xref: /freebsd/sys/kern/kern_thr.c (revision 6e778a7efdc0e804471750157f6bacd1ef7d1580)
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 <vm/vm_domain.h>
61 
62 #include <machine/frame.h>
63 
64 #include <security/audit/audit.h>
65 
66 static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0,
67     "thread allocation");
68 
69 static int max_threads_per_proc = 1500;
70 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
71     &max_threads_per_proc, 0, "Limit on threads per proc");
72 
73 static int max_threads_hits;
74 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
75     &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count");
76 
77 #ifdef COMPAT_FREEBSD32
78 
79 static inline int
80 suword_lwpid(void *addr, lwpid_t lwpid)
81 {
82 	int error;
83 
84 	if (SV_CURPROC_FLAG(SV_LP64))
85 		error = suword(addr, lwpid);
86 	else
87 		error = suword32(addr, lwpid);
88 	return (error);
89 }
90 
91 #else
92 #define suword_lwpid	suword
93 #endif
94 
95 /*
96  * System call interface.
97  */
98 
99 struct thr_create_initthr_args {
100 	ucontext_t ctx;
101 	long *tid;
102 };
103 
104 static int
105 thr_create_initthr(struct thread *td, void *thunk)
106 {
107 	struct thr_create_initthr_args *args;
108 
109 	/* Copy out the child tid. */
110 	args = thunk;
111 	if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid))
112 		return (EFAULT);
113 
114 	return (set_mcontext(td, &args->ctx.uc_mcontext));
115 }
116 
117 int
118 sys_thr_create(struct thread *td, struct thr_create_args *uap)
119     /* ucontext_t *ctx, long *id, int flags */
120 {
121 	struct thr_create_initthr_args args;
122 	int error;
123 
124 	if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx))))
125 		return (error);
126 	args.tid = uap->id;
127 	return (thread_create(td, NULL, thr_create_initthr, &args));
128 }
129 
130 int
131 sys_thr_new(struct thread *td, struct thr_new_args *uap)
132     /* struct thr_param * */
133 {
134 	struct thr_param param;
135 	int error;
136 
137 	if (uap->param_size < 0 || uap->param_size > sizeof(param))
138 		return (EINVAL);
139 	bzero(&param, sizeof(param));
140 	if ((error = copyin(uap->param, &param, uap->param_size)))
141 		return (error);
142 	return (kern_thr_new(td, &param));
143 }
144 
145 static int
146 thr_new_initthr(struct thread *td, void *thunk)
147 {
148 	stack_t stack;
149 	struct thr_param *param;
150 
151 	/*
152 	 * Here we copy out tid to two places, one for child and one
153 	 * for parent, because pthread can create a detached thread,
154 	 * if parent wants to safely access child tid, it has to provide
155 	 * its storage, because child thread may exit quickly and
156 	 * memory is freed before parent thread can access it.
157 	 */
158 	param = thunk;
159 	if ((param->child_tid != NULL &&
160 	    suword_lwpid(param->child_tid, td->td_tid)) ||
161 	    (param->parent_tid != NULL &&
162 	    suword_lwpid(param->parent_tid, td->td_tid)))
163 		return (EFAULT);
164 
165 	/* Set up our machine context. */
166 	stack.ss_sp = param->stack_base;
167 	stack.ss_size = param->stack_size;
168 	/* Set upcall address to user thread entry function. */
169 	cpu_set_upcall(td, param->start_func, param->arg, &stack);
170 	/* Setup user TLS address and TLS pointer register. */
171 	return (cpu_set_user_tls(td, param->tls_base));
172 }
173 
174 int
175 kern_thr_new(struct thread *td, struct thr_param *param)
176 {
177 	struct rtprio rtp, *rtpp;
178 	int error;
179 
180 	rtpp = NULL;
181 	if (param->rtp != 0) {
182 		error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
183 		if (error)
184 			return (error);
185 		rtpp = &rtp;
186 	}
187 	return (thread_create(td, rtpp, thr_new_initthr, param));
188 }
189 
190 int
191 thread_create(struct thread *td, struct rtprio *rtp,
192     int (*initialize_thread)(struct thread *, void *), void *thunk)
193 {
194 	struct thread *newtd;
195 	struct proc *p;
196 	int error;
197 
198 	p = td->td_proc;
199 
