xref: /freebsd/sys/kern/kern_thr.c (revision 7d0d268b8a67f28ccefdd0b8ce6fb38acac78d80) 
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_compat.h"
31 #include "opt_posix.h"
32 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/priv.h>
37 #include <sys/proc.h>
38 #include <sys/posix4.h>
39 #include <sys/resourcevar.h>
40 #include <sys/sched.h>
41 #include <sys/sysctl.h>
42 #include <sys/smp.h>
43 #include <sys/syscallsubr.h>
44 #include <sys/sysent.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/signalvar.h>
48 #include <sys/ucontext.h>
49 #include <sys/thr.h>
50 #include <sys/rtprio.h>
51 #include <sys/umtx.h>
52 #include <sys/limits.h>
53 
54 #include <machine/frame.h>
55 
56 #include <security/audit/audit.h>
57 
58 #ifdef COMPAT_IA32
59 
60 static inline int
61 suword_lwpid(void *addr, lwpid_t lwpid)
62 {
63 	int error;
64 
65 	if (SV_CURPROC_FLAG(SV_LP64))
66 		error = suword(addr, lwpid);
67 	else
68 		error = suword32(addr, lwpid);
69 	return (error);
70 }
71 
72 #else
73 #define suword_lwpid	suword
74 #endif
75 
76 extern int max_threads_per_proc;
77 extern int max_threads_hits;
78 
79 static int create_thread(struct thread *td, mcontext_t *ctx,
80 			 void (*start_func)(void *), void *arg,
81 			 char *stack_base, size_t stack_size,
82 			 char *tls_base,
83 			 long *child_tid, long *parent_tid,
84 			 int flags, struct rtprio *rtp);
85 
86 /*
87  * System call interface.
88  */
89 int
90 thr_create(struct thread *td, struct thr_create_args *uap)
91     /* ucontext_t *ctx, long *id, int flags */
92 {
93 	ucontext_t ctx;
94 	int error;
95 
96 	if ((error = copyin(uap->ctx, &ctx, sizeof(ctx))))
97 		return (error);
98 
99 	error = create_thread(td, &ctx.uc_mcontext, NULL, NULL,
100 		NULL, 0, NULL, uap->id, NULL, uap->flags, NULL);
101 	return (error);
102 }
103 
104 int
105 thr_new(struct thread *td, struct thr_new_args *uap)
106     /* struct thr_param * */
107 {
108 	struct thr_param param;
109 	int error;
110 
111 	if (uap->param_size < 0 || uap->param_size > sizeof(param))
112 		return (EINVAL);
113 	bzero(&param, sizeof(param));
114 	if ((error = copyin(uap->param, &param, uap->param_size)))
115 		return (error);
116 	return (kern_thr_new(td, &param));
117 }
118 
119 int
120 kern_thr_new(struct thread *td, struct thr_param *param)
121 {
122 	struct rtprio rtp, *rtpp;
123 	int error;
124 
125 	rtpp = NULL;
126 	if (param->rtp != 0) {
127 		error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
128 		if (error)
129 			return (error);
130 		rtpp = &rtp;
131 	}
132 	error = create_thread(td, NULL, param->start_func, param->arg,
133 		param->stack_base, param->stack_size, param->tls_base,
134 		param->child_tid, param->parent_tid, param->flags,
135 		rtpp);
136 	return (error);
137 }
138 
139 static int
140 create_thread(struct thread *td, mcontext_t *ctx,
141 	    void (*start_func)(void *), void *arg,
142 	    char *stack_base, size_t stack_size,
143 	    char *tls_base,
144 	    long *child_tid, long *parent_tid,
145 	    int flags, struct rtprio *rtp)
146 {
147 	stack_t stack;
148 	struct thread *newtd;
149 	struct proc *p;
150 	int error;
151 
152 	error = 0;
153 	p = td->td_proc;
154 
155 	/* Have race condition but it is cheap. */
156 	if (p->p_numthreads >= max_threads_per_proc) {
157 		++max_threads_hits;
158 		return (EPROCLIM);
159 	}
160 
161 	if (rtp != NULL) {
162 		switch(rtp->type) {
163 		case RTP_PRIO_REALTIME:
164 		case RTP_PRIO_FIFO:
165 			/* Only root can set scheduler policy */
166 			if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
167 				return (EPERM);
168 			if (rtp->prio > RTP_PRIO_MAX)
169 				return (EINVAL);
170 			break;
171 		case RTP_PRIO_NORMAL:
172 			rtp->prio = 0;
173 			break;
174 		default:
175 			return (EINVAL);
176 		}
177 	}
178 
179 	/* Initialize our td */
180 	newtd = thread_alloc();
181 	if (newtd == NULL)
182 		return (ENOMEM);
183 
184 	/*
185 	 * Try the copyout as soon as we allocate the td so we don't
186 	 * have to tear things down in a failure case below.
