rxrpc: Allow the kernel to mark a call as being non-interruptible
authorDavid Howells <dhowells@redhat.com>
Thu, 9 May 2019 07:21:21 +0000 (08:21 +0100)
committerDavid Howells <dhowells@redhat.com>
Thu, 16 May 2019 15:25:20 +0000 (16:25 +0100)
Allow kernel services using AF_RXRPC to indicate that a call should be
non-interruptible.  This allows kafs to make things like lock-extension and
writeback data storage calls non-interruptible.

If this is set, signals will be ignored for operations on that call where
possible - such as waiting to get a call channel on an rxrpc connection.

It doesn't prevent UDP sendmsg from being interrupted, but that will be
handled by packet retransmission.

rxrpc_kernel_recv_data() isn't affected by this since that never waits,
preferring instead to return -EAGAIN and leave the waiting to the caller.

Userspace initiated calls can't be set to be uninterruptible at this time.

Signed-off-by: David Howells <dhowells@redhat.com>
Documentation/networking/rxrpc.txt
fs/afs/rxrpc.c
include/net/af_rxrpc.h
net/rxrpc/af_rxrpc.c
net/rxrpc/ar-internal.h
net/rxrpc/call_object.c
net/rxrpc/conn_client.c
net/rxrpc/sendmsg.c

index ff035a6418e382a0c4827bea276ed05308224b03..180e07d956a7049f119a8ebac7dfa2241b6c0a56 100644 (file)
@@ -796,7 +796,9 @@ The kernel interface functions are as follows:
                                s64 tx_total_len,
                                gfp_t gfp,
                                rxrpc_notify_rx_t notify_rx,
-                               bool upgrade);
+                               bool upgrade,
+                               bool intr,
+                               unsigned int debug_id);
 
      This allocates the infrastructure to make a new RxRPC call and assigns
      call and connection numbers.  The call will be made on the UDP port that
@@ -824,6 +826,13 @@ The kernel interface functions are as follows:
      the server upgrade the service to a better one.  The resultant service ID
      is returned by rxrpc_kernel_recv_data().
 
+     intr should be set to true if the call should be interruptible.  If this
+     is not set, this function may not return until a channel has been
+     allocated; if it is set, the function may return -ERESTARTSYS.
+
+     debug_id is the call debugging ID to be used for tracing.  This can be
+     obtained by atomically incrementing rxrpc_debug_id.
+
      If this function is successful, an opaque reference to the RxRPC call is
      returned.  The caller now holds a reference on this and it must be
      properly ended.
index 4974defb45924b0c726ede65c8ec7bff01130e1e..87763379952d3853b58cd663e7a8f9baf0ababcb 100644 (file)
@@ -417,6 +417,7 @@ void afs_make_call(struct afs_addr_cursor *ac, struct afs_call *call, gfp_t gfp)
                                          afs_wake_up_async_call :
                                          afs_wake_up_call_waiter),
                                         call->upgrade,
+                                        true,
                                         call->debug_id);
        if (IS_ERR(rxcall)) {
                ret = PTR_ERR(rxcall);
index c04602ca4a55c79d92f47a08444291c80d18d4b5..93358bfc0e1b997c1c58d536dd02ff452c26cb0b 100644 (file)
@@ -45,6 +45,7 @@ struct rxrpc_call *rxrpc_kernel_begin_call(struct socket *,
                                           gfp_t,
                                           rxrpc_notify_rx_t,
                                           bool,
+                                          bool,
                                           unsigned int);
 int rxrpc_kernel_send_data(struct socket *, struct rxrpc_call *,
                           struct msghdr *, size_t,
index 213935fbbbf770fc37a8eb7b944ee04d88b3f0fe..ffde5b187f5d19d9975594d1826a2ace1a9fe71c 100644 (file)
@@ -270,6 +270,7 @@ static int rxrpc_listen(struct socket *sock, int backlog)
  * @gfp: The allocation constraints
  * @notify_rx: Where to send notifications instead of socket queue
  * @upgrade: Request service upgrade for call
+ * @intr: The call is interruptible
  * @debug_id: The debug ID for tracing to be assigned to the call
  *
  * Allow a kernel service to begin a call on the nominated socket.  This just
@@ -287,6 +288,7 @@ struct rxrpc_call *rxrpc_kernel_begin_call(struct socket *sock,
                                           gfp_t gfp,
                                           rxrpc_notify_rx_t notify_rx,
                                           bool upgrade,
+                                          bool intr,
                                           unsigned int debug_id)
 {
        struct rxrpc_conn_parameters cp;
@@ -311,6 +313,7 @@ struct rxrpc_call *rxrpc_kernel_begin_call(struct socket *sock,
        memset(&p, 0, sizeof(p));
        p.user_call_ID = user_call_ID;
        p.tx_total_len = tx_total_len;
+       p.intr = intr;
 
