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: stable/10/sys/kern/kern_thr.c 337258 2018-08-03 14:45:53Z asomers $");
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/ptrace.h>
40 #include <sys/racct.h>
41 #include <sys/resourcevar.h>
42 #include <sys/rwlock.h>
43 #include <sys/sched.h>
44 #include <sys/sysctl.h>
45 #include <sys/smp.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysent.h>
48 #include <sys/systm.h>
49 #include <sys/sysproto.h>
50 #include <sys/signalvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/ucontext.h>
53 #include <sys/thr.h>
54 #include <sys/rtprio.h>
55 #include <sys/umtx.h>
56 #include <sys/limits.h>
57
58 #include <machine/frame.h>
59
60 #include <security/audit/audit.h>
61
62 static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0,
63 "thread allocation");
64
65 static int max_threads_per_proc = 1500;
66 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
67 &max_threads_per_proc, 0, "Limit on threads per proc");
68
69 static int max_threads_hits;
70 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
71 &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count");
72
73 #ifdef COMPAT_FREEBSD32
74
75 static inline int
suword_lwpid(void * addr,lwpid_t lwpid)76 suword_lwpid(void *addr, lwpid_t lwpid)
77 {
78 int error;
79
80 if (SV_CURPROC_FLAG(SV_LP64))
81 error = suword(addr, lwpid);
82 else
83 error = suword32(addr, lwpid);
84 return (error);
85 }
86
87 #else
88 #define suword_lwpid suword
89 #endif
90
91 /*
92 * System call interface.
93 */
94
95 struct thr_create_initthr_args {
96 ucontext_t ctx;
97 long *tid;
98 };
99
100 static int
thr_create_initthr(struct thread * td,void * thunk)101 thr_create_initthr(struct thread *td, void *thunk)
102 {
103 struct thr_create_initthr_args *args;
104
105 /* Copy out the child tid. */
106 args = thunk;
107 if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid))
108 return (EFAULT);
109
110 return (set_mcontext(td, &args->ctx.uc_mcontext));
111 }
112
113 int
sys_thr_create(struct thread * td,struct thr_create_args * uap)114 sys_thr_create(struct thread *td, struct thr_create_args *uap)
115 /* ucontext_t *ctx, long *id, int flags */
116 {
117 struct thr_create_initthr_args args;
118 int error;
119
120 if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx))))
121 return (error);
122 args.tid = uap->id;
123 return (thread_create(td, NULL, thr_create_initthr, &args));
124 }
125
126 int
sys_thr_new(struct thread * td,struct thr_new_args * uap)127 sys_thr_new(struct thread *td, struct thr_new_args *uap)
128 /* struct thr_param * */
129 {
130 struct thr_param param;
131 int error;
132
133 if (uap->param_size < 0 || uap->param_size > sizeof(param))
134 return (EINVAL);
135 bzero(¶m, sizeof(param));
136 if ((error = copyin(uap->param, ¶m, uap->param_size)))
137 return (error);
138 return (kern_thr_new(td, ¶m));
139 }
140
141 static int
thr_new_initthr(struct thread * td,void * thunk)142 thr_new_initthr(struct thread *td, void *thunk)
143 {
144 stack_t stack;
145 struct thr_param *param;
146
147 /*
148 * Here we copy out tid to two places, one for child and one
149 * for parent, because pthread can create a detached thread,
150 * if parent wants to safely access child tid, it has to provide
151 * its storage, because child thread may exit quickly and
152 * memory is freed before parent thread can access it.