187 	 * Here we copy out tid to two places, one for child and one
188 	 * for parent, because pthread can create a detached thread,
189 	 * if parent wants to safely access child tid, it has to provide
190 	 * its storage, because child thread may exit quickly and
191 	 * memory is freed before parent thread can access it.
192 	 */
193 	if ((child_tid != NULL &&
194 	    suword_lwpid(child_tid, newtd->td_tid)) ||
195 	    (parent_tid != NULL &&
196 	    suword_lwpid(parent_tid, newtd->td_tid))) {
197 		thread_free(newtd);
198 		return (EFAULT);
199 	}
200 
201 	bzero(&newtd->td_startzero,
202 	    __rangeof(struct thread, td_startzero, td_endzero));
203 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
204 	    __rangeof(struct thread, td_startcopy, td_endcopy));
205 	newtd->td_proc = td->td_proc;
206 	newtd->td_ucred = crhold(td->td_ucred);
207 
208 	cpu_set_upcall(newtd, td);
209 
210 	if (ctx != NULL) { /* old way to set user context */
211 		error = set_mcontext(newtd, ctx);
212 		if (error != 0) {
213 			thread_free(newtd);
214 			crfree(td->td_ucred);
215 			return (error);
216 		}
217 	} else {
218 		/* Set up our machine context. */
219 		stack.ss_sp = stack_base;
220 		stack.ss_size = stack_size;
221 		/* Set upcall address to user thread entry function. */
222 		cpu_set_upcall_kse(newtd, start_func, arg, &stack);
223 		/* Setup user TLS address and TLS pointer register. */
224 		error = cpu_set_user_tls(newtd, tls_base);
225 		if (error != 0) {
226 			thread_free(newtd);
227 			crfree(td->td_ucred);
228 			return (error);
229 		}
230 	}
231 
232 	PROC_LOCK(td->td_proc);
233 	td->td_proc->p_flag |= P_HADTHREADS;
234 	newtd->td_sigmask = td->td_sigmask;
235 	thread_link(newtd, p);
236 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
237 	thread_lock(td);
238 	/* let the scheduler know about these things. */
239 	sched_fork_thread(td, newtd);
240 	thread_unlock(td);
241 	if (P_SHOULDSTOP(p))
242 		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
243 	PROC_UNLOCK(p);
244 	thread_lock(newtd);
245 	if (rtp != NULL) {
246 		if (!(td->td_pri_class == PRI_TIMESHARE &&
247 		      rtp->type == RTP_PRIO_NORMAL)) {
248 			rtp_to_pri(rtp, newtd);
249 			sched_prio(newtd, newtd->td_user_pri);
250 		} /* ignore timesharing class */
251 	}
252 	TD_SET_CAN_RUN(newtd);
253 	sched_add(newtd, SRQ_BORING);
254 	thread_unlock(newtd);
255 
256 	return (error);
257 }
258 
259 int
260 thr_self(struct thread *td, struct thr_self_args *uap)
261     /* long *id */
262 {
263 	int error;
264 
265 	error = suword_lwpid(uap->id, (unsigned)td->td_tid);
266 	if (error == -1)
267 		return (EFAULT);
268 	return (0);
269 }
270 
271 int
272 thr_exit(struct thread *td, struct thr_exit_args *uap)
273     /* long *state */
274 {
275 	struct proc *p;
276 
277 	p = td->td_proc;
278 
279 	/* Signal userland that it can free the stack. */
280 	if ((void *)uap->state != NULL) {
281 		suword_lwpid(uap->state, 1);
282 		kern_umtx_wake(td, uap->state, INT_MAX, 0);
283 	}
284 
285 	PROC_LOCK(p);
286 	sigqueue_flush(&td->td_sigqueue);
287 	PROC_SLOCK(p);
288 
289 	/*
290 	 * Shutting down last thread in the proc.  This will actually
291 	 * call exit() in the trampoline when it returns.
