On Wed, May 22, 2024 at 7:38 PM Eric Biggers <ebiggers@xxxxxxxxxx> wrote: > > On Wed, May 22, 2024 at 09:28:53AM -0700, Darrick J. Wong wrote: > > > > Do the other *at() syscalls prohibit dfd + path pointing to a different > > filesystem? It seems odd to have this restriction that the rest don't, > > but perhaps documenting this in the ioctl_xfs_fsgetxattrat manpage is ok. > > No, but they are arbitrary syscalls so they can do that. ioctls traditionally > operate on the specific filesystem of the fd. To emphasize the absurdity think opening /dev/random and doing ioctl to set projid on some xfs file. It is ridiculous. > > It feels like these should be syscalls, not ioctls. > I bet whatever name you choose for syscalls it is going to be too close lexicographically to [gs]etxattrat(2) [1]. It is really crowded in the area of getattr/getfattr/fgetxattr/getxattr/getfileattr/getfsxattr... I think I would vote for [gs]etfsxattrat(2) following the uapi struct fsxattr. I guess we have officially spiralled. Thanks, Amir. [1] https://lore.kernel.org/linux-fsdevel/20240426162042.191916-1-cgoettsche@xxxxxxxxxxxxx/
