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_hwpmc_hooks.h"
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/11.0/sys/kern/kern_thread.c 302328 2016-07-03 18:19:48Z kib $");
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/rangelock.h>
42 #include <sys/resourcevar.h>
43 #include <sys/sdt.h>
44 #include <sys/smp.h>
45 #include <sys/sched.h>
46 #include <sys/sleepqueue.h>
47 #include <sys/selinfo.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysent.h>
50 #include <sys/turnstile.h>
51 #include <sys/ktr.h>
52 #include <sys/rwlock.h>
53 #include <sys/umtx.h>
54 #include <sys/cpuset.h>
55 #ifdef HWPMC_HOOKS
56 #include <sys/pmckern.h>
57 #endif
58
59 #include <security/audit/audit.h>
60
61 #include <vm/vm.h>
62 #include <vm/vm_extern.h>
63 #include <vm/uma.h>
64 #include <vm/vm_domain.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 umtx_thread_init(td);
215 td->td_kstack = 0;
216 td->td_sel = NULL;
217 return (0);
218 }
219
220 /*
221 * Tear down type-stable parts of a thread (just before being discarded).
222 */
223 static void
224 thread_fini(void *mem, int size)
225 {
226 struct thread *td;
227
228 td = (struct thread *)mem;
229 EVENTHANDLER_INVOKE(thread_fini, td);
230 rlqentry_free(td->td_rlqe);
231 turnstile_free(td->td_turnstile);
232 sleepq_free(td->td_sleepqueue);
233 umtx_thread_fini(td);
234 seltdfini(td);
235 }
236
237 /*
238 * For a newly created process,
239 * link up all the structures and its initial threads etc.
240 * called from:
241 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc.
242 * proc_dtor() (should go away)
243 * proc_init()
244 */
245 void
246 proc_linkup0(struct proc *p, struct thread *td)
247 {
248 TAILQ_INIT(&p->p_threads); /* all threads in proc */
249 proc_linkup(p, td);
250 }
251
252 void
253 proc_linkup(struct proc *p, struct thread *td)
254 {
255
256 sigqueue_init(&p->p_sigqueue, p);
257 p->p_ksi = ksiginfo_alloc(1);
258 if (p->p_ksi != NULL) {
259 /* XXX p_ksi may be null if ksiginfo zone is not ready */
260 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
261 }
262 LIST_INIT(&p->p_mqnotifier);
263 p->p_numthreads = 0;
264 thread_link(td, p);
265 }
266
267 /*
268 * Initialize global thread allocation resources.
269 */
270 void
271 threadinit(void)
272 {
273
274 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
275
276 /*
277 * pid_max cannot be greater than PID_MAX.
278 * leave one number for thread0.
279 */
280 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
281
282 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
283 thread_ctor, thread_dtor, thread_init, thread_fini,
284 16 - 1, UMA_ZONE_NOFREE);
285 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
286 rw_init(&tidhash_lock, "tidhash");
287 }
288
289 /*
290 * Place an unused thread on the zombie list.
291 * Use the slpq as that must be unused by now.
292 */
293 void
294 thread_zombie(struct thread *td)
295 {
296 mtx_lock_spin(&zombie_lock);
297 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
298 mtx_unlock_spin(&zombie_lock);
299 }
300
301 /*
302 * Release a thread that has exited after cpu_throw().
303 */
304 void
305 thread_stash(struct thread *td)
306 {
307 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
308 thread_zombie(td);
309 }
310
311 /*
312 * Reap zombie resources.
313 */
314 void
315 thread_reap(void)
316 {
317 struct thread *td_first, *td_next;
318
319 /*
320 * Don't even bother to lock if none at this instant,
321 * we really don't care about the next instant..
322 */
323 if (!TAILQ_EMPTY(&zombie_threads)) {
324 mtx_lock_spin(&zombie_lock);
325 td_first = TAILQ_FIRST(&zombie_threads);
326 if (td_first)
327 TAILQ_INIT(&zombie_threads);
328 mtx_unlock_spin(&zombie_lock);
329 while (td_first) {
330 td_next = TAILQ_NEXT(td_first, td_slpq);
331 thread_cow_free(td_first);
332 thread_free(td_first);
333 td_first = td_next;
334 }
335 }
336 }
337
338 /*
339 * Allocate a thread.
