xref: /freebsd/sys/kern/kern_thr.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
1 /*-
2  * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_posix.h"
31 #include <sys/param.h>
32 #include <sys/kernel.h>
33 #include <sys/lock.h>
34 #include <sys/mutex.h>
35 #include <sys/proc.h>
36 #include <sys/resourcevar.h>
37 #include <sys/sched.h>
38 #include <sys/sysctl.h>
39 #include <sys/smp.h>
40 #include <sys/sysent.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/signalvar.h>
44 #include <sys/ucontext.h>
45 #include <sys/thr.h>
46 #include <sys/rtprio.h>
47 #include <posix4/sched.h>
48 #include <posix4/posix4.h>
49 #include <sys/umtx.h>
50 #include <sys/limits.h>
51 
52 #include <machine/frame.h>
53 
54 extern int max_threads_per_proc;
55 
56 static int create_thread(struct thread *td, mcontext_t *ctx,
57 			 void (*start_func)(void *), void *arg,
58 			 char *stack_base, size_t stack_size,
59 			 char *tls_base,
60 			 long *child_tid, long *parent_tid,
61 			 int flags, struct thr_sched_param *sched);
62 
63 /*
64  * System call interface.
65  */
66 int
67 thr_create(struct thread *td, struct thr_create_args *uap)
68     /* ucontext_t *ctx, long *id, int flags */
69 {
70 	ucontext_t ctx;
71 	int error;
72 
73 	if ((error = copyin(uap->ctx, &ctx, sizeof(ctx))))
74 		return (error);
75 
76 	error = create_thread(td, &ctx.uc_mcontext, NULL, NULL,
77 		NULL, 0, NULL, uap->id, NULL, uap->flags, NULL);
78 	return (error);
79 }
80 
81 int
82 thr_new(struct thread *td, struct thr_new_args *uap)
83     /* struct thr_param * */
84 {
85 	struct thr_param param;
86 	struct thr_sched_param sched_param, *sched;
87 	int error;
88 
89 	if (uap->param_size < sizeof(param))
90 		return (EINVAL);
91 	if ((error = copyin(uap->param, &param, sizeof(param))))
92 		return (error);
93 	sched = NULL;
94 	if (param.sched_param != NULL) {
95 		if (param.sched_param_size != sizeof(struct thr_sched_param))
96 			return (EINVAL);
97 
98 		error = copyin(param.sched_param, &sched_param,
99 			sizeof(sched_param));
100 		if (error)
101 			return (error);
102 		sched = &sched_param;
103 	}
104 
105 	error = create_thread(td, NULL, param.start_func, param.arg,
106 		param.stack_base, param.stack_size, param.tls_base,
107 		param.child_tid, param.parent_tid, param.flags,
108 		sched);
109 	return (error);
110 }
111 
112 static int
113 create_thread(struct thread *td, mcontext_t *ctx,
114 	    void (*start_func)(void *), void *arg,
115 	    char *stack_base, size_t stack_size,
116 	    char *tls_base,
117 	    long *child_tid, long *parent_tid,
118 	    int flags, struct thr_sched_param *sched)
119 {
120 	stack_t stack;
121 	struct thread *newtd;
122 	struct ksegrp *kg, *newkg;
123 	struct proc *p;
124 	long id;
125 	int error;
126 
127 	error = 0;
128 	p = td->td_proc;
129 	kg = td->td_ksegrp;
130 
131 	/* Have race condition but it is cheap. */
132 	if (p->p_numthreads >= max_threads_per_proc)
133 		return (EPROCLIM);
134 
135 	if (sched != NULL) {
136 		switch(sched->policy) {
137 		case SCHED_FIFO:
138 		case SCHED_RR:
139 			/* Only root can set scheduler policy */
140 			if (suser(td) != 0)
141 				return (EPERM);
142 			if (sched->param.sched_priority < RTP_PRIO_MIN ||
143 			    sched->param.sched_priority > RTP_PRIO_MAX)
144 				return (EINVAL);
145 			break;
146 		case SCHED_OTHER:
147 			break;
148 		default:
149 			return (EINVAL);
150 		}
151 	}
152 
153 	/* Initialize our td and new ksegrp.. */
154 	newtd = thread_alloc();
155 
156 	/*
157 	 * Try the copyout as soon as we allocate the td so we don't
158 	 * have to tear things down in a failure case below.
