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/11.1/sys/kern/kern_thread.c 318743 2017-05-23 12:40:50Z badger $");
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 /*
68 * Asserts below verify the stability of struct thread and struct proc
69 * layout, as exposed by KBI to modules. On head, the KBI is allowed
70 * to drift, change to the structures must be accompanied by the
71 * assert update.
72 *
73 * On the stable branches after KBI freeze, conditions must not be
74 * violated. Typically new fields are moved to the end of the
75 * structures.
76 */
77 #ifdef __amd64__
78 _Static_assert(offsetof(struct thread, td_flags) == 0xe4,
79 "struct thread KBI td_flags");
80 _Static_assert(offsetof(struct thread, td_pflags) == 0xec,
81 "struct thread KBI td_pflags");
82 _Static_assert(offsetof(struct thread, td_frame) == 0x418,
83 "struct thread KBI td_frame");
84 _Static_assert(offsetof(struct thread, td_emuldata) == 0x4c0,
85 "struct thread KBI td_emuldata");
86 _Static_assert(offsetof(struct proc, p_flag) == 0xb0,
87 "struct proc KBI p_flag");
88 _Static_assert(offsetof(struct proc, p_pid) == 0xbc,
89 "struct proc KBI p_pid");
90 _Static_assert(offsetof(struct proc, p_filemon) == 0x3c0,
91 "struct proc KBI p_filemon");
92 _Static_assert(offsetof(struct proc, p_comm) == 0x3d0,
93 "struct proc KBI p_comm");
94 _Static_assert(offsetof(struct proc, p_emuldata) == 0x4a0,
95 "struct proc KBI p_emuldata");
96 #endif
97 #ifdef __i386__
98 _Static_assert(offsetof(struct thread, td_flags) == 0x8c,
99 "struct thread KBI td_flags");
100 _Static_assert(offsetof(struct thread, td_pflags) == 0x94,
101 "struct thread KBI td_pflags");
102 _Static_assert(offsetof(struct thread, td_frame) == 0x2c0,
103 "struct thread KBI td_frame");
104 _Static_assert(offsetof(struct thread, td_emuldata) == 0x30c,
105 "struct thread KBI td_emuldata");
106 _Static_assert(offsetof(struct proc, p_flag) == 0x68,
107 "struct proc KBI p_flag");
108 _Static_assert(offsetof(struct proc, p_pid) == 0x74,
109 "struct proc KBI p_pid");
110 _Static_assert(offsetof(struct proc, p_filemon) == 0x268,
111 "struct proc KBI p_filemon");
112 _Static_assert(offsetof(struct proc, p_comm) == 0x274,
113 "struct proc KBI p_comm");
114 _Static_assert(offsetof(struct proc, p_emuldata) == 0x2f4,
115 "struct proc KBI p_emuldata");
116 #endif
117
118 SDT_PROVIDER_DECLARE(proc);
119 SDT_PROBE_DEFINE(proc, , , lwp__exit);
120
121 /*
122 * thread related storage.
123 */
124 static uma_zone_t thread_zone;
125
126 TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
127 static struct mtx zombie_lock;
128 MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN);
129
130 static void thread_zombie(struct thread *);
131 static int thread_unsuspend_one(struct thread *td, struct proc *p,
132 bool boundary);
133
134 #define TID_BUFFER_SIZE 1024
135
136 struct mtx tid_lock;
137 static struct unrhdr *tid_unrhdr;
138 static lwpid_t tid_buffer[TID_BUFFER_SIZE];
139 static int tid_head, tid_tail;
140 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
141
142 struct tidhashhead *tidhashtbl;
143 u_long tidhash;
144 struct rwlock tidhash_lock;
145
146 static lwpid_t
147 tid_alloc(void)
148 {
149 lwpid_t tid;
150
151 tid = alloc_unr(tid_unrhdr);
152 if (tid != -1)
153 return (tid);
154 mtx_lock(&tid_lock);
155 if (tid_head == tid_tail) {
156 mtx_unlock(&tid_lock);
157 return (-1);
158 }
159 tid = tid_buffer[tid_head];
160 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
161 mtx_unlock(&tid_lock);
162 return (tid);
163 }
164
165 static void
166 tid_free(lwpid_t tid)
167 {
168 lwpid_t tmp_tid = -1;
169
170 mtx_lock(&tid_lock);
171 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) {
172 tmp_tid = tid_buffer[tid_head];
173 tid_head = (tid_head + 1) % TID_BUFFER_SIZE;
174 }
175 tid_buffer[tid_tail] = tid;
176 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE;
177 mtx_unlock(&tid_lock);
178 if (tmp_tid != -1)
179 free_unr(tid_unrhdr, tmp_tid);
180 }
181
182 /*
183 * Prepare a thread for use.