340 */
341 struct thread *
342 thread_alloc(int pages)
343 {
344 struct thread *td;
345
346 thread_reap(); /* check if any zombies to get */
347
348 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
349 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
350 if (!vm_thread_new(td, pages)) {
351 uma_zfree(thread_zone, td);
352 return (NULL);
353 }
354 cpu_thread_alloc(td);
355 vm_domain_policy_init(&td->td_vm_dom_policy);
356 return (td);
357 }
358
359 int
360 thread_alloc_stack(struct thread *td, int pages)
361 {
362
363 KASSERT(td->td_kstack == 0,
364 ("thread_alloc_stack called on a thread with kstack"));
365 if (!vm_thread_new(td, pages))
366 return (0);
367 cpu_thread_alloc(td);
368 return (1);
369 }
370
371 /*
372 * Deallocate a thread.
373 */
374 void
375 thread_free(struct thread *td)
376 {
377
378 lock_profile_thread_exit(td);
379 if (td->td_cpuset)
380 cpuset_rel(td->td_cpuset);
381 td->td_cpuset = NULL;
382 cpu_thread_free(td);
383 if (td->td_kstack != 0)
384 vm_thread_dispose(td);
385 vm_domain_policy_cleanup(&td->td_vm_dom_policy);
386 uma_zfree(thread_zone, td);
387 }
388
389 void
390 thread_cow_get_proc(struct thread *newtd, struct proc *p)
391 {
392
393 PROC_LOCK_ASSERT(p, MA_OWNED);
394 newtd->td_ucred = crhold(p->p_ucred);
395 newtd->td_limit = lim_hold(p->p_limit);
396 newtd->td_cowgen = p->p_cowgen;
397 }
398
399 void
400 thread_cow_get(struct thread *newtd, struct thread *td)
401 {
402
403 newtd->td_ucred = crhold(td->td_ucred);
404 newtd->td_limit = lim_hold(td->td_limit);
405 newtd->td_cowgen = td->td_cowgen;
406 }
407
408 void
409 thread_cow_free(struct thread *td)
410 {
411
412 if (td->td_ucred != NULL)
413 crfree(td->td_ucred);
414 if (td->td_limit != NULL)
415 lim_free(td->td_limit);
416 }
417
418 void
419 thread_cow_update(struct thread *td)
420 {
421 struct proc *p;
422 struct ucred *oldcred;
423 struct plimit *oldlimit;
424
425 p = td->td_proc;
426 oldcred = NULL;
427 oldlimit = NULL;
428 PROC_LOCK(p);
429 if (td->td_ucred != p->p_ucred) {
430 oldcred = td->td_ucred;
431 td->td_ucred = crhold(p->p_ucred);
432 }
433 if (td->td_limit != p->p_limit) {
434 oldlimit = td->td_limit;
435 td->td_limit = lim_hold(p->p_limit);
436 }
437 td->td_cowgen = p->p_cowgen;
438 PROC_UNLOCK(p);
439 if (oldcred != NULL)
440 crfree(oldcred);
441 if (oldlimit != NULL)
442 lim_free(oldlimit);
443 }
444
445 /*
446 * Discard the current thread and exit from its context.
447 * Always called with scheduler locked.
448 *
449 * Because we can't free a thread while we're operating under its context,
450 * push the current thread into our CPU's deadthread holder. This means
451 * we needn't worry about someone else grabbing our context before we
452 * do a cpu_throw().