159 	 * Here we copy out tid to two places, one for child and one
160 	 * for parent, because pthread can create a detached thread,
161 	 * if parent wants to safely access child tid, it has to provide
162 	 * its storage, because child thread may exit quickly and
163 	 * memory is freed before parent thread can access it.
164 	 */
165 	id = newtd->td_tid;
166 	if ((child_tid != NULL &&
167 	    (error = copyout(&id, child_tid, sizeof(long)))) ||
168 	    (parent_tid != NULL &&
169 	    (error = copyout(&id, parent_tid, sizeof(long))))) {
170 	    	thread_free(newtd);
171 		return (error);
172 	}
173 	bzero(&newtd->td_startzero,
174 	    __rangeof(struct thread, td_startzero, td_endzero));
175 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
176 	    __rangeof(struct thread, td_startcopy, td_endcopy));
177 	newtd->td_proc = td->td_proc;
178 	newtd->td_ucred = crhold(td->td_ucred);
179 
180 	cpu_set_upcall(newtd, td);
181 
182 	if (ctx != NULL) { /* old way to set user context */
183 		error = set_mcontext(newtd, ctx);
184 		if (error != 0) {
185 			thread_free(newtd);
186 			crfree(td->td_ucred);
187 			return (error);
188 		}
189 	} else {
190 		/* Set up our machine context. */
191 		stack.ss_sp = stack_base;
192 		stack.ss_size = stack_size;
193 		/* Set upcall address to user thread entry function. */
194 		cpu_set_upcall_kse(newtd, start_func, arg, &stack);
195 		/* Setup user TLS address and TLS pointer register. */
196 		error = cpu_set_user_tls(newtd, tls_base);
197 		if (error != 0) {
198 			thread_free(newtd);
199 			crfree(td->td_ucred);
200 			return (error);
201 		}
202 	}
203 
204 	newkg = ksegrp_alloc();
205 	bzero(&newkg->kg_startzero,
206 	    __rangeof(struct ksegrp, kg_startzero, kg_endzero));
207 	bcopy(&kg->kg_startcopy, &newkg->kg_startcopy,
208 	    __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
209 	sched_init_concurrency(newkg);
210 	PROC_LOCK(td->td_proc);
211 	td->td_proc->p_flag |= P_HADTHREADS;
212 	newtd->td_sigmask = td->td_sigmask;
213 	mtx_lock_spin(&sched_lock);
214 	ksegrp_link(newkg, p);
215 	thread_link(newtd, newkg);
216 	PROC_UNLOCK(p);
217 
218 	/* let the scheduler know about these things. */
219 	sched_fork_ksegrp(td, newkg);
220 	sched_fork_thread(td, newtd);
221 	if (sched != NULL) {
222 		struct rtprio rtp;
223 		switch (sched->policy) {
224 		case SCHED_FIFO:
225 			rtp.type = PRI_FIFO;
226 			rtp.prio = sched->param.sched_priority;
227 			rtp_to_pri(&rtp, newkg);
228 			sched_prio(newtd, newkg->kg_user_pri);
229 			break;
230 		case SCHED_RR:
231 			rtp.type = PRI_REALTIME;
232 			rtp.prio = sched->param.sched_priority;
233 			rtp_to_pri(&rtp, newkg);
234 			sched_prio(newtd, newkg->kg_user_pri);
235 			break;
236 		case SCHED_OTHER:
237 			if (newkg->kg_pri_class != PRI_TIMESHARE) {
