[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both
[powerpc.git] / net / rxrpc / ar-recvmsg.c
diff --git a/net/rxrpc/ar-recvmsg.c b/net/rxrpc/ar-recvmsg.c
new file mode 100644 (file)
index 0000000..e947d5c
--- /dev/null
@@ -0,0 +1,366 @@
+/* RxRPC recvmsg() implementation
+ *
+ * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/net.h>
+#include <linux/skbuff.h>
+#include <net/sock.h>
+#include <net/af_rxrpc.h>
+#include "ar-internal.h"
+
+/*
+ * removal a call's user ID from the socket tree to make the user ID available
+ * again and so that it won't be seen again in association with that call
+ */
+static void rxrpc_remove_user_ID(struct rxrpc_sock *rx, struct rxrpc_call *call)
+{
+       _debug("RELEASE CALL %d", call->debug_id);
+
+       if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
+               write_lock_bh(&rx->call_lock);
+               rb_erase(&call->sock_node, &call->socket->calls);
+               clear_bit(RXRPC_CALL_HAS_USERID, &call->flags);
+               write_unlock_bh(&rx->call_lock);
+       }
+
+       read_lock_bh(&call->state_lock);
+       if (!test_bit(RXRPC_CALL_RELEASED, &call->flags) &&
+           !test_and_set_bit(RXRPC_CALL_RELEASE, &call->events))
+               schedule_work(&call->processor);
+       read_unlock_bh(&call->state_lock);
+}
+
+/*
+ * receive a message from an RxRPC socket
+ * - we need to be careful about two or more threads calling recvmsg
+ *   simultaneously
+ */
+int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
+                 struct msghdr *msg, size_t len, int flags)
+{
+       struct rxrpc_skb_priv *sp;
+       struct rxrpc_call *call = NULL, *continue_call = NULL;
+       struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
+       struct sk_buff *skb;
+       long timeo;
+       int copy, ret, ullen, offset, copied = 0;
+       u32 abort_code;
+
+       DEFINE_WAIT(wait);
+
+       _enter(",,,%zu,%d", len, flags);
+
+       if (flags & (MSG_OOB | MSG_TRUNC))
+               return -EOPNOTSUPP;
+
+       ullen = msg->msg_flags & MSG_CMSG_COMPAT ? 4 : sizeof(unsigned long);
+
+       timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
+       msg->msg_flags |= MSG_MORE;
+
+       lock_sock(&rx->sk);
+
+       for (;;) {
+               /* return immediately if a client socket has no outstanding
+                * calls */
+               if (RB_EMPTY_ROOT(&rx->calls)) {
+                       if (copied)
+                               goto out;
+                       if (rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
+                               release_sock(&rx->sk);
+                               if (continue_call)
+                                       rxrpc_put_call(continue_call);
+                               return -ENODATA;
+                       }
+               }
+
+               /* get the next message on the Rx queue */
+               skb = skb_peek(&rx->sk.sk_receive_queue);
+               if (!skb) {
+                       /* nothing remains on the queue */
+                       if (copied &&
+                           (msg->msg_flags & MSG_PEEK || timeo == 0))
+                               goto out;
+
+                       /* wait for a message to turn up */
+                       release_sock(&rx->sk);
+                       prepare_to_wait_exclusive(rx->sk.sk_sleep, &wait,
+                                                 TASK_INTERRUPTIBLE);
+                       ret = sock_error(&rx->sk);
+                       if (ret)
+                               goto wait_error;
+
+                       if (skb_queue_empty(&rx->sk.sk_receive_queue)) {
+                               if (signal_pending(current))
+                                       goto wait_interrupted;
+                               timeo = schedule_timeout(timeo);
+                       }
+                       finish_wait(rx->sk.sk_sleep, &wait);
+                       lock_sock(&rx->sk);
+                       continue;
+               }
+
+       peek_next_packet:
+               sp = rxrpc_skb(skb);
+               call = sp->call;
+               ASSERT(call != NULL);
+
+               _debug("next pkt %s", rxrpc_pkts[sp->hdr.type]);
+
+               /* make sure we wait for the state to be updated in this call */
+               spin_lock_bh(&call->lock);
+               spin_unlock_bh(&call->lock);
+
+               if (test_bit(RXRPC_CALL_RELEASED, &call->flags)) {