453 */
454 void
455 thread_exit(void)
456 {
457 uint64_t runtime, new_switchtime;
458 struct thread *td;
459 struct thread *td2;
460 struct proc *p;
461 int wakeup_swapper;
462
463 td = curthread;
464 p = td->td_proc;
465
466 PROC_SLOCK_ASSERT(p, MA_OWNED);
467 mtx_assert(&Giant, MA_NOTOWNED);
468
469 PROC_LOCK_ASSERT(p, MA_OWNED);
470 KASSERT(p != NULL, ("thread exiting without a process"));
471 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
472 (long)p->p_pid, td->td_name);
473 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
474
475 #ifdef AUDIT
476 AUDIT_SYSCALL_EXIT(0, td);
477 #endif
478 /*
479 * drop FPU & debug register state storage, or any other
480 * architecture specific resources that
481 * would not be on a new untouched process.
482 */
483 cpu_thread_exit(td);
484
485 /*
486 * The last thread is left attached to the process
487 * So that the whole bundle gets recycled. Skip
488 * all this stuff if we never had threads.
489 * EXIT clears all sign of other threads when
490 * it goes to single threading, so the last thread always
491 * takes the short path.
492 */
493 if (p->p_flag & P_HADTHREADS) {
494 if (p->p_numthreads > 1) {
495 atomic_add_int(&td->td_proc->p_exitthreads, 1);
496 thread_unlink(td);
497 td2 = FIRST_THREAD_IN_PROC(p);
498 sched_exit_thread(td2, td);
499
500 /*
501 * The test below is NOT true if we are the
502 * sole exiting thread. P_STOPPED_SINGLE is unset
503 * in exit1() after it is the only survivor.
504 */
505 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
506 if (p->p_numthreads == p->p_suspcount) {
507 thread_lock(p->p_singlethread);
508 wakeup_swapper = thread_unsuspend_one(
509 p->p_singlethread, p, false);
510 thread_unlock(p->p_singlethread);
511 if (wakeup_swapper)
512 kick_proc0();
513 }
514 }
515
516 PCPU_SET(deadthread, td);
517 } else {
518 /*
519 * The last thread is exiting.. but not through exit()
520 */
521 panic ("thread_exit: Last thread exiting on its own");
522 }
523 }
524 #ifdef HWPMC_HOOKS
525 /*
526 * If this thread is part of a process that is being tracked by hwpmc(4),
527 * inform the module of the thread's impending exit.
528 */
529 if (PMC_PROC_IS_USING_PMCS(td->td_proc))
530 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
531 #endif
532 PROC_UNLOCK(p);
533 PROC_STATLOCK(p);
534 thread_lock(td);
535 PROC_SUNLOCK(p);
536
537 /* Do the same timestamp bookkeeping that mi_switch() would do. */
538 new_switchtime = cpu_ticks();
539 runtime = new_switchtime - PCPU_GET(switchtime);
540 td->td_runtime += runtime;
541 td->td_incruntime += runtime;
542 PCPU_SET(switchtime, new_switchtime);
543 PCPU_SET(switchticks, ticks);
544 PCPU_INC(cnt.v_swtch);
545
546 /* Save our resource usage in our process. */
547 td->td_ru.ru_nvcsw++;
548 ruxagg(p, td);
549 rucollect(&p->p_ru, &td->td_ru);
550 PROC_STATUNLOCK(p);
551
552 td->td_state = TDS_INACTIVE;
553 #ifdef WITNESS
554 witness_thread_exit(td);
555 #endif
556 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
557 sched_throw(td);
558 panic("I'm a teapot!");
559 /* NOTREACHED */
560 }
561
562 /*
563 * Do any thread specific cleanups that may be needed in wait()
564 * called with Giant, proc and schedlock not held.
565 */
566 void
567 thread_wait(struct proc *p)
568 {
569 struct thread *td;
570
571 mtx_assert(&Giant, MA_NOTOWNED);
572 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
573 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
574 td = FIRST_THREAD_IN_PROC(p);
575 /* Lock the last thread so we spin until it exits cpu_throw(). */
576 thread_lock(td);
577 thread_unlock(td);
578 lock_profile_thread_exit(td);
579 cpuset_rel(td->td_cpuset);
580 td->td_cpuset = NULL;
581 cpu_thread_clean(td);
582 thread_cow_free(td);
583 thread_reap(); /* check for zombie threads etc. */
584 }
585
586 /*
587 * Link a thread to a process.
