atm: lec: fix use-after-free in sock_def_readable()

A race condition exists between lec_atm_close() setting priv->lecd
to NULL and concurrent access to priv->lecd in send_to_lecd(),
lec_handle_bridge(), and lec_atm_send(). When the socket is freed
via RCU while another thread is still using it, a use-after-free
occurs in sock_def_readable() when accessing the socket's wait queue.

The root cause is that lec_atm_close() clears priv->lecd without
any synchronization, while callers dereference priv->lecd without
any protection against concurrent teardown.

Fix this by converting priv->lecd to an RCU-protected pointer:
- Mark priv->lecd as __rcu in lec.h
- Use rcu_assign_pointer() in lec_atm_close() and lecd_attach()
  for safe pointer assignment
- Use rcu_access_pointer() for NULL checks that do not dereference
  the pointer in lec_start_xmit(), lec_push(), send_to_lecd() and
  lecd_attach()
- Use rcu_read_lock/rcu_dereference/rcu_read_unlock in send_to_lecd(),
  lec_handle_bridge() and lec_atm_send() to safely access lecd
- Use rcu_assign_pointer() followed by synchronize_rcu() in
  lec_atm_close() to ensure all readers have completed before
  proceeding. This is safe since lec_atm_close() is called from
  vcc_release() which holds lock_sock(), a sleeping lock.
- Remove the manual sk_receive_queue drain from lec_atm_close()
  since vcc_destroy_socket() already drains it after lec_atm_close()
  returns.

v2: Switch from spinlock + sock_hold/put approach to RCU to properly
    fix the race. The v1 spinlock approach had two issues pointed out
    by Eric Dumazet:
    1. priv->lecd was still accessed directly after releasing the
       lock instead of using a local copy.
    2. The spinlock did not prevent packets being queued after
       lec_atm_close() drains sk_receive_queue since timer and
       workqueue paths bypass netif_stop_queue().

Note: Syzbot patch testing was attempted but the test VM terminated
    unexpectedly with "Connection to localhost closed by remote host",
    likely due to a QEMU AHCI emulation issue unrelated to this fix.
    Compile testing with "make W=1 net/atm/lec.o" passes cleanly.

Reported-by: syzbot+f50072212ab792c86925@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=f50072212ab792c86925
Link: https://lore.kernel.org/all/20260309093614.502094-1-kartikey406@gmail.com/T/ [v1]
Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260309155908.508768-1-kartikey406@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

0 files changed, 0 insertions, 0 deletions