153 */
154 param = thunk;
155 if ((param->child_tid != NULL &&
156 suword_lwpid(param->child_tid, td->td_tid)) ||
157 (param->parent_tid != NULL &&
158 suword_lwpid(param->parent_tid, td->td_tid)))
159 return (EFAULT);
160
161 /* Set up our machine context. */
162 stack.ss_sp = param->stack_base;
163 stack.ss_size = param->stack_size;
164 /* Set upcall address to user thread entry function. */
165 cpu_set_upcall_kse(td, param->start_func, param->arg, &stack);
166 /* Setup user TLS address and TLS pointer register. */
167 return (cpu_set_user_tls(td, param->tls_base));
168 }
169
170 int
kern_thr_new(struct thread * td,struct thr_param * param)171 kern_thr_new(struct thread *td, struct thr_param *param)
172 {
173 struct rtprio rtp, *rtpp;
174 int error;
175
176 rtpp = NULL;
177 if (param->rtp != 0) {
178 error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
179 if (error)
180 return (error);
181 rtpp = &rtp;
182 }
183 return (thread_create(td, rtpp, thr_new_initthr, param));
184 }
185
186 int
thread_create(struct thread * td,struct rtprio * rtp,int (* initialize_thread)(struct thread *,void *),void * thunk)187 thread_create(struct thread *td, struct rtprio *rtp,
188 int (*initialize_thread)(struct thread *, void *), void *thunk)
189 {
190 struct thread *newtd;
191 struct proc *p;
192 int error;
193
194 p = td->td_proc;
195
196 if (rtp != NULL) {
197 switch(rtp->type) {
198 case RTP_PRIO_REALTIME:
199 case RTP_PRIO_FIFO:
200 /* Only root can set scheduler policy */
201 if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
202 return (EPERM);
203 if (rtp->prio > RTP_PRIO_MAX)
204 return (EINVAL);
205 break;
206 case RTP_PRIO_NORMAL:
207 rtp->prio = 0;
208 break;
209 default:
210 return (EINVAL);
211 }
212 }
213
214 #ifdef RACCT
215 PROC_LOCK(td->td_proc);
216 error = racct_add(p, RACCT_NTHR, 1);
217 PROC_UNLOCK(td->td_proc);
218 if (error != 0)
219 return (EPROCLIM);
220 #endif
221
222 /* Initialize our td */
223 error = kern_thr_alloc(p, 0, &newtd);
224 if (error)
225 goto fail;
226
227 cpu_set_upcall(newtd, td);
228
229 bzero(&newtd->td_startzero,
230 __rangeof(struct thread, td_startzero, td_endzero));
231 newtd->td_su = NULL;
232 newtd->td_sleeptimo = 0;
233 bcopy(&td->td_startcopy, &newtd->td_startcopy,
234 __rangeof(struct thread, td_startcopy, td_endcopy));
235 newtd->td_proc = td->td_proc;
236 newtd->td_ucred = crhold(td->td_ucred);
237 newtd->td_dbg_sc_code = td->td_dbg_sc_code;
238 newtd->td_dbg_sc_narg = td->td_dbg_sc_narg;
239
240 error = initialize_thread(newtd, thunk);
241 if (error != 0) {
242 thread_free(newtd);
243 crfree(td->td_ucred);
244 goto fail;
245 }
246
247 PROC_LOCK(td->td_proc);
248 td->td_proc->p_flag |= P_HADTHREADS;
249 thread_link(newtd, p);
250 bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
251 thread_lock(td);
252 /* let the scheduler know about these things. */
253 sched_fork_thread(td, newtd);
254 thread_unlock(td);
255 if (P_SHOULDSTOP(p))
256 newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
257 if (p->p_ptevents & PTRACE_LWP)
258 newtd->td_dbgflags |= TDB_BORN;
259 PROC_UNLOCK(p);
260
261 tidhash_add(newtd);
262
263 thread_lock(newtd);
264 if (rtp != NULL) {
265 if (!(td->td_pri_class == PRI_TIMESHARE &&
266 rtp->type == RTP_PRIO_NORMAL)) {
267 rtp_to_pri(rtp, newtd);
268 sched_prio(newtd, newtd->td_user_pri);
269 } /* ignore timesharing class */
270 }
271 TD_SET_CAN_RUN(newtd);
272 sched_add(newtd, SRQ_BORING);
273 thread_unlock(newtd);
274
275 return (0);
276
277 fail:
278 #ifdef RACCT
279 if (racct_enable) {
280 PROC_LOCK(p);
281 racct_sub(p, RACCT_NTHR, 1);
282 PROC_UNLOCK(p);
283 }
284 #endif
285 return (error);
286 }
287
288 int
sys_thr_self(struct thread * td,struct thr_self_args * uap)289 sys_thr_self(struct thread *td, struct thr_self_args *uap)
290 /* long *id */
291 {
292 int error;
293
294 error = suword_lwpid(uap->id, (unsigned)td->td_tid);
295 if (error == -1)
296 return (EFAULT);
297 return (0);
298 }
299
300 int
sys_thr_exit(struct thread * td,struct thr_exit_args * uap)301 sys_thr_exit(struct thread *td, struct thr_exit_args *uap)
302 /* long *state */
303 {
304
305 /* Signal userland that it can free the stack. */
306 if ((void *)uap->state != NULL) {
307 suword_lwpid(uap->state, 1);
308 kern_umtx_wake(td, uap->state, INT_MAX, 0);
309 }
310
311 return (kern_thr_exit(td));
312 }
313
314 int
kern_thr_exit(struct thread * td)315 kern_thr_exit(struct thread *td)
316 {
317 struct proc *p;
318
319 p = td->td_proc;
320
321 /*
322 * If all of the threads in a process call this routine to
323 * exit (e.g. all threads call pthread_exit()), exactly one
324 * thread should return to the caller to terminate the process
325 * instead of the thread.