588 * set up anything that needs to be initialized for it to
589 * be used by the process.
590 */
591 void
592 thread_link(struct thread *td, struct proc *p)
593 {
594
595 /*
596 * XXX This can't be enabled because it's called for proc0 before
597 * its lock has been created.
598 * PROC_LOCK_ASSERT(p, MA_OWNED);
599 */
600 td->td_state = TDS_INACTIVE;
601 td->td_proc = p;
602 td->td_flags = TDF_INMEM;
603
604 LIST_INIT(&td->td_contested);
605 LIST_INIT(&td->td_lprof[0]);
606 LIST_INIT(&td->td_lprof[1]);
607 sigqueue_init(&td->td_sigqueue, p);
608 callout_init(&td->td_slpcallout, 1);
609 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
610 p->p_numthreads++;
611 }
612
613 /*
614 * Called from:
615 * thread_exit()
616 */
617 void
618 thread_unlink(struct thread *td)
619 {
620 struct proc *p = td->td_proc;
621
622 PROC_LOCK_ASSERT(p, MA_OWNED);
623 TAILQ_REMOVE(&p->p_threads, td, td_plist);
624 p->p_numthreads--;
625 /* could clear a few other things here */
626 /* Must NOT clear links to proc! */
627 }
628
629 static int
630 calc_remaining(struct proc *p, int mode)
631 {
632 int remaining;
633
634 PROC_LOCK_ASSERT(p, MA_OWNED);
635 PROC_SLOCK_ASSERT(p, MA_OWNED);
636 if (mode == SINGLE_EXIT)
637 remaining = p->p_numthreads;
638 else if (mode == SINGLE_BOUNDARY)
639 remaining = p->p_numthreads - p->p_boundary_count;
640 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
641 remaining = p->p_numthreads - p->p_suspcount;
642 else
643 panic("calc_remaining: wrong mode %d", mode);
644 return (remaining);
645 }
646
647 static int
648 remain_for_mode(int mode)
649 {
650
651 return (mode == SINGLE_ALLPROC ? 0 : 1);
652 }
653
654 static int
655 weed_inhib(int mode, struct thread *td2, struct proc *p)
656 {
657 int wakeup_swapper;
658
659 PROC_LOCK_ASSERT(p, MA_OWNED);
660 PROC_SLOCK_ASSERT(p, MA_OWNED);
661 THREAD_LOCK_ASSERT(td2, MA_OWNED);
662
663 wakeup_swapper = 0;
664 switch (mode) {
665 case SINGLE_EXIT:
666 if (TD_IS_SUSPENDED(td2))
667 wakeup_swapper |= thread_unsuspend_one(td2, p, true);
668 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
669 wakeup_swapper |= sleepq_abort(td2, EINTR);
670 break;
671 case SINGLE_BOUNDARY:
672 case SINGLE_NO_EXIT:
673 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0)
674 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
675 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
676 wakeup_swapper |= sleepq_abort(td2, ERESTART);
677 break;
678 case SINGLE_ALLPROC:
679 /*
680 * ALLPROC suspend tries to avoid spurious EINTR for
681 * threads sleeping interruptable, by suspending the
682 * thread directly, similarly to sig_suspend_threads().
683 * Since such sleep is not performed at the user
684 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
685 * is used to avoid immediate un-suspend.
686 */
687 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
688 TDF_ALLPROCSUSP)) == 0)
689 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
690 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) {
691 if ((td2->td_flags & TDF_SBDRY) == 0) {
692 thread_suspend_one(td2);
693 td2->td_flags |= TDF_ALLPROCSUSP;
694 } else {
695 wakeup_swapper |= sleepq_abort(td2, ERESTART);
696 }
697 }
698 break;
699 }
700 return (wakeup_swapper);
701 }
702
703 /*
704 * Enforce single-threading.
705 *
706 * Returns 1 if the caller must abort (another thread is waiting to
707 * exit the process or similar). Process is locked!
708 * Returns 0 when you are successfully the only thread running.