184 */
185 static int
186 thread_ctor(void *mem, int size, void *arg, int flags)
187 {
188 struct thread *td;
189
190 td = (struct thread *)mem;
191 td->td_state = TDS_INACTIVE;
192 td->td_oncpu = NOCPU;
193
194 td->td_tid = tid_alloc();
195
196 /*
197 * Note that td_critnest begins life as 1 because the thread is not
198 * running and is thereby implicitly waiting to be on the receiving
199 * end of a context switch.
200 */
201 td->td_critnest = 1;
202 td->td_lend_user_pri = PRI_MAX;
203 EVENTHANDLER_INVOKE(thread_ctor, td);
204 #ifdef AUDIT
205 audit_thread_alloc(td);
206 #endif
207 umtx_thread_alloc(td);
208 return (0);
209 }
210
211 /*
212 * Reclaim a thread after use.
213 */
214 static void
215 thread_dtor(void *mem, int size, void *arg)
216 {
217 struct thread *td;
218
219 td = (struct thread *)mem;
220
221 #ifdef INVARIANTS
222 /* Verify that this thread is in a safe state to free. */
223 switch (td->td_state) {
224 case TDS_INHIBITED:
225 case TDS_RUNNING:
226 case TDS_CAN_RUN:
227 case TDS_RUNQ:
228 /*
229 * We must never unlink a thread that is in one of
230 * these states, because it is currently active.
231 */
232 panic("bad state for thread unlinking");
233 /* NOTREACHED */
234 case TDS_INACTIVE:
235 break;
236 default:
237 panic("bad thread state");
238 /* NOTREACHED */
239 }
240 #endif
241 #ifdef AUDIT
242 audit_thread_free(td);
243 #endif
244 /* Free all OSD associated to this thread. */
245 osd_thread_exit(td);
246 td_softdep_cleanup(td);
247 MPASS(td->td_su == NULL);
248
249 EVENTHANDLER_INVOKE(thread_dtor, td);
250 tid_free(td->td_tid);
251 }
252
253 /*
254 * Initialize type-stable parts of a thread (when newly created).
255 */
256 static int
257 thread_init(void *mem, int size, int flags)
258 {
259 struct thread *td;
260
261 td = (struct thread *)mem;
262
263 td->td_sleepqueue = sleepq_alloc();
264 td->td_turnstile = turnstile_alloc();
265 td->td_rlqe = NULL;
266 EVENTHANDLER_INVOKE(thread_init, td);
267 umtx_thread_init(td);
268 td->td_kstack = 0;
269 td->td_sel = NULL;
270 return (0);
271 }
272
273 /*
274 * Tear down type-stable parts of a thread (just before being discarded).
275 */
276 static void
277 thread_fini(void *mem, int size)
278 {
279 struct thread *td;
280
281 td = (struct thread *)mem;
282 EVENTHANDLER_INVOKE(thread_fini, td);
283 rlqentry_free(td->td_rlqe);
284 turnstile_free(td->td_turnstile);
285 sleepq_free(td->td_sleepqueue);
286 umtx_thread_fini(td);
287 seltdfini(td);
288 }
289
290 /*
291 * For a newly created process,
292 * link up all the structures and its initial threads etc.
293 * called from:
294 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc.
295 * proc_dtor() (should go away)
296 * proc_init()
297 */
298 void
299 proc_linkup0(struct proc *p, struct thread *td)
300 {
301 TAILQ_INIT(&p->p_threads); /* all threads in proc */
302 proc_linkup(p, td);
303 }
304
305 void
306 proc_linkup(struct proc *p, struct thread *td)
307 {
308
309 sigqueue_init(&p->p_sigqueue, p);
310 p->p_ksi = ksiginfo_alloc(1);
311 if (p->p_ksi != NULL) {
312 /* XXX p_ksi may be null if ksiginfo zone is not ready */
313 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
314 }
315 LIST_INIT(&p->p_mqnotifier);
316 p->p_numthreads = 0;
317 thread_link(td, p);
318 }
319
320 /*
321 * Initialize global thread allocation resources.
322 */
323 void
324 threadinit(void)
325 {
326
327 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
328
329 /*
330 * pid_max cannot be greater than PID_MAX.
331 * leave one number for thread0.
332 */
333 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock);
334
335 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
336 thread_ctor, thread_dtor, thread_init, thread_fini,
337 32 - 1, UMA_ZONE_NOFREE);
338 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
339 rw_init(&tidhash_lock, "tidhash");
340 }
341
342 /*
343 * Place an unused thread on the zombie list.
344 * Use the slpq as that must be unused by now.
345 */
346 void
347 thread_zombie(struct thread *td)
348 {
349 mtx_lock_spin(&zombie_lock);
350 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq);
351 mtx_unlock_spin(&zombie_lock);
352 }
353
354 /*
355 * Release a thread that has exited after cpu_throw().
356 */
357 void
358 thread_stash(struct thread *td)
359 {
360 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
361 thread_zombie(td);
362 }
363
364 /*
365 * Reap zombie resources.
