Re: [PATCH] binder: implement binderfs

[Greg KH]

On Tue, Dec 04, 2018 at 02:12:39PM +0100, Christian Brauner wrote:
> As discussed at Linux Plumbers Conference 2018 in Vancouver [1] this is the
> implementation of binderfs. If you want to skip reading and just see how it
> works, please go to [2].

First off, thanks for doing this so quickly.  I think the overall idea
and implementation is great, I just have some minor issues with the user
api:

> /* binder-control */
> Each new binderfs instance comes with a binder-control device. No other
> devices will be present at first. The binder-control device can be used to
> dynamically allocate binder devices. All requests operate on the binderfs
> mount the binder-control device resides in:
> - BINDER_CTL_ADD
>   Allocate a new binder device.
> Assuming a new instance of binderfs has been mounted at /dev/binderfs via
> mount -t binderfs binderfs /dev/binderfs. Then a request to create a new
> binder device can be made via:
> 
> struct binderfs_device device = {0};
> int fd = open("/dev/binderfs/binder-control", O_RDWR);
> ioctl(fd, BINDER_CTL_ADD, &device);
> 
> The struct binderfs_device will be used to return the major and minor
> number, as well as the index used as the new name for the device.
> Binderfs devices can simply be removed via unlink().

I think you should provide a name in the BINDER_CTL_ADD command.  That
way you can easily emulate the existing binder queues, and it saves you
a create/rename sequence that you will be forced to do otherwise.  Why
not do it just in a single command?

That way also you don't need to care about the major/minor number at
all.  Userspace should never need to worry about that, use a name,
that's the best thing.  Also, it allows you to drop the use of the idr,
making the kernel code simpler overall.

> /* Implementation details */
> - When binderfs is registered as a new filesystem it will dynamically
>   allocate a new major number. The allocated major number will be returned
>   in struct binderfs_device when a new binder device is allocated.

Why does userspace care about major/minor numbers at all?  You should
just be able to deal with the binder "names", that's all that userspace
uses normally as you are not calling mknod() yourself.

>   Minor numbers that have been given out are tracked in a global idr struct
>   that is capped at BINDERFS_MAX_MINOR. The minor number tracker is
>   protected by a global mutex. This is the only point of contention between
>   binderfs mounts.

I doubt this will be any real contention given that setting up / tearing
down binder mounts is going to be rare, right?  Well, hopefully, who
knows with some container systems...

> - The naming scheme for binder devices is binder%d. Each binderfs mount
>   starts numbering of new binder devices at 0 up to n. The indeces used in
>   constructing the name are tracked in a struct idr that is per-binderfs
>   super block.

Again, let userspace pick the name, as you will have to rename it anyway
to get userspace to work properly with it.

I'll stop repeating myself now :)

thanks,

greg k-h