summaryrefslogtreecommitdiffstats
path: root/net/rxrpc/local_object.c
diff options
context:
space:
mode:
authorDavid Howells <dhowells@redhat.com>2018-10-05 14:05:35 +0100
committerDavid Howells <dhowells@redhat.com>2018-10-05 14:21:59 +0100
commit2cfa2271604bb26e75b828d38f357ed084464795 (patch)
treea86cf7c8b12a35b9a1ffd6e28ccc64bf855256aa /net/rxrpc/local_object.c
parent5e33a23ba4b56c109b732d57a0a76558a37d9ec5 (diff)
downloadlinux-2cfa2271604bb26e75b828d38f357ed084464795.tar.gz
linux-2cfa2271604bb26e75b828d38f357ed084464795.tar.bz2
linux-2cfa2271604bb26e75b828d38f357ed084464795.zip
rxrpc: Fix the data_ready handler
Fix the rxrpc_data_ready() function to pick up all packets and to not miss any. There are two problems: (1) The sk_data_ready pointer on the UDP socket is set *after* it is bound. This means that it's open for business before we're ready to dequeue packets and there's a tiny window exists in which a packet can sneak onto the receive queue, but we never know about it. Fix this by setting the pointers on the socket prior to binding it. (2) skb_recv_udp() will return an error (such as ENETUNREACH) if there was an error on the transmission side, even though we set the sk_error_report hook. Because rxrpc_data_ready() returns immediately in such a case, it never actually removes its packet from the receive queue. Fix this by abstracting out the UDP dequeuing and checksumming into a separate function that keeps hammering on skb_recv_udp() until it returns -EAGAIN, passing the packets extracted to the remainder of the function. and two potential problems: (3) It might be possible in some circumstances or in the future for packets to be being added to the UDP receive queue whilst rxrpc is running consuming them, so the data_ready() handler might get called less often than once per packet. Allow for this by fully draining the queue on each call as (2). (4) If a packet fails the checksum check, the code currently returns after discarding the packet without checking for more. Allow for this by fully draining the queue on each call as (2). Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both") Signed-off-by: David Howells <dhowells@redhat.com> Acked-by: Paolo Abeni <pabeni@redhat.com>
Diffstat (limited to 'net/rxrpc/local_object.c')
-rw-r--r--net/rxrpc/local_object.c11
1 files changed, 6 insertions, 5 deletions
diff --git a/net/rxrpc/local_object.c b/net/rxrpc/local_object.c
index 94d234e9c685..30862f44c9f1 100644
--- a/net/rxrpc/local_object.c
+++ b/net/rxrpc/local_object.c
@@ -122,6 +122,12 @@ static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
return ret;
}
+ /* set the socket up */
+ sock = local->socket->sk;
+ sock->sk_user_data = local;
+ sock->sk_data_ready = rxrpc_data_ready;
+ sock->sk_error_report = rxrpc_error_report;
+
/* if a local address was supplied then bind it */
if (local->srx.transport_len > sizeof(sa_family_t)) {
_debug("bind");
@@ -191,11 +197,6 @@ static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
BUG();
}
- /* set the socket up */
- sock = local->socket->sk;
- sock->sk_user_data = local;
- sock->sk_data_ready = rxrpc_data_ready;
- sock->sk_error_report = rxrpc_error_report;
_leave(" = 0");
return 0;