366 */
367 void
368 thread_reap(void)
369 {
370 struct thread *td_first, *td_next;
371
372 /*
373 * Don't even bother to lock if none at this instant,
374 * we really don't care about the next instant.
375 */
376 if (!TAILQ_EMPTY(&zombie_threads)) {
377 mtx_lock_spin(&zombie_lock);
378 td_first = TAILQ_FIRST(&zombie_threads);
379 if (td_first)
380 TAILQ_INIT(&zombie_threads);
381 mtx_unlock_spin(&zombie_lock);
382 while (td_first) {
383 td_next = TAILQ_NEXT(td_first, td_slpq);
384 thread_cow_free(td_first);
385 thread_free(td_first);
386 td_first = td_next;
387 }
388 }
389 }
390
391 /*
392 * Allocate a thread.
393 */
394 struct thread *
395 thread_alloc(int pages)
396 {
397 struct thread *td;
398
399 thread_reap(); /* check if any zombies to get */
400
401 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK);
402 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
403 if (!vm_thread_new(td, pages)) {
404 uma_zfree(thread_zone, td);
405 return (NULL);
406 }
407 cpu_thread_alloc(td);
408 vm_domain_policy_init(&td->td_vm_dom_policy);
409 return (td);
410 }
411
412 int
413 thread_alloc_stack(struct thread *td, int pages)
414 {
415
416 KASSERT(td->td_kstack == 0,
417 ("thread_alloc_stack called on a thread with kstack"));
418 if (!vm_thread_new(td, pages))
419 return (0);
420 cpu_thread_alloc(td);
421 return (1);
422 }
423
424 /*
425 * Deallocate a thread.
426 */
427 void
428 thread_free(struct thread *td)
429 {
430
431 lock_profile_thread_exit(td);
432 if (td->td_cpuset)
433 cpuset_rel(td->td_cpuset);
434 td->td_cpuset = NULL;
435 cpu_thread_free(td);
436 if (td->td_kstack != 0)
437 vm_thread_dispose(td);
438 vm_domain_policy_cleanup(&td->td_vm_dom_policy);
439 callout_drain(&td->td_slpcallout);
440 uma_zfree(thread_zone, td);
441 }
442
443 void
444 thread_cow_get_proc(struct thread *newtd, struct proc *p)
445 {
446
447 PROC_LOCK_ASSERT(p, MA_OWNED);
448 newtd->td_ucred = crhold(p->p_ucred);
449 newtd->td_limit = lim_hold(p->p_limit);
450 newtd->td_cowgen = p->p_cowgen;
451 }
452
453 void
454 thread_cow_get(struct thread *newtd, struct thread *td)
455 {
456
457 newtd->td_ucred = crhold(td->td_ucred);
458 newtd->td_limit = lim_hold(td->td_limit);
459 newtd->td_cowgen = td->td_cowgen;
460 }
461
462 void
463 thread_cow_free(struct thread *td)
464 {
465
466 if (td->td_ucred != NULL)
467 crfree(td->td_ucred);
468 if (td->td_limit != NULL)
469 lim_free(td->td_limit);
470 }
471
472 void
473 thread_cow_update(struct thread *td)
474 {
475 struct proc *p;
476 struct ucred *oldcred;
477 struct plimit *oldlimit;
478
479 p = td->td_proc;
480 oldcred = NULL;
481 oldlimit = NULL;
482 PROC_LOCK(p);
483 if (td->td_ucred != p->p_ucred) {
484 oldcred = td->td_ucred;
485 td->td_ucred = crhold(p->p_ucred);
486 }
487 if (td->td_limit != p->p_limit) {
488 oldlimit = td->td_limit;
489 td->td_limit = lim_hold(p->p_limit);
490 }
491 td->td_cowgen = p->p_cowgen;
492 PROC_UNLOCK(p);
493 if (oldcred != NULL)
494 crfree(oldcred);
495 if (oldlimit != NULL)
496 lim_free(oldlimit);
497 }
498
499 /*
500 * Discard the current thread and exit from its context.
501 * Always called with scheduler locked.
502 *
503 * Because we can't free a thread while we're operating under its context,
504 * push the current thread into our CPU's deadthread holder. This means
505 * we needn't worry about someone else grabbing our context before we
506 * do a cpu_throw().
507 */
508 void
509 thread_exit(void)
510 {
511 uint64_t runtime, new_switchtime;
512 struct thread *td;
513 struct thread *td2;
514 struct proc *p;
515 int wakeup_swapper;
516
517 td = curthread;
518 p = td->td_proc;
519
520 PROC_SLOCK_ASSERT(p, MA_OWNED);
521 mtx_assert(&Giant, MA_NOTOWNED);
522
523 PROC_LOCK_ASSERT(p, MA_OWNED);
524 KASSERT(p != NULL, ("thread exiting without a process"));
525 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
526 (long)p->p_pid, td->td_name);
527 SDT_PROBE0(proc, , , lwp__exit);
528 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
529
530 #ifdef AUDIT
531 AUDIT_SYSCALL_EXIT(0, td);
532 #endif
533 /*
534 * drop FPU & debug register state storage, or any other
535 * architecture specific resources that
536 * would not be on a new untouched process.
