optimize-jbd2_log_do_checkpoint-a-bit

   1 jbd2: optimize jbd2_log_do_checkpoint() a bit
   2
   3 From: Jan Kara <jack@suse.cz>
   4
   5 When we discover written out buffer in transaction checkpoint list we
   6 don't have to recheck validity of a transaction. Either this is the
   7 last buffer in a transaction - and then we are done - or this isn't
   8 and then we can just take another buffer from the checkpoint list
   9 without dropping j_list_lock.
  10
  11 Signed-off-by: Jan Kara <jack@suse.cz>
  12 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
  13 ---
  14  fs/jbd2/checkpoint.c | 7 ++++---
  15  1 file changed, 4 insertions(+), 3 deletions(-)
  16
  17 diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
  18 index cb6e17c..7713f94 100644
  19 --- a/fs/jbd2/checkpoint.c
  20 +++ b/fs/jbd2/checkpoint.c
  21 @@ -289,9 +289,10 @@ restart:
  22                         if (unlikely(buffer_write_io_error(bh)) && !result)
  23                                 result = -EIO;
  24                         BUFFER_TRACE(bh, "remove from checkpoint");
  25 -                       __jbd2_journal_remove_checkpoint(jh);
  26 -                       spin_unlock(&journal->j_list_lock);
  27 -                       goto retry;
  28 +                       if (__jbd2_journal_remove_checkpoint(jh))
  29 +                               /* The transaction was released; we're done */
  30 +                               goto out;
  31 +                       continue;
  32                 }
  33                 /*
  34                  * Important: we are about to write the buffer, and