1 /*-
2 * Copyright (C) 2001 Julian Elischer <julian@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(s), this list of conditions and the following disclaimer as
10 * the first lines of this file unmodified other than the possible
11 * addition of one or more copyright notices.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice(s), this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
17 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
20 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
23 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
26 * DAMAGE.
27 */
28
29 #include "opt_witness.h"
30 #include "opt_kdtrace.h"
31 #include "opt_hwpmc_hooks.h"
32
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD: releng/10.4/sys/kern/kern_thread.c 315837 2017-03-23 08:02:29Z avg $");
35
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>
42 #include <sys/rangelock.h>
43 #include <sys/resourcevar.h>
44 #include <sys/sdt.h>
45 #include <sys/smp.h>
46 #include <sys/sched.h>
47 #include <sys/sleepqueue.h>
48 #include <sys/selinfo.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysent.h>
51 #include <sys/turnstile.h>
52 #include <sys/ktr.h>
53 #include <sys/rwlock.h>
54 #include <sys/umtx.h>
55 #include <sys/cpuset.h>
56 #ifdef HWPMC_HOOKS
57 #include <sys/pmckern.h>
58 #endif
59
60 #include <security/audit/audit.h>
61
62 #include <vm/vm.h>
63 #include <vm/vm_extern.h>
64 #include <vm/uma.h>
65 #include <sys/eventhandler.h>
66
67 SDT_PROVIDER_DECLARE(proc);
68 SDT_PROBE_DEFINE(proc, , , lwp__exit);
69
70 /*
71 * thread related storage.
72 */
73 static uma_zone_t thread_zone;
74
75 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
76 static struct mtx zombie_lock;
77 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
78
79 static void thread_zombie(struct thread *);
80 static int thread_unsuspend_one(struct thread *td, struct proc *p,
81 bool boundary);
82
83 #define TID_BUFFER_SIZE 1024
84
85 struct mtx tid_lock;
86 static struct unrhdr *tid_unrhdr;
87 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
88 static int tid_head, tid_tail;
89 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
90
91 struct tidhashhead *tidhashtbl;
92 u_long tidhash;
93 struct rwlock tidhash_lock;
94
95 static lwpid_t
96 tid_alloc(void)
97 {
98 lwpid_t tid;
99
100 tid = alloc_unr(tid_unrhdr);
101 if (tid != -1)
102 return (tid);
103 mtx_lock(&tid_lock);
104 if (tid_head == tid_tail) {
105 mtx_unlock(&tid_lock);
106 return (-1);
107 }
108 tid = tid_buffer[tid_head];
109 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
110 mtx_unlock(&tid_lock);
111 return (tid);
112 }
113
114 static void
115 tid_free(lwpid_t tid)
116 {
117 lwpid_t tmp_tid = -1;
118
119 mtx_lock(&tid_lock);
120 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
121 tmp_tid = tid_buffer[tid_head];
122 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
123 }
124 tid_buffer[tid_tail] = tid;
125 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
126 mtx_unlock(&tid_lock);
127 if (tmp_tid != -1)
128 free_unr(tid_unrhdr, tmp_tid);
129 }
130
131 /*
132 * Prepare a thread for use.
133 */
134 static int
135 thread_ctor(void *mem, int size, void *arg, int flags)
136 {
137 struct thread *td;
138
139 td = (struct thread *)mem;
140 td->td_state = TDS_INACTIVE;
141 td->td_oncpu = NOCPU;
142
143 td->td_tid = tid_alloc();
144
145 /*
146 * Note that td_critnest begins life as 1 because the thread is not
147 * running and is thereby implicitly waiting to be on the receiving
148 * end of a context switch.
149 */
150 td->td_critnest = 1;
151 td->td_lend_user_pri = PRI_MAX;
152 EVENTHANDLER_INVOKE(thread_ctor, td);
153 #ifdef AUDIT
154 audit_thread_alloc(td);
155 #endif
156 umtx_thread_alloc(td);
157 return (0);
158 }
159
160 /*
161 * Reclaim a thread after use.
162 */
163 static void
164 thread_dtor(void *mem, int size, void *arg)
165 {
166 struct thread *td;
167
168 td = (struct thread *)mem;
169
170 #ifdef INVARIANTS
171 /* Verify that this thread is in a safe state to free. */
172 switch (td->td_state) {
173 case TDS_INHIBITED:
174 case TDS_RUNNING:
175 case TDS_CAN_RUN:
176 case TDS_RUNQ:
177 /*
178 * We must never unlink a thread that is in one of
179 * these states, because it is currently active.