537 */
538 cpu_thread_exit(td);
539
540 /*
541 * The last thread is left attached to the process
542 * So that the whole bundle gets recycled. Skip
543 * all this stuff if we never had threads.
544 * EXIT clears all sign of other threads when
545 * it goes to single threading, so the last thread always
546 * takes the short path.
547 */
548 if (p->p_flag & P_HADTHREADS) {
549 if (p->p_numthreads > 1) {
550 atomic_add_int(&td->td_proc->p_exitthreads, 1);
551 thread_unlink(td);
552 td2 = FIRST_THREAD_IN_PROC(p);
553 sched_exit_thread(td2, td);
554
555 /*
556 * The test below is NOT true if we are the
557 * sole exiting thread. P_STOPPED_SINGLE is unset
558 * in exit1() after it is the only survivor.
559 */
560 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
561 if (p->p_numthreads == p->p_suspcount) {
562 thread_lock(p->p_singlethread);
563 wakeup_swapper = thread_unsuspend_one(
564 p->p_singlethread, p, false);
565 thread_unlock(p->p_singlethread);
566 if (wakeup_swapper)
567 kick_proc0();
568 }
569 }
570
571 PCPU_SET(deadthread, td);
572 } else {
573 /*
574 * The last thread is exiting.. but not through exit()
575 */
576 panic ("thread_exit: Last thread exiting on its own");
577 }
578 }
579 #ifdef HWPMC_HOOKS
580 /*
581 * If this thread is part of a process that is being tracked by hwpmc(4),
582 * inform the module of the thread's impending exit.
583 */
584 if (PMC_PROC_IS_USING_PMCS(td->td_proc))
585 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
586 #endif
587 PROC_UNLOCK(p);
588 PROC_STATLOCK(p);
589 thread_lock(td);
590 PROC_SUNLOCK(p);
591
592 /* Do the same timestamp bookkeeping that mi_switch() would do. */
593 new_switchtime = cpu_ticks();
594 runtime = new_switchtime - PCPU_GET(switchtime);
595 td->td_runtime += runtime;
596 td->td_incruntime += runtime;
597 PCPU_SET(switchtime, new_switchtime);
598 PCPU_SET(switchticks, ticks);
599 PCPU_INC(cnt.v_swtch);
600
601 /* Save our resource usage in our process. */
602 td->td_ru.ru_nvcsw++;
603 ruxagg(p, td);
604 rucollect(&p->p_ru, &td->td_ru);
605 PROC_STATUNLOCK(p);
606
607 td->td_state = TDS_INACTIVE;
608 #ifdef WITNESS
609 witness_thread_exit(td);
610 #endif
611 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
612 sched_throw(td);
613 panic("I'm a teapot!");
614 /* NOTREACHED */
615 }
616
617 /*
618 * Do any thread specific cleanups that may be needed in wait()
619 * called with Giant, proc and schedlock not held.
620 */
621 void
622 thread_wait(struct proc *p)
623 {
624 struct thread *td;
625
626 mtx_assert(&Giant, MA_NOTOWNED);
627 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
628 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
629 td = FIRST_THREAD_IN_PROC(p);
630 /* Lock the last thread so we spin until it exits cpu_throw(). */
631 thread_lock(td);
632 thread_unlock(td);
633 lock_profile_thread_exit(td);
634 cpuset_rel(td->td_cpuset);
635 td->td_cpuset = NULL;
636 cpu_thread_clean(td);
637 thread_cow_free(td);
638 callout_drain(&td->td_slpcallout);
639 thread_reap(); /* check for zombie threads etc. */
640 }
641
642 /*
643 * Link a thread to a process.
644 * set up anything that needs to be initialized for it to
645 * be used by the process.
646 */
647 void
648 thread_link(struct thread *td, struct proc *p)
649 {
650
651 /*
652 * XXX This can't be enabled because it's called for proc0 before
653 * its lock has been created.
