On Tue 10-01-12 13:50:22, Surbhi Palande wrote: > > > Hrm let me think through this a little more; we actually do: > > > > > > t16) ext4_journal_start() > > > t17) ext4_journal_start_sb() > > > t18) handle = ext4_journal_current_handle(); > > > t19) if (!handle) vfs_check_frozen() > > > t20) ... jbd2_journal_start() > > Ah, right. I forgot. > > > > > So actually we *do* block new handles, but let *existing* ones > > > continue (see commits 6b0310fbf087ad6e9e3b8392adca97cd77184084 > > > and be4f27d324e8ddd57cc0d4d604fe85ee0425cba9) > > > > > > So your assertion that a new handle is started is incorrect > > > in general, isn't it? So then does the fix seem necessary? > > > Or, at least, in the fashion below - maybe we need to just make > > > sure all started handles complete before the unlock_updates? > > > Or am I missing something...? > > Well, the problem with running operations and freezing is more > > fundamental I believe. See my email > > http://marc.info/?l=linux-fsdevel&m=132585911925796&w=2 > > > > So I believe we'll need some better exclusion mechanism already in VFS. > > > > Honza > > > > > If all the write operations were journaled, then this patch would not allow > ext4 filesystem to have any dirty data after its frozen. > (as journal_start() would block). > > I think the only one candidate that creates dirty data without calling > ext4_journal_start() is mmapped? No, the problem is in any write path. The problem is with operations that happen during the phase when s_frozen == SB_FREEZE_WRITE. These operations dirty the filesystem but running sync may easily miss them. During this phase journal is not frozen so that does not help you in any way. Honza -- Jan Kara <jack@xxxxxxx> SUSE Labs, CR -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
