[LTP] [PATCH] syscallls/sched_setscheduler: add sched_setscheduler03

[Chunyu Hu <chuhu@redhat.com>]

----- Original Message -----
> From: "Jan Stancek" <jstancek@redhat.com>
> To: "Chunyu Hu" <chuhu@redhat.com>
> Cc: ltp@lists.linux.it
> Sent: Friday, July 15, 2016 8:04:58 PM
> Subject: Re: [LTP] [PATCH] syscallls/sched_setscheduler: add sched_setscheduler03
> 
> 
> 
> 
> ----- Original Message -----
> > From: "Chunyu Hu" <chuhu@redhat.com>
> > To: ltp@lists.linux.it
> > Sent: Thursday, 14 July, 2016 11:25:29 AM
> > Subject: [LTP] [PATCH] syscallls/sched_setscheduler: add
> > sched_setscheduler03
> > +#define _GNU_SOURCE
> > +#include <stdio.h>
> > +#include <errno.h>
> > +#include <sched.h>
> > +#include <linux/capability.h>
> 
> Hi,
> 
> from doc/style-guide.txt:
> "Don't use +linux/+ headers if at all possible."

Ah, sure. thanks for this guide info..

> It does not compile on RHEL5.6 at the moment:
> sched_setscheduler03.c: In function ‘setup’:
> sched_setscheduler03.c:175: error: ‘_LINUX_CAPABILITY_VERSION_1’ undeclared
> (first use in this function)
> sched_setscheduler03.c:175: error: (Each undeclared identifier is reported
> only once
> sched_setscheduler03.c:175: error: for each function it appears in.)
> make: *** [sched_setscheduler03] Error 1
> 
> Are capabilities needed only to clear CAP_SYS_NICE?

Actually yes.

> Couldn't we just fork and seteuid to "nobody"?

Tried with method that fork and switch to nobody instead of calling the
cap* syscall. it looks good. 

2.6.18-238.el5
[root@intel-mahobay-02 sched_setscheduler]# ./sched_setscheduler03
tst_test.c:701: INFO: Timeout per run is 300s
sched_setscheduler03.c:79: INFO: rlimit rlim_cur=0
sched_setscheduler03.c:81: INFO: rlimit rlim_max=0
sched_setscheduler03.c:89: INFO: Setting rlim_cur to 19
sched_setscheduler03.c:91: INFO: Setting rlim_max to 19
sched_setscheduler03.c:79: INFO: rlimit rlim_cur=19
sched_setscheduler03.c:81: INFO: rlimit rlim_max=19
sched_setscheduler03.c:146: INFO: Setting init sched policy to SCHED_OTHER
sched_setscheduler03.c:154: INFO: Setting euid to nobody to drop privilege
sched_setscheduler03.c:172: INFO: forked pid is 11488
sched_setscheduler03.c:113: INFO: Verifying case[1]: policy = 0, priority = 0
sched_setscheduler03.c:122: PASS: case[1] succeeded
sched_setscheduler03.c:172: INFO: forked pid is 11489
sched_setscheduler03.c:113: INFO: Verifying case[2]: policy = 3, priority = 0
sched_setscheduler03.c:122: PASS: case[2] succeeded
sched_setscheduler03.c:164: INFO: cleanup() executed by pid 11487

Summary:
passed   2
failed   0
skipped  0
warnings 0



> > +#include <sys/time.h>
> > +#include <sys/resource.h>
> > +
> > +#include "tst_test.h"
> > +#include "linux_syscall_numbers.h"
> > +
> > +#define RLIMIT_NICE_NORMAL 20
> > +
> > +#ifndef SCHED_IDLE
> > +#define SCHED_IDLE 5
> > +#endif
> 
> This ifdef should rather disable testcase that needs it.
> Otherwise you'll get failure on older kernels:
> sched_setscheduler03.c:116: FAIL: unexpected error - 22 : Invalid argument -
> expected 0

My less consideration. I modify it to 
#ifdef SCHED_IDLE
{
 ...
}
#endif
And it can be skipped if not defined.


