syzbot: too eager to disable syscalls on the net instance

[a-nogikh]

We shouldn't have disabled bind and connect.

transitively disabled the following syscalls (missing resource [creating syscalls]):
bind                                        : nfc_dev_id [ioctl$IOCTL_GET_NCIDEV_IDX]
bind$nfc_llcp                               : nfc_dev_id [ioctl$IOCTL_GET_NCIDEV_IDX]
bind$vsock_dgram                            : sock_vsock_dgram [socket$vsock_dgram]
bpf$BPF_LINK_CREATE                         : fd_pidfd [openat$pidfd openat$thread_pidfd syz_clone3 syz_pidfd_open]
bpf$BPF_LINK_CREATE_XDP                     : fd_pidfd [openat$pidfd openat$thread_pidfd syz_clone3 syz_pidfd_open]
close$binfmt                                : fd_binfmt [openat$binfmt]
close$fd_v4l2_buffer                        : fd_v4l2_buffer [ioctl$VIDIOC_QUERYBUF_DMABUF]
close$ibv_device                            : fd_rdma [openat$uverbs0]
connect                                     : nfc_dev_id [ioctl$IOCTL_GET_NCIDEV_IDX]
connect$hf                                  : sock_hf [socket$hf]
connect$nfc_llcp                            : nfc_dev_id [ioctl$IOCTL_GET_NCIDEV_IDX]
connect$nfc_raw                             : nfc_dev_id [ioctl$IOCTL_GET_NCIDEV_IDX]
connect$vsock_dgram                         : sock_vsock_dgram [socket$vsock_dgram]

[a-nogikh]

bind() depends on nfc_dev_id via

syzkaller/sys/linux/socket.txt

Lines 83 to 84 in 956bd95

	# This sockaddr type corresponds to the sockaddr_storage type and is 128 bytes size.
	sockaddr_storage [

syzkaller/sys/linux/socket.txt

Line 107 in 956bd95

nfc sockaddr_nfc

Ideally, if at least one union's option has all its dependencies met, we should not disable all syscalls that referece the union. In this particular case, there would be big harm even if we just always let nfc_dev_id be 0.

Another concerning thing is that now it the disabling of bind goes completely unnoticed, even though we explicitly enable bind in enable_syscalls.

[a-nogikh]

This particular case (dependency on nfc) can be fixed up via

diff --git a/sys/linux/socket.txt b/sys/linux/socket.txt
index e8994fb4e..cf67dc16a 100644
--- a/sys/linux/socket.txt
+++ b/sys/linux/socket.txt
@@ -83,2 +83,3 @@ sockaddr [
 # This sockaddr type corresponds to the sockaddr_storage type and is 128 bytes size.
+# Note: nfc is not included here as it's used directly via bind$nfc_llcp.
 sockaddr_storage [
@@ -106,4 +107,2 @@ sockaddr_storage [
        alg             sockaddr_alg
-       nfc             sockaddr_nfc
-       nfc_llcp        sockaddr_nfc_llcp
        vsock           sockaddr_vm
diff --git a/sys/linux/socket_nfc.txt b/sys/linux/socket_nfc.txt
index c4e2d08fb..17f46ecc1 100644
--- a/sys/linux/socket_nfc.txt
+++ b/sys/linux/socket_nfc.txt
@@ -16,2 +16,3 @@ syz_init_net_socket$nfc_llcp(domain const[AF_NFC], type flags[nfc_llcp_type], pr
 bind$nfc_llcp(fd sock_nfc_llcp, addr ptr[in, sockaddr_nfc_llcp], addrlen len[addr])
+bind$nfc(fd sock, addr ptr[in, sockaddr_nfc], addrlen len[addr])
 connect$nfc_llcp(fd sock_nfc_llcp, addr ptr[in, sockaddr_nfc_llcp], addrlen len[addr])

But if/when something else is added to our descriptions or deprecated from the kernel, we will face the same problem.

[a-nogikh]

One more PR that is caused by big unions involving multiple (not always critical) resources:

#6270