180 */
181 panic("bad state for thread unlinking");
182 /* NOTREACHED */
183 case TDS_INACTIVE:
184 break;
185 default:
186 panic("bad thread state");
187 /* NOTREACHED */
188 }
189 #endif
190 #ifdef AUDIT
191 audit_thread_free(td);
192 #endif
193 /* Free all OSD associated to this thread. */
194 osd_thread_exit(td);
195
196 EVENTHANDLER_INVOKE(thread_dtor, td);
197 tid_free(td->td_tid);
198 }
199
200 /*
201 * Initialize type-stable parts of a thread (when newly created).
202 */
203 static int
204 thread_init(void *mem, int size, int flags)
205 {
206 struct thread *td;
207
208 td = (struct thread *)mem;
209
210 td->td_sleepqueue = sleepq_alloc();
211 td->td_turnstile = turnstile_alloc();
212 td->td_rlqe = NULL;
213 EVENTHANDLER_INVOKE(thread_init, td);
214 td->td_sched = (struct td_sched *)&td[1];
215 umtx_thread_init(td);
216 td->td_kstack = 0;
217 td->td_sel = NULL;
218 return (0);
219 }
220
221 /*
222 * Tear down type-stable parts of a thread (just before being discarded).
223 */
224 static void
225 thread_fini(void *mem, int size)
226 {
227 struct thread *td;
228
229 td = (struct thread *)mem;
230 EVENTHANDLER_INVOKE(thread_fini, td);
231 rlqentry_free(td->td_rlqe);
232 turnstile_free(td->td_turnstile);
233 sleepq_free(td->td_sleepqueue);
234 umtx_thread_fini(td);
235 seltdfini(td);
236 }
237
238 /*
239 * For a newly created process,
240 * link up all the structures and its initial threads etc.
241 * called from:
242 * {arch}/{arch}/machdep.c ia64_init(), init386() etc.
243 * proc_dtor() (should go away)
244 * proc_init()
245 */
246 void
247 proc_linkup0(struct proc *p, struct thread *td)
248 {
249 TAILQ_INIT(&p->p_threads); /* all threads in proc */
250 proc_linkup(p, td);
251 }
252
253 void
254 proc_linkup(struct proc *p, struct thread *td)
255 {
256
257 sigqueue_init(&p->p_sigqueue, p);
258 p->p_ksi = ksiginfo_alloc(1);
259 if (p->p_ksi != NULL) {
260 /* XXX p_ksi may be null if ksiginfo zone is not ready */
261 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
262 }
263 LIST_INIT(&p->p_mqnotifier);
264 p->p_numthreads = 0;
265 thread_link(td, p);
266 }
267
268 /*
269 * Initialize global thread allocation resources.
270 */
271 void
272 threadinit(void)
273 {
274
275 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
276
277 /*
278 * pid_max cannot be greater than PID_MAX.
279 * leave one number for thread0.
280 */
281 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
282
283 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
284 thread_ctor, thread_dtor, thread_init, thread_fini,
285 16 - 1, UMA_ZONE_NOFREE);
286 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
287 rw_init(&tidhash_lock, "tidhash");
288 }
289
290 /*
291 * Place an unused thread on the zombie list.
292 * Use the slpq as that must be unused by now.
293 */
294 void
295 thread_zombie(struct thread *td)
296 {
297 mtx_lock_spin(&zombie_lock);
298 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
299 mtx_unlock_spin(&zombie_lock);
300 }
301
302 /*
303 * Release a thread that has exited after cpu_throw().
304 */
305 void
306 thread_stash(struct thread *td)
307 {
308 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
309 thread_zombie(td);
310 }
311
312 /*
313 * Reap zombie resources.
314 */
315 void
316 thread_reap(void)
317 {
318 struct thread *td_first, *td_next;
319
320 /*
321 * Don't even bother to lock if none at this instant,
322 * we really don't care about the next instant.
323 */
324 if (!TAILQ_EMPTY(&zombie_threads)) {
325 mtx_lock_spin(&zombie_lock);
326 td_first = TAILQ_FIRST(&zombie_threads);
327 if (td_first)
328 TAILQ_INIT(&zombie_threads);
329 mtx_unlock_spin(&zombie_lock);
330 while (td_first) {
331 td_next = TAILQ_NEXT(td_first, td_slpq);
332 if (td_first->td_ucred)
333 crfree(td_first->td_ucred);
334 thread_free(td_first);
335 td_first = td_next;
336 }
337 }
338 }
339
340 /*
341 * Allocate a thread.