709 * A process has successfully single threaded in the suspend mode when
710 * There are no threads in user mode. Threads in the kernel must be
711 * allowed to continue until they get to the user boundary. They may even
712 * copy out their return values and data before suspending. They may however be
713 * accelerated in reaching the user boundary as we will wake up
714 * any sleeping threads that are interruptable. (PCATCH).
715 */
716 int
717 thread_single(struct proc *p, int mode)
718 {
719 struct thread *td;
720 struct thread *td2;
721 int remaining, wakeup_swapper;
722
723 td = curthread;
724 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
725 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
726 ("invalid mode %d", mode));
727 /*
728 * If allowing non-ALLPROC singlethreading for non-curproc
729 * callers, calc_remaining() and remain_for_mode() should be
730 * adjusted to also account for td->td_proc != p. For now
731 * this is not implemented because it is not used.
732 */
733 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
734 (mode != SINGLE_ALLPROC && td->td_proc == p),
735 ("mode %d proc %p curproc %p", mode, p, td->td_proc));
736 mtx_assert(&Giant, MA_NOTOWNED);
737 PROC_LOCK_ASSERT(p, MA_OWNED);
738
739 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
740 return (0);
741
742 /* Is someone already single threading? */
743 if (p->p_singlethread != NULL && p->p_singlethread != td)
744 return (1);
745
746 if (mode == SINGLE_EXIT) {
747 p->p_flag |= P_SINGLE_EXIT;
748 p->p_flag &= ~P_SINGLE_BOUNDARY;
749 } else {
750 p->p_flag &= ~P_SINGLE_EXIT;
751 if (mode == SINGLE_BOUNDARY)
752 p->p_flag |= P_SINGLE_BOUNDARY;
753 else
754 p->p_flag &= ~P_SINGLE_BOUNDARY;
755 }
756 if (mode == SINGLE_ALLPROC)
757 p->p_flag |= P_TOTAL_STOP;
758 p->p_flag |= P_STOPPED_SINGLE;
759 PROC_SLOCK(p);
760 p->p_singlethread = td;
761 remaining = calc_remaining(p, mode);
762 while (remaining != remain_for_mode(mode)) {
763 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
764 goto stopme;
765 wakeup_swapper = 0;
766 FOREACH_THREAD_IN_PROC(p, td2) {
767 if (td2 == td)
768 continue;
769 thread_lock(td2);
770 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
771 if (TD_IS_INHIBITED(td2)) {
772 wakeup_swapper |= weed_inhib(mode, td2, p);
773 #ifdef SMP
774 } else if (TD_IS_RUNNING(td2) && td != td2) {
775 forward_signal(td2);
776 #endif
777 }
778 thread_unlock(td2);
779 }
780 if (wakeup_swapper)
781 kick_proc0();
782 remaining = calc_remaining(p, mode);
783
784 /*
785 * Maybe we suspended some threads.. was it enough?
786 */
787 if (remaining == remain_for_mode(mode))
788 break;
789
790 stopme:
791 /*
792 * Wake us up when everyone else has suspended.
793 * In the mean time we suspend as well.
794 */
795 thread_suspend_switch(td, p);
796 remaining = calc_remaining(p, mode);
797 }
798 if (mode == SINGLE_EXIT) {
799 /*
800 * Convert the process to an unthreaded process. The
801 * SINGLE_EXIT is called by exit1() or execve(), in
802 * both cases other threads must be retired.
803 */
804 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
805 p->p_singlethread = NULL;
806 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
807
808 /*
809 * Wait for any remaining threads to exit cpu_throw().
810 */
811 while (p->p_exitthreads != 0) {
812 PROC_SUNLOCK(p);
813 PROC_UNLOCK(p);
814 sched_relinquish(td);
815 PROC_LOCK(p);
816 PROC_SLOCK(p);
817 }
818 } else if (mode == SINGLE_BOUNDARY) {
819 /*
820 * Wait until all suspended threads are removed from
821 * the processors. The thread_suspend_check()
822 * increments p_boundary_count while it is still
823 * running, which makes it possible for the execve()
824 * to destroy vmspace while our other threads are
825 * still using the address space.
