netlink

README.adoc

@@ -2439,7 +2439,7 @@ Can also be activated with the `panic_on_warn` boot parameter.
 
 ==== CONFIG_PROC_EVENTS
 
-Logs proc events such as process creation to a link:https://en.wikipedia.org/wiki/Netlink[netlink socket].
+Logs proc events such as process creation to a link:kernel_module/netlink.c[netlink socket].
 
 We then have a userland program that listens to the events and prints them out:
 

build-all

@@ -16,6 +16,6 @@ shift "$(($OPTIND - 1))"
 for arch in x86_64 arm aarch64; do
   ./build -a "$arch" -klq
   if "$gem5"; then
-    ./build -a "$arch" -g -Gkl
+    ./build -a "$arch" -Gkl
   fi
 done

kernel_module/README.adoc

@@ -10,37 +10,39 @@
 .. link:oops.c[]
 .. link:warn_on.c[]
 .. link:warn_on.c[]
-. Module utils
+. Modules
 .. link:params.c[]
 .. link:vermagic.c[]
 .. link:vermagic_fail.c[]
 .. link:module_info.c[]
+.. Module dependencies
+... link:dep.c[]
+... link:dep2.c[]
 . Pseudo filesystems
+.. link:anonymous_inode.c[]
+.. link:character_device.c[]
+.. link:character_device_create.c[]
 .. link:debugfs.c[]
-.. link:procfs.c[]
-.. link:sysfs.c[]
 .. link:fops.c[]
 .. link:ioctl.c[]
-.. link:poll.c[]
 .. link:mmap.c[]
-.. link:anonymous_inode.c[]
+.. link:poll.c[]
+.. link:procfs.c[]
 .. link:seq_file.c[]
 .. link:seq_file_inode.c[]
+.. link:sysfs.c[]
 . Asynchronous
-.. link:workqueue_cheat.c[]
-.. link:sleep.c[]
+.. link:irq.c[]
 .. link:kthread.c[]
 .. link:kthreads.c[]
 .. link:schedule.c[]
+.. link:sleep.c[]
 .. link:timer.c[]
 .. link:work_from_work.c[]
-.. link:irq.c[]
-. Module dependencies
-.. link:dep.c[]
-.. link:dep2.c[]
-.. link:character_device.c[]
-.. link:character_device_create.c[]
+.. link:workqueue_cheat.c[]
+. Misc
 .. link:virt_to_phys.c[]
+.. link:netlink.c[]
 . Utilities
 .. link:kstrto.c[]
 . Arch

kernel_module/netlink.c

@@ -0,0 +1,78 @@
+/*
+https://en.wikipedia.org/wiki/Netlink
+
+https://stackoverflow.com/questions/3299386/how-to-use-netlink-socket-to-communicate-with-a-kernel-module
+*/
+
+#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
+
+struct sock *nl_sk = NULL;
+
+static void callback(struct sk_buff *skb)
+{
+	char readbuf[1024];
+	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);
+    pid = nlh->nlmsg_pid;
+    skb_out = nlmsg_new(readbuflen, 0);
+    if (!skb_out) {
+        pr_err("nlmsg_new\n");
+        return;
+    }
+    nlh = nlmsg_put(skb_out, 0, 0, NLMSG_DONE, readbuflen, 0);
+    NETLINK_CB(skb_out).dst_group = 0;
+    strncpy(nlmsg_data(nlh), readbuf, readbuflen);
+    res = nlmsg_unicast(nl_sk, skb_out, pid);
+    if (res < 0)
+        pr_info("nlmsg_unicast\n");
+}
+
+static int myinit(void)
+{
+    struct netlink_kernel_cfg cfg = {
+        .input = callback,
+    };
+    nl_sk = netlink_kernel_create(&init_net, NETLINK_USER, &cfg);
+    if (!nl_sk) {
+        pr_err("netlink_kernel_create\n");
+        return -10;
+    }
+    return 0;
+}
+
+static void myexit(void)
+{
+    netlink_kernel_release(nl_sk);
+}
+
+module_init(myinit);
+module_exit(myexit);
+MODULE_LICENSE("GPL");

kernel_module/user/README.adoc

@@ -25,5 +25,6 @@ These programs can also be compiled and used on host.
 ... link:ring0.c[]
 . Module tests
 .. link:anonymous_inode.c[]
-.. link:poll.c[]
 .. link:ioctl.c[]
+.. link:netlink.c[]
+.. link:poll.c[]

kernel_module/user/netlink.c

@@ -0,0 +1,49 @@
+#include <linux/netlink.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+#define MAX_PAYLOAD 1024
+#define NETLINK_USER 31
+
+/* Some of these structs fields must be zeroed.
+ * We could brute force memset them, but
+ * TODO determine exactly which, and move into main. */
+int sock_fd;
+struct iovec iov;
+struct msghdr msg;
+struct nlmsghdr *nlh;
+struct sockaddr_nl src_addr, dest_addr;
+
+int main()
+{
+	sock_fd = socket(PF_NETLINK, SOCK_RAW, NETLINK_USER);
+	if (sock_fd < 0) {
+		perror("socket");
+		return EXIT_FAILURE;
+	}
+	src_addr.nl_family = AF_NETLINK;
+	src_addr.nl_pid = getpid();
+	bind(sock_fd, (struct sockaddr *)&src_addr, sizeof(src_addr));
+	dest_addr.nl_family = AF_NETLINK;
+	nlh = (struct nlmsghdr *)malloc(NLMSG_SPACE(MAX_PAYLOAD));
+	memset(nlh, 0, NLMSG_SPACE(MAX_PAYLOAD));
+	nlh->nlmsg_len = NLMSG_SPACE(MAX_PAYLOAD);
+	nlh->nlmsg_pid = getpid();
+	nlh->nlmsg_flags = 0;
+	strcpy(NLMSG_DATA(nlh), "user request");
+	iov.iov_base = (void *)nlh;
+	iov.iov_len = nlh->nlmsg_len;
+	msg.msg_name = (void *)&dest_addr;
+	msg.msg_namelen = sizeof(dest_addr);
+	msg.msg_iov = &iov;
+	msg.msg_iovlen = 1;
+	printf("before sendmsg\n");
+	sendmsg(sock_fd, &msg, 0);
+	printf("after sendmsg\n");
+	recvmsg(sock_fd, &msg, 0);
+	printf("userland received: %s\n", (char *)NLMSG_DATA(nlh));
+	close(sock_fd);
+}