xref: /freebsd/sys/kern/kern_thr.c (revision eb69d1f144a6fcc765d1b9d44a5ae8082353e70b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.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 unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 #include "opt_posix.h"
34 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/priv.h>
39 #include <sys/proc.h>
40 #include <sys/posix4.h>
41 #include <sys/ptrace.h>
42 #include <sys/racct.h>
43 #include <sys/resourcevar.h>
44 #include <sys/rwlock.h>
45 #include <sys/sched.h>
46 #include <sys/sysctl.h>
47 #include <sys/smp.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysent.h>
50 #include <sys/systm.h>
51 #include <sys/sysproto.h>
52 #include <sys/signalvar.h>
53 #include <sys/sysctl.h>
54 #include <sys/ucontext.h>
55 #include <sys/thr.h>
56 #include <sys/rtprio.h>
57 #include <sys/umtx.h>
58 #include <sys/limits.h>
59 
60 #include <machine/frame.h>
61 
62 #include <security/audit/audit.h>
63 
64 static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0,
65     "thread allocation");
66 
67 static int max_threads_per_proc = 1500;
68 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
69     &max_threads_per_proc, 0, "Limit on threads per proc");
70 
71 static int max_threads_hits;
72 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
73     &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count");
74 
75 #ifdef COMPAT_FREEBSD32
76 
77 static inline int
78 suword_lwpid(void *addr, lwpid_t lwpid)
79 {
80 	int error;
81 
82 	if (SV_CURPROC_FLAG(SV_LP64))
83 		error = suword(addr, lwpid);
84 	else
85 		error = suword32(addr, lwpid);
86 	return (error);
87 }
88 
89 #else
90 #define suword_lwpid	suword
91 #endif
92 
93 /*
94  * System call interface.
95  */
96 
97 struct thr_create_initthr_args {
98 	ucontext_t ctx;
99 	long *tid;
100 };
101 
102 static int
103 thr_create_initthr(struct thread *td, void *thunk)
104 {
105 	struct thr_create_initthr_args *args;
106 
107 	/* Copy out the child tid. */
108 	args = thunk;
109 	if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid))
110 		return (EFAULT);
111 
112 	return (set_mcontext(td, &args->ctx.uc_mcontext));
113 }
114 
115 int
116 sys_thr_create(struct thread *td, struct thr_create_args *uap)
117     /* ucontext_t *ctx, long *id, int flags */
118 {
119 	struct thr_create_initthr_args args;
120 	int error;
121 
122 	if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx))))
123 		return (error);
124 	args.tid = uap->id;
125 	return (thread_create(td, NULL, thr_create_initthr, &args));
126 }
127 
128 int
129 sys_thr_new(struct thread *td, struct thr_new_args *uap)
130     /* struct thr_param * */
131 {
132 	struct thr_param param;
133 	int error;
134 
135 	if (uap->param_size < 0 || uap->param_size > sizeof(param))
136 		return (EINVAL);
137 	bzero(&param, sizeof(param));
138 	if ((error = copyin(uap->param, &param, uap->param_size)))
139 		return (error);
140 	return (kern_thr_new(td, &param));
141 }
142 
143 static int
144 thr_new_initthr(struct thread *td, void *thunk)
145 {
146 	stack_t stack;
147 	struct thr_param *param;
148 
149 	/*
150 	 * Here we copy out tid to two places, one for child and one
151 	 * for parent, because pthread can create a detached thread,
152 	 * if parent wants to safely access child tid, it has to provide
153 	 * its storage, because child thread may exit quickly and
154 	 * memory is freed before parent thread can access it.
