kernel code is inherently more difficult to extend than user space code, especially because we can re-use user space shared libs in user space code. For example, writing a user space filesystem (Gnome-VFS, libferris...) should be easier than writing a kernel one, but those are not "true" filesystems...
1) is it planned to integrate in the "official" kernel something like fuse or lufs in order to write true filesystems in user space ?
2) would it be theoretically possible to move major parts of the kernel (ip stack, netfilter, device drivers, VFS...) into userland shared libs in order to make the kernel part lighter & more real-time and the "user part" more easily extensible ? (without loss of performance, for exemple that would avoid context switches for any system call belonging to this "user part").
# Kernel & user space libs
Posté par karteum59 (site web personnel) . En réponse au message Interview Alan Cox. Évalué à 1.
1) is it planned to integrate in the "official" kernel something like fuse or lufs in order to write true filesystems in user space ?
2) would it be theoretically possible to move major parts of the kernel (ip stack, netfilter, device drivers, VFS...) into userland shared libs in order to make the kernel part lighter & more real-time and the "user part" more easily extensible ? (without loss of performance, for exemple that would avoid context switches for any system call belonging to this "user part").