826 *
827 * We lock the thread, which is only allowed to
828 * succeed after context switch code finished using
829 * the address space.
830 */
831 FOREACH_THREAD_IN_PROC(p, td2) {
832 if (td2 == td)
833 continue;
834 thread_lock(td2);
835 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
836 ("td %p not on boundary", td2));
837 KASSERT(TD_IS_SUSPENDED(td2),
838 ("td %p is not suspended", td2));
839 thread_unlock(td2);
840 }
841 }
842 PROC_SUNLOCK(p);
843 return (0);
844 }
845
846 bool
847 thread_suspend_check_needed(void)
848 {
849 struct proc *p;
850 struct thread *td;
851
852 td = curthread;
853 p = td->td_proc;
854 PROC_LOCK_ASSERT(p, MA_OWNED);
855 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
856 (td->td_dbgflags & TDB_SUSPEND) != 0));
857 }
858
859 /*
860 * Called in from locations that can safely check to see
861 * whether we have to suspend or at least throttle for a
862 * single-thread event (e.g. fork).
863 *
864 * Such locations include userret().
865 * If the "return_instead" argument is non zero, the thread must be able to
866 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
867 *
868 * The 'return_instead' argument tells the function if it may do a
869 * thread_exit() or suspend, or whether the caller must abort and back
870 * out instead.
871 *
872 * If the thread that set the single_threading request has set the
873 * P_SINGLE_EXIT bit in the process flags then this call will never return
874 * if 'return_instead' is false, but will exit.
875 *
876 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
877 *---------------+--------------------+---------------------
878 * 0 | returns 0 | returns 0 or 1
879 * | when ST ends | immediately
880 *---------------+--------------------+---------------------
881 * 1 | thread exits | returns 1
882 * | | immediately
883 * 0 = thread_exit() or suspension ok,
884 * other = return error instead of stopping the thread.
885 *
886 * While a full suspension is under effect, even a single threading
887 * thread would be suspended if it made this call (but it shouldn't).
888 * This call should only be made from places where
889 * thread_exit() would be safe as that may be the outcome unless
890 * return_instead is set.
891 */
892 int
893 thread_suspend_check(int return_instead)
894 {
895 struct thread *td;
896 struct proc *p;
897 int wakeup_swapper;
898
899 td = curthread;
900 p = td->td_proc;
901 mtx_assert(&Giant, MA_NOTOWNED);
902 PROC_LOCK_ASSERT(p, MA_OWNED);
903 while (thread_suspend_check_needed()) {
904 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
905 KASSERT(p->p_singlethread != NULL,
906 ("singlethread not set"));
907 /*
908 * The only suspension in action is a
909 * single-threading. Single threader need not stop.
910 * It is safe to access p->p_singlethread unlocked
911 * because it can only be set to our address by us.
912 */
913 if (p->p_singlethread == td)
914 return (0); /* Exempt from stopping. */
915 }
916 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
917 return (EINTR);
918
919 /* Should we goto user boundary if we didn't come from there? */
920 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
921 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
922 return (ERESTART);
923
924 /*
925 * Ignore suspend requests if they are deferred.
926 */
927 if ((td->td_flags & TDF_SBDRY) != 0) {
928 KASSERT(return_instead,
929 ("TDF_SBDRY set for unsafe thread_suspend_check"));
930 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
931 (TDF_SEINTR | TDF_SERESTART),
932 ("both TDF_SEINTR and TDF_SERESTART"));
933 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
934 }
935
936 /*
937 * If the process is waiting for us to exit,
938 * this thread should just suicide.
939 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
940 */
941 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
942 PROC_UNLOCK(p);
943
944 /*
945 * Allow Linux emulation layer to do some work
946 * before thread suicide.
