Re: [Ext2-devel] [RFC] probably bug in current ext3/jbd

[Alex Tomas]

aha. now it's clear. thank you. I catched J_ASSERT(b_committed_data != NULL)
in ext3_free_blocks() with de-BKL'ed JBD. hence, my solution is to have a
tid journal_head indicating which transaction uses b_committed_data.

I don't want to look intrusive, but .. what do you think about new locking
schema I'm trying to implement?

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

 >> access for
 >> b_committed_data == NULL ?

 SCT> Not with BKL.  Without it, yes, that's definitely a risk, and you need
 SCT> some locking for the access to b_committed_data.  Without that, even if
 SCT> you keep the jh->b_committed_data field valid, you risk freeing the old
 SCT> copy that another thread is using.

 SCT> Cheers,
 SCT>  Stephen