238 				rtp.type = PRI_TIMESHARE;
239 				rtp.prio = 0;
240 				rtp_to_pri(&rtp, newkg);
241 				sched_prio(newtd, newkg->kg_user_pri);
242 			}
243 			break;
244 		default:
245 			panic("sched policy");
246 		}
247 	}
248 	TD_SET_CAN_RUN(newtd);
249 	/* if ((flags & THR_SUSPENDED) == 0) */
250 		setrunqueue(newtd, SRQ_BORING);
251 	mtx_unlock_spin(&sched_lock);
252 
253 	return (error);
254 }
255 
256 int
257 thr_self(struct thread *td, struct thr_self_args *uap)
258     /* long *id */
259 {
260 	long id;
261 	int error;
262 
263 	id = td->td_tid;
264 	if ((error = copyout(&id, uap->id, sizeof(long))))
265 		return (error);
266 
267 	return (0);
268 }
269 
270 int
271 thr_exit(struct thread *td, struct thr_exit_args *uap)
272     /* long *state */
273 {
274 	struct proc *p;
275 
276 	p = td->td_proc;
277 
278 	/* Signal userland that it can free the stack. */
279 	if ((void *)uap->state != NULL) {
280 		suword((void *)uap->state, 1);
281 		kern_umtx_wake(td, uap->state, INT_MAX);
282 	}
283 
284 	PROC_LOCK(p);
285 	sigqueue_flush(&td->td_sigqueue);
286 	mtx_lock_spin(&sched_lock);
287 
288 	/*
289 	 * Shutting down last thread in the proc.  This will actually
290 	 * call exit() in the trampoline when it returns.
291 	 */
292 	if (p->p_numthreads != 1) {
293 		thread_stopped(p);
294 		thread_exit();
295 		/* NOTREACHED */
296 	}
297 	mtx_unlock_spin(&sched_lock);
298 	PROC_UNLOCK(p);
299 	return (0);
300 }
301 
302 int
303 thr_kill(struct thread *td, struct thr_kill_args *uap)
304     /* long id, int sig */
305 {
306 	struct thread *ttd;
307 	struct proc *p;
308 	int error;
309 
310 	p = td->td_proc;
311 	error = 0;
312 	PROC_LOCK(p);
313 	if (uap->id == -1) {
314 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
315 			error = EINVAL;
316 		} else {
317 			error = ESRCH;
318 			FOREACH_THREAD_IN_PROC(p, ttd) {
319 				if (ttd != td) {
320 					error = 0;
321 					if (uap->sig == 0)
322 						break;
323 					tdsignal(p, ttd, uap->sig, NULL);
324 				}
325 			}
326 		}
327 	} else {
328 		if (uap->id != td->td_tid)
329 			ttd = thread_find(p, uap->id);
330 		else
331 			ttd = td;
332 		if (ttd == NULL)
333 			error = ESRCH;
334 		else if (uap->sig == 0)
335 			;
336 		else if (!_SIG_VALID(uap->sig))
337 			error = EINVAL;
338 		else
339 			tdsignal(p, ttd, uap->sig, NULL);
340 	}
341 	PROC_UNLOCK(p);
342 	return (error);
343 }
344 
345 int
346 thr_suspend(struct thread *td, struct thr_suspend_args *uap)
347 	/* const struct timespec *timeout */
348 {
349 	struct timespec ts;
350 	struct timeval	tv;
351 	int error;
352 	int hz;
353 
354 	hz = 0;
355 	error = 0;
356 	if (uap->timeout != NULL) {
357 		error = copyin((const void *)uap->timeout, (void *)&ts,
358 		    sizeof(struct timespec));
359 		if (error != 0)
360 			return (error);
