ioctl_readwrite_ptr missing

[Christiaan676]

Fairly new with LKM's but I have a IOCtl function that is defined as _IORW(MAGIC, 1, struct mystruct *) and the LKM uses the pointer to read and to write to a field in the struct. According to [1] that means the function should be declared are RW.

_IO/_IOR/_IOW/_IOWR
The macro name specifies how the argument will be used. Â It may be a
pointer to data to be passed into the kernel (_IOW), out of the kernel
(_IOR), or both (_IOWR). Â _IO can indicate either commands with no
argument or those passing an integer value instead of a pointer.
It is recommended to only use _IO for commands without arguments,
and use pointers for passing data.

Looking at the issues, this is mentioned a couple of times and people give an implementation but it was never added. Is there any reason its not implemented? Considering the description in [1] I would expect all three variants:

• ioctl_write_ptr
• ioctl_read_ptr
• ioctl_readwrite_ptr

but currently only ioctl_write_ptr is implemented.

Merge requests that add's it: #965
Question with someone offering a implementation: #1386 (comment)

[1] https://www.kernel.org/doc/Documentation/core-api/ioctl.rst

[Christiaan676]

Ok took some time. But think I under stand it now.

The read and readwrite functions always use a pointer, as else there is no other way to get a value back from the kernal driver. For the write this is different, as you can write a value like char/int/long as a direct parameters, or as a pointer to the value. So the distinction between the two is needed there.

[asomers]

So is your problem solved, or is there still something else that you need?

[Christiaan676]

@asomers My problem is fixed. Was planning on reading the documentation again to see if it could use an update to explain the situation with the write and write_ptr a bit better. But haven't had time for that....

Wat thrown me of is that the write_ptr function suggest that there should be a pointer variant for all the other types as well. But this is not the case, as they already use a pointer. (Maybe it would have been better to make the write variant accept a pointer, and to create a write_value version to keep the write macro in line with the other macro's).