947 */
948 if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
949 (p->p_sysent->sv_thread_detach)(td);
950 umtx_thread_exit(td);
951 kern_thr_exit(td);
952 panic("stopped thread did not exit");
953 }
954
955 PROC_SLOCK(p);
956 thread_stopped(p);
957 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
958 if (p->p_numthreads == p->p_suspcount + 1) {
959 thread_lock(p->p_singlethread);
960 wakeup_swapper = thread_unsuspend_one(
961 p->p_singlethread, p, false);
962 thread_unlock(p->p_singlethread);
963 if (wakeup_swapper)
964 kick_proc0();
965 }
966 }
967 PROC_UNLOCK(p);
968 thread_lock(td);
969 /*
970 * When a thread suspends, it just
971 * gets taken off all queues.
972 */
973 thread_suspend_one(td);
974 if (return_instead == 0) {
975 p->p_boundary_count++;
976 td->td_flags |= TDF_BOUNDARY;
977 }
978 PROC_SUNLOCK(p);
979 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
980 thread_unlock(td);
981 PROC_LOCK(p);
982 }
983 return (0);
984 }
985
986 void
987 thread_suspend_switch(struct thread *td, struct proc *p)
988 {
989
990 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
991 PROC_LOCK_ASSERT(p, MA_OWNED);
992 PROC_SLOCK_ASSERT(p, MA_OWNED);
993 /*
994 * We implement thread_suspend_one in stages here to avoid
995 * dropping the proc lock while the thread lock is owned.
996 */
997 if (p == td->td_proc) {
998 thread_stopped(p);
999 p->p_suspcount++;
1000 }
1001 PROC_UNLOCK(p);
1002 thread_lock(td);
1003 td->td_flags &= ~TDF_NEEDSUSPCHK;
1004 TD_SET_SUSPENDED(td);
1005 sched_sleep(td, 0);
1006 PROC_SUNLOCK(p);
1007 DROP_GIANT();
1008 mi_switch(SW_VOL | SWT_SUSPEND, NULL);
1009 thread_unlock(td);
1010 PICKUP_GIANT();
1011 PROC_LOCK(p);
1012 PROC_SLOCK(p);
1013 }
1014
1015 void
1016 thread_suspend_one(struct thread *td)
1017 {
1018 struct proc *p;
1019
1020 p = td->td_proc;
1021 PROC_SLOCK_ASSERT(p, MA_OWNED);
1022 THREAD_LOCK_ASSERT(td, MA_OWNED);
1023 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1024 p->p_suspcount++;
1025 td->td_flags &= ~TDF_NEEDSUSPCHK;
1026 TD_SET_SUSPENDED(td);
1027 sched_sleep(td, 0);
1028 }
1029
1030 static int
1031 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1032 {
1033
1034 THREAD_LOCK_ASSERT(td, MA_OWNED);
1035 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1036 TD_CLR_SUSPENDED(td);
1037 td->td_flags &= ~TDF_ALLPROCSUSP;
1038 if (td->td_proc == p) {
1039 PROC_SLOCK_ASSERT(p, MA_OWNED);
1040 p->p_suspcount--;
1041 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1042 td->td_flags &= ~TDF_BOUNDARY;
1043 p->p_boundary_count--;
1044 }
1045 }
1046 return (setrunnable(td));
1047 }
1048
1049 /*
1050 * Allow all threads blocked by single threading to continue running.
1051 */
1052 void
1053 thread_unsuspend(struct proc *p)
1054 {
1055 struct thread *td;
1056 int wakeup_swapper;
1057
1058 PROC_LOCK_ASSERT(p, MA_OWNED);
1059 PROC_SLOCK_ASSERT(p, MA_OWNED);
1060 wakeup_swapper = 0;
1061 if (!P_SHOULDSTOP(p)) {
1062 FOREACH_THREAD_IN_PROC(p, td) {
1063 thread_lock(td);
1064 if (TD_IS_SUSPENDED(td)) {
1065 wakeup_swapper |= thread_unsuspend_one(td, p,
1066 true);
1067 }
1068 thread_unlock(td);
1069 }
1070 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1071 p->p_numthreads == p->p_suspcount) {
1072 /*
1073 * Stopping everything also did the job for the single
1074 * threading request. Now we've downgraded to single-threaded,
1075 * let it continue.
