Why does the kernel hang with this use of netlink sockets?

Akshay Narayan akshayn at mit.edu
Tue May 2 17:57:34 EDT 2017


I have written a kernel module and userspace program which use a netlink
socket to communicate (full source here: https://github.com/
akshayknarayan/netlink-test). The kernel module periodically multicasts a
message, and the userspace program listens on the socket.

Relevant source snippets:
Kernel module:
 43     res = nlmsg_multicast(
 44             nl_sk,     // @sk: netlink socket to spread messages to
 45             skb_out,   // @skb: netlink message as socket buffer
 46             0,         // @portid: own netlink portid to avoid sending
to yourself
 47             MYMGRP,    // @group: multicast group id
 48             GFP_KERNEL // @flags: allocation flags
 49     );
 50     if (res < 0) {
 51         printk(KERN_INFO "Error while sending to user: %d\n", res);
 52     } else {
 53         mod_timer(&timer, jiffies + msecs_to_jiffies(1));
 54         printk(KERN_INFO "Send ok\n");
 55     }

Userspace program:
 39 void nl_recv(int sock) {
...
 54     ret = recvmsg(sock, &msg, 0);
 55     if (ret < 0)
 56         printf("ret < 0.\n");
 57     else
 58         printf("received message payload: %s\n", (char*)
NLMSG_DATA((struct nlmsghdr *) &buffer));
 59 }
 60
 61 int main(int argc, char *argv[]) {
...
 69     while (1) {
 70         nl_recv(nls);
 71         sleep(1);
 72         count++;
 73     }

If the call to sleep is removed, there is no issue. However, with the call,
on kernel 4.10 the kernel hangs after the kernel module is loaded. There's
no relevant information in /var/log/syslog after reboot.

It seems that some netlink buffer is getting filled, but in this case I
would expect nlmsg_multicast to return ENOBUFS.

Thanks,
Akshay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20170502/b5a7d3d7/attachment.html 


More information about the Kernelnewbies mailing list