[Devel] [PATCH VZ8 4/5] netlink: allow to set peeking offset for sockets

[Pavel Tikhomirov]

From: Stanislav Kinsburskiy <skinsbursky at virtuozzo.com>

Patchset description:

netlink: prepare to dump and restore data from a receive queue

CRIU can dump queued data for unix and tcp sockets,
now it's time for netlink sockets.

Here are there questions.
* How to dump data from a receive queue
  We can set peeking offset like we do for unix sockets.

* How to restore data back to a receive queue
  I suggest to add a repair mode like we do for tcp sockets.

* When we can dump data from a receive queue.
  I think we can do this only if a socket doesn't have a running callback.

  Andrey Vagin (3):
    netlink: allow to set peeking offset for sockets
    netlink: add an ability to restore messages in a receive queue
    netlink/diag: report flags for netlink sockets

https://jira.sw.ru/browse/PSBM-28386

khorenko@: there is no locking right now, but while we are the only
user for this interface, this is not essential at the moment, we'll add
locking on top later in the scope of:

https://jira.sw.ru/browse/PSBM-48484

===========================================================
This patch description:

This allows us to read socket's queue without removing skbs from it.

The same logic was implemented for unix and inet sockets and we use this
to dump and restore sockets in CRIU.

Here is a question whether sk_peek_off has to be protected by locks.
Currently it isn't protected and an user who uses sk_peek_off has to be
sure that nobody calls recvmsg for a socket except him.

https://jira.sw.ru/browse/PSBM-28386

Signed-off-by: Andrey Vagin <avagin at virtuozzo.com>
Reviewed-by: Cyrill Gorcunov <gorcunov at virtuozzo.com>

===========================================================

netlink: Don't manipulate @sk_peek_off if data fetching failed

When skb_copy_datagram_iovec called to fetch queued data
it may fail with EFAULT and if MSG_PEEK set by a caller
the position get advanced even if data hasn't been read.
So we might loose data bits here on subsequent recvmsg
calls. Instead lets exit early with error.

In sake of https://jira.sw.ru/browse/PSBM-57921

Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
Acked-by: Andrey Vagin <avagin at virtuozzo.com>

https://jira.sw.ru/browse/PSBM-101289

Signed-off-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
---
 net/netlink/af_netlink.c | 25 ++++++++++++++++++++-----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index ea497a739298..15d1f2b1e339 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -1943,17 +1943,18 @@ static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 	struct scm_cookie scm;
 	struct sock *sk = sock->sk;
 	struct netlink_sock *nlk = nlk_sk(sk);
-	int noblock = flags&MSG_DONTWAIT;
 	size_t copied;
 	struct sk_buff *skb, *data_skb;
+	int peeked, skip;
 	int err, ret;
 
 	if (flags&MSG_OOB)
 		return -EOPNOTSUPP;
 
 	copied = 0;
+	skip = sk_peek_offset(sk, flags);
 
-	skb = skb_recv_datagram(sk, flags, noblock, &err);
+	skb = __skb_recv_datagram(sk, flags, NULL, &peeked, &skip, &err);
 	if (skb == NULL)
 		goto out;
 
@@ -1981,14 +1982,20 @@ static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
 				     SKB_WITH_OVERHEAD(32768));
 
-	copied = data_skb->len;
+	copied = data_skb->len - skip;
 	if (len < copied) {
 		msg->msg_flags |= MSG_TRUNC;
 		copied = len;
 	}
 
 	skb_reset_transport_header(data_skb);
-	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
+	err = skb_copy_datagram_msg(data_skb, skip, msg, copied);
+	if (!err) {
+		if (flags & MSG_PEEK)
+			sk_peek_offset_fwd(sk, copied);
+		else
+			sk_peek_offset_bwd(sk, skb->len);
+	}
 
 	if (msg->msg_name) {
 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
@@ -2007,7 +2014,7 @@ static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 	memset(&scm, 0, sizeof(scm));
 	scm.creds = *NETLINK_CREDS(skb);
 	if (flags & MSG_TRUNC)
-		copied = data_skb->len;
+		copied = data_skb->len - skip;
 
 	skb_free_datagram(sk, skb);
 
@@ -2681,6 +2688,13 @@ int netlink_unregister_notifier(struct notifier_block *nb)
 }
 EXPORT_SYMBOL(netlink_unregister_notifier);
 
+static int netlink_set_peek_off(struct sock *sk, int val)
+{
+	sk->sk_peek_off = val;
+
+	return 0;
+}
+
 static const struct proto_ops netlink_ops = {
 	.family =	PF_NETLINK,
 	.owner =	THIS_MODULE,
@@ -2700,6 +2714,7 @@ static const struct proto_ops netlink_ops = {
 	.recvmsg =	netlink_recvmsg,
 	.mmap =		sock_no_mmap,
 	.sendpage =	sock_no_sendpage,
+	.set_peek_off = netlink_set_peek_off,
 };
 
 static const struct net_proto_family netlink_family_ops = {
-- 
2.24.1