[LTP] [PATCH] syscalls/sendto02.c: add new testcase

Xiao Yang yangx.jy@cn.fujitsu.com
Wed Nov 2 04:35:06 CET 2016


Hi cyril

I find that system would load sctp module when calling scoket(..., 
IPPROTO_SCTP).
If sctp module is not supported by system, it returns EPROTONOSUPPORT.
I will remove loading/unloading code.

Thanks,
Xiao Yang

On 2016/11/02 0:20, Cyril Hrubis wrote:
> Hi!
>> +#include<errno.h>
>> +#include<sys/types.h>
>> +#include<sys/socket.h>
>> +
>> +#include "tst_test.h"
>> +
>> +static int sockfd;
>> +static int rds_flag;
>> +static struct sockaddr_in sa;
>> +
>> +static void setup(void)
>> +{
>> +	int acc_res, load_res;
>> +	const char *cmd[] = {"modprobe", "sctp", NULL};
>> +
>> +	acc_res = access("/proc/sys/net/sctp", F_OK);
>> +	if (acc_res == -1&&  errno != ENOENT)
>> +		tst_brk(TFAIL | TERRNO, "failed to check stcp module");
>                                           ^
> I would make the error message as specific as possible here, so I would
> print what exactly has failed:
>
> tst_brk(TBROK | TERRNO, "access(/proc/sys/net/sctp, F_OK)");
>
>> +	if (acc_res == -1&&  errno == ENOENT) {
>> +		load_res = tst_run_cmd(cmd, NULL, NULL, 1);
>> +		if (load_res) {
>> +			tst_brk(TCONF, "failed to loaded sctp module, "
>                                                      ^
> 						    load
>> +				"so sctp modeule was not support by system");
>                                             ^
> 					  module
>
> Also I would say that all that needs to be printed here is the first
> part of the string, i.e. "failed to load sctp module".
>
>> +		}
>> +
>> +		tst_res(TINFO, "succeeded to load sctp module");
>> +		rds_flag = 1;
>                   ^
> 		 It would be a bit better to name it
> 		 rds_module_loaded or just module_loaded
>> +	}
>> +
>> +	tst_res(TINFO, "sctp module was supported by system");
> Just omit this message, it's misleading anyway.
>
>> +	sockfd = SAFE_SOCKET(AF_INET, SOCK_STREAM, 132 /*IPPROTO_SCTP*/);
>                                                      ^
> 						    This is way too ugly.
>
> If IPPROTO_SCTP is not supported on you system you should add a fallback
> definition such as:
>
> #ifndef IPPROTO_SCTP
> # define IPPROTO_SCTP 132
> #endif
>
> Also you didn't even try to include the<netinet/in.h>  header that
> contains this definition, so it's quite likely that all you need to do
> is to include this header as well.
>
>> +	memset(&sa, 0, sizeof(sa));
>> +	sa.sin_family = AF_INET;
>> +	sa.sin_addr.s_addr = inet_addr("127.0.0.1");
>> +	sa.sin_port = htons(11111);
>> +}
>> +
>> +static void cleanup(void)
>> +{
>> +	int unload_res;
>> +	const char *cmd[] = {"modprobe", "-r", "sctp", NULL};
>> +
>> +	if (rds_flag == 1) {
>> +		unload_res = tst_run_cmd(cmd, NULL, NULL, 1);
>> +		if (unload_res) {
>> +			/* We failed to unload sctp module(modprobe -r sctp)
>> +			 * and the operation failed with "FATAL: Module sctp is
>> +			 * in use." lsmod shows a reference count of 2 for sctp.
>> +			 * If we rebuild kernel with CONFIG_MODULE_FORCE_UNLOAD
>> +			 * enabled, we can succeed to unload sctp module
>> +			 * (rmmod -f sctp).
>> +			 */
>> +			tst_res(TINFO, "failed to unload sctp module, you can"
>> +				" reboot system to unload sctp after testing");
> Isn't the problem here that you have to close the sctp socket *berofe*
> you try to remove the module?
>
> It's pretty clear that the module would be in use at least until the
> socked is closed.
>
>> +		} else {
>> +			tst_res(TINFO, "succeeded to unload sctp modules");
>> +		}
>> +	}
>> +
>> +	if (sockfd>  0&&  close(sockfd))
>> +		tst_res(TWARN | TERRNO, "failed to close file");
>> +}
>> +
>> +static void verify_sendto(void)
>> +{
>> +	TEST(sendto(sockfd, NULL, 1, 0, (struct sockaddr *)&sa, sizeof(sa)));
>> +	if (TEST_RETURN != -1) {
>> +		tst_res(TFAIL, "sendto() succeeded unexpectedly");
>> +		return;
>> +	}
>> +
>> +	if (TEST_ERRNO == ENOMEM) {
>> +		tst_res(TFAIL | TTERRNO, "sendto() got unrepaired errno with"
>> +			" invalid buffer when sctp is selected in socket()");
> This message is absurdly long as well. Be short and to the point.
>
> Something as:
>
> 	tst_res(TFAIL | TTERRNO, "sendto(fd, NULL, ...) failed unexpectedly");
>
>> +		return;
>> +	}
>> +
>> +	if (TEST_ERRNO == EFAULT) {
>> +		tst_res(TPASS | TTERRNO, "sendto() got repaired errno with"
>> +			" invalid buffer when sctp is selected in socket()");
> Here as well. Shorten the message to something more reasonable.
>
>> +		return;
>> +	}
>> +
>> +	tst_res(TFAIL | TTERRNO, "sendto() got unexpected errno with invalid"
>> +		" buffer when sctp is selected in socket(), expected ENOMEM "
>> +		"or EFAULT");
> Why do we have special case for errno != ENOMEM&&  errno != EFAULT. We
> fail the testcase if the errno is not EFAULT anyway. There is no reason
> to complicate the code like this.
>
>> +}
>> +
>> +static struct tst_test test = {
>> +	.tid = "sendto02",
>> +	.setup = setup,
>> +	.cleanup = cleanup,
>> +	.test_all = verify_sendto,
>> +};
>> -- 
>> 1.8.3.1
>>
>>
>>
>>
>> -- 
>> Mailing list info: https://lists.linux.it/listinfo/ltp





More information about the ltp mailing list