342 */
343 struct thread *
344 thread_alloc(int pages)
345 {
346 struct thread *td;
347
348 thread_reap(); /* check if any zombies to get */
349
350 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
351 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
352 if (!vm_thread_new(td, pages)) {
353 uma_zfree(thread_zone, td);
354 return (NULL);
355 }
356 cpu_thread_alloc(td);
357 return (td);
358 }
359
360 int
361 thread_alloc_stack(struct thread *td, int pages)
362 {
363
364 KASSERT(td->td_kstack == 0,
365 ("thread_alloc_stack called on a thread with kstack"));
366 if (!vm_thread_new(td, pages))
367 return (0);
368 cpu_thread_alloc(td);
369 return (1);
370 }
371
372 /*
373 * Deallocate a thread.
374 */
375 void
376 thread_free(struct thread *td)
377 {
378
379 lock_profile_thread_exit(td);
380 if (td->td_cpuset)
381 cpuset_rel(td->td_cpuset);
382 td->td_cpuset = NULL;
383 cpu_thread_free(td);
384 if (td->td_kstack != 0)
385 vm_thread_dispose(td);
386 callout_drain(&td->td_slpcallout);
387 uma_zfree(thread_zone, td);
388 }
389
390 /*
391 * Discard the current thread and exit from its context.
392 * Always called with scheduler locked.
393 *
394 * Because we can't free a thread while we're operating under its context,
395 * push the current thread into our CPU's deadthread holder. This means
396 * we needn't worry about someone else grabbing our context before we
397 * do a cpu_throw().
398 */
399 void
400 thread_exit(void)
401 {
402 uint64_t runtime, new_switchtime;
403 struct thread *td;
404 struct thread *td2;
405 struct proc *p;
406 int wakeup_swapper;
407
408 td = curthread;
409 p = td->td_proc;
410
411 PROC_SLOCK_ASSERT(p, MA_OWNED);
412 mtx_assert(&Giant, MA_NOTOWNED);
413
414 PROC_LOCK_ASSERT(p, MA_OWNED);
415 KASSERT(p != NULL, ("thread exiting without a process"));
416 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
417 (long)p->p_pid, td->td_name);
418 SDT_PROBE0(proc, , , lwp__exit);
419 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
420
421 #ifdef AUDIT
422 AUDIT_SYSCALL_EXIT(0, td);
423 #endif
424 /*
425 * drop FPU & debug register state storage, or any other
426 * architecture specific resources that
427 * would not be on a new untouched process.
428 */
429 cpu_thread_exit(td);
430
431 /*
432 * The last thread is left attached to the process
433 * So that the whole bundle gets recycled. Skip
434 * all this stuff if we never had threads.
435 * EXIT clears all sign of other threads when
436 * it goes to single threading, so the last thread always
437 * takes the short path.
438 */
439 if (p->p_flag & P_HADTHREADS) {
440 if (p->p_numthreads > 1) {
441 atomic_add_int(&td->td_proc->p_exitthreads, 1);
442 thread_unlink(td);
443 td2 = FIRST_THREAD_IN_PROC(p);
444 sched_exit_thread(td2, td);
445
446 /*
447 * The test below is NOT true if we are the
448 * sole exiting thread. P_STOPPED_SINGLE is unset
449 * in exit1() after it is the only survivor.
450 */
451 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
452 if (p->p_numthreads == p->p_suspcount) {
453 thread_lock(p->p_singlethread);
454 wakeup_swapper = thread_unsuspend_one(
455 p->p_singlethread, p, false);
456 thread_unlock(p->p_singlethread);
457 if (wakeup_swapper)
458 kick_proc0();
459 }
460 }
461
462 PCPU_SET(deadthread, td);
463 } else {
464 /*
465 * The last thread is exiting.. but not through exit()
466 */
467 panic ("thread_exit: Last thread exiting on its own");
468 }
469 }
470 #ifdef HWPMC_HOOKS
471 /*
472 * If this thread is part of a process that is being tracked by hwpmc(4),
473 * inform the module of the thread's impending exit.
