xref: /netbsd-src/sys/kern/sys_sched.c (revision 2de962bd804263c16657f586aa00f1704045df8e)
1 /*	$NetBSD: sys_sched.c,v 1.21 2008/04/24 18:39:24 ad Exp $	*/
2 
3 /*
4  * Copyright (c) 2008, Mindaugas Rasiukevicius <rmind at NetBSD org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * System calls relating to the scheduler.
31  *
32  * TODO:
33  *  - Handle pthread_setschedprio() as defined by POSIX;
34  *  - Handle sched_yield() case for SCHED_FIFO as defined by POSIX;
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: sys_sched.c,v 1.21 2008/04/24 18:39:24 ad Exp $");
39 
40 #include <sys/param.h>
41 
42 #include <sys/cpu.h>
43 #include <sys/kauth.h>
44 #include <sys/kmem.h>
45 #include <sys/lwp.h>
46 #include <sys/mutex.h>
47 #include <sys/proc.h>
48 #include <sys/pset.h>
49 #include <sys/sched.h>
50 #include <sys/syscallargs.h>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
53 #include <sys/types.h>
54 #include <sys/unistd.h>
55 
56 /*
57  * Convert user priority or the in-kernel priority or convert the current
58  * priority to the appropriate range according to the policy change.
59  */
60 static pri_t
61 convert_pri(lwp_t *l, int policy, pri_t pri)
62 {
63 	int delta = 0;
64 
65 	switch (policy) {
66 	case SCHED_OTHER:
67 		delta = PRI_USER;
68 		break;
69 	case SCHED_FIFO:
70 	case SCHED_RR:
71 		delta = PRI_USER_RT;
72 		break;
73 	default:
74 		panic("upri_to_kpri");
75 	}
76 
77 	if (pri != PRI_NONE) {
78 		/* Convert user priority to the in-kernel */
79 		KASSERT(pri >= SCHED_PRI_MIN && pri <= SCHED_PRI_MAX);
80 		return pri + delta;
81 	}
82 	if (l->l_class == policy)
83 		return l->l_priority;
84 
85 	/* Change the current priority to the appropriate range */
86 	if (l->l_class == SCHED_OTHER) {
87 		KASSERT(policy == SCHED_FIFO || policy == SCHED_RR);
88 		return l->l_priority + delta;
89 	}
90 	if (policy == SCHED_OTHER) {
91 		KASSERT(l->l_class == SCHED_FIFO || l->l_class == SCHED_RR);
92 		return l->l_priority - delta;
93 	}
94 	KASSERT(l->l_class != SCHED_OTHER && policy != SCHED_OTHER);
95 	return l->l_class;
96 }
97 
98 int
99 do_sched_setparam(pid_t pid, lwpid_t lid, int policy,
100     const struct sched_param *params)
101 {
102 	struct proc *p;
103 	struct lwp *t;
104 	pri_t pri;
105 	u_int lcnt;
106 	int error;
107 
108 	error = 0;
109 
110 	pri = params->sched_priority;
111 
112 	/* If no parameters specified, just return (this should not happen) */
113 	if (pri == PRI_NONE && policy == SCHED_NONE)
114 		return 0;
115 
116 	/* Validate scheduling class */
117 	if (policy != SCHED_NONE && (policy < SCHED_OTHER || policy > SCHED_RR))
118 		return EINVAL;
119 
120 	/* Validate priority */
121 	if (pri != PRI_NONE && (pri < SCHED_PRI_MIN || pri > SCHED_PRI_MAX))
122 		return EINVAL;
123 
124 	if (pid != 0) {
125 		/* Find the process */
126 		mutex_enter(proc_lock);
127 		p = p_find(pid, PFIND_LOCKED);
128 		if (p == NULL) {
129 			mutex_exit(proc_lock);
130 			return ESRCH;
131 		}
132 		mutex_enter(p->p_lock);
133 		mutex_exit(proc_lock);
134 		/* Disallow modification of system processes */
135 		if ((p->p_flag & PK_SYSTEM) != 0) {
136 			mutex_exit(p->p_lock);
137 			return EPERM;
138 		}
139 	} else {
140 		/* Use the calling process */
141 		p = curlwp->l_proc;
142 		mutex_enter(p->p_lock);
143 	}
144 
145 	/* Find the LWP(s) */
146 	lcnt = 0;
147 	LIST_FOREACH(t, &p->p_lwps, l_sibling) {
148 		pri_t kpri;
149 		int lpolicy;
150 
151 		if (lid && lid != t->l_lid)
152 			continue;
153 		lcnt++;
154 		KASSERT(pri != PRI_NONE || policy != SCHED_NONE);
155 		lwp_lock(t);
156 
157 		if (policy == SCHED_NONE)
158 			lpolicy = t->l_class;
159 		else
160 			lpolicy = policy;
161 
162 		/*
163 		 * Note that, priority may need to be changed to get into
164 		 * the correct priority range of the new scheduling class.