200 	if (rtp != NULL) {
201 		switch(rtp->type) {
202 		case RTP_PRIO_REALTIME:
203 		case RTP_PRIO_FIFO:
204 			/* Only root can set scheduler policy */
205 			if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
206 				return (EPERM);
207 			if (rtp->prio > RTP_PRIO_MAX)
208 				return (EINVAL);
209 			break;
210 		case RTP_PRIO_NORMAL:
211 			rtp->prio = 0;
212 			break;
213 		default:
214 			return (EINVAL);
215 		}
216 	}
217 
218 #ifdef RACCT
219 	if (racct_enable) {
220 		PROC_LOCK(p);
221 		error = racct_add(p, RACCT_NTHR, 1);
222 		PROC_UNLOCK(p);
223 		if (error != 0)
224 			return (EPROCLIM);
225 	}
226 #endif
227 
228 	/* Initialize our td */
229 	error = kern_thr_alloc(p, 0, &newtd);
230 	if (error)
231 		goto fail;
232 
233 	cpu_copy_thread(newtd, td);
234 
235 	bzero(&newtd->td_startzero,
236 	    __rangeof(struct thread, td_startzero, td_endzero));
237 	bcopy(&td->td_startcopy, &newtd->td_startcopy,
238 	    __rangeof(struct thread, td_startcopy, td_endcopy));
239 	newtd->td_proc = td->td_proc;
240 	newtd->td_rb_list = newtd->td_rbp_list = newtd->td_rb_inact = 0;
241 	thread_cow_get(newtd, td);
242 
243 	error = initialize_thread(newtd, thunk);
244 	if (error != 0) {
245 		thread_cow_free(newtd);
246 		thread_free(newtd);
247 		goto fail;
248 	}
249 
250 	PROC_LOCK(p);
251 	p->p_flag |= P_HADTHREADS;
252 	thread_link(newtd, p);
253 	bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
254 	thread_lock(td);
255 	/* let the scheduler know about these things. */
256 	sched_fork_thread(td, newtd);
257 	thread_unlock(td);
258 	if (P_SHOULDSTOP(p))
259 		newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
260 	if (p->p_ptevents & PTRACE_LWP)
261 		newtd->td_dbgflags |= TDB_BORN;
262 
263 	/*
264 	 * Copy the existing thread VM policy into the new thread.
265 	 */
266 	vm_domain_policy_localcopy(&newtd->td_vm_dom_policy,
267 	    &td->td_vm_dom_policy);
268 
269 	PROC_UNLOCK(p);
270 
271 	tidhash_add(newtd);
272 
273 	thread_lock(newtd);
274 	if (rtp != NULL) {
275 		if (!(td->td_pri_class == PRI_TIMESHARE &&
276 		      rtp->type == RTP_PRIO_NORMAL)) {
277 			rtp_to_pri(rtp, newtd);
278 			sched_prio(newtd, newtd->td_user_pri);
279 		} /* ignore timesharing class */
280 	}
281 	TD_SET_CAN_RUN(newtd);
282 	sched_add(newtd, SRQ_BORING);
283 	thread_unlock(newtd);
284 
285 	return (0);
286 
287 fail:
288 #ifdef RACCT
289 	if (racct_enable) {
290 		PROC_LOCK(p);
291 		racct_sub(p, RACCT_NTHR, 1);
292 		PROC_UNLOCK(p);
293 	}
294 #endif
295 	return (error);
296 }
297 
298 int
299 sys_thr_self(struct thread *td, struct thr_self_args *uap)
300     /* long *id */
301 {
302 	int error;
303 
304 	error = suword_lwpid(uap->id, (unsigned)td->td_tid);
305 	if (error == -1)
306 		return (EFAULT);
307 	return (0);
308 }
309 
310 int
311 sys_thr_exit(struct thread *td, struct thr_exit_args *uap)
312     /* long *state */
313 {
314 
315 	umtx_thread_exit(td);
316 
317 	/* Signal userland that it can free the stack. */
318 	if ((void *)uap->state != NULL) {
319 		suword_lwpid(uap->state, 1);
320 		kern_umtx_wake(td, uap->state, INT_MAX, 0);
321 	}
322 
323 	return (kern_thr_exit(td));
324 }
325 
326 int
327 kern_thr_exit(struct thread *td)
328 {
329 	struct proc *p;
330 
331 	p = td->td_proc;
332 
333 	/*
334 	 * If all of the threads in a process call this routine to
335 	 * exit (e.g. all threads call pthread_exit()), exactly one
336 	 * thread should return to the caller to terminate the process
337 	 * instead of the thread.
338 	 *
339 	 * Checking p_numthreads alone is not sufficient since threads
340 	 * might be committed to terminating while the PROC_LOCK is
341 	 * dropped in either ptracestop() or while removing this thread
342 	 * from the tidhash.  Instead, the p_pendingexits field holds
343 	 * the count of threads in either of those states and a thread
344 	 * is considered the "last" thread if all of the other threads
345 	 * in a process are already terminating.