474 */
475 if (PMC_PROC_IS_USING_PMCS(td->td_proc))
476 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
477 #endif
478 PROC_UNLOCK(p);
479 PROC_STATLOCK(p);
480 thread_lock(td);
481 PROC_SUNLOCK(p);
482
483 /* Do the same timestamp bookkeeping that mi_switch() would do. */
484 new_switchtime = cpu_ticks();
485 runtime = new_switchtime - PCPU_GET(switchtime);
486 td->td_runtime += runtime;
487 td->td_incruntime += runtime;
488 PCPU_SET(switchtime, new_switchtime);
489 PCPU_SET(switchticks, ticks);
490 PCPU_INC(cnt.v_swtch);
491
492 /* Save our resource usage in our process. */
493 td->td_ru.ru_nvcsw++;
494 ruxagg(p, td);
495 rucollect(&p->p_ru, &td->td_ru);
496 PROC_STATUNLOCK(p);
497
498 td->td_state = TDS_INACTIVE;
499 #ifdef WITNESS
500 witness_thread_exit(td);
501 #endif
502 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
503 sched_throw(td);
504 panic("I'm a teapot!");
505 /* NOTREACHED */
506 }
507
508 /*
509 * Do any thread specific cleanups that may be needed in wait()
510 * called with Giant, proc and schedlock not held.
511 */
512 void
513 thread_wait(struct proc *p)
514 {
515 struct thread *td;
516
517 mtx_assert(&Giant, MA_NOTOWNED);
518 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
519 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
520 td = FIRST_THREAD_IN_PROC(p);
521 /* Lock the last thread so we spin until it exits cpu_throw(). */
522 thread_lock(td);
523 thread_unlock(td);
524 lock_profile_thread_exit(td);
525 cpuset_rel(td->td_cpuset);
526 td->td_cpuset = NULL;
527 cpu_thread_clean(td);
528 crfree(td->td_ucred);
529 callout_drain(&td->td_slpcallout);
530 thread_reap(); /* check for zombie threads etc. */
531 }
532
533 /*
534 * Link a thread to a process.
535 * set up anything that needs to be initialized for it to
536 * be used by the process.
537 */
538 void
539 thread_link(struct thread *td, struct proc *p)
540 {
541
542 /*
543 * XXX This can't be enabled because it's called for proc0 before
544 * its lock has been created.
545 * PROC_LOCK_ASSERT(p, MA_OWNED);
546 */
547 td->td_state = TDS_INACTIVE;
548 td->td_proc = p;
549 td->td_flags = TDF_INMEM;
550
551 LIST_INIT(&td->td_contested);
552 LIST_INIT(&td->td_lprof[0]);
553 LIST_INIT(&td->td_lprof[1]);
554 sigqueue_init(&td->td_sigqueue, p);
555 callout_init(&td->td_slpcallout, 1);
556 TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
557 p->p_numthreads++;
558 }
559
560 /*
561 * Called from:
562 * thread_exit()
563 */
564 void
565 thread_unlink(struct thread *td)
566 {
567 struct proc *p = td->td_proc;
568
569 PROC_LOCK_ASSERT(p, MA_OWNED);
570 TAILQ_REMOVE(&p->p_threads, td, td_plist);
571 p->p_numthreads--;
572 /* could clear a few other things here */
573 /* Must NOT clear links to proc! */
574 }
575
576 static int
577 calc_remaining(struct proc *p, int mode)
578 {
579 int remaining;
580
581 PROC_LOCK_ASSERT(p, MA_OWNED);
582 PROC_SLOCK_ASSERT(p, MA_OWNED);
583 if (mode == SINGLE_EXIT)
584 remaining = p->p_numthreads;
585 else if (mode == SINGLE_BOUNDARY)
586 remaining = p->p_numthreads - p->p_boundary_count;
587 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
588 remaining = p->p_numthreads - p->p_suspcount;
589 else
590 panic("calc_remaining: wrong mode %d", mode);
591 return (remaining);
592 }
593
594 static int
595 remain_for_mode(int mode)
596 {
597
598 return (mode == SINGLE_ALLPROC ? 0 : 1);
599 }
600
601 static int
602 weed_inhib(int mode, struct thread *td2, struct proc *p)
603 {
604 int wakeup_swapper;
605
606 PROC_LOCK_ASSERT(p, MA_OWNED);
607 PROC_SLOCK_ASSERT(p, MA_OWNED);
608 THREAD_LOCK_ASSERT(td2, MA_OWNED);
609
610 wakeup_swapper = 0;
611 switch (mode) {
612 case SINGLE_EXIT:
613 if (TD_IS_SUSPENDED(td2))
614 wakeup_swapper |= thread_unsuspend_one(td2, p, true);
615 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
616 wakeup_swapper |= sleepq_abort(td2, EINTR);
617 break;
618 case SINGLE_BOUNDARY:
619 case SINGLE_NO_EXIT:
620 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0)
621 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
622 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
623 wakeup_swapper |= sleepq_abort(td2, ERESTART);
624 break;
625 case SINGLE_ALLPROC:
626 /*
627 * ALLPROC suspend tries to avoid spurious EINTR for
628 * threads sleeping interruptable, by suspending the
629 * thread directly, similarly to sig_suspend_threads().