361 		if (ts.tv_nsec < 0 || ts.tv_nsec > 1000000000)
362 			return (EINVAL);
363 		if (ts.tv_sec == 0 && ts.tv_nsec == 0)
364 			return (ETIMEDOUT);
365 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
366 		hz = tvtohz(&tv);
367 	}
368 	PROC_LOCK(td->td_proc);
369 	if ((td->td_flags & TDF_THRWAKEUP) == 0)
370 		error = msleep((void *)td, &td->td_proc->p_mtx, PCATCH, "lthr",
371 		    hz);
372 	if (td->td_flags & TDF_THRWAKEUP) {
373 		mtx_lock_spin(&sched_lock);
374 		td->td_flags &= ~TDF_THRWAKEUP;
375 		mtx_unlock_spin(&sched_lock);
376 		PROC_UNLOCK(td->td_proc);
377 		return (0);
378 	}
379 	PROC_UNLOCK(td->td_proc);
380 	if (error == EWOULDBLOCK)
381 		error = ETIMEDOUT;
382 	else if (error == ERESTART) {
383 		if (hz != 0)
384 			error = EINTR;
385 	}
386 	return (error);
387 }
388 
389 int
390 thr_wake(struct thread *td, struct thr_wake_args *uap)
391 	/* long id */
392 {
393 	struct proc *p;
394 	struct thread *ttd;
395 
396 	p = td->td_proc;
397 	PROC_LOCK(p);
398 	ttd = thread_find(p, uap->id);
399 	if (ttd == NULL) {
400 		PROC_UNLOCK(p);
401 		return (ESRCH);
402 	}
403 	mtx_lock_spin(&sched_lock);
404 	ttd->td_flags |= TDF_THRWAKEUP;
405 	mtx_unlock_spin(&sched_lock);
406 	wakeup((void *)ttd);
407 	PROC_UNLOCK(p);
408 	return (0);
409 }
410 
411 int
412 thr_set_name(struct thread *td, struct thr_set_name_args *uap)
413 {
414 	struct proc *p = td->td_proc;
415 	char name[MAXCOMLEN + 1];
416 	struct thread *ttd;
417 	int error;
418 
419 	error = 0;
420 	name[0] = '\0';
421 	if (uap->name != NULL) {
422 		error = copyinstr(uap->name, name, sizeof(name),
423 			NULL);
424 		if (error)
425 			return (error);
426 	}
427 	PROC_LOCK(p);
428 	if (uap->id == td->td_tid)
429 		ttd = td;
430 	else
431 		ttd = thread_find(p, uap->id);
432 	if (ttd != NULL)
433 		strcpy(ttd->td_name, name);
434 	else
435 		error = ESRCH;
436 	PROC_UNLOCK(p);
437 	return (error);
438 }
439 
440 int
441 thr_setscheduler(struct thread *td, struct thr_setscheduler_args *uap)
442 {
443 	struct proc *p;
444 	struct thread *ttd;
445 	struct rtprio rtp;
446 	struct sched_param param;
447 	int ret;
448 
449 	if (uap->param_size != sizeof(struct sched_param))
450 		return (EINVAL);
451 
452 	ret = copyin(uap->param, &param, sizeof(struct sched_param));
453 	if (ret)
454 		return (ret);
455 
456 	switch(uap->policy) {
457 	case SCHED_FIFO:
458 		if (suser(td) != 0)
459 			return (EPERM);
460 		rtp.type = PRI_FIFO;
461 		break;
462 	case SCHED_RR:
463 		if (suser(td) != 0)
464 			return (EPERM);
465 		rtp.type = PRI_REALTIME;
466 		break;
467 	case SCHED_OTHER:
468 		rtp.type = PRI_TIMESHARE;
469 		break;
470 	default:
471 		return (EINVAL);
472 	}
473 	rtp.prio = param.sched_priority;
474 
475 	p = td->td_proc;
476 	PROC_LOCK(p);
477 	ret = p_cansched(td, p);