292 	 */
293 	if (p->p_numthreads != 1) {
294 		thread_stopped(p);
295 		thread_exit();
296 		/* NOTREACHED */
297 	}
298 	PROC_SUNLOCK(p);
299 	PROC_UNLOCK(p);
300 	return (0);
301 }
302 
303 int
304 thr_kill(struct thread *td, struct thr_kill_args *uap)
305     /* long id, int sig */
306 {
307 	struct thread *ttd;
308 	struct proc *p;
309 	int error;
310 
311 	p = td->td_proc;
312 	error = 0;
313 	PROC_LOCK(p);
314 	if (uap->id == -1) {
315 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
316 			error = EINVAL;
317 		} else {
318 			error = ESRCH;
319 			FOREACH_THREAD_IN_PROC(p, ttd) {
320 				if (ttd != td) {
321 					error = 0;
322 					if (uap->sig == 0)
323 						break;
324 					tdsignal(p, ttd, uap->sig, NULL);
325 				}
326 			}
327 		}
328 	} else {
329 		if (uap->id != td->td_tid)
330 			ttd = thread_find(p, uap->id);
331 		else
332 			ttd = td;
333 		if (ttd == NULL)
334 			error = ESRCH;
335 		else if (uap->sig == 0)
336 			;
337 		else if (!_SIG_VALID(uap->sig))
338 			error = EINVAL;
339 		else
340 			tdsignal(p, ttd, uap->sig, NULL);
341 	}
342 	PROC_UNLOCK(p);
343 	return (error);
344 }
345 
346 int
347 thr_kill2(struct thread *td, struct thr_kill2_args *uap)
348     /* pid_t pid, long id, int sig */
349 {
350 	struct thread *ttd;
351 	struct proc *p;
352 	int error;
353 
354 	AUDIT_ARG(signum, uap->sig);
355 
356 	if (uap->pid == td->td_proc->p_pid) {
357 		p = td->td_proc;
358 		PROC_LOCK(p);
359 	} else if ((p = pfind(uap->pid)) == NULL) {
360 		return (ESRCH);
361 	}
362 	AUDIT_ARG(process, p);
363 
364 	error = p_cansignal(td, p, uap->sig);
365 	if (error == 0) {
366 		if (uap->id == -1) {
367 			if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
368 				error = EINVAL;
369 			} else {
370 				error = ESRCH;
371 				FOREACH_THREAD_IN_PROC(p, ttd) {
372 					if (ttd != td) {
373 						error = 0;
374 						if (uap->sig == 0)
375 							break;
376 						tdsignal(p, ttd, uap->sig, NULL);
377 					}
378 				}
379 			}
380 		} else {
381 			if (uap->id != td->td_tid)
382 				ttd = thread_find(p, uap->id);
383 			else
384 				ttd = td;
385 			if (ttd == NULL)
386 				error = ESRCH;
387 			else if (uap->sig == 0)
388 				;
389 			else if (!_SIG_VALID(uap->sig))
390 				error = EINVAL;
391 			else
392 				tdsignal(p, ttd, uap->sig, NULL);
393 		}
394 	}
395 	PROC_UNLOCK(p);
396 	return (error);
397 }
398 
399 int
400 thr_suspend(struct thread *td, struct thr_suspend_args *uap)
401 	/* const struct timespec *timeout */
402 {
403 	struct timespec ts, *tsp;
404 	int error;