630 * Since such sleep is not performed at the user
631 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
632 * is used to avoid immediate un-suspend.
633 */
634 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
635 TDF_ALLPROCSUSP)) == 0)
636 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
637 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) {
638 if ((td2->td_flags & TDF_SBDRY) == 0) {
639 thread_suspend_one(td2);
640 td2->td_flags |= TDF_ALLPROCSUSP;
641 } else {
642 wakeup_swapper |= sleepq_abort(td2, ERESTART);
643 }
644 }
645 break;
646 }
647 return (wakeup_swapper);
648 }
649
650 /*
651 * Enforce single-threading.
652 *
653 * Returns 1 if the caller must abort (another thread is waiting to
654 * exit the process or similar). Process is locked!
655 * Returns 0 when you are successfully the only thread running.
656 * A process has successfully single threaded in the suspend mode when
657 * There are no threads in user mode. Threads in the kernel must be
658 * allowed to continue until they get to the user boundary. They may even
659 * copy out their return values and data before suspending. They may however be
660 * accelerated in reaching the user boundary as we will wake up
661 * any sleeping threads that are interruptable. (PCATCH).
662 */
663 int
664 thread_single(struct proc *p, int mode)
665 {
666 struct thread *td;
667 struct thread *td2;
668 int remaining, wakeup_swapper;
669
670 td = curthread;
671 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
672 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
673 ("invalid mode %d", mode));
674 /*
675 * If allowing non-ALLPROC singlethreading for non-curproc
676 * callers, calc_remaining() and remain_for_mode() should be
677 * adjusted to also account for td->td_proc != p. For now
678 * this is not implemented because it is not used.
679 */
680 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
681 (mode != SINGLE_ALLPROC && td->td_proc == p),
682 ("mode %d proc %p curproc %p", mode, p, td->td_proc));
683 mtx_assert(&Giant, MA_NOTOWNED);
684 PROC_LOCK_ASSERT(p, MA_OWNED);
685
686 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
687 return (0);
688
689 /* Is someone already single threading? */
690 if (p->p_singlethread != NULL && p->p_singlethread != td)
691 return (1);
692
693 if (mode == SINGLE_EXIT) {
694 p->p_flag |= P_SINGLE_EXIT;
695 p->p_flag &= ~P_SINGLE_BOUNDARY;
696 } else {
697 p->p_flag &= ~P_SINGLE_EXIT;
698 if (mode == SINGLE_BOUNDARY)
699 p->p_flag |= P_SINGLE_BOUNDARY;
700 else
701 p->p_flag &= ~P_SINGLE_BOUNDARY;
702 }
703 if (mode == SINGLE_ALLPROC)
704 p->p_flag |= P_TOTAL_STOP;
705 p->p_flag |= P_STOPPED_SINGLE;
706 PROC_SLOCK(p);
707 p->p_singlethread = td;
708 remaining = calc_remaining(p, mode);
709 while (remaining != remain_for_mode(mode)) {
710 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
711 goto stopme;
712 wakeup_swapper = 0;
713 FOREACH_THREAD_IN_PROC(p, td2) {
714 if (td2 == td)
715 continue;
716 thread_lock(td2);
717 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
718 if (TD_IS_INHIBITED(td2)) {
719 wakeup_swapper |= weed_inhib(mode, td2, p);
720 #ifdef SMP
721 } else if (TD_IS_RUNNING(td2) && td != td2) {
722 forward_signal(td2);
723 #endif
724 }
725 thread_unlock(td2);
726 }
727 if (wakeup_swapper)
728 kick_proc0();
729 remaining = calc_remaining(p, mode);
730
731 /*
732 * Maybe we suspended some threads.. was it enough?
