Re: write_out_data in JBD

[Jan Kara]

  Hello,

> Stephen,
  I'm not Stephen but I guess my answer would suffice ;)

> Here is a code fragment starting at "write_out_data:" in
> "journal_commit_transaction()":
> 
> Let's assume that there is a single "jh" on the list.
  OK.

> 
>   while (commit_transaction->t_sync_datalist) {
> 
>       jh = commit_transaction->t_sync_datalist;
>       commit_transaction->t_sync_datalist = jh->b_tnext;
> 
>       // "commit_transaction->t_sync_datalist" happens always
>       // to point at our single "jh"
> 
>       bh = jh2bh(jh);
> 
>       // Assume not locked
>       // Assume dirty
> 
>       if (buffer_dirty(bh)) {
>           get_bh(bh);
>           wbuf[bufs++] = bh;
>           if (bufs == journal->j_wbufsize) {
  Now this would not happen as j_wbufsize is larger than 1. But still
even if it would happen ll_rw_block() would be called, that would clear
a dirty bit and and the next time we see the buffer we take a different
route.
  But actually you are right that if we fail with
bufs==journal->j_wbufsize we would go into the next iteration of the
loop and queue the buffer again. Anyway we agreed to rewrite this code
in another thread :). I just have to make sure we don't have a similar
problem in the new version.

>               ...
>               goto write_out_data;
>           }
>       } else ...
>   }
> 
> I think our single "jh" will be put on "wbuf[]" repeatedly
> ("journal->j_wbufsize" times).


								Honza
-- 
Jan Kara <jack@xxxxxxx>
SuSE CR Labs
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/