Skip to contentSkip to conversation

net/tap: simplfication and servicabilty improvements

PATCH In review Stephen Hemminger 4 participants 8 messages Apr 24, 2020 – May 6, 2020 Network drivers

lore
Share & tools
lore
#0 lore raw

cover net/tap: simplfication and servicabilty improvements

These are a couple of small fixes to the TAP driver. The first makes it more robust to random signals, and the second one adds better error reporting.

Stephen Hemminger (2):
net/tap: simplify netlink send/receive functions net/tap: use netlink extended ack support

 drivers/net/tap/tap_netlink.c | 123 +++++++++++++++++++++++++---------
 1 file changed, 92 insertions(+), 31 deletions(-)
Signature · 1 line
2.20.1
#1 lore raw

PATCH 1/2 net/tap: simplify netlink send/receive functions

The tap_nl_recv() function does not need to use the full complex recvmsg() system call, basic recv() will work here.

Ditto for tap_nl_send() full sendmsg is not needed.

Add logic to retry in case EINTR rather than forcing error handling back in driver or worse to ethdev API.

---
 drivers/net/tap/tap_netlink.c | 45 +++++++++++------------------------
 1 file changed, 14 insertions(+), 31 deletions(-)
drivers/net/tap/tap_netlink.c +14 -31
index 14bbbec754f6..64220178a6ba 100644
@@ -101,26 +101,17 @@ tap_nl_final(int nlsk_fd)
 int
 tap_nl_send(int nlsk_fd, struct nlmsghdr *nh)
 {
-	/* man 7 netlink EXAMPLE */
-	struct sockaddr_nl sa = {
-		.nl_family = AF_NETLINK,
-	};
-	struct iovec iov = {
-		.iov_base = nh,
-		.iov_len = nh->nlmsg_len,
-	};
-	struct msghdr msg = {
-		.msg_name = &sa,
-		.msg_namelen = sizeof(sa),
-		.msg_iov = &iov,
-		.msg_iovlen = 1,
-	};
 	int send_bytes;
 
 	nh->nlmsg_pid = 0; /* communication with the kernel uses pid 0 */
 	nh->nlmsg_seq = (uint32_t)rte_rand();
-	send_bytes = sendmsg(nlsk_fd, &msg, 0);
+
+retry:
+	send_bytes = send(nlsk_fd, nh, nh->nlmsg_len, 0);
 	if (send_bytes < 0) {
+		if (errno == EINTR)
+			goto retry;
+
 		TAP_LOG(ERR, "Failed to send netlink message: %s (%d)",
 			strerror(errno), errno);
 		return -1;
@@ -161,30 +152,22 @@ tap_nl_recv_ack(int nlsk_fd)
 int
 tap_nl_recv(int nlsk_fd, int (*cb)(struct nlmsghdr *, void *arg), void *arg)
 {
-	/* man 7 netlink EXAMPLE */
-	struct sockaddr_nl sa;
 	char buf[BUF_SIZE];
-	struct iovec iov = {
-		.iov_base = buf,
-		.iov_len = sizeof(buf),
-	};
-	struct msghdr msg = {
-		.msg_name = &sa,
-		.msg_namelen = sizeof(sa),
-		.msg_iov = &iov,
-		/* One message at a time */
-		.msg_iovlen = 1,
-	};
 	int multipart = 0;
 	int ret = 0;
 
 	do {
 		struct nlmsghdr *nh;
-		int recv_bytes = 0;
+		int recv_bytes;
 
-		recv_bytes = recvmsg(nlsk_fd, &msg, 0);
-		if (recv_bytes < 0)
+retry:
+		recv_bytes = recv(nlsk_fd, buf, sizeof(buf), 0);
+		if (recv_bytes < 0) {
+			if (errno == EINTR)
+				goto retry;
 			return -1;
+		}
+
 		for (nh = (struct nlmsghdr *)buf;
 		     NLMSG_OK(nh, (unsigned int)recv_bytes);
 		     nh = NLMSG_NEXT(nh, recv_bytes)) {
Signature · 1 line
2.20.1
#2 lore raw

PATCH 2/2 net/tap: use netlink extended ack support

In reply to Stephen Hemminger

In recent Linux kernels, there is support for extended acknowledgement to netlink messages. This is quite useful for diagnosing errors in configuration in the kernel with TAP.

