tcp: implement mmap() for zero copy receive

Some networks can make sure TCP payload can exactly fit 4KB pages,
with well chosen MSS/MTU and architectures.

Implement mmap() system call so that applications can avoid
copying data without complex splice() games.

Note that a successful mmap( X bytes) on TCP socket is consuming
bytes, as if recvmsg() has been done. (tp->copied += X)

Only PROT_READ mappings are accepted, as skb page frags
are fundamentally shared and read only.

If tcp_mmap() finds data that is not a full page, or a patch of
urgent data, -EINVAL is returned, no bytes are consumed.

Application must fallback to recvmsg() to read the problematic sequence.

mmap() wont block,  regardless of socket being in blocking or
non-blocking mode. If not enough bytes are in receive queue,
mmap() would return -EAGAIN, or -EIO if socket is in a state
where no other bytes can be added into receive queue.

An application might use SO_RCVLOWAT, poll() and/or ioctl( FIONREAD)
to efficiently use mmap()

On the sender side, MSG_EOR might help to clearly separate unaligned
headers and 4K-aligned chunks if necessary.

Tested:

mlx4 (cx-3) 40Gbit NIC, with tcp_mmap program provided in following patch.
MTU set to 4168  (4096 TCP payload, 40 bytes IPv6 header, 32 bytes TCP header)

Without mmap() (tcp_mmap -s)

received 32768 MB (0 % mmap'ed) in 8.13342 s, 33.7961 Gbit,
  cpu usage user:0.034 sys:3.778, 116.333 usec per MB, 63062 c-switches
received 32768 MB (0 % mmap'ed) in 8.14501 s, 33.748 Gbit,
  cpu usage user:0.029 sys:3.997, 122.864 usec per MB, 61903 c-switches
received 32768 MB (0 % mmap'ed) in 8.11723 s, 33.8635 Gbit,
  cpu usage user:0.048 sys:3.964, 122.437 usec per MB, 62983 c-switches
received 32768 MB (0 % mmap'ed) in 8.39189 s, 32.7552 Gbit,
  cpu usage user:0.038 sys:4.181, 128.754 usec per MB, 55834 c-switches

With mmap() on receiver (tcp_mmap -s -z)

received 32768 MB (100 % mmap'ed) in 8.03083 s, 34.2278 Gbit,
  cpu usage user:0.024 sys:1.466, 45.4712 usec per MB, 65479 c-switches
received 32768 MB (100 % mmap'ed) in 7.98805 s, 34.4111 Gbit,
  cpu usage user:0.026 sys:1.401, 43.5486 usec per MB, 65447 c-switches
received 32768 MB (100 % mmap'ed) in 7.98377 s, 34.4296 Gbit,
  cpu usage user:0.028 sys:1.452, 45.166 usec per MB, 65496 c-switches
received 32768 MB (99.9969 % mmap'ed) in 8.01838 s, 34.281 Gbit,
  cpu usage user:0.02 sys:1.446, 44.7388 usec per MB, 65505 c-switches

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 117 insertions, 2 deletions

diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0ee85c47c185..833154e3df17 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -404,6 +404,8 @@ int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
 		int flags, int *addr_len);
 int tcp_set_rcvlowat(struct sock *sk, int val);
 void tcp_data_ready(struct sock *sk);
+int tcp_mmap(struct file *file, struct socket *sock,
+	     struct vm_area_struct *vma);
 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
 		       struct tcp_options_received *opt_rx,
 		       int estab, struct tcp_fastopen_cookie *foc);
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index f5c562aaef35..3ebf599cebae 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -994,7 +994,7 @@ const struct proto_ops inet_stream_ops = {
 	.getsockopt	   = sock_common_getsockopt,
 	.sendmsg	   = inet_sendmsg,
 	.recvmsg	   = inet_recvmsg,
-	.mmap		   = sock_no_mmap,
+	.mmap		   = tcp_mmap,
 	.sendpage	   = inet_sendpage,
 	.splice_read	   = tcp_splice_read,
 	.read_sock	   = tcp_read_sock,
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index c768d306b657..438fbca96cd3 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -1726,6 +1726,119 @@ int tcp_set_rcvlowat(struct sock *sk, int val)
 }
 EXPORT_SYMBOL(tcp_set_rcvlowat);
 