346 	 */
347 	PROC_LOCK(p);
348 	if (p->p_numthreads == p->p_pendingexits + 1) {
349 		/*
350 		 * Ignore attempts to shut down last thread in the
351 		 * proc.  This will actually call _exit(2) in the
352 		 * usermode trampoline when it returns.
353 		 */
354 		PROC_UNLOCK(p);
355 		return (0);
356 	}
357 
358 	p->p_pendingexits++;
359 	td->td_dbgflags |= TDB_EXIT;
360 	if (p->p_ptevents & PTRACE_LWP)
361 		ptracestop(td, SIGTRAP, NULL);
362 	PROC_UNLOCK(p);
363 	tidhash_remove(td);
364 	PROC_LOCK(p);
365 	p->p_pendingexits--;
366 
367 	/*
368 	 * The check above should prevent all other threads from this
369 	 * process from exiting while the PROC_LOCK is dropped, so
370 	 * there must be at least one other thread other than the
371 	 * current thread.
372 	 */
373 	KASSERT(p->p_numthreads > 1, ("too few threads"));
374 	racct_sub(p, RACCT_NTHR, 1);
375 	tdsigcleanup(td);
376 	PROC_SLOCK(p);
377 	thread_stopped(p);
378 	thread_exit();
379 	/* NOTREACHED */
380 }
381 
382 int
383 sys_thr_kill(struct thread *td, struct thr_kill_args *uap)
384     /* long id, int sig */
385 {
386 	ksiginfo_t ksi;
387 	struct thread *ttd;
388 	struct proc *p;
389 	int error;
390 
391 	p = td->td_proc;
392 	ksiginfo_init(&ksi);
393 	ksi.ksi_signo = uap->sig;
394 	ksi.ksi_code = SI_LWP;
395 	ksi.ksi_pid = p->p_pid;
396 	ksi.ksi_uid = td->td_ucred->cr_ruid;
397 	if (uap->id == -1) {
398 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
399 			error = EINVAL;
400 		} else {
401 			error = ESRCH;
402 			PROC_LOCK(p);
403 			FOREACH_THREAD_IN_PROC(p, ttd) {
404 				if (ttd != td) {
405 					error = 0;
406 					if (uap->sig == 0)
407 						break;
408 					tdksignal(ttd, uap->sig, &ksi);
409 				}
410 			}
411 			PROC_UNLOCK(p);
412 		}
413 	} else {
414 		error = 0;
415 		ttd = tdfind((lwpid_t)uap->id, p->p_pid);
416 		if (ttd == NULL)
417 			return (ESRCH);
418 		if (uap->sig == 0)
419 			;
420 		else if (!_SIG_VALID(uap->sig))
421 			error = EINVAL;
422 		else
423 			tdksignal(ttd, uap->sig, &ksi);
424 		PROC_UNLOCK(ttd->td_proc);
425 	}
426 	return (error);
427 }
428 
429 int
430 sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap)
431     /* pid_t pid, long id, int sig */
432 {
433 	ksiginfo_t ksi;
434 	struct thread *ttd;
435 	struct proc *p;
436 	int error;
437 
438 	AUDIT_ARG_SIGNUM(uap->sig);
439 
440 	ksiginfo_init(&ksi);
441 	ksi.ksi_signo = uap->sig;
442 	ksi.ksi_code = SI_LWP;
443 	ksi.ksi_pid = td->td_proc->p_pid;
444 	ksi.ksi_uid = td->td_ucred->cr_ruid;
445 	if (uap->id == -1) {
446 		if ((p = pfind(uap->pid)) == NULL)
447 			return (ESRCH);
448 		AUDIT_ARG_PROCESS(p);
449 		error = p_cansignal(td, p, uap->sig);
450 		if (error) {
451 			PROC_UNLOCK(p);
452 			return (error);
453 		}
454 		if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
455 			error = EINVAL;
456 		} else {
457 			error = ESRCH;
458 			FOREACH_THREAD_IN_PROC(p, ttd) {
459 				if (ttd != td) {
460 					error = 0;
461 					if (uap->sig == 0)
462 						break;
463 					tdksignal(ttd, uap->sig, &ksi);
464 				}
465 			}
466 		}
467 		PROC_UNLOCK(p);
468 	} else {
469 		ttd = tdfind((lwpid_t)uap->id, uap->pid);
470 		if (ttd == NULL)
471 			return (ESRCH);
472 		p = ttd->td_proc;
473 		AUDIT_ARG_PROCESS(p);
474 		error = p_cansignal(td, p, uap->sig);