733 */
734 if (remaining == remain_for_mode(mode))
735 break;
736
737 stopme:
738 /*
739 * Wake us up when everyone else has suspended.
740 * In the mean time we suspend as well.
741 */
742 thread_suspend_switch(td, p);
743 remaining = calc_remaining(p, mode);
744 }
745 if (mode == SINGLE_EXIT) {
746 /*
747 * Convert the process to an unthreaded process. The
748 * SINGLE_EXIT is called by exit1() or execve(), in
749 * both cases other threads must be retired.
750 */
751 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
752 p->p_singlethread = NULL;
753 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
754
755 /*
756 * Wait for any remaining threads to exit cpu_throw().
757 */
758 while (p->p_exitthreads != 0) {
759 PROC_SUNLOCK(p);
760 PROC_UNLOCK(p);
761 sched_relinquish(td);
762 PROC_LOCK(p);
763 PROC_SLOCK(p);
764 }
765 } else if (mode == SINGLE_BOUNDARY) {
766 /*
767 * Wait until all suspended threads are removed from
768 * the processors. The thread_suspend_check()
769 * increments p_boundary_count while it is still
770 * running, which makes it possible for the execve()
771 * to destroy vmspace while our other threads are
772 * still using the address space.
773 *
774 * We lock the thread, which is only allowed to
775 * succeed after context switch code finished using
776 * the address space.
777 */
778 FOREACH_THREAD_IN_PROC(p, td2) {
779 if (td2 == td)
780 continue;
781 thread_lock(td2);
782 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
783 ("td %p not on boundary", td2));
784 KASSERT(TD_IS_SUSPENDED(td2),
785 ("td %p is not suspended", td2));
786 thread_unlock(td2);
787 }
788 }
789 PROC_SUNLOCK(p);
790 return (0);
791 }
792
793 bool
794 thread_suspend_check_needed(void)
795 {
796 struct proc *p;
797 struct thread *td;
798
799 td = curthread;
800 p = td->td_proc;
801 PROC_LOCK_ASSERT(p, MA_OWNED);
802 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
803 (td->td_dbgflags & TDB_SUSPEND) != 0));
804 }
805
806 /*
807 * Called in from locations that can safely check to see
808 * whether we have to suspend or at least throttle for a
809 * single-thread event (e.g. fork).
810 *
811 * Such locations include userret().
812 * If the "return_instead" argument is non zero, the thread must be able to
813 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
814 *
815 * The 'return_instead' argument tells the function if it may do a
816 * thread_exit() or suspend, or whether the caller must abort and back
817 * out instead.
818 *
819 * If the thread that set the single_threading request has set the
820 * P_SINGLE_EXIT bit in the process flags then this call will never return
821 * if 'return_instead' is false, but will exit.
822 *
823 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
824 *---------------+--------------------+---------------------
825 * 0 | returns 0 | returns 0 or 1
826 * | when ST ends | immediately
827 *---------------+--------------------+---------------------
828 * 1 | thread exits | returns 1
829 * | | immediately
830 * 0 = thread_exit() or suspension ok,
831 * other = return error instead of stopping the thread.
832 *
833 * While a full suspension is under effect, even a single threading
834 * thread would be suspended if it made this call (but it shouldn't).
835 * This call should only be made from places where
836 * thread_exit() would be safe as that may be the outcome unless
837 * return_instead is set.
838 */
839 int
840 thread_suspend_check(int return_instead)
841 {
842 struct thread *td;
843 struct proc *p;
844 int wakeup_swapper;
845
846 td = curthread;
847 p = td->td_proc;
848 mtx_assert(&Giant, MA_NOTOWNED);
849 PROC_LOCK_ASSERT(p, MA_OWNED);
850 while (thread_suspend_check_needed()) {
851 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
852 KASSERT(p->p_singlethread != NULL,
853 ("singlethread not set"));
854 /*
855 * The only suspension in action is a
856 * single-threading. Single threader need not stop.
857 * It is safe to access p->p_singlethread unlocked
858 * because it can only be set to our address by us.
859 */
860 if (p->p_singlethread == td)
861 return (0); /* Exempt from stopping. */
862 }
863 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
864 return (EINTR);
865
866 /* Should we goto user boundary if we didn't come from there? */
867 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
868 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
869 return (ERESTART);
870
871 /*
872 * Ignore suspend requests if they are deferred.
