[PATCH] connector: fix out-of-order cn_proc...

[PATCH] connector: fix out-of-order cn_proc netlink message delivery

From: Aaron Campbell <hidden>
Date: 2016-06-24 13:14:02

The proc connector messages include a sequence number, allowing userspace
programs to detect lost messages.  However, performing this detection is
currently more difficult than necessary, since netlink messages can be
delivered to the application out-of-order.  To fix this, leave pre-emption
disabled during cn_netlink_send(), and use GFP_NOWAIT.

The following was written as a test case.  Building the kernel w/ make -j32
proved a reliable way to generate out-of-order cn_proc messages.

int
main(int argc, char *argv[])
{
	static uint32_t last_seq[CPU_SETSIZE], seq;
	int cpu, fd;
	struct sockaddr_nl sa;
	struct __attribute__((aligned(NLMSG_ALIGNTO))) {
		struct nlmsghdr nl_hdr;
		struct __attribute__((__packed__)) {
			struct cn_msg cn_msg;
			struct proc_event cn_proc;
		};
	} rmsg;
	struct __attribute__((aligned(NLMSG_ALIGNTO))) {
		struct nlmsghdr nl_hdr;
		struct __attribute__((__packed__)) {
			struct cn_msg cn_msg;
			enum proc_cn_mcast_op cn_mcast;
		};
	} smsg;

	fd = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_CONNECTOR);
	if (fd < 0) {
		perror("socket");
	}

	sa.nl_family = AF_NETLINK;
	sa.nl_groups = CN_IDX_PROC;
	sa.nl_pid = getpid();
	if (bind(fd, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
		perror("bind");
	}

	memset(&smsg, 0, sizeof(smsg));
	smsg.nl_hdr.nlmsg_len = sizeof(smsg);
	smsg.nl_hdr.nlmsg_pid = getpid();
	smsg.nl_hdr.nlmsg_type = NLMSG_DONE;
	smsg.cn_msg.id.idx = CN_IDX_PROC;
	smsg.cn_msg.id.val = CN_VAL_PROC;
	smsg.cn_msg.len = sizeof(enum proc_cn_mcast_op);
	smsg.cn_mcast = PROC_CN_MCAST_LISTEN;
	if (send(fd, &smsg, sizeof(smsg), 0) != sizeof(smsg)) {
		perror("send");
	}

	while (recv(fd, &rmsg, sizeof(rmsg), 0) == sizeof(rmsg)) {
		cpu = rmsg.cn_proc.cpu;
		if (cpu < 0) {
			continue;
		}
		seq = rmsg.cn_msg.seq;
		if ((last_seq[cpu] != 0) && (seq != last_seq[cpu] + 1)) {
			printf("out-of-order seq=%d on cpu=%d\n", seq, cpu);
		}
		last_seq[cpu] = seq;
	}

	/* NOTREACHED */

	perror("recv");

	return -1;
}

Signed-off-by: Aaron Campbell <redacted>
---
 drivers/connector/cn_proc.c | 43 ++++++++++++++++++++++---------------------
 1 file changed, 22 insertions(+), 21 deletions(-)

diff --git a/drivers/connector/cn_proc.c b/drivers/connector/cn_proc.c
index 15d06fc..b02f9c6 100644
--- a/drivers/connector/cn_proc.c
+++ b/drivers/connector/cn_proc.c

@@ -56,11 +56,21 @@ static struct cb_id cn_proc_event_id = { CN_IDX_PROC, CN_VAL_PROC };
 /* proc_event_counts is used as the sequence number of the netlink message */
 static DEFINE_PER_CPU(__u32, proc_event_counts) = { 0 };
 
-static inline void get_seq(__u32 *ts, int *cpu)
+static inline void send_msg(struct cn_msg *msg)
 {
 	preempt_disable();
-	*ts = __this_cpu_inc_return(proc_event_counts) - 1;
-	*cpu = smp_processor_id();
+
+	msg->seq = __this_cpu_inc_return(proc_event_counts) - 1;
+	((struct proc_event *)msg->data)->cpu = smp_processor_id();
+
+	/*
+	 * Preemption remains disabled during send to ensure the messages are
+	 * ordered according to their sequence numbers.
+	 *
+	 * If cn_netlink_send() fails, the data is not sent.
+	 */
+	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_NOWAIT);
+
 	preempt_enable();
 }

@@ -77,7 +87,6 @@ void proc_fork_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_FORK;
 	rcu_read_lock();