475 		if (uap->sig == 0)
476 			;
477 		else if (!_SIG_VALID(uap->sig))
478 			error = EINVAL;
479 		else
480 			tdksignal(ttd, uap->sig, &ksi);
481 		PROC_UNLOCK(p);
482 	}
483 	return (error);
484 }
485 
486 int
487 sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap)
488 	/* const struct timespec *timeout */
489 {
490 	struct timespec ts, *tsp;
491 	int error;
492 
493 	tsp = NULL;
494 	if (uap->timeout != NULL) {
495 		error = umtx_copyin_timeout(uap->timeout, &ts);
496 		if (error != 0)
497 			return (error);
498 		tsp = &ts;
499 	}
500 
501 	return (kern_thr_suspend(td, tsp));
502 }
503 
504 int
505 kern_thr_suspend(struct thread *td, struct timespec *tsp)
506 {
507 	struct proc *p = td->td_proc;
508 	struct timeval tv;
509 	int error = 0;
510 	int timo = 0;
511 
512 	if (td->td_pflags & TDP_WAKEUP) {
513 		td->td_pflags &= ~TDP_WAKEUP;
514 		return (0);
515 	}
516 
517 	if (tsp != NULL) {
518 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
519 			error = EWOULDBLOCK;
520 		else {
521 			TIMESPEC_TO_TIMEVAL(&tv, tsp);
522 			timo = tvtohz(&tv);
523 		}
524 	}
525 
526 	PROC_LOCK(p);
527 	if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0)
528 		error = msleep((void *)td, &p->p_mtx,
529 			 PCATCH, "lthr", timo);
530 
531 	if (td->td_flags & TDF_THRWAKEUP) {
532 		thread_lock(td);
533 		td->td_flags &= ~TDF_THRWAKEUP;
534 		thread_unlock(td);
535 		PROC_UNLOCK(p);
536 		return (0);
537 	}
538 	PROC_UNLOCK(p);
539 	if (error == EWOULDBLOCK)
540 		error = ETIMEDOUT;
541 	else if (error == ERESTART) {
542 		if (timo != 0)
543 			error = EINTR;
544 	}
545 	return (error);
546 }
547 
548 int
549 sys_thr_wake(struct thread *td, struct thr_wake_args *uap)
550 	/* long id */
551 {
552 	struct proc *p;
553 	struct thread *ttd;
554 
555 	if (uap->id == td->td_tid) {
556 		td->td_pflags |= TDP_WAKEUP;
557 		return (0);
558 	}
559 
560 	p = td->td_proc;
561 	ttd = tdfind((lwpid_t)uap->id, p->p_pid);
562 	if (ttd == NULL)
563 		return (ESRCH);
564 	thread_lock(ttd);
565 	ttd->td_flags |= TDF_THRWAKEUP;
566 	thread_unlock(ttd);
567 	wakeup((void *)ttd);
568 	PROC_UNLOCK(p);
569 	return (0);
570 }
571 
572 int
573 sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap)
574 {
575 	struct proc *p;
576 	char name[MAXCOMLEN + 1];
577 	struct thread *ttd;
578 	int error;
579 
580 	error = 0;
581 	name[0] = '\0';
582 	if (uap->name != NULL) {
583 		error = copyinstr(uap->name, name, sizeof(name), NULL);
584 		if (error == ENAMETOOLONG) {
585 			error = copyin(uap->name, name, sizeof(name) - 1);
586 			name[sizeof(name) - 1] = '\0';
587 		}
588 		if (error)
589 			return (error);
590 	}
591 	p = td->td_proc;
592 	ttd = tdfind((lwpid_t)uap->id, p->p_pid);
593 	if (ttd == NULL)
594 		return (ESRCH);
595 	strcpy(ttd->td_name, name);
596 #ifdef KTR
597 	sched_clear_tdname(ttd);
598 #endif
599 	PROC_UNLOCK(p);
600 	return (error);
601 }
602 
603 int
604 kern_thr_alloc(struct proc *p, int pages, struct thread **ntd)
605 {
606 
607 	/* Have race condition but it is cheap. */
608 	if (p->p_numthreads >= max_threads_per_proc) {
609 		++max_threads_hits;
610 		return (EPROCLIM);
611 	}
612 
613 	*ntd = thread_alloc(pages);
614 	if (*ntd == NULL)
615 		return (ENOMEM);
616 
617 	return (0);
618 }
619