netlink: move docs to README

Author: cirosantilli

README.adoc

@@ -3920,6 +3920,42 @@ Reads to that inode return the sequence: `1`, `10`, `100`, ... `10000000`, `1`,
 
 Bibliography: https://stackoverflow.com/questions/4508998/what-is-an-anonymous-inode-in-linux
 
+==== netlink sockets
+
+Netlink sockets offer a socket API for kernel / userland communication:
+
+....
+/netlink.sh
+echo $?
+....
+
+Outcome: the test passes:
+
+....
+0
+....
+
+Sources:
+
+* link:kernel_module/netlink.c[]
+* link:kernel_module/netlink.h[]
+* link:kernel_module/user/netlink.c[]
+* link:rootfs_overlay/netlink.sh[]
+
+Launch multiple user requests in parallel to stress our socket:
+
+....
+insmod /netlink.ko sleep=1
+for i in `seq 16`; do /netlink.out & done
+....
+
+TODO: what is the advantage over `read`, `write` and `poll`? https://stackoverflow.com/questions/16727212/how-netlink-socket-in-linux-kernel-is-different-from-normal-polling-done-by-appl
+
+Bibliography:
+
+* https://stackoverflow.com/questions/3299386/how-to-use-netlink-socket-to-communicate-with-a-kernel-module
+* https://en.wikipedia.org/wiki/Netlink
+
 === Linux kernel asynchronous APIs
 
 In this section we will document asynchronous APIs of Linux kernel, especially kthread-related scheduled events.

kernel_module/README.adoc

@@ -9,8 +9,6 @@ Our kernel modules!
 .. link:timer.c[]
 .. link:work_from_work.c[]
 .. link:workqueue_cheat.c[]
-. Misc
-.. link:netlink.c[]
 . Hardening
 .. link:strlen_overflow.c[]
 . Tracing

kernel_module/netlink.c

@@ -1,46 +1,34 @@
-/*
-https://en.wikipedia.org/wiki/Netlink
-
-https://stackoverflow.com/questions/3299386/how-to-use-netlink-socket-to-communicate-with-a-kernel-module
-*/
+/* https://github.com/cirosantilli/linux-kernel-module-cheat#netlink-sockets */
 
 #include <linux/delay.h> /* usleep_range */
-#include <linux/jiffies.h>
 #include <linux/module.h>
 #include <linux/netlink.h>
 #include <linux/skbuff.h>
 #include <net/sock.h>
 
-/* Socket identifier, matches userland. TODO can be anything?
- * Is there a more scalable way to do it? E.g. ioctl device,
- * kernel generates one on the fly, then give it back and connect?
- * https://stackoverflow.com/questions/32898173/can-i-have-more-than-32-netlink-sockets-in-kernelspace */
-#define NETLINK_USER 31
+#include "netlink.h"
 
 struct sock *nl_sk = NULL;
 
+static u32 count;
+static u32 sleep;
+module_param(sleep, int, S_IRUSR | S_IWUSR);
+
 static void callback(struct sk_buff *skb)
 {
-	char readbuf[1024];
+	char readbuf[9];
 	size_t readbuflen;
     int pid;
     int res;
     struct nlmsghdr *nlh;
     struct sk_buff *skb_out;
 
-	/* Read user message. */
     nlh = (struct nlmsghdr *)skb->data;
     pr_info("kernel received: %s\n", (char *)nlmsg_data(nlh));
-
-	/* Add an artificial sleep to see what happens when
-	 * multiple requests come in at the same time.
-	 *
-	 * Try this out (it works):
-	 * for i in `seq 16`; do /netlink.out & done */
-	usleep_range(1000000, 1000001);
-
-	/* Reply with jiffies. */
-	readbuflen = snprintf(readbuf, sizeof(readbuf), "%llu", (unsigned long long)jiffies);
+	if (sleep)
+		usleep_range(1000000, 1000001);
+	readbuflen = snprintf(readbuf, sizeof(readbuf), "%x", count);
+    count++;
     pid = nlh->nlmsg_pid;
     skb_out = nlmsg_new(readbuflen, 0);
     if (!skb_out) {

kernel_module/netlink.h

@@ -0,0 +1,10 @@
+#ifndef NETLINK_H
+#define NETLINK_H
+
+/* Socket identifier, matches userland. TODO can be anything?
+ * Is there a more scalable way to do it? E.g. ioctl device,
+ * kernel generates one on the fly, then give it back and connect?
+ * https://stackoverflow.com/questions/32898173/can-i-have-more-than-32-netlink-sockets-in-kernelspace */
+#define NETLINK_USER 31
+
+#endif

kernel_module/user/netlink.c

@@ -1,12 +1,15 @@
+/* https://github.com/cirosantilli/linux-kernel-module-cheat#netlink-sockets */
+
 #include <linux/netlink.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/socket.h>
 #include <unistd.h>
 
+#include "../netlink.h"
+
 #define MAX_PAYLOAD 1024
-#define NETLINK_USER 31
 
 /* Some of these structs fields must be zeroed.
  * We could brute force memset them, but
@@ -40,10 +43,10 @@ int main()
 	msg.msg_namelen = sizeof(dest_addr);
 	msg.msg_iov = &iov;
 	msg.msg_iovlen = 1;
-	printf("before sendmsg\n");
+	fprintf(stderr, "before sendmsg\n");
 	sendmsg(sock_fd, &msg, 0);
-	printf("after sendmsg\n");
+	fprintf(stderr, "after sendmsg\n");
 	recvmsg(sock_fd, &msg, 0);
-	printf("userland received: %s\n", (char *)NLMSG_DATA(nlh));
+	printf("%s\n", (char *)NLMSG_DATA(nlh));
 	close(sock_fd);
 }

rootfs_overlay/netlink.sh

@@ -0,0 +1,7 @@
+#!/bin/sh
+set -e
+insmod /netlink.ko
+[ "$(/netlink.out)" = 0 ]
+[ "$(/netlink.out)" = 1 ]
+[ "$(/netlink.out)" = 2 ]
+rmmod netlink

rootfs_overlay/test_all.sh

@@ -10,6 +10,7 @@ for test in \
   /ioctl.sh \
   /kstrto.sh \
   /mmap.sh \
+  /netlink.sh \
   /params.sh \
   /procfs.sh \
   /seq_file.sh \