net: move .getsockopt away from __user buffers

Currently, the .getsockopt callback requires __user pointers:

  int (*getsockopt)(struct socket *sock, int level,
                    int optname, char __user *optval, int __user *optlen);

This prevents kernel callers (io_uring, BPF) from using getsockopt on
levels other than SOL_SOCKET, since they pass kernel pointers.

Following Linus' suggestion [0], this series introduces sockopt_t, a
type-safe wrapper around iov_iter, and a getsockopt_iter callback that
works with both user and kernel buffers. AF_PACKET and CAN raw are
converted as initial users, with selftests covering the trickiest
conversion patterns.

[0] https://lore.kernel.org/all/CAHk-=whmzrO-BMU=uSVXbuoLi-3t...

Below are some questions raised during the RFC discussion:

1) Should optlen be an iov_iter as well?
  No. optlen can remain a plain kernel int since do_sock_getsockopt_iter() syncs
  it back to userspace on both success and failure. The existing callback
  patterns all work with this approach:

  a) Most callbacks (roughly 2/3) always write back optlen.

  b) Some callbacks read optlen but never update it. The original
     value is written back unchanged.

  c) CAN raw updates optlen even on error (-ERANGE) to report the
     required buffer size:

          err = -ERANGE;
          if (put_user(fsize, optlen))
                  err = -EFAULT;

     No regression, since opt.optlen is always written back to
     userspace by the wrapper.

  d) Bluetooth uses put_user() with mixed sizes (u32, u16, u8) but
     never updates optlen. Same as case (b).

2) Can callbacks change iov_iter direction mid-flight?

  Yes. Some protocols read from and then write back to optval in the same
  getsockopt call. For example, PACKET_HDRLEN reads a tpacket version from optval
  and writes back the corresponding header size.

  The converted callback handles this by temporarily flipping the iter direction,
  reverting the position, and writing back:

          case PACKET_HDRLEN:
                  // opt->iter.data_source is ITER_SOURCE;
                  if (copy_from_iter(&val, len, &opt->iter) != len)
                          return -EFAULT;
		  // unroll the bytes
                  iov_iter_revert(&opt->iter, len);
                  opt->iter.data_source = ITER_DEST;
                  // ... update val ...
                  if (copy_to_iter(&val, len, &opt->iter) != len)
                          return -EFAULT;

  The callback needs to handle two things after reading from the iter:
  reset the position with iov_iter_revert(), and flip data_source back
  to ITER_DEST before writing.

  - ITER_DEST — the iter is a destination (kernel writes to it).
		copy_to_iter() works, copy_from_iter() refuses.
  - ITER_SOURCE — the iter is a source (kernel reads from it).
		copy_from_iter() works, copy_to_iter() refuses.

3) In which case iov_iter_revert() needs to be called?

  When a callback needs to read from and then write back to the same
  buffer in a single getsockopt call. The iter advances its position on
  copy_from_iter(), so you need iov_iter_revert() to reset the position
  back to the start before you can copy_to_iter() into the same location.

  Without the revert, copy_to_iter() would write past the end of the
  buffer since the iter already advanced during the read.

4) Do we have any selftest for this change?

  Yes, I've created a commit that I am using to test it, but, I am not
  sure how useful it is rigth now, so, not appending it here.

  You can find it at
  https://github.com/leitao/linux/commit/2d9311947061f1baa4...

Note: The dance regarding changes to iov_iter_revert() (2) and
opt->iter.data_source (3) is a bit fragile. It will not be a bad idea to
creaet a helper (e.g., sockopt_read_val()) would be safer to prevent
others from getting it wrong.

I am not adding it now, so, it is easier to read the bare bones of the
change and helpers can come later.

Link: https://lore.kernel.org/all/CAHk-=whmzrO-BMU=uSVXbuoLi-3t... [0]
---
Changes in v2:
- Restore optlen even on error path (getsockopt_iter fails)
- Move af_packet.c and can instead of netlink (given these are the most
  complicate ones).
- Link to v1: https://patch.msgid.link/20260130-getsockopt-v1-0-9154fcf...

---
Breno Leitao (4):
      net: add getsockopt_iter callback to proto_ops
      net: call getsockopt_iter if available
      af_packet: convert to getsockopt_iter
      can: raw: convert to getsockopt_iter

 include/linux/net.h    | 19 +++++++++++++++++++
 net/can/raw.c          | 28 +++++++++++++---------------
 net/packet/af_packet.c | 18 ++++++++++--------
 net/socket.c           | 48 +++++++++++++++++++++++++++++++++++++++++++++---
 4 files changed, 87 insertions(+), 26 deletions(-)
---
base-commit: 2d9311947061f1baa43858f597dd6c54d7ccc5d2
change-id: 20260130-getsockopt-9f36625eedcb

Best regards,
--  
Breno Leitao <leitao@debian.org>