+                       _debug("packet from released call");
+                       if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
+                               BUG();
+                       rxrpc_free_skb(skb);
+                       continue;
+               }
+
+               /* determine whether to continue last data receive */
+               if (continue_call) {
+                       _debug("maybe cont");
+                       if (call != continue_call ||
+                           skb->mark != RXRPC_SKB_MARK_DATA) {
+                               release_sock(&rx->sk);
+                               rxrpc_put_call(continue_call);
+                               _leave(" = %d [noncont]", copied);
+                               return copied;
+                       }
+               }
+
+               rxrpc_get_call(call);
+
+               /* copy the peer address and timestamp */
+               if (!continue_call) {
+                       if (msg->msg_name && msg->msg_namelen > 0)
+                               memcpy(&msg->msg_name, &call->conn->trans->peer->srx,
+                                      sizeof(call->conn->trans->peer->srx));
+                       sock_recv_timestamp(msg, &rx->sk, skb);
+               }
+
+               /* receive the message */
+               if (skb->mark != RXRPC_SKB_MARK_DATA)
+                       goto receive_non_data_message;
+
+               _debug("recvmsg DATA #%u { %d, %d }",
+                      ntohl(sp->hdr.seq), skb->len, sp->offset);
+
+               if (!continue_call) {
+                       /* only set the control data once per recvmsg() */
+                       ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
+                                      ullen, &call->user_call_ID);
+                       if (ret < 0)
+                               goto copy_error;
+                       ASSERT(test_bit(RXRPC_CALL_HAS_USERID, &call->flags));
+               }
+
+               ASSERTCMP(ntohl(sp->hdr.seq), >=, call->rx_data_recv);
+               ASSERTCMP(ntohl(sp->hdr.seq), <=, call->rx_data_recv + 1);
+               call->rx_data_recv = ntohl(sp->hdr.seq);
+
+               ASSERTCMP(ntohl(sp->hdr.seq), >, call->rx_data_eaten);
+
+               offset = sp->offset;
+               copy = skb->len - offset;
+               if (copy > len - copied)
+                       copy = len - copied;
+
+               if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
+                       ret = skb_copy_datagram_iovec(skb, offset,
+                                                     msg->msg_iov, copy);
+               } else {
+                       ret = skb_copy_and_csum_datagram_iovec(skb, offset,
+                                                              msg->msg_iov);
+                       if (ret == -EINVAL)
+                               goto csum_copy_error;
+               }
+
+               if (ret < 0)
+                       goto copy_error;
+
+               /* handle piecemeal consumption of data packets */
+               _debug("copied %d+%d", copy, copied);
+
+               offset += copy;
+               copied += copy;
+
+               if (!(flags & MSG_PEEK))
+                       sp->offset = offset;
+
+               if (sp->offset < skb->len) {
+                       _debug("buffer full");
+                       ASSERTCMP(copied, ==, len);
+                       break;
+               }
+
+               /* we transferred the whole data packet */
+               if (sp->hdr.flags & RXRPC_LAST_PACKET) {
+                       _debug("last");
+                       if (call->conn->out_clientflag) {
+                                /* last byte of reply received */
+                               ret = copied;
+                               goto terminal_message;
+                       }
+
+                       /* last bit of request received */
+                       if (!(flags & MSG_PEEK)) {
+                               _debug("eat packet");
+                               if (skb_dequeue(&rx->sk.sk_receive_queue) !=
+                                   skb)
+                                       BUG();
+                               rxrpc_free_skb(skb);
+                       }
+                       msg->msg_flags &= ~MSG_MORE;
+                       break;
+               }
+
+               /* move on to the next data message */
+               _debug("next");
