On Tue, Nov 27, 2012 at 11:19:41AM -0500, Jeff Moyer wrote: > Christoph Hellwig <hch@xxxxxxxxxxxxx> writes: > > > Call generic_write_sync from the deferred I/O completion handler if > > O_DSYNC is set for a write request. Also make sure various callers > > don't call generic_write_sync if the direct I/O code returns > > -EIOCBQUEUED. > > > > Note: this currently breaks ext4 due to it's convoluted unwritten > > extent conversion code. I've tried to understand it and as far > > as I can see it's a workaround for the fact that ext4 marks page > > writeback as completed before converting unwritten extents. > > Ext4 should follow xfs on this and only mark writeback as completed > > when it really is and at that point can remove the big hairy mess to > > force unwritten extent conversions from fsync, truncate and a few other > > places. > > Well, I've tested the xfs bits, and there are no regressions. I won't > bother testing ext4 until a full solution is in place. Jan, is this > something you have the time for and/or interest in doing? I tested the raw block device use case and it seems to be ok too. cc'ing the ext4 list, since this affects ext4 pretty heavily. --D > > Cheers, > Jeff -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