155 	 */
156 	param = thunk;
157 	if ((param->child_tid != NULL &&
158 	    suword_lwpid(param->child_tid, td->td_tid)) ||
159 	    (param->parent_tid != NULL &&
160 	    suword_lwpid(param->parent_tid, td->td_tid)))
161 		return (EFAULT);
162 
163 	/* Set up our machine context. */
164 	stack.ss_sp = param->stack_base;
165 	stack.ss_size = param->stack_size;
166 	/* Set upcall address to user thread entry function. */
167 	cpu_set_upcall(td, param->start_func, param->arg, &stack);
168 	/* Setup user TLS address and TLS pointer register. */
169 	return (cpu_set_user_tls(td, param->tls_base));
170 }
171 
172 int
173 kern_thr_new(struct thread *td, struct thr_param *param)
174 {
175 	struct rtprio rtp, *rtpp;
176 	int error;
177 
178 	rtpp = NULL;
179 	if (param->rtp != 0) {
180 		error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
181 		if (error)
182 			return (error);
183 		rtpp = &rtp;
184 	}
185 	return (thread_create(td, rtpp, thr_new_initthr, param));
186 }
187 
188 int
189 thread_create(struct thread *td, struct rtprio *rtp,
190     int (*initialize_thread)(struct thread *, void *), void *thunk)
191 {
192 	struct thread *newtd;
193 	struct proc *p;
194 	int error;
195 
196 	p = td->td_proc;
197 
198 	if (rtp != NULL) {
199 		switch(rtp->type) {
200 		case RTP_PRIO_REALTIME:
201 		case RTP_PRIO_FIFO:
202 			/* Only root can set scheduler policy */
203 			if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
204 				return (EPERM);
205 			if (rtp->prio > RTP_PRIO_MAX)
206 				return (EINVAL);
207 			break;
208 		case RTP_PRIO_NORMAL:
209 			rtp->prio = 0;
210 			break;
211 		default:
212 			return (EINVAL);
213 		}
214 	}
215 
216 #ifdef RACCT
217 	if (racct_enable) {
218 		PROC_LOCK(p);
219 		error = racct_add(p, RACCT_NTHR, 1);
220 		PROC_UNLOCK(p);
221 		if (error != 0)
222 			return (EPROCLIM);
223 	}
224 #endif
225 
226 	/* Initialize our td */
227 	error = kern_thr_alloc(p, 0, &newtd);
228 	if (error)
229 		goto fail;
230 
231 	cpu_copy_thread(newtd, td);
232 
233 	bzero(&newtd->td_startzero,
234 	    __rangeof(struct thread, td_startzero, td_endzero));
235 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
236 	    __rangeof(struct thread, td_startcopy, td_endcopy));
237 	newtd->td_proc = td->td_proc;
238 	newtd->td_rb_list = newtd->td_rbp_list = newtd->td_rb_inact = 0;
239 	thread_cow_get(newtd, td);
240 
241 	error = initialize_thread(newtd, thunk);
242 	if (error != 0) {
243 		thread_cow_free(newtd);
244 		thread_free(newtd);
245 		goto fail;
246 	}
247 
248 	PROC_LOCK(p);
249 	p->p_flag |= P_HADTHREADS;
250 	thread_link(newtd, p);
251 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
252 	thread_lock(td);
253 	/* let the scheduler know about these things. */
254 	sched_fork_thread(td, newtd);
255 	thread_unlock(td);
256 	if (P_SHOULDSTOP(p))
257 		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
258 	if (p->p_ptevents & PTRACE_LWP)
259 		newtd->td_dbgflags |= TDB_BORN;
260 
261 	PROC_UNLOCK(p);
262 
263 	tidhash_add(newtd);
264 
265 	thread_lock(newtd);
266 	if (rtp != NULL) {
267 		if (!(td->td_pri_class == PRI_TIMESHARE &&
268 		      rtp->type == RTP_PRIO_NORMAL)) {
269 			rtp_to_pri(rtp, newtd);
270 			sched_prio(newtd, newtd->td_user_pri);
271 		} /* ignore timesharing class */
272 	}
273 	TD_SET_CAN_RUN(newtd);
274 	sched_add(newtd, SRQ_BORING);
275 	thread_unlock(newtd);
276 
277 	return (0);
278 
279 fail:
280 #ifdef RACCT
281 	if (racct_enable) {
282 		PROC_LOCK(p);
283 		racct_sub(p, RACCT_NTHR, 1);
284 		PROC_UNLOCK(p);
285 	}
286 #endif
287 	return (error);
288 }
289 
290 int
291 sys_thr_self(struct thread *td, struct thr_self_args *uap)
292     /* long *id */
293 {
294 	int error;
295 
296 	error = suword_lwpid(uap->id, (unsigned)td->td_tid);
297 	if (error == -1)
298 		return (EFAULT);
299 	return (0);
300 }
301 
302 int
303 sys_thr_exit(struct thread *td, struct thr_exit_args *uap)
304     /* long *state */
305 {
306 
307 	umtx_thread_exit(td);
308 
309 	/* Signal userland that it can free the stack. */
310 	if ((void *)uap->state != NULL) {
311 		suword_lwpid(uap->state, 1);
312 		kern_umtx_wake(td, uap->state, INT_MAX, 0);
313 	}
314 
315 	return (kern_thr_exit(td));
316 }
317 
318 int
319 kern_thr_exit(struct thread *td)
320 {
321 	struct proc *p;
322 
323 	p = td->td_proc;
324 
325 	/*
326 	 * If all of the threads in a process call this routine to
327 	 * exit (e.g. all threads call pthread_exit()), exactly one
328 	 * thread should return to the caller to terminate the process
329 	 * instead of the thread.