405 
406 	error = 0;
407 	tsp = NULL;
408 	if (uap->timeout != NULL) {
409 		error = copyin((const void *)uap->timeout, (void *)&ts,
410 		    sizeof(struct timespec));
411 		if (error != 0)
412 			return (error);
413 		tsp = &ts;
414 	}
415 
416 	return (kern_thr_suspend(td, tsp));
417 }
418 
419 int
420 kern_thr_suspend(struct thread *td, struct timespec *tsp)
421 {
422 	struct timeval tv;
423 	int error = 0, hz = 0;
424 
425 	if (tsp != NULL) {
426 		if (tsp->tv_nsec < 0 || tsp->tv_nsec > 1000000000)
427 			return (EINVAL);
428 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
429 			return (ETIMEDOUT);
430 		TIMESPEC_TO_TIMEVAL(&tv, tsp);
431 		hz = tvtohz(&tv);
432 	}
433 
434 	if (td->td_pflags & TDP_WAKEUP) {
435 		td->td_pflags &= ~TDP_WAKEUP;
436 		return (0);
437 	}
438 
439 	PROC_LOCK(td->td_proc);
440 	if ((td->td_flags & TDF_THRWAKEUP) == 0)
441 		error = msleep((void *)td, &td->td_proc->p_mtx, PCATCH, "lthr",
442 		    hz);
443 	if (td->td_flags & TDF_THRWAKEUP) {
444 		thread_lock(td);
445 		td->td_flags &= ~TDF_THRWAKEUP;
446 		thread_unlock(td);
447 		PROC_UNLOCK(td->td_proc);
448 		return (0);
449 	}
450 	PROC_UNLOCK(td->td_proc);
451 	if (error == EWOULDBLOCK)
452 		error = ETIMEDOUT;
453 	else if (error == ERESTART) {
454 		if (hz != 0)
455 			error = EINTR;
456 	}
457 	return (error);
458 }
459 
460 int
461 thr_wake(struct thread *td, struct thr_wake_args *uap)
462 	/* long id */
463 {
464 	struct proc *p;
465 	struct thread *ttd;
466 
467 	if (uap->id == td->td_tid) {
468 		td->td_pflags |= TDP_WAKEUP;
469 		return (0);
470 	}
471 
472 	p = td->td_proc;
473 	PROC_LOCK(p);
474 	ttd = thread_find(p, uap->id);
475 	if (ttd == NULL) {
476 		PROC_UNLOCK(p);
477 		return (ESRCH);
478 	}
479 	thread_lock(ttd);
480 	ttd->td_flags |= TDF_THRWAKEUP;
481 	thread_unlock(ttd);
482 	wakeup((void *)ttd);
483 	PROC_UNLOCK(p);
484 	return (0);
485 }
486 
487 int
488 thr_set_name(struct thread *td, struct thr_set_name_args *uap)
489 {
490 	struct proc *p = td->td_proc;
491 	char name[MAXCOMLEN + 1];
492 	struct thread *ttd;
493 	int error;
494 
495 	error = 0;
496 	name[0] = '\0';
497 	if (uap->name != NULL) {
498 		error = copyinstr(uap->name, name, sizeof(name),
499 			NULL);
500 		if (error)
501 			return (error);
502 	}
503 	PROC_LOCK(p);
504 	if (uap->id == td->td_tid)
505 		ttd = td;
506 	else
507 		ttd = thread_find(p, uap->id);
508 	if (ttd != NULL)
509 		strcpy(ttd->td_name, name);
510 	else
511 		error = ESRCH;
512 	PROC_UNLOCK(p);
513 	return (error);
514 }
515