---
 drivers/net/tap/tap_netlink.c | 78 +++++++++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/net/tap/tap_netlink.c +78 -0
index 64220178a6ba..2365678a9c13 100644
@@ -9,10 +9,12 @@
 #include <string.h>
 #include <sys/socket.h>
 #include <unistd.h>
+#include <stdbool.h>
 
 #include <rte_malloc.h>
 #include <tap_netlink.h>
 #include <rte_random.h>
+
 #include "tap_log.h"
 
 /* Must be quite large to support dumping a huge list of QDISC or filters. */
@@ -43,6 +45,7 @@ tap_nl_init(uint32_t nl_groups)
 		.nl_family = AF_NETLINK,
 		.nl_groups = nl_groups,
 	};
+	int one = 1;
 
 	fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE);
 	if (fd < 0) {
@@ -59,6 +62,12 @@ tap_nl_init(uint32_t nl_groups)
 		close(fd);
 		return -1;
 	}
+
+#ifdef NETLINK_EXT_ACK
+	/* Ask for extended ACK response. on older kernel will ignore request. */
+	setsockopt(fd, SOL_NETLINK, NETLINK_EXT_ACK, &one, sizeof(one));
+#endif
+
 	if (bind(fd, (struct sockaddr *)&local, sizeof(local)) < 0) {
 		TAP_LOG(ERR, "Unable to bind to the netlink socket");
 		close(fd);
@@ -119,6 +128,74 @@ tap_nl_send(int nlsk_fd, struct nlmsghdr *nh)
 	return send_bytes;
 }
 
+#ifdef NETLINK_EXT_ACK
+static const struct nlattr *
+tap_nl_attr_first(const struct nlmsghdr *nh, size_t offset)
+{
+	return (const struct nlattr *)((const char *)nh + NLMSG_SPACE(offset));
+}
+
+static const struct nlattr *
+tap_nl_attr_next(const struct nlattr *attr)
+{
+	return (const struct nlattr *)((const char *)attr
+				       + NLMSG_ALIGN(attr->nla_len));
+}
+
+static bool
+tap_nl_attr_ok(const struct nlattr *attr, int len)
+{
+	if (len < (int)sizeof(struct nlattr))
+		return false; /* missing header */
+	if (attr->nla_len < sizeof(struct nlattr))
+		return false; /* attribute length should include itself */
+	if ((int)attr->nla_len  > len)
+		return false; /* attribute is truncated */
+	return true;
+}
+
+
+/* Decode extended errors from kernel */
+static void
+tap_nl_dump_ext_ack(const struct nlmsghdr *nh, const struct nlmsgerr *err)
+{
+	const struct nlattr *attr;
+	const char *tail = (const char *)nh + NLMSG_ALIGN(nh->nlmsg_len);
+	size_t hlen = sizeof(*err);
+
+	/* no TLVs, no extended response */
+	if (!(nh->nlmsg_flags & NLM_F_ACK_TLVS))
+		return;
+
+	if (!(nh->nlmsg_flags & NLM_F_CAPPED))
+		hlen += err->msg.nlmsg_len - NLMSG_HDRLEN;
+
+	for (attr = tap_nl_attr_first(nh, hlen);
+	     tap_nl_attr_ok(attr, tail - (const char *)attr);
+	     attr = tap_nl_attr_next(attr)) {
+		uint16_t type = attr->nla_type & NLA_TYPE_MASK;
+
+		if (type == NLMSGERR_ATTR_MSG) {
+			const char *msg = (const char *)attr
+				+ NLMSG_ALIGN(sizeof(*attr));
+
+			if (err->error)
+				TAP_LOG(ERR, "%s", msg);
+			else
+
+				TAP_LOG(WARNING, "%s", msg);
+			break;
+		}
+	}
+}
+#else
+/*
+ * External ACK support was added in Linux kernel 4.17
+ * on older kernels, just ignore that part of message
+ */
+#define tap_nl_dump_ext_ack(nh, err) do { } while();
+#endif
+
 /**
  * Check that the kernel sends an appropriate ACK in response
  * to an tap_nl_send().
@@ -174,6 +251,7 @@ tap_nl_recv(int nlsk_fd, int (*cb)(struct nlmsghdr *, void *arg), void *arg)
 			if (nh->nlmsg_type == NLMSG_ERROR) {
 				struct nlmsgerr *err_data = NLMSG_DATA(nh);
 
+				tap_nl_dump_ext_ack(nh, err_data);
 				if (err_data->error < 0) {
 					errno = -err_data->error;
 					return -1;
Signature · 1 line
2.20.1
#3 lore raw

Re: [dpdk-dev] [PATCH 2/2] net/tap: use netlink extended ack support

Quoted 6 lines · Stephen Hemminger

In recent Linux kernels, there is support for extended acknowledgement to netlink messages. This is quite useful for diagnosing errors in configuration in the kernel with TAP.