873 */
874 if ((td->td_flags & TDF_SBDRY) != 0) {
875 KASSERT(return_instead,
876 ("TDF_SBDRY set for unsafe thread_suspend_check"));
877 return (0);
878 }
879
880 /*
881 * If the process is waiting for us to exit,
882 * this thread should just suicide.
883 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
884 */
885 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
886 PROC_UNLOCK(p);
887
888 /*
889 * Allow Linux emulation layer to do some work
890 * before thread suicide.
891 */
892 if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
893 (p->p_sysent->sv_thread_detach)(td);
894 kern_thr_exit(td);
895 panic("stopped thread did not exit");
896 }
897
898 PROC_SLOCK(p);
899 thread_stopped(p);
900 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
901 if (p->p_numthreads == p->p_suspcount + 1) {
902 thread_lock(p->p_singlethread);
903 wakeup_swapper = thread_unsuspend_one(
904 p->p_singlethread, p, false);
905 thread_unlock(p->p_singlethread);
906 if (wakeup_swapper)
907 kick_proc0();
908 }
909 }
910 PROC_UNLOCK(p);
911 thread_lock(td);
912 /*
913 * When a thread suspends, it just
914 * gets taken off all queues.
915 */
916 thread_suspend_one(td);
917 if (return_instead == 0) {
918 p->p_boundary_count++;
919 td->td_flags |= TDF_BOUNDARY;
920 }
921 PROC_SUNLOCK(p);
922 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
923 thread_unlock(td);
924 PROC_LOCK(p);
925 }
926 return (0);
927 }
928
929 void
930 thread_suspend_switch(struct thread *td, struct proc *p)
931 {
932
933 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
934 PROC_LOCK_ASSERT(p, MA_OWNED);
935 PROC_SLOCK_ASSERT(p, MA_OWNED);
936 /*
937 * We implement thread_suspend_one in stages here to avoid
938 * dropping the proc lock while the thread lock is owned.
939 */
940 if (p == td->td_proc) {
941 thread_stopped(p);
942 p->p_suspcount++;
943 }
944 PROC_UNLOCK(p);
945 thread_lock(td);
946 td->td_flags &= ~TDF_NEEDSUSPCHK;
947 TD_SET_SUSPENDED(td);
948 sched_sleep(td, 0);
949 PROC_SUNLOCK(p);
950 DROP_GIANT();
951 mi_switch(SW_VOL | SWT_SUSPEND, NULL);
952 thread_unlock(td);
953 PICKUP_GIANT();
954 PROC_LOCK(p);
955 PROC_SLOCK(p);
956 }
957
958 void
959 thread_suspend_one(struct thread *td)
960 {
961 struct proc *p;
962
963 p = td->td_proc;
964 PROC_SLOCK_ASSERT(p, MA_OWNED);
965 THREAD_LOCK_ASSERT(td, MA_OWNED);
966 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
967 p->p_suspcount++;
968 td->td_flags &= ~TDF_NEEDSUSPCHK;
969 TD_SET_SUSPENDED(td);
970 sched_sleep(td, 0);
971 }
972
973 static int
974 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
975 {
976
977 THREAD_LOCK_ASSERT(td, MA_OWNED);
978 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
979 TD_CLR_SUSPENDED(td);
980 td->td_flags &= ~TDF_ALLPROCSUSP;
981 if (td->td_proc == p) {
982 PROC_SLOCK_ASSERT(p, MA_OWNED);
983 p->p_suspcount--;
984 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
985 td->td_flags &= ~TDF_BOUNDARY;
986 p->p_boundary_count--;
987 }
988 }
989 return (setrunnable(td));
990 }
991
992 /*
993 * Allow all threads blocked by single threading to continue running.
994 */
995 void
996 thread_unsuspend(struct proc *p)
997 {
998 struct thread *td;
999 int wakeup_swapper;
1000
1001 PROC_LOCK_ASSERT(p, MA_OWNED);
1002 PROC_SLOCK_ASSERT(p, MA_OWNED);
1003 wakeup_swapper = 0;
1004 if (!P_SHOULDSTOP(p)) {
1005 FOREACH_THREAD_IN_PROC(p, td) {
1006 thread_lock(td);
1007 if (TD_IS_SUSPENDED(td)) {
1008 wakeup_swapper |= thread_unsuspend_one(td, p,
1009 true);
1010 }
1011 thread_unlock(td);
1012 }
1013 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1014 p->p_numthreads == p->p_suspcount) {
1015 /*
1016 * Stopping everything also did the job for the single
1017 * threading request. Now we've downgraded to single-threaded,
1018 * let it continue.