@@ -92,8 +101,7 @@ void proc_fork_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	/*  If cn_netlink_send() failed, the data is not sent */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_exec_connector(struct task_struct *task)

@@ -108,7 +116,6 @@ void proc_exec_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_EXEC;
 	ev->event_data.exec.process_pid = task->pid;

@@ -118,7 +125,7 @@ void proc_exec_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_id_connector(struct task_struct *task, int which_id)

@@ -150,14 +157,13 @@ void proc_id_connector(struct task_struct *task, int which_id)
 		return;
 	}
 	rcu_read_unlock();
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 
 	memcpy(&msg->id, &cn_proc_event_id, sizeof(msg->id));
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_sid_connector(struct task_struct *task)

@@ -172,7 +178,6 @@ void proc_sid_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_SID;
 	ev->event_data.sid.process_pid = task->pid;

@@ -182,7 +187,7 @@ void proc_sid_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_ptrace_connector(struct task_struct *task, int ptrace_id)

@@ -197,7 +202,6 @@ void proc_ptrace_connector(struct task_struct *task, int ptrace_id)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_PTRACE;
 	ev->event_data.ptrace.process_pid  = task->pid;

@@ -215,7 +219,7 @@ void proc_ptrace_connector(struct task_struct *task, int ptrace_id)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_comm_connector(struct task_struct *task)

@@ -230,7 +234,6 @@ void proc_comm_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_COMM;
 	ev->event_data.comm.process_pid  = task->pid;

@@ -241,7 +244,7 @@ void proc_comm_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_coredump_connector(struct task_struct *task)

@@ -256,7 +259,6 @@ void proc_coredump_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_COREDUMP;
 	ev->event_data.coredump.process_pid = task->pid;

@@ -266,7 +268,7 @@ void proc_coredump_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 void proc_exit_connector(struct task_struct *task)

@@ -281,7 +283,6 @@ void proc_exit_connector(struct task_struct *task)
 	msg = buffer_to_cn_msg(buffer);
 	ev = (struct proc_event *)msg->data;
 	memset(&ev->event_data, 0, sizeof(ev->event_data));
-	get_seq(&msg->seq, &ev->cpu);
 	ev->timestamp_ns = ktime_get_ns();
 	ev->what = PROC_EVENT_EXIT;
 	ev->event_data.exit.process_pid = task->pid;

@@ -293,7 +294,7 @@ void proc_exit_connector(struct task_struct *task)
 	msg->ack = 0; /* not used */
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 /*

@@ -325,7 +326,7 @@ static void cn_proc_ack(int err, int rcvd_seq, int rcvd_ack)
 	msg->ack = rcvd_ack + 1;
 	msg->len = sizeof(*ev);
 	msg->flags = 0; /* not used */
-	cn_netlink_send(msg, 0, CN_IDX_PROC, GFP_KERNEL);
+	send_msg(msg);
 }
 
 /**

-- 
2.7.4

Re: [PATCH] connector: fix out-of-order cn_proc netlink message delivery

From: David Miller <davem@davemloft.net>
Date: 2016-06-28 12:49:51

From: Aaron Campbell <redacted>
Date: Fri, 24 Jun 2016 10:05:32 -0300

The proc connector messages include a sequence number, allowing userspace
programs to detect lost messages.  However, performing this detection is
currently more difficult than necessary, since netlink messages can be
delivered to the application out-of-order.  To fix this, leave pre-emption
disabled during cn_netlink_send(), and use GFP_NOWAIT.

The following was written as a test case.  Building the kernel w/ make -j32
proved a reliable way to generate out-of-order cn_proc messages.

...

Signed-off-by: Aaron Campbell <redacted>

Applied.

Re: [PATCH] connector: fix out-of-order cn_proc netlink message delivery

From: Evgeniy Polyakov <hidden>
Date: 2016-06-28 13:03:40

Hi Aaron

24.06.2016, 16:07, "Aaron Campbell" [off-list ref]:

The proc connector messages include a sequence number, allowing userspace
programs to detect lost messages. However, performing this detection is
currently more difficult than necessary, since netlink messages can be
delivered to the application out-of-order. To fix this, leave pre-emption
disabled during cn_netlink_send(), and use GFP_NOWAIT.

The following was written as a test case. Building the kernel w/ make -j32
proved a reliable way to generate out-of-order cn_proc messages.

This is not actually about out-of-order sending which is impossible iirc,
but the way fork pushes messages into socket queue in parallel. What you've done
is syncing one more layer higher.

I'm not against this patch if you think it does fix some issues, but wording is not correct imo.

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help