330 	 *
331 	 * Checking p_numthreads alone is not sufficient since threads
332 	 * might be committed to terminating while the PROC_LOCK is
333 	 * dropped in either ptracestop() or while removing this thread
334 	 * from the tidhash.  Instead, the p_pendingexits field holds
335 	 * the count of threads in either of those states and a thread
336 	 * is considered the "last" thread if all of the other threads
337 	 * in a process are already terminating.
338 	 */
339 	PROC_LOCK(p);
340 	if (p->p_numthreads == p->p_pendingexits + 1) {
341 		/*
342 		 * Ignore attempts to shut down last thread in the
343 		 * proc.  This will actually call _exit(2) in the
344 		 * usermode trampoline when it returns.
345 		 */
346 		PROC_UNLOCK(p);
347 		return (0);
348 	}
349 
350 	p->p_pendingexits++;
351 	td->td_dbgflags |= TDB_EXIT;
352 	if (p->p_ptevents & PTRACE_LWP)
353 		ptracestop(td, SIGTRAP, NULL);
354 	PROC_UNLOCK(p);
355 	tidhash_remove(td);
356 	PROC_LOCK(p);
357 	p->p_pendingexits--;
358 
359 	/*
360 	 * The check above should prevent all other threads from this
361 	 * process from exiting while the PROC_LOCK is dropped, so
362 	 * there must be at least one other thread other than the
363 	 * current thread.
364 	 */
365 	KASSERT(p->p_numthreads > 1, ("too few threads"));
366 	racct_sub(p, RACCT_NTHR, 1);
367 	tdsigcleanup(td);
368 	PROC_SLOCK(p);
369 	thread_stopped(p);
370 	thread_exit();
371 	/* NOTREACHED */
372 }
373 
374 int
375 sys_thr_kill(struct thread *td, struct thr_kill_args *uap)
376     /* long id, int sig */
377 {
378 	ksiginfo_t ksi;
379 	struct thread *ttd;
380 	struct proc *p;
381 	int error;
382 
383 	p = td->td_proc;
384 	ksiginfo_init(&ksi);
385 	ksi.ksi_signo = uap->sig;
386 	ksi.ksi_code = SI_LWP;
387 	ksi.ksi_pid = p->p_pid;
388 	ksi.ksi_uid = td->td_ucred->cr_ruid;
389 	if (uap->id == -1) {
390 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
391 			error = EINVAL;
392 		} else {
393 			error = ESRCH;
394 			PROC_LOCK(p);
395 			FOREACH_THREAD_IN_PROC(p, ttd) {
396 				if (ttd != td) {
397 					error = 0;
398 					if (uap->sig == 0)
399 						break;
400 					tdksignal(ttd, uap->sig, &ksi);
401 				}
402 			}
403 			PROC_UNLOCK(p);
404 		}
405 	} else {
406 		error = 0;
407 		ttd = tdfind((lwpid_t)uap->id, p->p_pid);
408 		if (ttd == NULL)
409 			return (ESRCH);
410 		if (uap->sig == 0)
411 			;
412 		else if (!_SIG_VALID(uap->sig))
413 			error = EINVAL;
414 		else
415 			tdksignal(ttd, uap->sig, &ksi);
416 		PROC_UNLOCK(ttd->td_proc);
417 	}
418 	return (error);
419 }
420 
421 int
422 sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap)
423     /* pid_t pid, long id, int sig */
424 {
425 	ksiginfo_t ksi;
426 	struct thread *ttd;
427 	struct proc *p;
428 	int error;
429 
430 	AUDIT_ARG_SIGNUM(uap->sig);
431 
432 	ksiginfo_init(&ksi);
433 	ksi.ksi_signo = uap->sig;
434 	ksi.ksi_code = SI_LWP;
435 	ksi.ksi_pid = td->td_proc->p_pid;
436 	ksi.ksi_uid = td->td_ucred->cr_ruid;
437 	if (uap->id == -1) {
438 		if ((p = pfind(uap->pid)) == NULL)
439 			return (ESRCH);
440 		AUDIT_ARG_PROCESS(p);
441 		error = p_cansignal(td, p, uap->sig);
442 		if (error) {
443 			PROC_UNLOCK(p);
444 			return (error);
445 		}
446 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
447 			error = EINVAL;
448 		} else {
449 			error = ESRCH;
450 			FOREACH_THREAD_IN_PROC(p, ttd) {
451 				if (ttd != td) {
452 					error = 0;
453 					if (uap->sig == 0)
454 						break;
455 					tdksignal(ttd, uap->sig, &ksi);
456 				}
