On Wed, 2025-06-25 at 04:21 -0700, Christoph Hellwig wrote: > On Wed, Jun 25, 2025 at 06:40:07AM -0400, Jeff Layton wrote: > > Another option: > > > > We could expose this functionality in preadv2() with a new RWF_WBERR > > flag (better names welcome). That way applications could opt-in to > > checking for writeback errors like this. With that, the application is > > at least explicitly saying that it wants this behavior. > > That sounds like a really strange interface to me. > > I have to admit I don't fully understand the use case where an > application cares about these errors, but also doesn't use f(data)sync > or sync(fs) to actually persist the data. If we can come up with a > coherent use case for that we should simply add a new syscall or fcntl > to query the delayed writeback errors instead of overloading other > interfaces. It is weird, but I do sort of get the motivation. In a some cases you want to be able to stream writes as fast as possible and let the kernel lazily write that back (because you don't want to block a thread), but knowing if a prior writeback error has occurred is a good thing too. Another idea: add a new generic ioctl() that checks for writeback errors without syncing anything. That would be fairly simple to do and sounds like it would be useful, but I'd want to hear a better description of the use-case before we did anything like that. -- Jeff Layton <jlayton@xxxxxxxxxx>
