Jens Axboe [off-list ref] wrote:

One concern here is that we're using IOCB_WRITE here to tell if
sb_start_write() has been done or not, and hence whether
kiocb_end_write() needs to be called. You know set it earlier, which
means if we get a failure if we need to setup async data, then we know
have IOCB_WRITE set at that point even though we did not call
sb_start_write().

Hmmm...  It's set earlier in a number of places anyway - __cachefiles_write()
for example.

Btw, can you please put some comments on the IOCB_* constants?  I have to
guess at what they mean and how they're meant to be used.  Or better still,
get Christoph to write Documentation/core-api/iocb.rst describing the API? ;-)

David