478 	if (ret != 0) {
479 		PROC_UNLOCK(p);
480 		return (ret);
481 	}
482 
483 	ttd = thread_find(p, uap->id);
484 	if (ttd == NULL) {
485 		PROC_UNLOCK(p);
486 		return (ESRCH);
487 	}
488 	mtx_lock_spin(&sched_lock);
489 	ret = rtp_to_pri(&rtp, ttd->td_ksegrp);
490 	if (ret == 0) {
491 		if (TD_IS_RUNNING(ttd))
492 			ttd->td_flags |= TDF_NEEDRESCHED;
493 		else if (ttd->td_priority > ttd->td_ksegrp->kg_user_pri)
494 			sched_prio(ttd, ttd->td_ksegrp->kg_user_pri);
495 	}
496 	mtx_unlock_spin(&sched_lock);
497 	PROC_UNLOCK(p);
498 	return (ret);
499 }
500 
501 int
502 thr_getscheduler(struct thread *td, struct thr_getscheduler_args *uap)
503 {
504 	struct proc *p;
505 	struct thread *ttd;
506 	struct rtprio rtp;
507 	struct sched_param param;
508 	int policy;
509 	int ret;
510 
511 	if (uap->param_size != sizeof(struct sched_param))
512 		return (EINVAL);
513 
514 	p = td->td_proc;
515 	PROC_LOCK(p);
516 	ttd = thread_find(p, uap->id);
517 	if (ttd == NULL) {
518 		PROC_UNLOCK(p);
519 		return (ESRCH);
520 	}
521 	mtx_lock_spin(&sched_lock);
522 	switch(ttd->td_ksegrp->kg_pri_class) {
523 	case PRI_TIMESHARE:
524 		policy = SCHED_OTHER;
525 		break;
526 	case PRI_FIFO:
527 		policy = SCHED_FIFO;
528 		break;
529 	case PRI_REALTIME:
530 		policy = SCHED_RR;
531 		break;
532 	default:
533 		policy = SCHED_OTHER; /* XXX SCHED_IDLE */
534 	}
535 	pri_to_rtp(ttd->td_ksegrp, &rtp);
536 	mtx_unlock_spin(&sched_lock);
537 	PROC_UNLOCK(p);
538 
539 	param.sched_priority = rtp.prio;
540 	ret = copyout(&policy, uap->policy, sizeof(policy));
541 	if (ret == 0)
542 		ret = copyout(&param, uap->param, sizeof(param));
543 	return (ret);
544 }
545 
546 int
547 thr_setschedparam(struct thread *td, struct thr_setschedparam_args *uap)
548 {
549 	struct proc *p;
550 	struct thread *ttd;
551 	struct rtprio rtp;
552 	struct sched_param param;
553 	int ret;
554 
555 	if (uap->param_size != sizeof(struct sched_param))
556 		return (EINVAL);
557 
558 	ret = copyin(uap->param, &param, sizeof(struct sched_param));
559 	if (ret)
560 		return (ret);
561 
562 	p = td->td_proc;
563 	PROC_LOCK(p);
564 	ret = p_cansched(td, p);
565 	if (ret != 0) {
566 		PROC_UNLOCK(p);
567 		return (ret);
568 	}
569 
570 	ttd = thread_find(p, uap->id);
571 	if (ttd == NULL) {
572 		PROC_UNLOCK(p);
573 		return (ESRCH);
574 	}
575 
576 	mtx_lock_spin(&sched_lock);
577 	pri_to_rtp(ttd->td_ksegrp, &rtp);
578 	rtp.prio = param.sched_priority;
579 	ret = rtp_to_pri(&rtp, ttd->td_ksegrp);
580 	if (ret == 0) {
581 		if (TD_IS_RUNNING(ttd))
582 			ttd->td_flags |= TDF_NEEDRESCHED;
583 		else if (ttd->td_priority > ttd->td_ksegrp->kg_user_pri)
584 			sched_prio(ttd, ttd->td_ksegrp->kg_user_pri);
585 	}
586 	mtx_unlock_spin(&sched_lock);
587 	PROC_UNLOCK(p);
588 	return (ret);
589 }
590