[CRIU] [RFC] net: netlink -- Add waiting for RX fillup

[Cyrill Gorcunov]

In case if RX buffer of netlink socket is receiving data
when we're trying to dump it, the CRIU will refuse to
proceed simply because we can't fetch the data which
is placed that deep inside kernel code without modifying
kernel itself. Lets try to wait until receiving is complete,
if it takes too long -- exit out with error.

CC: Andrew Vagin <avagin at parallels.com>
CC: Igor <igor at parallels.com>
CC: Pavel Emelyanov <xemul at parallels.com>
Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
---
 sk-netlink.c | 35 ++++++++++++++++++++++++++++++-----
 1 file changed, 30 insertions(+), 5 deletions(-)

diff --git a/sk-netlink.c b/sk-netlink.c
index 83a20905400c..b0e1465d6146 100644
--- a/sk-netlink.c
+++ b/sk-netlink.c
@@ -2,6 +2,7 @@
 #include <linux/netlink.h>
 #include <linux/rtnetlink.h>
 #include <poll.h>
+#include <time.h>
 
 #include "fdset.h"
 #include "files.h"
@@ -68,12 +69,36 @@ int netlink_receive_one(struct nlmsghdr *hdr, void *arg)
 static bool can_dump_netlink_sk(int lfd)
 {
 	struct pollfd pfd = {lfd, POLLIN, 0};
-	int ret;
+	int ret = 1, num;
+
+	for (num = 1; ret && num < 11; num++) {
+		ret = poll(&pfd, 1, 0);
+		if (ret < 0) {
+			pr_perror("poll() failed");
+			break;
+		} else if (ret == 1) {
+			struct timespec ts = {
+				.tv_nsec = 50000 * num,
+			};
+
+			pr_debug("nelink RX buffer is busy, "
+				 "waiting for %lu nanoseconds\n",
+				 (unsigned long)ts.tv_nsec);
+			/*
+			 * The netlink socket has some data in
+			 * RX ring which are not yet complete.
+			 * Lets wait some time until sending
+			 * side finish transmission but don't
+			 * wait here forewer.
+			 */
+			if (nanosleep(&ts, NULL)) {
+				pr_perror("nelink nanosleep failed");
+				break;
+			}
+		}
+	}
 
-	ret = poll(&pfd, 1, 0);
-	if (ret < 0) {
-		pr_perror("poll() failed");
-	} else if (ret == 1)
+	if (ret)
 		pr_err("The socket has data to read\n");
 
 	return ret == 0;
-- 
1.9.3