feat: add support for ReadMulti opcode (6.7+)

[kp-omer-shamash]

Simple addition, already have the CODE in the /sys/

[kp-omer-shamash]

Right now I'm doing what feels like a crime against humanity by implementing like this and having tons of structures copy-pasta'd from the /sys/ part of the crate, would really appreciate if we can merge so I don't use this abomination :

pub struct ReadMulti {
    fd: i32,
    buf_group: u16,
    flags: i32,
}

impl ReadMulti {
    #[inline]
    pub fn new(fd: i32, buf_group: u16) -> Self {
        ReadMulti {
            fd,
            buf_group,
            flags: 0,
        }
    }

    #[inline]
    pub fn build(self) -> Entry {
        let sqe = CustomSQE {
            opcode: IORING_OP_READ_MULTISHOT as _,
            flags: io_uring::squeue::Flags::BUFFER_SELECT.bits(),
            fd: self.fd,
            buf_index: PackedU16 {
                buf_index: self.buf_group,
            },
            msg_flags: Union3 {
                msg_flags: self.flags as _,
            },
            ..Default::default()
        };

        // Safety: CustomSQE has identical memory layout to io_uring_sqe
        #[allow(unsafe_code)]
        unsafe {
            std::mem::transmute(sqe)
        }
    }
}

[quininer]

This looks good, would you like to add a test?

[kp-omer-shamash]

@quininer done
also, not sure what your release-plans are, but would appreciate if it's possible to bump to 0.7.5 soon after

[quininer]

Do we need to provide a way to specify len and offset?

[kp-omer-shamash]

No, it should act the exact same as RecvMulti - you cannot specify anything other then which buffer-group to use, fd and flags (which I think none are not used at kernel-side right now)

[quininer]

I think this is not true, ReadMulti is more similar to Read than to RecvMulti.

see https://github.com/torvalds/linux/blob/v6.13/io_uring/rw.c#L366 and https://github.com/axboe/liburing/blob/liburing-2.9/src/include/liburing.h#L801

[quininer]

What does this flags mean and why does it write to msg_flags?

[kp-omer-shamash]

same as above - the API is similar to RecvMulti
flags are here in case kernel maintainers support any custom flags and we would like to use them, but default to 0

[quininer]

I think the kernel doesn't read msg_flags, right?

[kp-omer-shamash]

Sorry for late reply, I'm currently no longer available in this account, I hope this can be merged

[kp-omer-shamash]

No, it should act the exact same as RecvMulti - you cannot specify anything other then which buffer-group to use, fd and flags (which I think none are not used at kernel-side right now)

[kp-omer-shamash]

same as above - the API is similar to RecvMulti
flags are here in case kernel maintainers support any custom flags and we would like to use them, but default to 0