)) ;
    }
  }

  if (atomic_read(&(jl->j_commit_left)) != 1) { /* just the commit_bh left, flush it without calling getblk for everyone */
    if (retry_count < 2) {
      printk("journal-582: flush_commit_list, not all log blocks on disk yet, trying again\n") ;
      retry_count++ ;
      goto retry;
    }
    reiserfs_panic(s, "journal-563: flush_commit_list: BAD, j_commit_left is %u, should be 1\n", 
		   atomic_read(&(jl->j_commit_left)));
  }

  mark_buffer_dirty(jl->j_commit_bh) ;
  ll_rw_block(WRITE, 1, &(jl->j_commit_bh)) ;
  wait_on_buffer(jl->j_commit_bh) ;
  if (!buffer_uptodate(jl->j_commit_bh)) {