Updated CIFSD In-Kernel SMB3 File Sharing Server Patches Published

[phoronix]

Phoronix: Updated CIFSD In-Kernel SMB3 File Sharing Server Patches Published

The Samsung-led CIFSD as an in-kernel SMB3 file-sharing server continues on its trajectory toward the mainline Linux kernel...

https://www.phoronix.com/scan.php?page=news_item&px=CIFSD-v2-Linux-Server-Patches

[Azrael5]

Is there the possibility to share files between two specific clients of the same LAN?

[skeevy420]

I just tried these patches out and if you hold the down arrow during network initialization you'll get dropped into a special flute shell. It takes about 5 seconds for it to kick in.

[User42]

The target is to provide optimized performance, GPLv2 SMB server, better lease handling (distributed caching). The bigger goal is to add new features more rapidly (e.g. RDMA aka "smbdirect", and recent encryption and signing improvements to the protocol) which are easier to develop on a smaller, more tightly optimized kernel server than for example in Samba.

For 32k new lines of code, I wonder if that really is worthwhile... Would be glad to see this compared to user-space samba to quantify the performance improvement... I suppose that will come in time.

[Mangix]

Samba is much larger.

[Guest]

For 32k new lines of code, I wonder if that really is worthwhile... Would be glad to see this compared to user-space samba to quantify the performance improvement... I suppose that will come in time.

Yeah, the kernel's main purpose is arbitrating hardware access for processes and users. Having a file server in there seems like a stupid idea.

It's so much easier to debug, test, sandbox and use modern OOP languages in userspace.

[User42]

Samba is much larger.

I understand that but Samba is also userspace and I tend to think like sandy8925: having a big file server there looks like a bad idea... Though I'm not working in the kernel, hence my surprise and my comment in hope somebody would make it clearer.

From what I quoted, I understand the RDMA bit though that could be handle as a smaller kernel module + userspace code. The rest is dubious (until there are benchmark) or not convincing (for me) like being able to make changes quicker. I mean, I do believe them that making changes now is quicker, as it is the case for most projects that restart from scratch. It's just that after a surprising short while it will become as hard to contribute to that module as in Samba.

All I see now is that any bug in 32k lines of code could lead to exploitation and could introduce new ways to escape containers.

Anyway not my call, still surprised this is really considered though.

[Mangix]

Well, it does have a user space component which is necessary for the whole thing to work. Probably to reduce the surface a bit.

[hoohoo]

If it allows SMB over RDMA then I'm all for it. NFS over RDMA is much superior to NFS over IP, I assume SMB would exhibit similar behavior.