1019 */
1020 if (p->p_singlethread->td_proc == p) {
1021 thread_lock(p->p_singlethread);
1022 wakeup_swapper = thread_unsuspend_one(
1023 p->p_singlethread, p, false);
1024 thread_unlock(p->p_singlethread);
1025 }
1026 }
1027 if (wakeup_swapper)
1028 kick_proc0();
1029 }
1030
1031 /*
1032 * End the single threading mode..
1033 */
1034 void
1035 thread_single_end(struct proc *p, int mode)
1036 {
1037 struct thread *td;
1038 int wakeup_swapper;
1039
1040 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1041 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1042 ("invalid mode %d", mode));
1043 PROC_LOCK_ASSERT(p, MA_OWNED);
1044 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1045 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1046 ("mode %d does not match P_TOTAL_STOP", mode));
1047 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1048 ("thread_single_end from other thread %p %p",
1049 curthread, p->p_singlethread));
1050 KASSERT(mode != SINGLE_BOUNDARY ||
1051 (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1052 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1053 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1054 P_TOTAL_STOP);
1055 PROC_SLOCK(p);
1056 p->p_singlethread = NULL;
1057 wakeup_swapper = 0;
1058 /*
1059 * If there are other threads they may now run,
1060 * unless of course there is a blanket 'stop order'
1061 * on the process. The single threader must be allowed
1062 * to continue however as this is a bad place to stop.
1063 */
1064 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1065 FOREACH_THREAD_IN_PROC(p, td) {
1066 thread_lock(td);
1067 if (TD_IS_SUSPENDED(td)) {
1068 wakeup_swapper |= thread_unsuspend_one(td, p,
1069 mode == SINGLE_BOUNDARY);
1070 }
1071 thread_unlock(td);
1072 }
1073 }
1074 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1075 ("inconsistent boundary count %d", p->p_boundary_count));
1076 PROC_SUNLOCK(p);
1077 if (wakeup_swapper)
1078 kick_proc0();
1079 }
1080
1081 struct thread *
1082 thread_find(struct proc *p, lwpid_t tid)
1083 {
1084 struct thread *td;
1085
1086 PROC_LOCK_ASSERT(p, MA_OWNED);
1087 FOREACH_THREAD_IN_PROC(p, td) {
1088 if (td->td_tid == tid)
1089 break;
1090 }
1091 return (td);
1092 }
1093
1094 /* Locate a thread by number; return with proc lock held. */
1095 struct thread *
1096 tdfind(lwpid_t tid, pid_t pid)
1097 {
1098 #define RUN_THRESH 16
1099 struct thread *td;
1100 int run = 0;
1101
1102 rw_rlock(&tidhash_lock);
1103 LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1104 if (td->td_tid == tid) {
1105 if (pid != -1 && td->td_proc->p_pid != pid) {
1106 td = NULL;
1107 break;
1108 }
1109 PROC_LOCK(td->td_proc);
1110 if (td->td_proc->p_state == PRS_NEW) {
1111 PROC_UNLOCK(td->td_proc);
1112 td = NULL;
1113 break;
1114 }
1115 if (run > RUN_THRESH) {
1116 if (rw_try_upgrade(&tidhash_lock)) {
1117 LIST_REMOVE(td, td_hash);
1118 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1119 td, td_hash);
1120 rw_wunlock(&tidhash_lock);
1121 return (td);
1122 }
1123 }
1124 break;
1125 }
1126 run++;
1127 }
1128 rw_runlock(&tidhash_lock);
1129 return (td);
1130 }
1131
1132 void
1133 tidhash_add(struct thread *td)
1134 {
1135 rw_wlock(&tidhash_lock);
1136 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1137 rw_wunlock(&tidhash_lock);
1138 }
1139
1140 void
1141 tidhash_remove(struct thread *td)
1142 {
1143 rw_wlock(&tidhash_lock);
1144 LIST_REMOVE(td, td_hash);
1145 rw_wunlock(&tidhash_lock);
1146 }
Cache object: 2d8b32e6aa57f6af4398a08b41dc016d
|