tipc: redesign connection-level flow control
There are two flow control mechanisms in TIPC; one at link level that
handles network congestion, burst control, and retransmission, and one
at connection level which' only remaining task is to prevent overflow
in the receiving socket buffer. In TIPC, the latter task has to be
solved end-to-end because messages can not be thrown away once they
have been accepted and delivered upwards from the link layer, i.e, we
can never permit the receive buffer to overflow.
Currently, this algorithm is message based. A counter in the receiving
socket keeps track of number of consumed messages, and sends a dedicated
acknowledge message back to the sender for each 256 consumed message.
A counter at the sending end keeps track of the sent, not yet
acknowledged messages, and blocks the sender if this number ever reaches
512 unacknowledged messages. When the missing acknowledge arrives, the
socket is then woken up for renewed transmission. This works well for
keeping the message flow running, as it almost never happens that a
sender socket is blocked this way.
A problem with the current mechanism is that it potentially is very
memory consuming. Since we don't distinguish between small and large
messages, we have to dimension the socket receive buffer according
to a worst-case of both. I.e., the window size must be chosen large
enough to sustain a reasonable throughput even for the smallest
messages, while we must still consider a scenario where all messages
are of maximum size. Hence, the current fix window size of 512 messages
and a maximum message size of 66k results in a receive buffer of 66 MB
when truesize(66k) = 131k is taken into account. It is possible to do
much better.
This commit introduces an algorithm where we instead use 1024-byte
blocks as base unit. This unit, always rounded upwards from the
actual message size, is used when we advertise windows as well as when
we count and acknowledge transmitted data. The advertised window is
based on the configured receive buffer size in such a way that even
the worst-case truesize/msgsize ratio always is covered. Since the
smallest possible message size (from a flow control viewpoint) now is
1024 bytes, we can safely assume this ratio to be less than four, which
is the value we are now using.
This way, we have been able to reduce the default receive buffer size
from 66 MB to 2 MB with maintained performance.
In order to keep this solution backwards compatible, we introduce a
new capability bit in the discovery protocol, and use this throughout
the message sending/reception path to always select the right unit.
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/tipc/socket.c b/net/tipc/socket.c
index 94bd286..1262889 100644
--- a/net/tipc/socket.c
+++ b/net/tipc/socket.c
@@ -96,9 +96,11 @@
uint conn_timeout;
atomic_t dupl_rcvcnt;
bool link_cong;
- uint sent_unacked;
- uint rcv_unacked;
+ u16 snt_unacked;
+ u16 snd_win;
u16 peer_caps;
+ u16 rcv_unacked;
+ u16 rcv_win;
struct sockaddr_tipc remote;
struct rhash_head node;
struct rcu_head rcu;
@@ -228,9 +230,29 @@
return container_of(sk, struct tipc_sock, sk);
}
-static int tsk_conn_cong(struct tipc_sock *tsk)
+static bool tsk_conn_cong(struct tipc_sock *tsk)
{
- return tsk->sent_unacked >= TIPC_FLOWCTRL_WIN;
+ return tsk->snt_unacked >= tsk->snd_win;
+}
+
+/* tsk_blocks(): translate a buffer size in bytes to number of
+ * advertisable blocks, taking into account the ratio truesize(len)/len
+ * We can trust that this ratio is always < 4 for len >= FLOWCTL_BLK_SZ
+ */
+static u16 tsk_adv_blocks(int len)
