Re: barriers off by default?

[Eric Sandeen <sandeen@redhat.com>]

Andreas Dilger wrote:
> On May 15, 2008  11:00 -0500, Eric Sandeen wrote:
>> ... the fsync change seems sane too...
>>
>> @@ -85,7 +86,10 @@ int ext3_sync_file(struct file * file, s
>>                         .sync_mode = WB_SYNC_ALL,
>>                         .nr_to_write = 0, /* sys_fsync did this */
>>                 };
>> +               journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
>>                 ret = sync_inode(inode, &wbc);
>> +               if (journal && (journal->j_flags & JFS_BARRIER))
>> +                       blkdev_issue_flush(inode->i_sb->s_bdev, NULL);
>>         }
>>  out:
>>
>> I assume we need that for some power-plug-pull scenarios ... in fact I
>> had just been meaning to do something similar after reading an old
>> thread on barriers.  reiserfs & xfs do this already in their sync paths.
> 
> Wouldn't it make more sense to add this into the generic do_fsync()
> routine?

blkdev_issue_flush() depends on barrier support:

        submit_bio(1 << BIO_RW_BARRIER, bio);

and I think only the filesystem can know for sure if it has barriers
currently in use and available?

-Eric