654 * PROC_LOCK_ASSERT(p, MA_OWNED);
655 */
656 td->td_state = TDS_INACTIVE;
657 td->td_proc = p;
658 td->td_flags = TDF_INMEM;
659
660 LIST_INIT(&td->td_contested);
661 LIST_INIT(&td->td_lprof[0]);
662 LIST_INIT(&td->td_lprof[1]);
663 sigqueue_init(&td->td_sigqueue, p);
664 callout_init(&td->td_slpcallout, 1);
665 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
666 p->p_numthreads++;
667 }
668
669 /*
670 * Called from:
671 * thread_exit()
672 */
673 void
674 thread_unlink(struct thread *td)
675 {
676 struct proc *p = td->td_proc;
677
678 PROC_LOCK_ASSERT(p, MA_OWNED);
679 TAILQ_REMOVE(&p->p_threads, td, td_plist);
680 p->p_numthreads--;
681 /* could clear a few other things here */
682 /* Must NOT clear links to proc! */
683 }
684
685 static int
686 calc_remaining(struct proc *p, int mode)
687 {
688 int remaining;
689
690 PROC_LOCK_ASSERT(p, MA_OWNED);
691 PROC_SLOCK_ASSERT(p, MA_OWNED);
692 if (mode == SINGLE_EXIT)
693 remaining = p->p_numthreads;
694 else if (mode == SINGLE_BOUNDARY)
695 remaining = p->p_numthreads - p->p_boundary_count;
696 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
697 remaining = p->p_numthreads - p->p_suspcount;
698 else
699 panic("calc_remaining: wrong mode %d", mode);
700 return (remaining);
701 }
702
703 static int
704 remain_for_mode(int mode)
705 {
706
707 return (mode == SINGLE_ALLPROC ? 0 : 1);
708 }
709
710 static int
711 weed_inhib(int mode, struct thread *td2, struct proc *p)
712 {
713 int wakeup_swapper;
714
715 PROC_LOCK_ASSERT(p, MA_OWNED);
716 PROC_SLOCK_ASSERT(p, MA_OWNED);
717 THREAD_LOCK_ASSERT(td2, MA_OWNED);
718
719 wakeup_swapper = 0;
720 switch (mode) {
721 case SINGLE_EXIT:
722 if (TD_IS_SUSPENDED(td2))
723 wakeup_swapper |= thread_unsuspend_one(td2, p, true);
724 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
725 wakeup_swapper |= sleepq_abort(td2, EINTR);
726 break;
727 case SINGLE_BOUNDARY:
728 case SINGLE_NO_EXIT:
729 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0)
730 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
731 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0)
732 wakeup_swapper |= sleepq_abort(td2, ERESTART);
733 break;
734 case SINGLE_ALLPROC:
735 /*
736 * ALLPROC suspend tries to avoid spurious EINTR for
737 * threads sleeping interruptable, by suspending the
738 * thread directly, similarly to sig_suspend_threads().
739 * Since such sleep is not performed at the user
740 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
741 * is used to avoid immediate un-suspend.
742 */
743 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
744 TDF_ALLPROCSUSP)) == 0)
745 wakeup_swapper |= thread_unsuspend_one(td2, p, false);
746 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) {
747 if ((td2->td_flags & TDF_SBDRY) == 0) {
748 thread_suspend_one(td2);
749 td2->td_flags |= TDF_ALLPROCSUSP;
750 } else {
751 wakeup_swapper |= sleepq_abort(td2, ERESTART);
752 }
753 }
754 break;
755 }
756 return (wakeup_swapper);
757 }
758
759 /*
760 * Enforce single-threading.
761 *
762 * Returns 1 if the caller must abort (another thread is waiting to
763 * exit the process or similar). Process is locked!
764 * Returns 0 when you are successfully the only thread running.
765 * A process has successfully single threaded in the suspend mode when
766 * There are no threads in user mode. Threads in the kernel must be
767 * allowed to continue until they get to the user boundary. They may even
768 * copy out their return values and data before suspending. They may however be
769 * accelerated in reaching the user boundary as we will wake up
770 * any sleeping threads that are interruptable. (PCATCH).
771 */
772 int
773 thread_single(struct proc *p, int mode)
774 {
775 struct thread *td;
776 struct thread *td2;
777 int remaining, wakeup_swapper;
778
779 td = curthread;
780 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
781 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
782 ("invalid mode %d", mode));
783 /*
784 * If allowing non-ALLPROC singlethreading for non-curproc
785 * callers, calc_remaining() and remain_for_mode() should be
786 * adjusted to also account for td->td_proc != p. For now
787 * this is not implemented because it is not used.