326 *
327 * Checking p_numthreads alone is not sufficient since threads
328 * might be committed to terminating while the PROC_LOCK is
329 * dropped in either ptracestop() or while removing this thread
330 * from the tidhash. Instead, the p_pendingexits field holds
331 * the count of threads in either of those states and a thread
332 * is considered the "last" thread if all of the other threads
333 * in a process are already terminating.
334 */
335 PROC_LOCK(p);
336 if (p->p_numthreads == p->p_pendingexits + 1) {
337 /*
338 * Ignore attempts to shut down last thread in the
339 * proc. This will actually call _exit(2) in the
340 * usermode trampoline when it returns.
341 */
342 PROC_UNLOCK(p);
343 return (0);
344 }
345
346 p->p_pendingexits++;
347 td->td_dbgflags |= TDB_EXIT;
348 if (p->p_ptevents & PTRACE_LWP)
349 ptracestop(td, SIGTRAP, NULL);
350 PROC_UNLOCK(p);
351 tidhash_remove(td);
352 PROC_LOCK(p);
353 p->p_pendingexits--;
354
355 /*
356 * The check above should prevent all other threads from this
357 * process from exiting while the PROC_LOCK is dropped, so
358 * there must be at least one other thread other than the
359 * current thread.
360 */
361 KASSERT(p->p_numthreads > 1, ("too few threads"));
362 racct_sub(p, RACCT_NTHR, 1);
363 tdsigcleanup(td);
364 umtx_thread_exit(td);
365
366 #ifdef AUDIT
367 AUDIT_SYSCALL_EXIT(0, td);
368 #endif
369
370 PROC_SLOCK(p);
371 thread_stopped(p);
372 thread_exit();
373 /* NOTREACHED */
374 }
375
376 int
sys_thr_kill(struct thread * td,struct thr_kill_args * uap)377 sys_thr_kill(struct thread *td, struct thr_kill_args *uap)
378 /* long id, int sig */
379 {
380 ksiginfo_t ksi;
381 struct thread *ttd;
382 struct proc *p;
383 int error;
384
385 p = td->td_proc;
386 ksiginfo_init(&ksi);
387 ksi.ksi_signo = uap->sig;
388 ksi.ksi_code = SI_LWP;
389 ksi.ksi_pid = p->p_pid;
390 ksi.ksi_uid = td->td_ucred->cr_ruid;
391 if (uap->id == -1) {
392 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
393 error = EINVAL;
394 } else {
395 error = ESRCH;
396 PROC_LOCK(p);
397 FOREACH_THREAD_IN_PROC(p, ttd) {
398 if (ttd != td) {
399 error = 0;
400 if (uap->sig == 0)
401 break;
402 tdksignal(ttd, uap->sig, &ksi);
403 }
404 }
405 PROC_UNLOCK(p);
406 }
407 } else {
408 error = 0;
409 ttd = tdfind((lwpid_t)uap->id, p->p_pid);
410 if (ttd == NULL)
411 return (ESRCH);
412 if (uap->sig == 0)
413 ;
414 else if (!_SIG_VALID(uap->sig))
415 error = EINVAL;
416 else
417 tdksignal(ttd, uap->sig, &ksi);
418 PROC_UNLOCK(ttd->td_proc);
419 }
420 return (error);
421 }
422
423 int
sys_thr_kill2(struct thread * td,struct thr_kill2_args * uap)424 sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap)
425 /* pid_t pid, long id, int sig */
426 {
427 ksiginfo_t ksi;
428 struct thread *ttd;
429 struct proc *p;
430 int error;
431
432 AUDIT_ARG_SIGNUM(uap->sig);
433
434 ksiginfo_init(&ksi);
435 ksi.ksi_signo = uap->sig;
436 ksi.ksi_code = SI_LWP;
437 ksi.ksi_pid = td->td_proc->p_pid;
438 ksi.ksi_uid = td->td_ucred->cr_ruid;
439 if (uap->id == -1) {
440 if ((p = pfind(uap->pid)) == NULL)
441 return (ESRCH);
442 AUDIT_ARG_PROCESS(p);
443 error = p_cansignal(td, p, uap->sig);
444 if (error) {
445 PROC_UNLOCK(p);
446 return (error);
447 }
448 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
449 error = EINVAL;
450 } else {
451 error = ESRCH;
452 FOREACH_THREAD_IN_PROC(p, ttd) {
453 if (ttd != td) {
454 error = 0;
455 if (uap->sig == 0)
456 break;
457 tdksignal(ttd, uap->sig, &ksi);
458 }
459 }
460 }
461 PROC_UNLOCK(p);
462 } else {
463 ttd = tdfind((lwpid_t)uap->id, uap->pid);