> # uname -r
> 2.6.18-238.el5
> 
> > +
> > +/* cap_user_header_t is a pointer to __user_cap_header_struct */
> > +static struct __user_cap_header_struct cap_header;
> > +
> > +/*
> > + * cap_user_data_t is a pointer to to __user_cap_data_struct.
> > + * there are three versions of header, on v2 and v3, the calls
> > + * need two data struct, here we use only v1 is enough, but
> > + * leave it as a two item array for deprecate consideration.
> > + */
> > +static struct __user_cap_data_struct cap_data[2];
> > +
> > +static struct rlimit limit;
> > +static pid_t zero_pid;
> > +static struct sched_param param[1] = {{0}};
> > +
> > +struct test_case_t {
> > +	pid_t *pid;
> > +	int policy;
> > +	struct sched_param *sched_param;
> > +	int error;
> > +};
> > +
> > +struct test_case_t cases[] = {
> > +	{
> > +		.pid = &zero_pid,
> > +		.policy = SCHED_OTHER,
> > +		.sched_param = &param[0]
> > +	},
> > +	{
> > +		.pid = &zero_pid,
> > +		.policy = SCHED_BATCH,
> > +		.sched_param = &param[0]
> > +	},
> > +	{
> > +		.pid = &zero_pid,
> > +		.policy = SCHED_IDLE,
> > +		.sched_param = &param[0]
> > +	}
> > +};
> > +
> <snip>
> > +
> > +static void verify_fn(unsigned int i)
> > +{
> > +
> 
> Here could be some message saying what test is going to run.

OK. add it in v2.

> 
> > +	TEST(sched_setscheduler(*cases[i].pid, cases[i].policy,
> > +					cases[i].sched_param));
> > +	if (TEST_RETURN)
> > +		tst_res(TFAIL, "unexpected error - %d : %s - "
> 
> TFAIL | TTERRNO
> 
> > +			 "expected %d", TEST_ERRNO,
> > +			 strerror(TEST_ERRNO), cases[i].error);
> > +	else
> > +		tst_res(TPASS, "Succeed - %d : %s - "
> > +			 "expected %d", TEST_ERRNO,
> > +			 strerror(TEST_ERRNO), cases[i].error);
> 
> No need to print errno if call succeeded and that was expected.

Remove this in V2. Thanks for suggestion.

> > +}
> > +
> > +static void setup(void)
> > +{
> > +
> > +	cap_header.version = _LINUX_CAPABILITY_VERSION_1;
> > +
> > +	l_rlimit_show(RLIMIT_NICE, &limit);
> > +
> > +	/*
> > +	* nice rlimit ranges from 1 to 40, mapping to real nice
> > +	* value from 19 to -20. We set it to 19, as the default priority
> > +	* of process with fair policy is 120, which will be translated
> > +	* into nice 20, we make this RLIMIT_NICE smaller than that, to
> > +	* verify the can_nice usage issue.
> > +	*/
> > +	limit.rlim_cur = (RLIMIT_NICE_NORMAL - 1);
> > +	limit.rlim_max = (RLIMIT_NICE_NORMAL - 1);
> > +
> > +	l_rlimit_setup(RLIMIT_NICE, &limit);
> > +
> > +	tst_res(TINFO, "Setting init sched policy to SCHED_OTHER");
> > +	if (sched_setscheduler(0, SCHED_OTHER, &param[0]) != 0)
> > +		tst_brk(TBROK | TERRNO,
> > +			 "ERROR sched_setscheduler: (0, SCHED_OTHER, param)");
> > +
> > +	if (sched_getscheduler(0) != SCHED_OTHER)
> > +		tst_brk(TBROK | TERRNO, "ERROR sched_tetscheduler");
> > +
> > +	l_cap_show(&cap_header, &cap_data[0]);
> > +
> > +	cap_data[0].effective &= (~(1 << CAP_SYS_NICE));
> > +	cap_data[0].permitted &= (~(1 << CAP_SYS_NICE));
> > +	cap_data[0].inheritable = 1;
> > +
> > +	l_cap_setup(&cap_header, &cap_data[0]);
> > +}
> > +
> > +static void cleanup(void)
> > +{
> > +	tst_res(TINFO, "cleanup() executed by pid %i", getpid());
> > +}
> > +
> > +static void do_test(unsigned int i)
> > +{
> > +	verify_fn(i);
> > +}
> > +
> > +static struct tst_test test = {
> > +	.tid = "sched_setscheduler03",
> > +	.tcnt = 3,
> 
> ARRAY_SIZE(cases)
> 
> Also needs_root seems to be missing here.

Modify this in v2, thanks.


> Regards,
> Jan
> 
> > +	.test = do_test,
> > +	.setup = setup,
> > +	.cleanup = cleanup,
> > +	.forks_child = 1,
> > +};
> > +
> > --
> > 1.8.3.1
> > 
> > 
> > --
> > Mailing list info: https://lists.linux.it/listinfo/ltp
> > 
> 

-- 
Regards,
Chunyu Hu