netlink_unicast 返回 -11的问题

jinxinxin163

我在一个驱动的write函数中调用netlink_unicast发送数据给应用层程序。
刚开始的时候，是ok的，但过了一会儿之后，netlink_unicast失败，
返回值为  -11。
在include/asm-generic/errno-base.h文件中查到,11表示“try again”
大家有碰到过这个问题么？

emmoblin

你发向用户态的应用程序填写了pid吗？还是直接填0，进行广播

emmoblin

是不是用户太的进程死掉了？

jinxinxin163

回复 3# emmoblin


    没有死，刚开始还好好的，其实就是不停得发数据，但是刚开始是ok的，发了一会就发不出去了

emmoblin

看看是不是需要设置socket的buffer大小。是否每次都发送同样多的数据后就不行了呢？

jinxinxin163

回复 5# emmoblin


count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
send_to_user: have sent=1040
advtty_write: 1024
count = 2048
send_to_user: have sent=1040
advtty_write: 1024
count = 1024
advtty:can not unicast skb(-11)
count = 2048
advtty:can not unicast skb(-11)
count = 2048
advtty:can not unicast skb(-11)

刚开始发送是ok的，但后来就发不出去了

jinxinxin163

我在
http://lists.netfilter.org/piper ... ptember/009317.html
看到以前有个人也碰到了和我一样的问题，但是他没有解决。

I have a problem when using netlink_unicast(...) in
net/netlink/af_netlink.c. When calling netlink_unicast(...) multiple times,
e.g. in a loop, the internal buffer in the socket will get filled up, then
it will refuse to send any more netlink msg and return -EAGAIN.

Is there anyway to force netlink socket to send/flush whatever in the buffer
thus leave more room for the next netlink_unicast call?

Any advice is appreciated. thank you very much.

jinxinxin163

问题的原因找到了
其实就是netlink_unicast发送的比接收端快很多，以至于"the internal buffer in the socket will get filled up"。
我尝试在驱动的write函数里在调用netlink_unicast前msleep(100),就正常了。

jinxinxin163

http://linux.chinaunix.net/bbs/thread-1153854-1-1.html

nehcc

回复 8# jinxinxin163


    你前面有提到 netlink 发给用户层的数据非常多
    然后 后面 却又 sleep(100) , 这样会不会延迟、堵塞 ？