2023-08-08 19:09:54

by Breno Leitao

Subject: [PATCH v2 8/8] io_uring/cmd: BPF hook for setsockopt cmd

Add support for BPF hooks for io_uring setsockopts command.

This implementation follows a similar approach to what
__sys_setsockopt() does, but, operates only on kernel memory instead of
user memory (which is also possible, but not preferred since the kernel
memory is already available)

Signed-off-by: Breno Leitao <[email protected]>
---
io_uring/uring_cmd.c | 23 +++++++++++++++++++++--
1 file changed, 21 insertions(+), 2 deletions(-)

diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
index 3693e5779229..b7b27e4dbddd 100644
--- a/io_uring/uring_cmd.c
+++ b/io_uring/uring_cmd.c
@@ -205,23 +205,42 @@ static inline int io_uring_cmd_setsockopt(struct socket *sock,
{
void __user *optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
int optname = READ_ONCE(cmd->sqe->optname);
+ sockptr_t optval_s = USER_SOCKPTR(optval);
int optlen = READ_ONCE(cmd->sqe->optlen);
int level = READ_ONCE(cmd->sqe->level);
+ char *kernel_optval = NULL;
int err;

err = security_socket_setsockopt(sock, level, optname);
if (err)
return err;

+ if (!in_compat_syscall()) {
+ err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level,
+ &optname,
+ USER_SOCKPTR(optval),
+ &optlen,
+ &kernel_optval);
+ if (err < 0)
+ return err;
+ if (err > 0)
+ return 0;
+
+ /* Replace optval by the one returned by BPF */
+ if (kernel_optval)
+ optval_s = KERNEL_SOCKPTR(kernel_optval);
+ }
+
if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
err = sock_setsockopt(sock, level, optname,
- USER_SOCKPTR(optval), optlen);
+ optval_s, optlen);
else if (unlikely(!sock->ops->setsockopt))
err = -EOPNOTSUPP;
else
err = sock->ops->setsockopt(sock, level, optname,
- USER_SOCKPTR(koptval), optlen);
+ optval_s, optlen);

+ kfree(kernel_optval);
return err;
}

--
2.34.1

2023-08-09 23:08:11

by Martin KaFai Lau

Subject: Re: [PATCH v2 8/8] io_uring/cmd: BPF hook for setsockopt cmd

On 8/8/23 6:40 AM, Breno Leitao wrote:
> Add support for BPF hooks for io_uring setsockopts command.
>
> This implementation follows a similar approach to what
> __sys_setsockopt() does, but, operates only on kernel memory instead of
> user memory (which is also possible, but not preferred since the kernel
> memory is already available)
>
> Signed-off-by: Breno Leitao <[email protected]>
> ---
> io_uring/uring_cmd.c | 23 +++++++++++++++++++++--
> 1 file changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/io_uring/uring_cmd.c b/io_uring/uring_cmd.c
> index 3693e5779229..b7b27e4dbddd 100644
> --- a/io_uring/uring_cmd.c
> +++ b/io_uring/uring_cmd.c
> @@ -205,23 +205,42 @@ static inline int io_uring_cmd_setsockopt(struct socket *sock,
> {
> void __user *optval = u64_to_user_ptr(READ_ONCE(cmd->sqe->optval));
> int optname = READ_ONCE(cmd->sqe->optname);
> + sockptr_t optval_s = USER_SOCKPTR(optval);
> int optlen = READ_ONCE(cmd->sqe->optlen);
> int level = READ_ONCE(cmd->sqe->level);
> + char *kernel_optval = NULL;
> int err;
>
> err = security_socket_setsockopt(sock, level, optname);
> if (err)
> return err;
>
> + if (!in_compat_syscall()) {
> + err = BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock->sk, &level,
> + &optname,
> + USER_SOCKPTR(optval),
> + &optlen,
> + &kernel_optval);
> + if (err < 0)
> + return err;
> + if (err > 0)
> + return 0;
> +
> + /* Replace optval by the one returned by BPF */
> + if (kernel_optval)
> + optval_s = KERNEL_SOCKPTR(kernel_optval);
> + }
> +
> if (level == SOL_SOCKET && !sock_use_custom_sol_socket(sock))
> err = sock_setsockopt(sock, level, optname,
> - USER_SOCKPTR(optval), optlen);
> + optval_s, optlen);
> else if (unlikely(!sock->ops->setsockopt))
> err = -EOPNOTSUPP;
> else
> err = sock->ops->setsockopt(sock, level, optname,
> - USER_SOCKPTR(koptval), optlen);
> + optval_s, optlen);

The bpf side changes make sense. Thanks.

With all the bpf pieces in place, __sys_{get,set}sockopt() is looking very
similar to io_uring_cmd_{get,set}sockopt(). There are small differences like one
takes fd and another already has a sock ptr, and io_uring_cmd_getsockopt() is
SOL_SOCKET only. In general, can they be refactored somehow such that future
changes don't have to be made in multiple places?