Skip to contentSkip to conversation

Socket hangs in read() when putting interface down

DISCUSSION chris schlund 2 participants 3 messages Jun 5, 2007 – Jun 6, 2007

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Using read() on a socket(AF_INET, SOCK_STREAM) connected via any network interface (e.g eth0) and putting this interface down (ifconfig) while the socket is in read() produces a hanger (the read() will not return anymore).
Even if I install sigaction signal handler I got no signal when the interface is removed.

This may be ok on a major interface like eth0 but it is a problem when using a ppp interface via GSM data or GPRS connection. Killing the pppd while the socket waits in read() produces the same behaviour. Also a broken connection looks the same.
It is reproducable on several 2.6 kernel versions including the lastest stable 2.6.21.

On my understanding "ifconfig down" should lead into a read() error.

Unfortunately I am not very familiar with in-depth kernel programming and I am still wondering about the magic inside the kernel:

ifconfig -> net/dev.c:dev_close() notifies a call chain which probably could/should handle this.
But I miss the dependency to net/ipv4/tcp.c: tcp_recvmsg() function. This is where I think my receive is blocking.

Can someone give me a hint where to look?

Thanks and kind regards
Chris

p.s.
This happens only, when socket is in blocking mode. I am using a simple example like:
s = socket();
connect(s);
while(){
read(s)
}

#1 lore raw
Quoted 6 lines · chris schlund

Using read() on a socket(AF_INET, SOCK_STREAM) connected via any network interface (e.g eth0) and putting this interface down (ifconfig) while the socket is in read() produces a hanger (the read() will not return anymore).
Even if I install sigaction signal handler I got no signal when the interface is removed.

I'd not think there was a direct tie-in between an application signal handler and an interface being downed.

Quoted 8 lines

This may be ok on a major interface like eth0 but it is a problem when using a ppp interface via GSM data or GPRS connection. Killing the pppd while the socket waits in read() produces the same behaviour. Also a broken connection looks the same.
It is reproducable on several 2.6 kernel versions including the lastest stable 2.6.21.

On my understanding "ifconfig down" should lead into a read() error.

Why? Can't traffic for the endpoint arrive via some other interface on the system? I would think that getting out of a blocking read() requires the application to either have a timer running (and so get out of read() with EINTR) or to have keepalives enabled on the TCP connection (I'm assuming TCP for an AF_INET, SOCK_STREAM although I suppose it could be SCTP)

rick jones

#2 lore raw
Quoted 9 lines
Quoted 8 lines

This may be ok on a major interface like eth0 but it is a problem when using a ppp interface via GSM data or GPRS connection. Killing the pppd while the socket waits in read() produces the same behaviour. Also a broken connection looks the same.
It is reproducable on several 2.6 kernel versions including the lastest stable 2.6.21.

On my understanding "ifconfig down" should lead into a read() error.

Quoted 8 lines

Why? Can't traffic for the endpoint arrive via some other interface on the system? I would think that getting out of a blocking read() requires the application to either have a timer running (and so get out of read() with EINTR) or to have keepalives enabled on the TCP connection (I'm assuming TCP for an AF_INET, SOCK_STREAM although I suppose it could be SCTP)

rick jones

Of course, you are right.
I am so focused to my small embedded world, that I simply have forgotten how IP-traffic works in big systems (to clarify my first post: I'm using TCP).

Your hint with the self produced signal to get out of read() was very helpful for me. Thats what I was missing.

I also learned another think, while reading the list FAQ:
PEBCAC (=Problem exists between chair and computer) Obviously, thats what best described me!

Thank you very much!

Chris Schlund