Re: Question about REQ_FLUSH and bios with data

From: scameron
Date: Fri Aug 09 2013 - 16:34:48 EST

On Fri, Aug 09, 2013 at 09:18:15PM +0100, Steven Whitehouse wrote:
> Hi,
> On Fri, 2013-08-09 at 14:09 -0500, scameron@xxxxxxxxxxxxxxxxxx wrote:
> > So, I'm working on a block driver using the make_request_fn
> > interface, and have to handle a bio that comes in with
> > (bi_rw & REQ_FLUSH) set AND data to transfer.
> >
> > According to Documentation/block/writeback_cache_control.txt:
> >
> > The REQ_FLUSH flag can be OR ed into the r/w flags of a
> > bio submitted from the filesystem and will make sure the
> > volatile cache of the storage device has been flushed
> > before the actual I/O operation is started.
> >
> > So I've written code that handles that case, along with the
> > various error cases I might encounter, I think, which leads
> > to my question:
> >
> > How do I get such a bio with a data transfer AND the REQ_FLUSH bit
> > set to come into the driver? Just wondering how to test this case.
> >
> GFS2 does this... if you look at the log flush code and the function
> log_write_header() in particular, you'll see that it sets this on each
> log header that gets written. You don't need a cluster to generate this
> kind of i/o, just supply mkfs.gfs2 with -p lock_nolock and mount it as a
> local filesystem. The combination of touch foo; sync should be enough to
> generate a log flush writing a log header with this flag set,


Turns out ext3 seems to do it as well (now I feel kind of dumb
for asking -- had tried with ext2, and had tried some writes
with a file descriptor with O_SYNC set, but I was kind of shooting
in the dark.)

And my code appears to work too.

-- steve

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at