+{
+ return len / FLOWCTL_BLK_SZ / 4;
+}
+
+/* tsk_inc(): increment counter for sent or received data
+ * - If block based flow control is not supported by peer we
+ * fall back to message based ditto, incrementing the counter
+ */
+static u16 tsk_inc(struct tipc_sock *tsk, int msglen)
+{
+ if (likely(tsk->peer_caps & TIPC_BLOCK_FLOWCTL))
+ return ((msglen / FLOWCTL_BLK_SZ) + 1);
+ return 1;
}
/**
@@ -378,9 +400,12 @@
sk->sk_write_space = tipc_write_space;
sk->sk_destruct = tipc_sock_destruct;
tsk->conn_timeout = CONN_TIMEOUT_DEFAULT;
- tsk->sent_unacked = 0;
atomic_set(&tsk->dupl_rcvcnt, 0);
+ /* Start out with safe limits until we receive an advertised window */
+ tsk->snd_win = tsk_adv_blocks(RCVBUF_MIN);
+ tsk->rcv_win = tsk->snd_win;
+
if (sock->state == SS_READY) {
tsk_set_unreturnable(tsk, true);
if (sock->type == SOCK_DGRAM)
@@ -776,7 +801,7 @@
struct sock *sk = &tsk->sk;
struct tipc_msg *hdr = buf_msg(skb);
int mtyp = msg_type(hdr);
- int conn_cong;
+ bool conn_cong;
/* Ignore if connection cannot be validated: */
if (!tsk_peer_msg(tsk, hdr))
@@ -790,7 +815,9 @@
return;
} else if (mtyp == CONN_ACK) {
conn_cong = tsk_conn_cong(tsk);
- tsk->sent_unacked -= msg_msgcnt(hdr);
+ tsk->snt_unacked -= msg_conn_ack(hdr);
+ if (tsk->peer_caps & TIPC_BLOCK_FLOWCTL)
+ tsk->snd_win = msg_adv_win(hdr);
if (conn_cong)
sk->sk_write_space(sk);
} else if (mtyp != CONN_PROBE_REPLY) {
@@ -1021,12 +1048,14 @@
u32 dnode;
uint mtu, send, sent = 0;
struct iov_iter save;
+ int hlen = MIN_H_SIZE;
/* Handle implied connection establishment */
if (unlikely(dest)) {
rc = __tipc_sendmsg(sock, m, dsz);
+ hlen = msg_hdr_sz(mhdr);
if (dsz && (dsz == rc))
- tsk->sent_unacked = 1;
+ tsk->snt_unacked = tsk_inc(tsk, dsz + hlen);
return rc;
}
if (dsz > (uint)INT_MAX)
@@ -1055,7 +1084,7 @@
if (likely(!tsk_conn_cong(tsk))) {
rc = tipc_node_xmit(net, &pktchain, dnode, portid);
if (likely(!rc)) {
- tsk->sent_unacked++;
+ tsk->snt_unacked += tsk_inc(tsk, send + hlen);
sent += send;
if (sent == dsz)
return dsz;
@@ -1120,6 +1149,12 @@
tipc_node_add_conn(net, peer_node, tsk->portid, peer_port);
tsk->max_pkt = tipc_node_get_mtu(net, peer_node, tsk->portid);
tsk->peer_caps = tipc_node_get_capabilities(net, peer_node);
+ if (tsk->peer_caps & TIPC_BLOCK_FLOWCTL)
+ return;
+
+ /* Fall back to message based flow control */
+ tsk->rcv_win = FLOWCTL_MSG_WIN;
+ tsk->snd_win = FLOWCTL_MSG_WIN;
}
/**
@@ -1216,7 +1251,7 @@
return 0;
}
-static void tipc_sk_send_ack(struct tipc_sock *tsk, uint ack)
+static void tipc_sk_send_ack(struct tipc_sock *tsk)
{
struct net *net = sock_net(&tsk->sk);
struct sk_buff *skb = NULL;
@@ -1232,7 +1267,14 @@
if (!skb)
return;
msg = buf_msg(skb);
- msg_set_msgcnt(msg, ack);
+ msg_set_conn_ack(msg, tsk->rcv_unacked);
+ tsk->rcv_unacked = 0;
+
+ /* Adjust to and advertize the correct window limit */
+ if (tsk->peer_caps & TIPC_BLOCK_FLOWCTL) {
+ tsk->rcv_win = tsk_adv_blocks(tsk->sk.sk_rcvbuf);
+ msg_set_adv_win(msg, tsk->rcv_win);
+ }
tipc_node_xmit_skb(net, skb, dnode, msg_link_selector(msg));
}
@@ -1290,7 +1332,7 @@
long timeo;
unsigned int sz;
u32 err;
- int res;
+ int res, hlen;
/* Catch invalid receive requests */
if (unlikely(!buf_len))
@@ -1315,6 +1357,7 @@
buf = skb_peek(&sk->sk_receive_queue);
msg = buf_msg(buf);
sz = msg_data_sz(msg);
+ hlen = msg_hdr_sz(msg);
err = msg_errcode(msg);
/* Discard an empty non-errored message & try again */
@@ -1337,7 +1380,7 @@
sz = buf_len;
m->msg_flags |= MSG_TRUNC;
}
- res = skb_copy_datagram_msg(buf, msg_hdr_sz(msg), m, sz);
+ res = skb_copy_datagram_msg(buf, hlen, m, sz);
if (res)