788 */
789 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
790 (mode != SINGLE_ALLPROC && td->td_proc == p),
791 ("mode %d proc %p curproc %p", mode, p, td->td_proc));
792 mtx_assert(&Giant, MA_NOTOWNED);
793 PROC_LOCK_ASSERT(p, MA_OWNED);
794
795 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
796 return (0);
797
798 /* Is someone already single threading? */
799 if (p->p_singlethread != NULL && p->p_singlethread != td)
800 return (1);
801
802 if (mode == SINGLE_EXIT) {
803 p->p_flag |= P_SINGLE_EXIT;
804 p->p_flag &= ~P_SINGLE_BOUNDARY;
805 } else {
806 p->p_flag &= ~P_SINGLE_EXIT;
807 if (mode == SINGLE_BOUNDARY)
808 p->p_flag |= P_SINGLE_BOUNDARY;
809 else
810 p->p_flag &= ~P_SINGLE_BOUNDARY;
811 }
812 if (mode == SINGLE_ALLPROC)
813 p->p_flag |= P_TOTAL_STOP;
814 p->p_flag |= P_STOPPED_SINGLE;
815 PROC_SLOCK(p);
816 p->p_singlethread = td;
817 remaining = calc_remaining(p, mode);
818 while (remaining != remain_for_mode(mode)) {
819 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
820 goto stopme;
821 wakeup_swapper = 0;
822 FOREACH_THREAD_IN_PROC(p, td2) {
823 if (td2 == td)
824 continue;
825 thread_lock(td2);
826 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
827 if (TD_IS_INHIBITED(td2)) {
828 wakeup_swapper |= weed_inhib(mode, td2, p);
829 #ifdef SMP
830 } else if (TD_IS_RUNNING(td2) && td != td2) {
831 forward_signal(td2);
832 #endif
833 }
834 thread_unlock(td2);
835 }
836 if (wakeup_swapper)
837 kick_proc0();
838 remaining = calc_remaining(p, mode);
839
840 /*
841 * Maybe we suspended some threads.. was it enough?
842 */
843 if (remaining == remain_for_mode(mode))
844 break;
845
846 stopme:
847 /*
848 * Wake us up when everyone else has suspended.
849 * In the mean time we suspend as well.
850 */
851 thread_suspend_switch(td, p);
852 remaining = calc_remaining(p, mode);
853 }
854 if (mode == SINGLE_EXIT) {
855 /*
856 * Convert the process to an unthreaded process. The
857 * SINGLE_EXIT is called by exit1() or execve(), in
858 * both cases other threads must be retired.
859 */
860 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
861 p->p_singlethread = NULL;
862 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
863
864 /*
865 * Wait for any remaining threads to exit cpu_throw().
866 */
867 while (p->p_exitthreads != 0) {
868 PROC_SUNLOCK(p);
869 PROC_UNLOCK(p);
870 sched_relinquish(td);
871 PROC_LOCK(p);
872 PROC_SLOCK(p);
873 }
874 } else if (mode == SINGLE_BOUNDARY) {
875 /*
876 * Wait until all suspended threads are removed from
877 * the processors. The thread_suspend_check()
878 * increments p_boundary_count while it is still
879 * running, which makes it possible for the execve()
880 * to destroy vmspace while our other threads are
881 * still using the address space.
882 *
883 * We lock the thread, which is only allowed to
884 * succeed after context switch code finished using
885 * the address space.
886 */
887 FOREACH_THREAD_IN_PROC(p, td2) {
888 if (td2 == td)
889 continue;
890 thread_lock(td2);
891 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
892 ("td %p not on boundary", td2));
893 KASSERT(TD_IS_SUSPENDED(td2),
894 ("td %p is not suspended", td2));
895 thread_unlock(td2);
896 }
897 }
898 PROC_SUNLOCK(p);
899 return (0);
900 }
901
902 bool
903 thread_suspend_check_needed(void)
904 {
905 struct proc *p;
906 struct thread *td;
907
908 td = curthread;
909 p = td->td_proc;
910 PROC_LOCK_ASSERT(p, MA_OWNED);
911 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
912 (td->td_dbgflags & TDB_SUSPEND) != 0));
913 }
914
915 /*
916 * Called in from locations that can safely check to see
917 * whether we have to suspend or at least throttle for a
918 * single-thread event (e.g. fork).
919 *
920 * Such locations include userret().
921 * If the "return_instead" argument is non zero, the thread must be able to
922 * accept 0 (caller may continue), or 1 (caller must abort) as a result.
923 *
924 * The 'return_instead' argument tells the function if it may do a
925 * thread_exit() or suspend, or whether the caller must abort and back
926 * out instead.
927 *
928 * If the thread that set the single_threading request has set the
929 * P_SINGLE_EXIT bit in the process flags then this call will never return
930 * if 'return_instead' is false, but will exit.
931 *
932 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
933 *---------------+--------------------+---------------------
934 * 0 | returns 0 | returns 0 or 1
935 * | when ST ends | immediately
936 *---------------+--------------------+---------------------
937 * 1 | thread exits | returns 1
938 * | | immediately
939 * 0 = thread_exit() or suspension ok,
940 * other = return error instead of stopping the thread.
941 *
942 * While a full suspension is under effect, even a single threading
943 * thread would be suspended if it made this call (but it shouldn't).
944 * This call should only be made from places where
945 * thread_exit() would be safe as that may be the outcome unless
946 * return_instead is set.
947 */
948 int
949 thread_suspend_check(int return_instead)
950 {
951 struct thread *td;
952 struct proc *p;
953 int wakeup_swapper;
954
955 td = curthread;
956 p = td->td_proc;
957 mtx_assert(&Giant, MA_NOTOWNED);
958 PROC_LOCK_ASSERT(p, MA_OWNED);
959 while (thread_suspend_check_needed()) {
960 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
961 KASSERT(p->p_singlethread != NULL,
962 ("singlethread not set"));
963 /*
964 * The only suspension in action is a
965 * single-threading. Single threader need not stop.