+               if (!continue_call)
+                       continue_call = sp->call;
+               else
+                       rxrpc_put_call(call);
+               call = NULL;
+
+               if (flags & MSG_PEEK) {
+                       _debug("peek next");
+                       skb = skb->next;
+                       if (skb == (struct sk_buff *) &rx->sk.sk_receive_queue)
+                               break;
+                       goto peek_next_packet;
+               }
+
+               _debug("eat packet");
+               if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
+                       BUG();
+               rxrpc_free_skb(skb);
+       }
+
+       /* end of non-terminal data packet reception for the moment */
+       _debug("end rcv data");
+out:
+       release_sock(&rx->sk);
+       if (call)
+               rxrpc_put_call(call);
+       if (continue_call)
+               rxrpc_put_call(continue_call);
+       _leave(" = %d [data]", copied);
+       return copied;
+
+       /* handle non-DATA messages such as aborts, incoming connections and
+        * final ACKs */
+receive_non_data_message:
+       _debug("non-data");
+
+       if (skb->mark == RXRPC_SKB_MARK_NEW_CALL) {
+               _debug("RECV NEW CALL");
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NEW_CALL, 0, &abort_code);
+               if (ret < 0)
+                       goto copy_error;
+               if (!(flags & MSG_PEEK)) {
+                       if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
+                               BUG();
+                       rxrpc_free_skb(skb);
+               }
+               goto out;
+       }
+
+       ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
+                      ullen, &call->user_call_ID);
+       if (ret < 0)
+               goto copy_error;
+       ASSERT(test_bit(RXRPC_CALL_HAS_USERID, &call->flags));
+
+       switch (skb->mark) {
+       case RXRPC_SKB_MARK_DATA:
+               BUG();
+       case RXRPC_SKB_MARK_FINAL_ACK:
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &abort_code);
+               break;
+       case RXRPC_SKB_MARK_BUSY:
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_BUSY, 0, &abort_code);
+               break;
+       case RXRPC_SKB_MARK_REMOTE_ABORT:
+               abort_code = call->abort_code;
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &abort_code);
+               break;
+       case RXRPC_SKB_MARK_NET_ERROR:
+               _debug("RECV NET ERROR %d", sp->error);
+               abort_code = sp->error;
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &abort_code);
+               break;
+       case RXRPC_SKB_MARK_LOCAL_ERROR:
+               _debug("RECV LOCAL ERROR %d", sp->error);
+               abort_code = sp->error;
+               ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4,
+                              &abort_code);
+               break;
+       default:
+               BUG();
+               break;
+       }
+
+       if (ret < 0)
+               goto copy_error;
+
+terminal_message:
+       _debug("terminal");
+       msg->msg_flags &= ~MSG_MORE;
+       msg->msg_flags |= MSG_EOR;
+
+       if (!(flags & MSG_PEEK)) {
+               _net("free terminal skb %p", skb);
+               if (skb_dequeue(&rx->sk.sk_receive_queue) != skb)
+                       BUG();
+               rxrpc_free_skb(skb);
+               rxrpc_remove_user_ID(rx, call);
+       }
+
+       release_sock(&rx->sk);
+       rxrpc_put_call(call);
+       if (continue_call)
+               rxrpc_put_call(continue_call);
+       _leave(" = %d", ret);
+       return ret;
+
+copy_error:
+       _debug("copy error");
+       release_sock(&rx->sk);
+       rxrpc_put_call(call);
+       if (continue_call)
+               rxrpc_put_call(continue_call);
+       _leave(" = %d", ret);
+       return ret;
+
+csum_copy_error:
+       _debug("csum error");
+       release_sock(&rx->sk);
+       if (continue_call)
+               rxrpc_put_call(continue_call);
+       rxrpc_kill_skb(skb);
+       skb_kill_datagram(&rx->sk, skb, flags);
+       rxrpc_put_call(call);
+       return -EAGAIN;
+
+wait_interrupted:
+       ret = sock_intr_errno(timeo);
+wait_error:
+       finish_wait(rx->sk.sk_sleep, &wait);
+       if (continue_call)
+               rxrpc_put_call(continue_call);
+       if (copied)
+               copied = ret;
+       _leave(" = %d [waitfail %d]", copied, ret);
+       return copied;
+
+}