457 			}
458 		}
459 		PROC_UNLOCK(p);
460 	} else {
461 		ttd = tdfind((lwpid_t)uap->id, uap->pid);
462 		if (ttd == NULL)
463 			return (ESRCH);
464 		p = ttd->td_proc;
465 		AUDIT_ARG_PROCESS(p);
466 		error = p_cansignal(td, p, uap->sig);
467 		if (uap->sig == 0)
468 			;
469 		else if (!_SIG_VALID(uap->sig))
470 			error = EINVAL;
471 		else
472 			tdksignal(ttd, uap->sig, &ksi);
473 		PROC_UNLOCK(p);
474 	}
475 	return (error);
476 }
477 
478 int
479 sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap)
480 	/* const struct timespec *timeout */
481 {
482 	struct timespec ts, *tsp;
483 	int error;
484 
485 	tsp = NULL;
486 	if (uap->timeout != NULL) {
487 		error = umtx_copyin_timeout(uap->timeout, &ts);
488 		if (error != 0)
489 			return (error);
490 		tsp = &ts;
491 	}
492 
493 	return (kern_thr_suspend(td, tsp));
494 }
495 
496 int
497 kern_thr_suspend(struct thread *td, struct timespec *tsp)
498 {
499 	struct proc *p = td->td_proc;
500 	struct timeval tv;
501 	int error = 0;
502 	int timo = 0;
503 
504 	if (td->td_pflags & TDP_WAKEUP) {
505 		td->td_pflags &= ~TDP_WAKEUP;
506 		return (0);
507 	}
508 
509 	if (tsp != NULL) {
510 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
511 			error = EWOULDBLOCK;
512 		else {
513 			TIMESPEC_TO_TIMEVAL(&tv, tsp);
514 			timo = tvtohz(&tv);
515 		}
516 	}
517 
518 	PROC_LOCK(p);
519 	if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0)
520 		error = msleep((void *)td, &p->p_mtx,
521 			 PCATCH, "lthr", timo);
522 
523 	if (td->td_flags & TDF_THRWAKEUP) {
524 		thread_lock(td);
525 		td->td_flags &= ~TDF_THRWAKEUP;
526 		thread_unlock(td);
527 		PROC_UNLOCK(p);
528 		return (0);
529 	}
530 	PROC_UNLOCK(p);
531 	if (error == EWOULDBLOCK)
532 		error = ETIMEDOUT;
533 	else if (error == ERESTART) {
534 		if (timo != 0)
535 			error = EINTR;
536 	}
537 	return (error);
538 }
539 
540 int
541 sys_thr_wake(struct thread *td, struct thr_wake_args *uap)
542 	/* long id */
543 {
544 	struct proc *p;
545 	struct thread *ttd;
546 
547 	if (uap->id == td->td_tid) {
548 		td->td_pflags |= TDP_WAKEUP;
549 		return (0);
550 	}
551 
552 	p = td->td_proc;
553 	ttd = tdfind((lwpid_t)uap->id, p->p_pid);
554 	if (ttd == NULL)
555 		return (ESRCH);
556 	thread_lock(ttd);
557 	ttd->td_flags |= TDF_THRWAKEUP;
558 	thread_unlock(ttd);
559 	wakeup((void *)ttd);
560 	PROC_UNLOCK(p);
561 	return (0);
562 }
563 
564 int
565 sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap)
566 {
567 	struct proc *p;
568 	char name[MAXCOMLEN + 1];
569 	struct thread *ttd;
570 	int error;
571 
572 	error = 0;
573 	name[0] = '\0';
574 	if (uap->name != NULL) {
575 		error = copyinstr(uap->name, name, sizeof(name), NULL);
576 		if (error == ENAMETOOLONG) {
577 			error = copyin(uap->name, name, sizeof(name) - 1);
578 			name[sizeof(name) - 1] = '\0';
579 		}
580 		if (error)
581 			return (error);
582 	}
583 	p = td->td_proc;
584 	ttd = tdfind((lwpid_t)uap->id, p->p_pid);
585 	if (ttd == NULL)
586 		return (ESRCH);
587 	strcpy(ttd->td_name, name);
588 #ifdef KTR
589 	sched_clear_tdname(ttd);
590 #endif
591 	PROC_UNLOCK(p);
592 	return (error);
593 }
594 
595 int
596 kern_thr_alloc(struct proc *p, int pages, struct thread **ntd)
597 {
598 
599 	/* Have race condition but it is cheap. */
600 	if (p->p_numthreads >= max_threads_per_proc) {
601 		++max_threads_hits;
602 		return (EPROCLIM);
603 	}
604 
605 	*ntd = thread_alloc(pages);
606 	if (*ntd == NULL)
607 		return (ENOMEM);
608 
609 	return (0);
610 }
611