1076 */
1077 if (p->p_singlethread->td_proc == p) {
1078 thread_lock(p->p_singlethread);
1079 wakeup_swapper = thread_unsuspend_one(
1080 p->p_singlethread, p, false);
1081 thread_unlock(p->p_singlethread);
1082 }
1083 }
1084 if (wakeup_swapper)
1085 kick_proc0();
1086 }
1087
1088 /*
1089 * End the single threading mode..
1090 */
1091 void
1092 thread_single_end(struct proc *p, int mode)
1093 {
1094 struct thread *td;
1095 int wakeup_swapper;
1096
1097 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1098 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1099 ("invalid mode %d", mode));
1100 PROC_LOCK_ASSERT(p, MA_OWNED);
1101 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1102 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1103 ("mode %d does not match P_TOTAL_STOP", mode));
1104 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1105 ("thread_single_end from other thread %p %p",
1106 curthread, p->p_singlethread));
1107 KASSERT(mode != SINGLE_BOUNDARY ||
1108 (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1109 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1110 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1111 P_TOTAL_STOP);
1112 PROC_SLOCK(p);
1113 p->p_singlethread = NULL;
1114 wakeup_swapper = 0;
1115 /*
1116 * If there are other threads they may now run,
1117 * unless of course there is a blanket 'stop order'
1118 * on the process. The single threader must be allowed
1119 * to continue however as this is a bad place to stop.
1120 */
1121 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1122 FOREACH_THREAD_IN_PROC(p, td) {
1123 thread_lock(td);
1124 if (TD_IS_SUSPENDED(td)) {
1125 wakeup_swapper |= thread_unsuspend_one(td, p,
1126 mode == SINGLE_BOUNDARY);
1127 }
1128 thread_unlock(td);
1129 }
1130 }
1131 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1132 ("inconsistent boundary count %d", p->p_boundary_count));
1133 PROC_SUNLOCK(p);
1134 if (wakeup_swapper)
1135 kick_proc0();
1136 }
1137
1138 struct thread *
1139 thread_find(struct proc *p, lwpid_t tid)
1140 {
1141 struct thread *td;
1142
1143 PROC_LOCK_ASSERT(p, MA_OWNED);
1144 FOREACH_THREAD_IN_PROC(p, td) {
1145 if (td->td_tid == tid)
1146 break;
1147 }
1148 return (td);
1149 }
1150
1151 /* Locate a thread by number; return with proc lock held. */
1152 struct thread *
1153 tdfind(lwpid_t tid, pid_t pid)
1154 {
1155 #define RUN_THRESH 16
1156 struct thread *td;
1157 int run = 0;
1158
1159 rw_rlock(&tidhash_lock);
1160 LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1161 if (td->td_tid == tid) {
1162 if (pid != -1 && td->td_proc->p_pid != pid) {
1163 td = NULL;
1164 break;
1165 }
1166 PROC_LOCK(td->td_proc);
1167 if (td->td_proc->p_state == PRS_NEW) {
1168 PROC_UNLOCK(td->td_proc);
1169 td = NULL;
1170 break;
1171 }
1172 if (run > RUN_THRESH) {
1173 if (rw_try_upgrade(&tidhash_lock)) {
1174 LIST_REMOVE(td, td_hash);
1175 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1176 td, td_hash);
1177 rw_wunlock(&tidhash_lock);
1178 return (td);
1179 }
1180 }
1181 break;
1182 }
1183 run++;
1184 }
1185 rw_runlock(&tidhash_lock);
1186 return (td);
1187 }
1188
1189 void
1190 tidhash_add(struct thread *td)
1191 {
1192 rw_wlock(&tidhash_lock);
1193 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1194 rw_wunlock(&tidhash_lock);
1195 }
1196
1197 void
1198 tidhash_remove(struct thread *td)
1199 {
1200 rw_wlock(&tidhash_lock);
1201 LIST_REMOVE(td, td_hash);
1202 rw_wunlock(&tidhash_lock);
1203 }
Cache object: f999de9e216526fdd76bc3bb9453ced0
|