This is a note to let you know that I've just added the patch titled
mptcp: implement TCP_NOTSENT_LOWAT support
to the 6.8-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
mptcp-implement-tcp_notsent_lowat-support.patch
and it can be found in the queue-6.8 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 2a46eda8186692bbbf798c05afd9481c5d708887
Author: Paolo Abeni <pabeni@xxxxxxxxxx>
Date: Fri Mar 1 18:43:46 2024 +0100
mptcp: implement TCP_NOTSENT_LOWAT support
[ Upstream commit 29b5e5ef87397963ca38d3eec0d296ad1c979bbc ]
Add support for such socket option storing the user-space provided
value in a new msk field, and using such data to implement the
_mptcp_stream_memory_free() helper, similar to the TCP one.
To avoid adding more indirect calls in the fast path, open-code
a variant of sk_stream_memory_free() in mptcp_sendmsg() and add
direct calls to the mptcp stream memory free helper where possible.
Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/464
Signed-off-by: Paolo Abeni <pabeni@xxxxxxxxxx>
Reviewed-by: Mat Martineau <martineau@xxxxxxxxxx>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@xxxxxxxxxx>
Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>
Stable-dep-of: bd11dc4fb969 ("mptcp: fix full TCP keep-alive support")
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index fcd85adc621c1..54e29ab911f0d 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1757,6 +1757,30 @@ static int do_copy_data_nocache(struct sock *sk, int copy,
return 0;
}
+/* open-code sk_stream_memory_free() plus sent limit computation to
+ * avoid indirect calls in fast-path.
+ * Called under the msk socket lock, so we can avoid a bunch of ONCE
+ * annotations.
+ */
+static u32 mptcp_send_limit(const struct sock *sk)
+{
+ const struct mptcp_sock *msk = mptcp_sk(sk);
+ u32 limit, not_sent;
+
+ if (sk->sk_wmem_queued >= READ_ONCE(sk->sk_sndbuf))
+ return 0;
+
+ limit = mptcp_notsent_lowat(sk);
+ if (limit == UINT_MAX)
+ return UINT_MAX;
+
+ not_sent = msk->write_seq - msk->snd_nxt;
+ if (not_sent >= limit)
+ return 0;
+
+ return limit - not_sent;
+}
+
static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
{
struct mptcp_sock *msk = mptcp_sk(sk);
@@ -1801,6 +1825,12 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
struct mptcp_data_frag *dfrag;
bool dfrag_collapsed;
size_t psize, offset;
+ u32 copy_limit;
+
+ /* ensure fitting the notsent_lowat() constraint */
+ copy_limit = mptcp_send_limit(sk);
+ if (!copy_limit)
+ goto wait_for_memory;
/* reuse tail pfrag, if possible, or carve a new one from the
* page allocator
@@ -1808,9 +1838,6 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
dfrag = mptcp_pending_tail(sk);
dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
if (!dfrag_collapsed) {
- if (!sk_stream_memory_free(sk))
- goto wait_for_memory;
-
if (!mptcp_page_frag_refill(sk, pfrag))
goto wait_for_memory;
@@ -1825,6 +1852,7 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
offset = dfrag->offset + dfrag->data_len;
psize = pfrag->size - offset;
psize = min_t(size_t, psize, msg_data_left(msg));
+ psize = min_t(size_t, psize, copy_limit);
total_ts = psize + frag_truesize;
if (!sk_wmem_schedule(sk, total_ts))
@@ -3761,6 +3789,7 @@ static struct proto mptcp_prot = {
.unhash = mptcp_unhash,
.get_port = mptcp_get_port,
.forward_alloc_get = mptcp_forward_alloc_get,
+ .stream_memory_free = mptcp_stream_memory_free,
.sockets_allocated = &mptcp_sockets_allocated,
.memory_allocated = &tcp_memory_allocated,
@@ -3932,12 +3961,12 @@ static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
{
struct sock *sk = (struct sock *)msk;
- if (sk_stream_is_writeable(sk))
+ if (__mptcp_stream_is_writeable(sk, 1))
return EPOLLOUT | EPOLLWRNORM;
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
smp_mb__after_atomic(); /* NOSPACE is changed by mptcp_write_space() */
- if (sk_stream_is_writeable(sk))
+ if (__mptcp_stream_is_writeable(sk, 1))
return EPOLLOUT | EPOLLWRNORM;
return 0;
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 6b83869ef7938..2f17f295d7c8b 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -305,6 +305,7 @@ struct mptcp_sock {
in_accept_queue:1,
free_first:1,
rcvspace_init:1;
+ u32 notsent_lowat;
struct work_struct work;
struct sk_buff *ooo_last_skb;
struct rb_root out_of_order_queue;
@@ -789,11 +790,36 @@ static inline bool mptcp_data_fin_enabled(const struct mptcp_sock *msk)
READ_ONCE(msk->write_seq) == READ_ONCE(msk->snd_nxt);
}
+static inline u32 mptcp_notsent_lowat(const struct sock *sk)
+{
+ struct net *net = sock_net(sk);
+ u32 val;
+
+ val = READ_ONCE(mptcp_sk(sk)->notsent_lowat);
+ return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
+}
+
+static inline bool mptcp_stream_memory_free(const struct sock *sk, int wake)
+{
+ const struct mptcp_sock *msk = mptcp_sk(sk);
+ u32 notsent_bytes;
+
+ notsent_bytes = READ_ONCE(msk->write_seq) - READ_ONCE(msk->snd_nxt);
+ return (notsent_bytes << wake) < mptcp_notsent_lowat(sk);
+}
+
+static inline bool __mptcp_stream_is_writeable(const struct sock *sk, int wake)
+{
+ return mptcp_stream_memory_free(sk, wake) &&
+ __sk_stream_is_writeable(sk, wake);
+}
+
static inline void mptcp_write_space(struct sock *sk)
{
/* pairs with memory barrier in mptcp_poll */
smp_mb();
- sk_stream_write_space(sk);
+ if (mptcp_stream_memory_free(sk, 1))
+ sk_stream_write_space(sk);
}
static inline void __mptcp_sync_sndbuf(struct sock *sk)
diff --git a/net/mptcp/sockopt.c b/net/mptcp/sockopt.c
index 82d0cd0819f09..f2fe28a3912a9 100644
--- a/net/mptcp/sockopt.c
+++ b/net/mptcp/sockopt.c
@@ -810,6 +810,16 @@ static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
return 0;
case TCP_ULP:
return -EOPNOTSUPP;
+ case TCP_NOTSENT_LOWAT:
+ ret = mptcp_get_int_option(msk, optval, optlen, &val);
+ if (ret)
+ return ret;
+
+ lock_sock(sk);
+ WRITE_ONCE(msk->notsent_lowat, val);
+ mptcp_write_space(sk);
+ release_sock(sk);
+ return 0;
case TCP_CONGESTION:
return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen);
case TCP_CORK:
@@ -1343,6 +1353,8 @@ static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname,
return mptcp_put_int_option(msk, optval, optlen, msk->cork);
case TCP_NODELAY:
return mptcp_put_int_option(msk, optval, optlen, msk->nodelay);
+ case TCP_NOTSENT_LOWAT:
+ return mptcp_put_int_option(msk, optval, optlen, msk->notsent_lowat);
}
return -EOPNOTSUPP;
}