+/* When user wants to mmap X pages, we first need to perform the mapping
+ * before freeing any skbs in receive queue, otherwise user would be unable
+ * to fallback to standard recvmsg(). This happens if some data in the
+ * requested block is not exactly fitting in a page.
+ *
+ * We only support order-0 pages for the moment.
+ * mmap() on TCP is very strict, there is no point
+ * trying to accommodate with pathological layouts.
+ */
+int tcp_mmap(struct file *file, struct socket *sock,
+	     struct vm_area_struct *vma)
+{
+	unsigned long size = vma->vm_end - vma->vm_start;
+	unsigned int nr_pages = size >> PAGE_SHIFT;
+	struct page **pages_array = NULL;
+	u32 seq, len, offset, nr = 0;
+	struct sock *sk = sock->sk;
+	const skb_frag_t *frags;
+	struct tcp_sock *tp;
+	struct sk_buff *skb;
+	int ret;
+
+	if (vma->vm_pgoff || !nr_pages)
+		return -EINVAL;
+
+	if (vma->vm_flags & VM_WRITE)
+		return -EPERM;
+	/* TODO: Maybe the following is not needed if pages are COW */
+	vma->vm_flags &= ~VM_MAYWRITE;
+
+	lock_sock(sk);
+
+	ret = -ENOTCONN;
+	if (sk->sk_state == TCP_LISTEN)
+		goto out;
+
+	sock_rps_record_flow(sk);
+
+	if (tcp_inq(sk) < size) {
+		ret = sock_flag(sk, SOCK_DONE) ? -EIO : -EAGAIN;
+		goto out;
+	}
+	tp = tcp_sk(sk);
+	seq = tp->copied_seq;
+	/* Abort if urgent data is in the area */
+	if (unlikely(tp->urg_data)) {
+		u32 urg_offset = tp->urg_seq - seq;
+
+		ret = -EINVAL;
+		if (urg_offset < size)
+			goto out;
+	}
+	ret = -ENOMEM;
+	pages_array = kvmalloc_array(nr_pages, sizeof(struct page *),
+				     GFP_KERNEL);
+	if (!pages_array)
+		goto out;
+	skb = tcp_recv_skb(sk, seq, &offset);
+	ret = -EINVAL;
+skb_start:
+	/* We do not support anything not in page frags */
+	offset -= skb_headlen(skb);
+	if ((int)offset < 0)
+		goto out;
+	if (skb_has_frag_list(skb))
+		goto out;
+	len = skb->data_len - offset;
+	frags = skb_shinfo(skb)->frags;
+	while (offset) {
+		if (frags->size > offset)
+			goto out;
+		offset -= frags->size;
+		frags++;
+	}
+	while (nr < nr_pages) {
+		if (len) {
+			if (len < PAGE_SIZE)
+				goto out;
+			if (frags->size != PAGE_SIZE || frags->page_offset)
+				goto out;
+			pages_array[nr++] = skb_frag_page(frags);
+			frags++;
+			len -= PAGE_SIZE;
+			seq += PAGE_SIZE;
+			continue;
+		}
+		skb = skb->next;
+		offset = seq - TCP_SKB_CB(skb)->seq;
+		goto skb_start;
+	}
+	/* OK, we have a full set of pages ready to be inserted into vma */
+	for (nr = 0; nr < nr_pages; nr++) {
+		ret = vm_insert_page(vma, vma->vm_start + (nr << PAGE_SHIFT),
+				     pages_array[nr]);
+		if (ret)
+			goto out;
+	}
+	/* operation is complete, we can 'consume' all skbs */
+	tp->copied_seq = seq;
+	tcp_rcv_space_adjust(sk);
+
+	/* Clean up data we have read: This will do ACK frames. */
+	tcp_recv_skb(sk, seq, &offset);
+	tcp_cleanup_rbuf(sk, size);
+
+	ret = 0;
+out:
+	release_sock(sk);
+	kvfree(pages_array);
+	return ret;
+}
+EXPORT_SYMBOL(tcp_mmap);
+
 static void tcp_update_recv_tstamps(struct sk_buff *skb,
 				    struct scm_timestamping *tss)
 {
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index e70d59fb26e1..2c694912df2e 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -579,7 +579,7 @@ const struct proto_ops inet6_stream_ops = {
 	.getsockopt	   = sock_common_getsockopt,	/* ok		*/
 	.sendmsg	   = inet_sendmsg,		/* ok		*/
 	.recvmsg	   = inet_recvmsg,		/* ok		*/
-	.mmap		   = sock_no_mmap,
+	.mmap		   = tcp_mmap,
 	.sendpage	   = inet_sendpage,
 	.sendmsg_locked    = tcp_sendmsg_locked,
 	.sendpage_locked   = tcp_sendpage_locked,