32 bit (socket layer) ioctl emulation for 64 bit kernels- Question regarding...

32 bit (socket layer) ioctl emulation for 64 bit kernels- Question regarding...

[spereira]

Hi
I am trying to implement 32 bit userland support for existing x25
socket layer ioctls in a x86_64 kernel for use in a project for a
telco.

 It used to be possible to extend the  contents of the
ioctl32_hash_table(fs/compat.c) for ioctl commands and associated
handlers (= 64 bit converstion functions) by calling the
register_ioctl32_conversion() function.
However this function has now been scheduled for removal. The
suggested replacement seems to be available only for inode structures
created for "actual" file systems, where the i_fop points to
operations for "real" files. When using "socket type" inodes
i_fop(inode/fs.h) points to operations for sockets which does not
provide a hook like compat_ioctl for handling such conversions.

Is there currently an alternative to register_ioctl32_conversion that
would help achive 32 bit ioctl emulation at the socket layer?
Any suggestions/advice whould be much appreciated.
Many Thanks
Shaun

Re: 32 bit (socket layer) ioctl emulation for 64 bit kernels- Question regarding...

[Arnd Bergmann]

On Maandag 10 Oktober 2005 06:21, spereira wrote:
>
> Is there currently an alternative to register_ioctl32_conversion that
> would help achive 32 bit ioctl emulation at the socket layer?
> Any suggestions/advice whould be much appreciated.

The correct solution would be to add the missing functionality to
net/socket.c and move over the implementation of SIOC* from
fs/compat_ioctl.c. Getting the code path right is a little tricky,
but I think a patch to fix this up would be appreciated.

As a start, you could define a compat_sock_ioctl along the
lines of compat_blkdev_ioctl and add your own handlers to the
x25_proto_ops, but IMHO it would makes sense to get rid of stuff
like dev_ifsioc from fs/compat_ioctl.c at the same time by
introducing a new compat_dev_ioctl called from compat_sock_ioctl.

	Arnd <><

Re: 32 bit (socket layer) ioctl emulation for 64 bit kernels- Question regarding...

[spereira]

Thanks Arnd
If I have understood correctly the following would be the changes I
would have to make in the kernel...Please correct me if I am wrong.

It seems to me step (d) would be as you say a little tricky but I
could attempt it
depending on the amount of time I have in the project I am on at the moment.
Proposed modifications...
a) Struct file_operations(include/linux/fs.h) has a compat_ioctl hook for file
ioctls. Include a compat_proto_ioctl hook in proto_ops(include/net.h) for
modular socket ioctls.

b) socket.c has a 'file_operations' type struct socket_file_ops where the
.compat_ioctl member is currently not used. Define compat_sock_ioctl in
socket.c and assign to .compat_ioctl member of socket_file_ops
This is ifdef'd with CONFIG_COMPAT

c) compat_sock_ioctl will by default call the protocol's ioctl
 currently supporting the socket interface when handling modular socket ioctls,
like so. socket->ops->compat_proto_ioctl(...)
which in this case is x25_compat_ioctl.

d)If compat_sock_ioctl has to perform a function similar to
sock_ioctl(socket.c)
where in SIOC* commands are handled, then introduce a newly defined
compat_dev_ioctl that would have stuff like dev_ifsioc that would have
previously
been removed from fs/compat_ioctl.c.
Then the protocol's ioctl (called by default in the step c above) would itself
default to compat_dev_ioctl ensuring other socket layer
ioctls are handled by the device layer function, compat_dev_ioctl

regards,
Shaun


On 10/10/05, Arnd Bergmann <arnd@arndb.de> wrote:
> On Maandag 10 Oktober 2005 06:21, spereira wrote:
> >
> > Is there currently an alternative to register_ioctl32_conversion that
> > would help achive 32 bit ioctl emulation at the socket layer?
> > Any suggestions/advice whould be much appreciated.
>
> The correct solution would be to add the missing functionality to
> net/socket.c and move over the implementation of SIOC* from
> fs/compat_ioctl.c. Getting the code path right is a little tricky,
> but I think a patch to fix this up would be appreciated.
>
> As a start, you could define a compat_sock_ioctl along the
> lines of compat_blkdev_ioctl and add your own handlers to the
> x25_proto_ops, but IMHO it would makes sense to get rid of stuff
> like dev_ifsioc from fs/compat_ioctl.c at the same time by
> introducing a new compat_dev_ioctl called from compat_sock_ioctl.
>
>         Arnd <><
>

Re: 32 bit (socket layer) ioctl emulation for 64 bit kernels- Question regarding...

[Arnd Bergmann]

On Dinsdag 11 Oktober 2005 09:09, spereira wrote:
> If I have understood correctly the following would be the changes I
> would have to make in the kernel...Please correct me if I am wrong.
> 
> It seems to me step (d) would be as you say a little tricky but I
> could attempt it
> depending on the amount of time I have in the project I am on at the moment.
> Proposed modifications...
> a) Struct file_operations(include/linux/fs.h) has a compat_ioctl hook for file
> ioctls. Include a compat_proto_ioctl hook in proto_ops(include/net.h) for
> modular socket ioctls.

Yes, but the hook would be called 'compat_ioctl' by convention, not
'compat_proto_ioctl'.

> b) socket.c has a 'file_operations' type struct socket_file_ops where the
> .compat_ioctl member is currently not used. Define compat_sock_ioctl in
> socket.c and assign to .compat_ioctl member of socket_file_ops
> This is ifdef'd with CONFIG_COMPAT

Yes
 
> c) compat_sock_ioctl will by default call the protocol's ioctl
>  currently supporting the socket interface when handling modular
>  socket ioctls, like so. socket->ops->compat_proto_ioctl(...)
> which in this case is x25_compat_ioctl.

Yes, and return -ENOIOCTLCMD if there is no socket->ops->compat_ioctl()
handler installed. That will cause the current mechanism to be used as
a fallback.

> d)If compat_sock_ioctl has to perform a function similar to
> sock_ioctl(socket.c)
> where in SIOC* commands are handled, then introduce a newly defined
> compat_dev_ioctl that would have stuff like dev_ifsioc that would have
> previously
> been removed from fs/compat_ioctl.c.
> Then the protocol's ioctl (called by default in the step c above) would itself
> default to compat_dev_ioctl ensuring other socket layer
> ioctls are handled by the device layer function, compat_dev_ioctl

I realized that it's even more complicated than this. For the ipv4 and ipv6
protocols, there is also an ioctl operation in struct proto, not only one
in struct proto_ops, so that one would also need a compat_ioctl that can
be called from inet{,6}_ioctl in the default path before calling dev_ioctl.
Similarly, the struct net_device contains a do_ioctl method that would
need a new compat_do_ioctl companion.

Maybe you can leave the current dev_ifsioc in place for now (though
I believe it should go away eventually) and just put the infrastructure in
place to have compat_ioctl functions in struct proto_ops, struct proto and 
struct net_device that are called when !NULL or otherwise return -ENOIOCTLCMD.

	Arnd <><