When use-gss-proxy is set to 1, write_gssp() creates a rpc client in
gssp_rpc_create(), this increases netns refcount by 2 [1], these
refcounts are supposed to be released in rpcsec_gss_exit_net(), but
it will never happen because rpcsec_gss_exit_net() is triggered only
when netns refcount gets to 0, specifically:
refcount=0 -> cleanup_net() -> ops_exit_list -> rpcsec_gss_exit_net
It is a deadlock situation here, refcount will never get to 0 unless
rpcsec_gss_exit_net() is called.
[1]
SyS_write
vfs_write
proc_reg_write
write_gssp
set_gssp_clnt
gssp_rpc_create
rpc_create
xprt_create_transport
xs_setup_local
xs_setup_xprt
xprt_alloc // get net refcount
xs_local_setup_socket
unix_create
kernel_connect // get net refcount
In this case, the net refcount shouldn't be increased when creating rpc
client, otherwise it will lead to deadlock.
This patchset removes the increased netns reference count.
Wang Hai (2):
net: Modify unix_stream_connect to not reference count the netns of
kernel sockets
auth_gss: Fix deadlock that blocks rpcsec_gss_exit_net when
use-gss-proxy==1
include/linux/sunrpc/clnt.h | 1 +
include/linux/sunrpc/xprt.h | 6 ++++--
net/sunrpc/auth_gss/gss_rpc_upcall.c | 3 ++-
net/sunrpc/clnt.c | 2 ++
net/sunrpc/xprt.c | 13 +++++++++----
net/sunrpc/xprtrdma/svc_rdma_backchannel.c | 2 +-
net/sunrpc/xprtrdma/transport.c | 2 +-
net/sunrpc/xprtsock.c | 4 +++-
net/unix/af_unix.c | 3 ++-
9 files changed, 25 insertions(+), 11 deletions(-)
--
2.17.1