165 		 */
166 		kpri = convert_pri(t, lpolicy, pri);
167 
168 		/* Check the permission */
169 		error = kauth_authorize_process(kauth_cred_get(),
170 		    KAUTH_PROCESS_SCHEDULER_SETPARAM, p, t, KAUTH_ARG(lpolicy),
171 		    KAUTH_ARG(kpri));
172 		if (error) {
173 			lwp_unlock(t);
174 			break;
175 		}
176 
177 		/* Set the scheduling class */
178 		if (policy != SCHED_NONE)
179 			t->l_class = policy;
180 
181 		/* Change the priority */
182 		if (t->l_priority != kpri)
183 			lwp_changepri(t, kpri);
184 
185 		lwp_unlock(t);
186 	}
187 	mutex_exit(p->p_lock);
188 	return (lcnt == 0) ? ESRCH : error;
189 }
190 
191 /*
192  * Set scheduling parameters.
193  */
194 int
195 sys__sched_setparam(struct lwp *l, const struct sys__sched_setparam_args *uap,
196     register_t *retval)
197 {
198 	/* {
199 		syscallarg(pid_t) pid;
200 		syscallarg(lwpid_t) lid;
201 		syscallarg(int) policy;
202 		syscallarg(const struct sched_param *) params;
203 	} */
204 	struct sched_param params;
205 	int error;
206 
207 	/* Get the parameters from the user-space */
208 	error = copyin(SCARG(uap, params), &params, sizeof(params));
209 	if (error)
210 		goto out;
211 
212 	error = do_sched_setparam(SCARG(uap, pid), SCARG(uap, lid),
213 	    SCARG(uap, policy), &params);
214 
215  out:
216 	return (error);
217 }
218 
219 int
220 do_sched_getparam(pid_t pid, lwpid_t lid, int *policy,
221     struct sched_param *params)
222 {
223 	struct sched_param lparams;
224 	struct lwp *t;
225 	int error, lpolicy;
226 
227 	/* Locks the LWP */
228 	t = lwp_find2(pid, lid);
229 	if (t == NULL)
230 		return ESRCH;
231 
232 	/* Check the permission */
233 	error = kauth_authorize_process(kauth_cred_get(),
234 	    KAUTH_PROCESS_SCHEDULER_GETPARAM, t->l_proc, NULL, NULL, NULL);
235 	if (error != 0) {
236 		mutex_exit(t->l_proc->p_lock);
237 		return error;
238 	}
239 
240 	lwp_lock(t);
241 	lparams.sched_priority = t->l_priority;
242 	lpolicy = t->l_class;
243 
244 	switch (lpolicy) {
245 	case SCHED_OTHER:
246 		lparams.sched_priority -= PRI_USER;
247 		break;
248 	case SCHED_RR:
249 	case SCHED_FIFO:
250 		lparams.sched_priority -= PRI_USER_RT;
251 		break;
252 	}
253 
254 	if (policy != NULL)
255 		*policy = lpolicy;
256 
257 	if (params != NULL)
258 		*params = lparams;
259 
260 	lwp_unlock(t);
261 	mutex_exit(t->l_proc->p_lock);
262 	return error;
263 }
264 
265 /*
266  * Get scheduling parameters.