goto exit;
res = sz;
@@ -1349,15 +1392,15 @@
res = -ECONNRESET;
}
- /* Consume received message (optional) */
- if (likely(!(flags & MSG_PEEK))) {
- if ((sock->state != SS_READY) &&
- (++tsk->rcv_unacked >= TIPC_CONNACK_INTV)) {
- tipc_sk_send_ack(tsk, tsk->rcv_unacked);
- tsk->rcv_unacked = 0;
- }
- tsk_advance_rx_queue(sk);
+ if (unlikely(flags & MSG_PEEK))
+ goto exit;
+
+ if (likely(sock->state != SS_READY)) {
+ tsk->rcv_unacked += tsk_inc(tsk, hlen + sz);
+ if (unlikely(tsk->rcv_unacked >= (tsk->rcv_win / 4)))
+ tipc_sk_send_ack(tsk);
}
+ tsk_advance_rx_queue(sk);
exit:
release_sock(sk);
return res;
@@ -1386,7 +1429,7 @@
int sz_to_copy, target, needed;
int sz_copied = 0;
u32 err;
- int res = 0;
+ int res = 0, hlen;
/* Catch invalid receive attempts */
if (unlikely(!buf_len))
@@ -1412,6 +1455,7 @@
buf = skb_peek(&sk->sk_receive_queue);
msg = buf_msg(buf);
sz = msg_data_sz(msg);
+ hlen = msg_hdr_sz(msg);
err = msg_errcode(msg);
/* Discard an empty non-errored message & try again */
@@ -1436,8 +1480,7 @@
needed = (buf_len - sz_copied);
sz_to_copy = (sz <= needed) ? sz : needed;
- res = skb_copy_datagram_msg(buf, msg_hdr_sz(msg) + offset,
- m, sz_to_copy);
+ res = skb_copy_datagram_msg(buf, hlen + offset, m, sz_to_copy);
if (res)
goto exit;
@@ -1459,20 +1502,18 @@
res = -ECONNRESET;
}
- /* Consume received message (optional) */
- if (likely(!(flags & MSG_PEEK))) {
- if (unlikely(++tsk->rcv_unacked >= TIPC_CONNACK_INTV)) {
- tipc_sk_send_ack(tsk, tsk->rcv_unacked);
- tsk->rcv_unacked = 0;
- }
- tsk_advance_rx_queue(sk);
- }
+ if (unlikely(flags & MSG_PEEK))
+ goto exit;
+
+ tsk->rcv_unacked += tsk_inc(tsk, hlen + sz);
+ if (unlikely(tsk->rcv_unacked >= (tsk->rcv_win / 4)))
+ tipc_sk_send_ack(tsk);
+ tsk_advance_rx_queue(sk);
/* Loop around if more data is required */
if ((sz_copied < buf_len) && /* didn't get all requested data */
(!skb_queue_empty(&sk->sk_receive_queue) ||
(sz_copied < target)) && /* and more is ready or required */
- (!(flags & MSG_PEEK)) && /* and aren't just peeking at data */
(!err)) /* and haven't reached a FIN */
goto restart;
@@ -1604,30 +1645,33 @@
/**
* rcvbuf_limit - get proper overload limit of socket receive queue
* @sk: socket
- * @buf: message
+ * @skb: message
*
- * For all connection oriented messages, irrespective of importance,
- * the default overload value (i.e. 67MB) is set as limit.
+ * For connection oriented messages, irrespective of importance,
+ * default queue limit is 2 MB.
*
- * For all connectionless messages, by default new queue limits are
- * as belows:
+ * For connectionless messages, queue limits are based on message
+ * importance as follows:
*
- * TIPC_LOW_IMPORTANCE (4 MB)
- * TIPC_MEDIUM_IMPORTANCE (8 MB)
- * TIPC_HIGH_IMPORTANCE (16 MB)
- * TIPC_CRITICAL_IMPORTANCE (32 MB)
+ * TIPC_LOW_IMPORTANCE (2 MB)
+ * TIPC_MEDIUM_IMPORTANCE (4 MB)
+ * TIPC_HIGH_IMPORTANCE (8 MB)
+ * TIPC_CRITICAL_IMPORTANCE (16 MB)
*
* Returns overload limit according to corresponding message importance
*/
-static unsigned int rcvbuf_limit(struct sock *sk, struct sk_buff *buf)
+static unsigned int rcvbuf_limit(struct sock *sk, struct sk_buff *skb)
{
- struct tipc_msg *msg = buf_msg(buf);
+ struct tipc_sock *tsk = tipc_sk(sk);
+ struct tipc_msg *hdr = buf_msg(skb);
- if (msg_connected(msg))
- return sysctl_tipc_rmem[2];
+ if (unlikely(!msg_connected(hdr)))
+ return sk->sk_rcvbuf << msg_importance(hdr);
- return sk->sk_rcvbuf >> TIPC_CRITICAL_IMPORTANCE <<
- msg_importance(msg);
+ if (likely(tsk->peer_caps & TIPC_BLOCK_FLOWCTL))
+ return sk->sk_rcvbuf;
+
+ return FLOWCTL_MSG_LIM;
}
/**