This is an RFC patch because I'm not sure if this is
the right way to do this... There might be a better way.
Recently I've had some problems with upcalls, during
mounts, hanging. The reason the hang is happening is one
problem but it occurred to me the kernel should not
hang forever for a daemon that may never come back.
Now a couple thoughts...
1) How should the mount be failed. If ETIMEDOUT
is used, the mount will be retied via the
nfs4_discover_server_trunking() code.
But, if the daemon is not responding why keep trying?
We could error out the mount with EPIPE which cause
the mount fail immediately and logging this error
message
NFS: nfs4_discover_server_trunking unhandled error 32. Exiting with error EIO
2) Is there a better way to do this? That timeout code
looks a bit crusty :-)
3) Is the 10 sec timeout to short? Maybe we could bump
it up to a 30 sec timeout? If do that I would suggest
we error out the mount immediately with EPIPE.
4) Am I missing something?
Steve Dickson (1):
auth_rpcgss: Add a timer to the gss upcall.
net/sunrpc/auth_gss/auth_gss.c | 43 ++++++++++++++++++++++++++++++++++++++++++
1 file changed, 43 insertions(+)
--
2.14.3
So mounts will not hang forever, waiting for a user land
daemon that never come back, set a 10 sec timer
before the upcall is made. If the timer pops,
error out the mount with a timeout error.
Signed-off-by: Steve Dickson <[email protected]>
---
net/sunrpc/auth_gss/auth_gss.c | 43 ++++++++++++++++++++++++++++++++++++++++++
1 file changed, 43 insertions(+)
diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index 9463af4b32e8..9c866812bc2d 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -66,6 +66,9 @@ static unsigned int gss_expired_cred_retry_delay = GSS_RETRY_EXPIRED;
#define GSS_KEY_EXPIRE_TIMEO 240
static unsigned int gss_key_expire_timeo = GSS_KEY_EXPIRE_TIMEO;
+#define GSS_MSG_TIMEO 10 /* seconds */
+static unsigned int gss_msg_timeo = GSS_MSG_TIMEO;
+
#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
# define RPCDBG_FACILITY RPCDBG_AUTH
#endif
@@ -295,6 +298,7 @@ struct gss_upcall_msg {
struct rpc_pipe *pipe;
struct rpc_wait_queue rpc_waitqueue;
wait_queue_head_t waitqueue;
+ struct timer_list timeout;
struct gss_cl_ctx *ctx;
char databuf[UPCALL_BUF_LEN];
};
@@ -399,6 +403,36 @@ gss_unhash_msg(struct gss_upcall_msg *gss_msg)
spin_unlock(&pipe->lock);
}
+static void
+gss_msg_timeout(struct timer_list *t)
+{
+ struct gss_upcall_msg *gss_msg = from_timer(gss_msg, t, timeout);
+ struct rpc_pipe *pipe = gss_msg->pipe;
+
+ dprintk("RPC: %s gss_msg 0x%p\n", __func__, gss_msg);
+ spin_lock(&pipe->lock);
+ gss_msg->msg.errno = -ETIMEDOUT;
+ __gss_unhash_msg(gss_msg);
+ spin_unlock(&pipe->lock);
+}
+
+static inline void
+gss_msg_timer(struct gss_upcall_msg *gss_msg)
+{
+ if (!timer_pending(&gss_msg->timeout)) {
+ timer_setup(&gss_msg->timeout, gss_msg_timeout, 0);
+ mod_timer(&gss_msg->timeout, (jiffies + (gss_msg_timeo * HZ)));
+ }
+}
+
+static inline void
+gss_msg_del_timer(struct gss_upcall_msg *gss_msg)
+{
+ if (timer_pending(&gss_msg->timeout))
+ del_timer(&gss_msg->timeout);
+}
+
static void
gss_handle_downcall_result(struct gss_cred *gss_cred, struct gss_upcall_msg *gss_msg)
{
@@ -587,6 +621,10 @@ gss_refresh_upcall(struct rpc_task *task)
err = PTR_ERR(gss_msg);
goto out;
}
+
+ /* Set a timer so this upcall does not get stuck in userland */
+ gss_msg_timer(gss_msg);
+
pipe = gss_msg->pipe;
spin_lock(&pipe->lock);
if (gss_cred->gc_upcall != NULL)
@@ -646,6 +684,9 @@ gss_create_upcall(struct gss_auth *gss_auth, struct gss_cred *gss_cred)
err = PTR_ERR(gss_msg);
goto out;
}
+ /* Set a timer so this upcall does not get stuck in userland */
+ gss_msg_timer(gss_msg);
+
pipe = gss_msg->pipe;
for (;;) {
prepare_to_wait(&gss_msg->waitqueue, &wait, TASK_KILLABLE);
@@ -726,6 +767,8 @@ gss_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
goto err_put_ctx;
}
list_del_init(&gss_msg->list);
+ /* Cancel the timer */
+ gss_msg_del_timer(gss_msg);
spin_unlock(&pipe->lock);
p = gss_fill_context(p, end, ctx, gss_msg->auth->mech);
--
2.14.3