Re: [PATCH bpf-next v1 1/3] bpf, sockmap: avoid using sk_socket after free

On 2025-02-27 11:45:53, Cong Wang wrote:

On Wed, Feb 26, 2025 at 09:22:40PM +0800, Jiayuan Chen wrote:

quoted

Use RCU lock to protect sk_socket, preventing concurrent close and release
by another thread.

Because TCP/UDP are already within a relatively large critical section:
'''
ip_local_deliver_finish
  rcu_read_lock
  ip_protocol_deliver_rcu
      tcp_rcv/udp_rcv
  rcu_read_unlock
'''

Adding rcu_read_{un}lock() at the entrance and exit of sk_data_ready
will not increase performance overhead.

Reported-by: syzbot+dd90a702f518e0eac072@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/bpf/6734c033.050a0220.2a2fcc.0015.GAE@google.com/ (local)
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>

sock_def_readable() already acquires RCU read lock anyway.

Reviewed-by: Cong Wang <redacted>

Thanks.

Reviewed-by: John Fastabend <john.fastabend@gmail.com>