966 * It is safe to access p->p_singlethread unlocked
967 * because it can only be set to our address by us.
968 */
969 if (p->p_singlethread == td)
970 return (0); /* Exempt from stopping. */
971 }
972 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
973 return (EINTR);
974
975 /* Should we goto user boundary if we didn't come from there? */
976 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
977 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
978 return (ERESTART);
979
980 /*
981 * Ignore suspend requests if they are deferred.
982 */
983 if ((td->td_flags & TDF_SBDRY) != 0) {
984 KASSERT(return_instead,
985 ("TDF_SBDRY set for unsafe thread_suspend_check"));
986 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
987 (TDF_SEINTR | TDF_SERESTART),
988 ("both TDF_SEINTR and TDF_SERESTART"));
989 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
990 }
991
992 /*
993 * If the process is waiting for us to exit,
994 * this thread should just suicide.
995 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
996 */
997 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
998 PROC_UNLOCK(p);
999
1000 /*
1001 * Allow Linux emulation layer to do some work
1002 * before thread suicide.
1003 */
1004 if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
1005 (p->p_sysent->sv_thread_detach)(td);
1006 umtx_thread_exit(td);
1007 kern_thr_exit(td);
1008 panic("stopped thread did not exit");
1009 }
1010
1011 PROC_SLOCK(p);
1012 thread_stopped(p);
1013 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1014 if (p->p_numthreads == p->p_suspcount + 1) {
1015 thread_lock(p->p_singlethread);
1016 wakeup_swapper = thread_unsuspend_one(
1017 p->p_singlethread, p, false);
1018 thread_unlock(p->p_singlethread);
1019 if (wakeup_swapper)
1020 kick_proc0();
1021 }
1022 }
1023 PROC_UNLOCK(p);
1024 thread_lock(td);
1025 /*
1026 * When a thread suspends, it just
1027 * gets taken off all queues.
1028 */
1029 thread_suspend_one(td);
1030 if (return_instead == 0) {
1031 p->p_boundary_count++;
1032 td->td_flags |= TDF_BOUNDARY;
1033 }
1034 PROC_SUNLOCK(p);
1035 mi_switch(SW_INVOL | SWT_SUSPEND, NULL);
1036 thread_unlock(td);
1037 PROC_LOCK(p);
1038 }
1039 return (0);
1040 }
1041
1042 void
1043 thread_suspend_switch(struct thread *td, struct proc *p)
1044 {
1045
1046 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1047 PROC_LOCK_ASSERT(p, MA_OWNED);
1048 PROC_SLOCK_ASSERT(p, MA_OWNED);
1049 /*
1050 * We implement thread_suspend_one in stages here to avoid
1051 * dropping the proc lock while the thread lock is owned.
1052 */
1053 if (p == td->td_proc) {
1054 thread_stopped(p);
1055 p->p_suspcount++;
1056 }
1057 PROC_UNLOCK(p);
1058 thread_lock(td);
1059 td->td_flags &= ~TDF_NEEDSUSPCHK;
1060 TD_SET_SUSPENDED(td);
1061 sched_sleep(td, 0);
1062 PROC_SUNLOCK(p);
1063 DROP_GIANT();
1064 mi_switch(SW_VOL | SWT_SUSPEND, NULL);
1065 thread_unlock(td);
1066 PICKUP_GIANT();
1067 PROC_LOCK(p);
1068 PROC_SLOCK(p);
1069 }
1070
1071 void
1072 thread_suspend_one(struct thread *td)
1073 {
1074 struct proc *p;
1075
1076 p = td->td_proc;
1077 PROC_SLOCK_ASSERT(p, MA_OWNED);
1078 THREAD_LOCK_ASSERT(td, MA_OWNED);
1079 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1080 p->p_suspcount++;
1081 td->td_flags &= ~TDF_NEEDSUSPCHK;
1082 TD_SET_SUSPENDED(td);
1083 sched_sleep(td, 0);
1084 }
1085
1086 static int
1087 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1088 {
1089
1090 THREAD_LOCK_ASSERT(td, MA_OWNED);
1091 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1092 TD_CLR_SUSPENDED(td);
1093 td->td_flags &= ~TDF_ALLPROCSUSP;
1094 if (td->td_proc == p) {
1095 PROC_SLOCK_ASSERT(p, MA_OWNED);
1096 p->p_suspcount--;
1097 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1098 td->td_flags &= ~TDF_BOUNDARY;
1099 p->p_boundary_count--;
1100 }
1101 }
1102 return (setrunnable(td));
1103 }
1104
1105 /*
1106 * Allow all threads blocked by single threading to continue running.