267  */
268 int
269 sys__sched_getparam(struct lwp *l, const struct sys__sched_getparam_args *uap,
270     register_t *retval)
271 {
272 	/* {
273 		syscallarg(pid_t) pid;
274 		syscallarg(lwpid_t) lid;
275 		syscallarg(int *) policy;
276 		syscallarg(struct sched_param *) params;
277 	} */
278 	struct sched_param params;
279 	int error, policy;
280 
281 	error = do_sched_getparam(SCARG(uap, pid), SCARG(uap, lid), &policy,
282 	    &params);
283 	if (error)
284 		goto out;
285 
286 	error = copyout(&params, SCARG(uap, params), sizeof(params));
287 	if (error == 0 && SCARG(uap, policy) != NULL)
288 		error = copyout(&policy, SCARG(uap, policy), sizeof(int));
289 
290  out:
291 	return (error);
292 }
293 
294 /*
295  * Set affinity.
296  */
297 int
298 sys__sched_setaffinity(struct lwp *l,
299     const struct sys__sched_setaffinity_args *uap, register_t *retval)
300 {
301 	/* {
302 		syscallarg(pid_t) pid;
303 		syscallarg(lwpid_t) lid;
304 		syscallarg(size_t) size;
305 		syscallarg(void *) cpuset;
306 	} */
307 	cpuset_t *cpuset;
308 	struct cpu_info *ci = NULL;
309 	struct proc *p;
310 	struct lwp *t;
311 	CPU_INFO_ITERATOR cii;
312 	lwpid_t lid;
313 	u_int lcnt;
314 	int error;
315 
316 	/* Allocate the CPU set, and get it from userspace */
317 	cpuset = kmem_zalloc(sizeof(cpuset_t), KM_SLEEP);
318 	error = copyin(SCARG(uap, cpuset), cpuset,
319 	    min(SCARG(uap, size), sizeof(cpuset_t)));
320 	if (error)
321 		goto error;
322 
323 	/* Look for a CPU in the set */
324 	for (CPU_INFO_FOREACH(cii, ci))
325 		if (CPU_ISSET(cpu_index(ci), cpuset))
326 			break;
327 	if (ci == NULL) {
328 		/* Empty set */
329 		kmem_free(cpuset, sizeof(cpuset_t));
330 		cpuset = NULL;
331 	}
332 
333 	if (SCARG(uap, pid) != 0) {
334 		/* Find the process */
335 		mutex_enter(proc_lock);
336 		p = p_find(SCARG(uap, pid), PFIND_LOCKED);
337 		if (p == NULL) {
338 			mutex_exit(proc_lock);
339 			error = ESRCH;
340 			goto error;
341 		}
342 		mutex_enter(p->p_lock);
343 		mutex_exit(proc_lock);
344 		/* Disallow modification of system processes. */
345 		if ((p->p_flag & PK_SYSTEM) != 0) {
346 			mutex_exit(p->p_lock);
347 			error = EPERM;
348 			goto error;
349 		}
350 	} else {
351 		/* Use the calling process */
352 		p = l->l_proc;
353 		mutex_enter(p->p_lock);
354 	}
355 
356 	/*
357 	 * Check the permission.
358 	 */
359 	error = kauth_authorize_process(l->l_cred,
360 	    KAUTH_PROCESS_SCHEDULER_SETAFFINITY, p, NULL, NULL, NULL);
361 	if (error != 0) {
362 		mutex_exit(p->p_lock);
363 		goto error;
364 	}
365 
366 	/* Find the LWP(s) */
367 	lcnt = 0;
368 	lid = SCARG(uap, lid);
369 	LIST_FOREACH(t, &p->p_lwps, l_sibling) {
370 		if (lid && lid != t->l_lid)
371 			continue;
372 		lwp_lock(t);
373 		if (cpuset) {
374 			/* Set the affinity flag and new CPU set */
375 			t->l_flag |= LW_AFFINITY;
376 			memcpy(&t->l_affinity, cpuset, sizeof(cpuset_t));
377 			/* Migrate to another CPU, unlocks LWP */
378 			lwp_migrate(t, ci);
379 		} else {
380 			/* Unset the affinity flag */
381 			t->l_flag &= ~LW_AFFINITY;
382 			lwp_unlock(t);
383 		}
384 		lcnt++;
385 	}
386 	mutex_exit(p->p_lock);
387 	if (lcnt == 0)
388 		error = ESRCH;
389 error:
390 	if (cpuset != NULL)
391 		kmem_free(cpuset, sizeof(cpuset_t));
392 	return error;
393 }
394 
395 /*
396  * Get affinity.
