On Mon, Sep 30, 2024 at 05:43:24PM GMT, Miklos Szeredi wrote: > On Thu, 26 Sept 2024 at 14:24, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > > > Daniel took a different approach for averting the security issue > > in the FUSE BPF patches. > > The OPEN response itself was converted to use an ioctl instead of write: > > https://lore.kernel.org/linux-fsdevel/20240329015351.624249-6-drosen@xxxxxxxxxx/ > > as well as the LOOKUP response. > > > > Are there any negative performance or other implications in this approach? > > It would work, but I'd try to avoid adding more ioctls if possible. > Hence the io-uring suggestion. > > OTOH I'm not sure io_uring is the best interface for all cases, so it > might make sense to cherry pick some features from the io-uring API > (like COMMIT_AND_FETCH) to the regular synchronous API. And that > would need new ioctl commands anyway. A few years ago I vowed to not add ioctl()s anymore because of my dislike of multiplexers. Fast-foward to now and over the last few cycle I probably added around 10...
