Re: [patch iproute2 v5 1/3] lib/libnetlink: Add a function rtnl_talk_msg

[Chris Mi <chrism@mellanox.com>]

2018/1/3 12:08, David Ahern:
> On 1/2/18 7:55 PM, Chris Mi wrote:
>> diff --git a/lib/libnetlink.c b/lib/libnetlink.c
>> index 00e6ce0c..cc02a139 100644
>> --- a/lib/libnetlink.c
>> +++ b/lib/libnetlink.c
>> @@ -581,32 +581,34 @@ static void rtnl_talk_error(struct nlmsghdr *h, struct nlmsgerr *err,
>>   		strerror(-err->error));
>>   }
>>   
>> -static int __rtnl_talk(struct rtnl_handle *rtnl, struct nlmsghdr *n,
>> -		       struct nlmsghdr **answer,
>> -		       bool show_rtnl_err, nl_ext_ack_fn_t errfn)
>> +static int __rtnl_talk_msg(struct rtnl_handle *rtnl, struct msghdr *m,
>> +			   struct nlmsghdr **answer,
>> +			   bool show_rtnl_err, nl_ext_ack_fn_t errfn)
>>   {
>> -	int status;
>> -	unsigned int seq;
>> -	struct nlmsghdr *h;
>> +	int iovlen = m->msg_iovlen;
>> +	unsigned int seq = 0;
>> +	int i, status;
>> +	char *buf;
>> +
>>   	struct sockaddr_nl nladdr = { .nl_family = AF_NETLINK };
>> -	struct iovec iov = {
>> -		.iov_base = n,
>> -		.iov_len = n->nlmsg_len
>> -	};
>> +	struct iovec iov, *v;
>> +	struct nlmsghdr *h;
>>   	struct msghdr msg = {
>>   		.msg_name = &nladdr,
>>   		.msg_namelen = sizeof(nladdr),
>>   		.msg_iov = &iov,
>>   		.msg_iovlen = 1,
>>   	};
>> -	char *buf;
> Reverse xmas tree is the coding standard for net code. Please adhere to
> it. Only dependencies between variables are an acceptable exception.
OK, got it.
>
> Some of those (struct nlmsghdr *h and struct iovec *v) can be moved to
> the for loop which aligns with your intentions of grouping variables.
Done.
>
>>   
>> -	n->nlmsg_seq = seq = ++rtnl->seq;
>> -
>> -	if (answer == NULL)
>> -		n->nlmsg_flags |= NLM_F_ACK;
>> +	for (i = 0; i < iovlen; i++) {
>> +		v = &m->msg_iov[i];
>> +		h = v->iov_base;
>> +		h->nlmsg_seq = seq = ++rtnl->seq;
> doesn't seq need to track the recvmsg loop? I think for batching you
> want it to start at the first seq number and then in the recvmsg loop
> increment it.
Yes, it is a bug. Thanks for your test case.
>
> As it stands this file:
> $ cat tc.batch
> filter add dev eth2 ingress protocol ip pref 21 flower dst_ip
> 192.168.1.0/16 action drop
> filter add dev eth2 ingress protocol ip pref 22 flower dst_ip
> 192.168.2.0/16 action drop
> filter add dev eth2 ingress protocol ip pref 22 flower dst_ip
> 192.168.3.0/16 action drop
> filter add dev eth2 ingress protocol ip pref 24 flower dst_ip
> 192.168.4.0/16 action drop
> filter add dev eth2 ingress protocol ip pref 25 flower dst_ip
> 192.168.5.0/16 action drop
>
> does not give me an error message:
> $ tc -b tc.batch -bs 5
> <no output>
>
> Yet it failed to insert all filters:
> $ tc filter show dev eth2 ingress
> filter protocol ip pref 21 flower chain 0
> filter protocol ip pref 21 flower chain 0 handle 0x1
>    eth_type ipv4
>    dst_ip 192.168.1.0/16
>    not_in_hw
> 	action order 1: gact action drop
> 	 random type none pass val 0
> 	 index 1 ref 1 bind 1
>
> filter protocol ip pref 22 flower chain 0
> filter protocol ip pref 22 flower chain 0 handle 0x1
>    eth_type ipv4
>    dst_ip 192.168.2.0/16
>    not_in_hw
> 	action order 1: gact action drop
> 	 random type none pass val 0
> 	 index 2 ref 1 bind 1
>
> filter protocol ip pref 24 flower chain 0
> filter protocol ip pref 24 flower chain 0 handle 0x1
>    eth_type ipv4
>    dst_ip 192.168.4.0/16
>    not_in_hw
> 	action order 1: gact action drop
> 	 random type none pass val 0
> 	 index 3 ref 1 bind 1
>
> filter protocol ip pref 25 flower chain 0
> filter protocol ip pref 25 flower chain 0 handle 0x1
>    eth_type ipv4
>    dst_ip 192.168.5.0/16
>    not_in_hw
> 	action order 1: gact action drop
> 	 random type none pass val 0
> 	 index 4 ref 1 bind 1
>
After fixing it, the test result is:

# tc -b tc.batch -bs 5
RTNETLINK answers: File exists
We have an error talking to the kernel, -1
Command failed 1.txt:0-4

We can't tell exactly which command causes this error, so we give a 
range which is less than the batch size.