397  */
398 int
399 sys__sched_getaffinity(struct lwp *l,
400     const struct sys__sched_getaffinity_args *uap, register_t *retval)
401 {
402 	/* {
403 		syscallarg(pid_t) pid;
404 		syscallarg(lwpid_t) lid;
405 		syscallarg(size_t) size;
406 		syscallarg(void *) cpuset;
407 	} */
408 	struct lwp *t;
409 	void *cpuset;
410 	int error;
411 
412 	if (SCARG(uap, size) <= 0)
413 		return EINVAL;
414 	cpuset = kmem_zalloc(sizeof(cpuset_t), KM_SLEEP);
415 
416 	/* Locks the LWP */
417 	t = lwp_find2(SCARG(uap, pid), SCARG(uap, lid));
418 	if (t == NULL) {
419 		kmem_free(cpuset, sizeof(cpuset_t));
420 		return ESRCH;
421 	}
422 	/* Check the permission */
423 	if (kauth_authorize_process(l->l_cred,
424 	    KAUTH_PROCESS_SCHEDULER_GETAFFINITY, t->l_proc, NULL, NULL, NULL)) {
425 		mutex_exit(t->l_proc->p_lock);
426 		kmem_free(cpuset, sizeof(cpuset_t));
427 		return EPERM;
428 	}
429 	lwp_lock(t);
430 	if (t->l_flag & LW_AFFINITY)
431 		memcpy(cpuset, &t->l_affinity, sizeof(cpuset_t));
432 	lwp_unlock(t);
433 	mutex_exit(t->l_proc->p_lock);
434 
435 	error = copyout(cpuset, SCARG(uap, cpuset),
436 	    min(SCARG(uap, size), sizeof(cpuset_t)));
437 
438 	kmem_free(cpuset, sizeof(cpuset_t));
439 	return error;
440 }
441 
442 /*
443  * Yield.
444  */
445 int
446 sys_sched_yield(struct lwp *l, const void *v, register_t *retval)
447 {
448 
449 	yield();
450 	return 0;
451 }
452 
453 /*
454  * Sysctl nodes and initialization.
455  */
456 SYSCTL_SETUP(sysctl_sched_setup, "sysctl sched setup")
457 {
458 	const struct sysctlnode *node = NULL;
459 
460 	sysctl_createv(clog, 0, NULL, NULL,
461 		CTLFLAG_PERMANENT,
462 		CTLTYPE_NODE, "kern", NULL,
463 		NULL, 0, NULL, 0,
464 		CTL_KERN, CTL_EOL);
465 	sysctl_createv(clog, 0, NULL, NULL,
466 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
467 		CTLTYPE_INT, "posix_sched",
468 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
469 			     "Process Scheduling option to which the "
470 			     "system attempts to conform"),
471 		NULL, _POSIX_PRIORITY_SCHEDULING, NULL, 0,
472 		CTL_KERN, CTL_CREATE, CTL_EOL);
473 	sysctl_createv(clog, 0, NULL, &node,
474 		CTLFLAG_PERMANENT,
475 		CTLTYPE_NODE, "sched",
476 		SYSCTL_DESCR("Scheduler options"),
477 		NULL, 0, NULL, 0,
478 		CTL_KERN, CTL_CREATE, CTL_EOL);
479 
480 	if (node == NULL)
481 		return;
482 
483 	sysctl_createv(clog, 0, &node, NULL,
484 		CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
485 		CTLTYPE_INT, "pri_min",
486 		SYSCTL_DESCR("Minimal POSIX real-time priority"),
487 		NULL, SCHED_PRI_MIN, NULL, 0,
488 		CTL_CREATE, CTL_EOL);
489 	sysctl_createv(clog, 0, &node, NULL,
490 		CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
491 		CTLTYPE_INT, "pri_max",
492 		SYSCTL_DESCR("Maximal POSIX real-time priority"),
493 		NULL, SCHED_PRI_MAX, NULL, 0,
494 		CTL_CREATE, CTL_EOL);
495 }
496