        memset(&cp, 0, sizeof(cp));
        cp.local                = rx->local;
index 062ca9dc29b8ab2fa7381c606791d4fd39657962..07fc1dfa487890e800598755d2c2c9202be86ccf 100644 (file)
@@ -482,6 +482,7 @@ enum rxrpc_call_flag {
        RXRPC_CALL_BEGAN_RX_TIMER,      /* We began the expect_rx_by timer */
        RXRPC_CALL_RX_HEARD,            /* The peer responded at least once to this call */
        RXRPC_CALL_RX_UNDERRUN,         /* Got data underrun */
+       RXRPC_CALL_IS_INTR,             /* The call is interruptible */
 };
 
 /*
@@ -711,6 +712,7 @@ struct rxrpc_call_params {
                u32             normal;         /* Max time since last call packet (msec) */
        } timeouts;
        u8                      nr_timeouts;    /* Number of timeouts specified */
+       bool                    intr;           /* The call is interruptible */
 };
 
 struct rxrpc_send_params {
index fe96881a334daff644a1f9d01497771f745e9fc8..d0ca98d7aef57691664b2c91350bcae210252332 100644 (file)
@@ -241,6 +241,8 @@ struct rxrpc_call *rxrpc_new_client_call(struct rxrpc_sock *rx,
                return call;
        }
 
+       if (p->intr)
+               __set_bit(RXRPC_CALL_IS_INTR, &call->flags);
        call->tx_total_len = p->tx_total_len;
        trace_rxrpc_call(call, rxrpc_call_new_client, atomic_read(&call->usage),
                         here, (const void *)p->user_call_ID);
index 83797b3949e2f0122d87d15a48651b3b540ab156..5cf5595a14d8e73408de49bab6765b4c7057a69a 100644 (file)
@@ -656,10 +656,14 @@ static int rxrpc_wait_for_channel(struct rxrpc_call *call, gfp_t gfp)
 
                add_wait_queue_exclusive(&call->waitq, &myself);
                for (;;) {
-                       set_current_state(TASK_INTERRUPTIBLE);
+                       if (test_bit(RXRPC_CALL_IS_INTR, &call->flags))
+                               set_current_state(TASK_INTERRUPTIBLE);
+                       else
+                               set_current_state(TASK_UNINTERRUPTIBLE);
                        if (call->call_id)
                                break;
-                       if (signal_pending(current)) {
+                       if (test_bit(RXRPC_CALL_IS_INTR, &call->flags) &&
+                           signal_pending(current)) {
                                ret = -ERESTARTSYS;
                                break;
                        }
index bec64deb7b0a2794345c896827846fa8bac57e19..45a05d9a27fa122dc03440d6734b229be7c15991 100644 (file)
@@ -80,7 +80,8 @@ static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
                if (call->state >= RXRPC_CALL_COMPLETE)
                        return call->error;
 
-               if (timeout == 0 &&
+               if (test_bit(RXRPC_CALL_IS_INTR, &call->flags) &&
+                   timeout == 0 &&
                    tx_win == tx_start && signal_pending(current))
                        return -EINTR;
 
@@ -620,6 +621,7 @@ int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
                .call.tx_total_len      = -1,
                .call.user_call_ID      = 0,
                .call.nr_timeouts       = 0,
+               .call.intr              = true,
                .abort_code             = 0,
                .command                = RXRPC_CMD_SEND_DATA,
                .exclusive              = false,