---

[...]

drivers/net/tap/tap_netlink.c tap_nl_send(int nlsk_fd, struct nlmsghdr *nh) · line 196

Quoted 6 lines
drivers/net/tap/tap_netlink.c +6 -0
+#else
+/*
+ * External ACK support was added in Linux kernel 4.17
+ * on older kernels, just ignore that part of message
+ */
+#define tap_nl_dump_ext_ack(nh, err) do { } while();

Maybe "while(0)" here?

With regards
Andrzej Ostruszka

#4 lore raw

Re: [dpdk-dev] [PATCH 2/2] net/tap: use netlink extended ack support

Quoted 8 lines · Andrzej Ostruszka [C]
Quoted 6 lines · Stephen Hemminger

In recent Linux kernels, there is support for extended acknowledgement to netlink messages. This is quite useful for diagnosing errors in configuration in the kernel with TAP.

---

[...]

Quoted 9 lines

drivers/net/tap/tap_netlink.c tap_nl_send(int nlsk_fd, struct nlmsghdr *nh) · line 196

Quoted 6 lines
drivers/net/tap/tap_netlink.c +6 -0
+#else
+/*
+ * External ACK support was added in Linux kernel 4.17
+ * on older kernels, just ignore that part of message
+ */
+#define tap_nl_dump_ext_ack(nh, err) do { } while();

Maybe "while(0)" here?

+1, will fix while merging.

I assume since it is in #else leg it has not been tested but #if leg tested well.

#5 lore raw

Re: [dpdk-dev] [PATCH 2/2] net/tap: use netlink extended ack support

Quoted 19 lines · On Wed, 6 May 2020 19:41:28 +0100 Ferruh Yigit <[email protected]> wrote:
Quoted 8 lines · Andrzej Ostruszka [C]
Quoted 6 lines · Stephen Hemminger

In recent Linux kernels, there is support for extended acknowledgement to netlink messages. This is quite useful for diagnosing errors in configuration in the kernel with TAP.

---

[...]

Quoted 9 lines

drivers/net/tap/tap_netlink.c tap_nl_send(int nlsk_fd, struct nlmsghdr *nh) · line 196

Quoted 6 lines
drivers/net/tap/tap_netlink.c +6 -0
+#else
+/*
+ * External ACK support was added in Linux kernel 4.17
+ * on older kernels, just ignore that part of message
+ */
+#define tap_nl_dump_ext_ack(nh, err) do { } while();  

Maybe "while(0)" here?

Quoted 3 lines

+1, will fix while merging.

I assume since it is in #else leg it has not been tested but #if leg tested well.

Yes, haven't built on distros that are old enough to hit the #else :-(

Acked-by on the v1 series
#6 lore raw
In reply to Stephen Hemminger
Quoted 8 lines · Stephen Hemminger

These are a couple of small fixes to the TAP driver. The first makes it more robust to random signals, and the second one adds better error reporting.

Quoted 9 lines

Stephen Hemminger (2):
net/tap: simplify netlink send/receive functions
net/tap: use netlink extended ack support

drivers/net/tap/tap_netlink.c | 123 +++++++++++++++++++++++++---------

1 file changed, 92 insertions(+), 31 deletions(-)

2.20.1

#7 lore raw
Quoted 12 lines · Keith
Quoted 12 lines · Stephen Hemminger

These are a couple of small fixes to the TAP driver. The first makes it more robust to random signals, and the second one adds better error reporting.

Stephen Hemminger (2):
net/tap: simplify netlink send/receive functions
net/tap: use netlink extended ack support

Quoted 8 lines
Quoted 5 lines

drivers/net/tap/tap_netlink.c | 123 +++++++++++++++++++++++++---------

1 file changed, 92 insertions(+), 31 deletions(-)

2.20.1

Series applied to dpdk-next-net/master, thanks.