1107 */
1108 void
1109 thread_unsuspend(struct proc *p)
1110 {
1111 struct thread *td;
1112 int wakeup_swapper;
1113
1114 PROC_LOCK_ASSERT(p, MA_OWNED);
1115 PROC_SLOCK_ASSERT(p, MA_OWNED);
1116 wakeup_swapper = 0;
1117 if (!P_SHOULDSTOP(p)) {
1118 FOREACH_THREAD_IN_PROC(p, td) {
1119 thread_lock(td);
1120 if (TD_IS_SUSPENDED(td)) {
1121 wakeup_swapper |= thread_unsuspend_one(td, p,
1122 true);
1123 }
1124 thread_unlock(td);
1125 }
1126 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1127 p->p_numthreads == p->p_suspcount) {
1128 /*
1129 * Stopping everything also did the job for the single
1130 * threading request. Now we've downgraded to single-threaded,
1131 * let it continue.
1132 */
1133 if (p->p_singlethread->td_proc == p) {
1134 thread_lock(p->p_singlethread);
1135 wakeup_swapper = thread_unsuspend_one(
1136 p->p_singlethread, p, false);
1137 thread_unlock(p->p_singlethread);
1138 }
1139 }
1140 if (wakeup_swapper)
1141 kick_proc0();
1142 }
1143
1144 /*
1145 * End the single threading mode..
1146 */
1147 void
1148 thread_single_end(struct proc *p, int mode)
1149 {
1150 struct thread *td;
1151 int wakeup_swapper;
1152
1153 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1154 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1155 ("invalid mode %d", mode));
1156 PROC_LOCK_ASSERT(p, MA_OWNED);
1157 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1158 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1159 ("mode %d does not match P_TOTAL_STOP", mode));
1160 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1161 ("thread_single_end from other thread %p %p",
1162 curthread, p->p_singlethread));
1163 KASSERT(mode != SINGLE_BOUNDARY ||
1164 (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1165 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1166 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1167 P_TOTAL_STOP);
1168 PROC_SLOCK(p);
1169 p->p_singlethread = NULL;
1170 wakeup_swapper = 0;
1171 /*
1172 * If there are other threads they may now run,
1173 * unless of course there is a blanket 'stop order'
1174 * on the process. The single threader must be allowed
1175 * to continue however as this is a bad place to stop.
1176 */
1177 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1178 FOREACH_THREAD_IN_PROC(p, td) {
1179 thread_lock(td);
1180 if (TD_IS_SUSPENDED(td)) {
1181 wakeup_swapper |= thread_unsuspend_one(td, p,
1182 mode == SINGLE_BOUNDARY);
1183 }
1184 thread_unlock(td);
1185 }
1186 }
1187 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1188 ("inconsistent boundary count %d", p->p_boundary_count));
1189 PROC_SUNLOCK(p);
1190 if (wakeup_swapper)
1191 kick_proc0();
1192 }
1193
1194 struct thread *
1195 thread_find(struct proc *p, lwpid_t tid)
1196 {
1197 struct thread *td;
1198
1199 PROC_LOCK_ASSERT(p, MA_OWNED);
1200 FOREACH_THREAD_IN_PROC(p, td) {
1201 if (td->td_tid == tid)
1202 break;
1203 }
1204 return (td);
1205 }
1206
1207 /* Locate a thread by number; return with proc lock held. */
1208 struct thread *
1209 tdfind(lwpid_t tid, pid_t pid)
1210 {
1211 #define RUN_THRESH 16
1212 struct thread *td;
1213 int run = 0;
1214
1215 rw_rlock(&tidhash_lock);
1216 LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1217 if (td->td_tid == tid) {
1218 if (pid != -1 && td->td_proc->p_pid != pid) {
1219 td = NULL;
1220 break;
1221 }
1222 PROC_LOCK(td->td_proc);
1223 if (td->td_proc->p_state == PRS_NEW) {
1224 PROC_UNLOCK(td->td_proc);
1225 td = NULL;
1226 break;
1227 }
1228 if (run > RUN_THRESH) {
1229 if (rw_try_upgrade(&tidhash_lock)) {
1230 LIST_REMOVE(td, td_hash);
1231 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1232 td, td_hash);
1233 rw_wunlock(&tidhash_lock);
1234 return (td);
1235 }
1236 }
1237 break;
1238 }
1239 run++;
1240 }
1241 rw_runlock(&tidhash_lock);
1242 return (td);
1243 }
1244
1245 void
1246 tidhash_add(struct thread *td)
1247 {
1248 rw_wlock(&tidhash_lock);
1249 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1250 rw_wunlock(&tidhash_lock);
1251 }
1252
1253 void
1254 tidhash_remove(struct thread *td)
1255 {
1256 rw_wlock(&tidhash_lock);
1257 LIST_REMOVE(td, td_hash);
1258 rw_wunlock(&tidhash_lock);
1259 }
Cache object: 7da0336c8a83c5f05af03ffbabc917bb
|