Re: [Ext2-devel] [PATCH] JBD: journal_release_buffer()

[Alex Tomas <alex@clusterfs.com>]

>>>>> Stephen C Tweedie (SCT) writes:

 >> journal_dirty_metadata(handle, bh)
 >> {
 >>     transaction->t_reserved--;
 >>     handle->h_buffer_credits--;
 >>     if (jh->b_tcount > 0) {
 >>         /* modifed, no need to track it any more */
 >>          transaction-> t_outstanding_credits++;
 >>        jh-> b_tcount = -1;
 >>      }
 >> }

 SCT> Actually, the whole thing can be wrapped in if (jh->b_tcount > 0) {}, I
 SCT> think.  If we have already charged the transaction for this buffer, then
 SCT> there's no need to charge the handle for it again.  That's going to be
 SCT> particularly important for truncate(), where we are continually updating
 SCT> the same blocks (eg. bitmap, indirect blocks), so we want to make sure
 SCT> that after the first journal_dirty_metadata() call, no further charge is
 SCT> made.

the idea is:
1) the sooner we drop reservation, the higher probability to cover many
   changes by single transaction
1) having h_buffer_credits being decremented for each modification
   could help us to debug handle overflow situations

 SCT> If we do that, do we in fact need t_reserved at all?

hmm. if t_outstanding_credits holds number of modified buffers,
then we need sum of all running h_buffer_credits to protect
from transaction overflow. t_reserved is sum of h_buffer_credits.


thanks, Alex