464 if (ttd == NULL)
465 return (ESRCH);
466 p = ttd->td_proc;
467 AUDIT_ARG_PROCESS(p);
468 error = p_cansignal(td, p, uap->sig);
469 if (uap->sig == 0)
470 ;
471 else if (!_SIG_VALID(uap->sig))
472 error = EINVAL;
473 else
474 tdksignal(ttd, uap->sig, &ksi);
475 PROC_UNLOCK(p);
476 }
477 return (error);
478 }
479
480 int
sys_thr_suspend(struct thread * td,struct thr_suspend_args * uap)481 sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap)
482 /* const struct timespec *timeout */
483 {
484 struct timespec ts, *tsp;
485 int error;
486
487 tsp = NULL;
488 if (uap->timeout != NULL) {
489 error = umtx_copyin_timeout(uap->timeout, &ts);
490 if (error != 0)
491 return (error);
492 tsp = &ts;
493 }
494
495 return (kern_thr_suspend(td, tsp));
496 }
497
498 int
kern_thr_suspend(struct thread * td,struct timespec * tsp)499 kern_thr_suspend(struct thread *td, struct timespec *tsp)
500 {
501 struct proc *p = td->td_proc;
502 struct timeval tv;
503 int error = 0;
504 int timo = 0;
505
506 if (td->td_pflags & TDP_WAKEUP) {
507 td->td_pflags &= ~TDP_WAKEUP;
508 return (0);
509 }
510
511 if (tsp != NULL) {
512 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
513 error = EWOULDBLOCK;
514 else {
515 TIMESPEC_TO_TIMEVAL(&tv, tsp);
516 timo = tvtohz(&tv);
517 }
518 }
519
520 PROC_LOCK(p);
521 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0)
522 error = msleep((void *)td, &p->p_mtx,
523 PCATCH, "lthr", timo);
524
525 if (td->td_flags & TDF_THRWAKEUP) {
526 thread_lock(td);
527 td->td_flags &= ~TDF_THRWAKEUP;
528 thread_unlock(td);
529 PROC_UNLOCK(p);
530 return (0);
531 }
532 PROC_UNLOCK(p);
533 if (error == EWOULDBLOCK)
534 error = ETIMEDOUT;
535 else if (error == ERESTART) {
536 if (timo != 0)
537 error = EINTR;
538 }
539 return (error);
540 }
541
542 int
sys_thr_wake(struct thread * td,struct thr_wake_args * uap)543 sys_thr_wake(struct thread *td, struct thr_wake_args *uap)
544 /* long id */
545 {
546 struct proc *p;
547 struct thread *ttd;
548
549 if (uap->id == td->td_tid) {
550 td->td_pflags |= TDP_WAKEUP;
551 return (0);
552 }
553
554 p = td->td_proc;
555 ttd = tdfind((lwpid_t)uap->id, p->p_pid);
556 if (ttd == NULL)
557 return (ESRCH);
558 thread_lock(ttd);
559 ttd->td_flags |= TDF_THRWAKEUP;
560 thread_unlock(ttd);
561 wakeup((void *)ttd);
562 PROC_UNLOCK(p);
563 return (0);
564 }
565
566 int
sys_thr_set_name(struct thread * td,struct thr_set_name_args * uap)567 sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap)
568 {
569 struct proc *p;
570 char name[MAXCOMLEN + 1];
571 struct thread *ttd;
572 int error;
573
574 error = 0;
575 name[0] = '\0';
576 if (uap->name != NULL) {
577 error = copyinstr(uap->name, name, sizeof(name), NULL);
578 if (error == ENAMETOOLONG) {
579 error = copyin(uap->name, name, sizeof(name) - 1);
580 name[sizeof(name) - 1] = '\0';
581 }
582 if (error)
583 return (error);
584 }
585 p = td->td_proc;
586 ttd = tdfind((lwpid_t)uap->id, p->p_pid);
587 if (ttd == NULL)
588 return (ESRCH);
589 strcpy(ttd->td_name, name);
590 #ifdef KTR
591 sched_clear_tdname(ttd);
592 #endif
593 PROC_UNLOCK(p);
594 return (error);
595 }
596
597 int
kern_thr_alloc(struct proc * p,int pages,struct thread ** ntd)598 kern_thr_alloc(struct proc *p, int pages, struct thread **ntd)
599 {
600
601 /* Have race condition but it is cheap. */
602 if (p->p_numthreads >= max_threads_per_proc) {
603 ++max_threads_hits;
604 return (EPROCLIM);
605 }
606
607 *ntd = thread_alloc(pages);
608 if (*ntd == NULL)
609 return (ENOMEM);
610
611 return (0);
612 }
613