io_uring: allow tcp ancillary data for __sys_recvmsg_sock()
authorLuke Hsiao <lukehsiao@google.com>
Sat, 22 Aug 2020 04:41:04 +0000 (21:41 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 24 Aug 2020 23:16:06 +0000 (16:16 -0700)
commit583bbf0624dfd8fc45f1049be1d4980be59451ff
tree4ec39c88cd93f67918197f4c001c28a652cb5b98
parent37a319b5f9900a1716e140b3719296e6bb19d04c
io_uring: allow tcp ancillary data for __sys_recvmsg_sock()

For TCP tx zero-copy, the kernel notifies the process of completions by
queuing completion notifications on the socket error queue. This patch
allows reading these notifications via recvmsg to support TCP tx
zero-copy.

Ancillary data was originally disallowed due to privilege escalation
via io_uring's offloading of sendmsg() onto a kernel thread with kernel
credentials (https://crbug.com/project-zero/1975). So, we must ensure
that the socket type is one where the ancillary data types that are
delivered on recvmsg are plain data (no file descriptors or values that
are translated based on the identity of the calling process).

This was tested by using io_uring to call recvmsg on the MSG_ERRQUEUE
with tx zero-copy enabled. Before this patch, we received -EINVALID from
this specific code path. After this patch, we could read tcp tx
zero-copy completion notifications from the MSG_ERRQUEUE.

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Arjun Roy <arjunroy@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jann Horn <jannh@google.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Luke Hsiao <lukehsiao@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/net.h
net/ipv4/af_inet.c
net/ipv6/af_inet6.c
net/socket.c