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: stable/10/sys/kern/kern_thread.c 337258 2018-08-03 14:45:53Z asomers $"); 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 67SDT_PROVIDER_DECLARE(proc); 68SDT_PROBE_DEFINE(proc, , , lwp__exit); 69 70/* 71 * thread related storage. 72 */ 73static uma_zone_t thread_zone; 74 75TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); 76static struct mtx zombie_lock; 77MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); 78 79static void thread_zombie(struct thread *); 80static int thread_unsuspend_one(struct thread *td, struct proc *p, 81 bool boundary); 82 83#define TID_BUFFER_SIZE 1024 84 85struct mtx tid_lock; 86static struct unrhdr *tid_unrhdr; 87static lwpid_t tid_buffer[TID_BUFFER_SIZE]; 88static int tid_head, tid_tail; 89static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); 90 91struct tidhashhead *tidhashtbl; 92u_long tidhash; 93struct rwlock tidhash_lock; 94 95static lwpid_t 96tid_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 114static void 115tid_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 */ 134static int 135thread_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 */ 163static void 164thread_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 */ 203static int 204thread_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 */ 224static void 225thread_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 */ 246void 247proc_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 253void 254proc_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 */ 271void 272threadinit(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 */ 294void 295thread_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 */ 305void 306thread_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 */ 315void 316thread_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 */ 343struct thread * 344thread_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 360int 361thread_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 */ 375void 376thread_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 */ 399void 400thread_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 /* 422 * drop FPU & debug register state storage, or any other 423 * architecture specific resources that 424 * would not be on a new untouched process. 425 */ 426 cpu_thread_exit(td); 427 428 /* 429 * The last thread is left attached to the process 430 * So that the whole bundle gets recycled. Skip 431 * all this stuff if we never had threads. 432 * EXIT clears all sign of other threads when 433 * it goes to single threading, so the last thread always 434 * takes the short path. 435 */ 436 if (p->p_flag & P_HADTHREADS) { 437 if (p->p_numthreads > 1) { 438 atomic_add_int(&td->td_proc->p_exitthreads, 1); 439 thread_unlink(td); 440 td2 = FIRST_THREAD_IN_PROC(p); 441 sched_exit_thread(td2, td); 442 443 /* 444 * The test below is NOT true if we are the 445 * sole exiting thread. P_STOPPED_SINGLE is unset 446 * in exit1() after it is the only survivor. 447 */ 448 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 449 if (p->p_numthreads == p->p_suspcount) { 450 thread_lock(p->p_singlethread); 451 wakeup_swapper = thread_unsuspend_one( 452 p->p_singlethread, p, false); 453 thread_unlock(p->p_singlethread); 454 if (wakeup_swapper) 455 kick_proc0(); 456 } 457 } 458 459 PCPU_SET(deadthread, td); 460 } else { 461 /* 462 * The last thread is exiting.. but not through exit() 463 */ 464 panic ("thread_exit: Last thread exiting on its own"); 465 } 466 } 467#ifdef HWPMC_HOOKS 468 /* 469 * If this thread is part of a process that is being tracked by hwpmc(4), 470 * inform the module of the thread's impending exit. 471 */ 472 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 473 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 474#endif 475 PROC_UNLOCK(p); 476 PROC_STATLOCK(p); 477 thread_lock(td); 478 PROC_SUNLOCK(p); 479 480 /* Do the same timestamp bookkeeping that mi_switch() would do. */ 481 new_switchtime = cpu_ticks(); 482 runtime = new_switchtime - PCPU_GET(switchtime); 483 td->td_runtime += runtime; 484 td->td_incruntime += runtime; 485 PCPU_SET(switchtime, new_switchtime); 486 PCPU_SET(switchticks, ticks); 487 PCPU_INC(cnt.v_swtch); 488 489 /* Save our resource usage in our process. */ 490 td->td_ru.ru_nvcsw++; 491 ruxagg(p, td); 492 rucollect(&p->p_ru, &td->td_ru); 493 PROC_STATUNLOCK(p); 494 495 td->td_state = TDS_INACTIVE; 496#ifdef WITNESS 497 witness_thread_exit(td); 498#endif 499 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); 500 sched_throw(td); 501 panic("I'm a teapot!"); 502 /* NOTREACHED */ 503} 504 505/* 506 * Do any thread specific cleanups that may be needed in wait() 507 * called with Giant, proc and schedlock not held. 508 */ 509void 510thread_wait(struct proc *p) 511{ 512 struct thread *td; 513 514 mtx_assert(&Giant, MA_NOTOWNED); 515 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); 516 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); 517 td = FIRST_THREAD_IN_PROC(p); 518 /* Lock the last thread so we spin until it exits cpu_throw(). */ 519 thread_lock(td); 520 thread_unlock(td); 521 lock_profile_thread_exit(td); 522 cpuset_rel(td->td_cpuset); 523 td->td_cpuset = NULL; 524 cpu_thread_clean(td); 525 crfree(td->td_ucred); 526 callout_drain(&td->td_slpcallout); 527 thread_reap(); /* check for zombie threads etc. */ 528} 529 530/* 531 * Link a thread to a process. 532 * set up anything that needs to be initialized for it to 533 * be used by the process. 534 */ 535void 536thread_link(struct thread *td, struct proc *p) 537{ 538 539 /* 540 * XXX This can't be enabled because it's called for proc0 before 541 * its lock has been created. 542 * PROC_LOCK_ASSERT(p, MA_OWNED); 543 */ 544 td->td_state = TDS_INACTIVE; 545 td->td_proc = p; 546 td->td_flags = TDF_INMEM; 547 548 LIST_INIT(&td->td_contested); 549 LIST_INIT(&td->td_lprof[0]); 550 LIST_INIT(&td->td_lprof[1]); 551 sigqueue_init(&td->td_sigqueue, p); 552 callout_init(&td->td_slpcallout, 1); 553 TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist); 554 p->p_numthreads++; 555} 556 557/* 558 * Called from: 559 * thread_exit() 560 */ 561void 562thread_unlink(struct thread *td) 563{ 564 struct proc *p = td->td_proc; 565 566 PROC_LOCK_ASSERT(p, MA_OWNED); 567 TAILQ_REMOVE(&p->p_threads, td, td_plist); 568 p->p_numthreads--; 569 /* could clear a few other things here */ 570 /* Must NOT clear links to proc! */ 571} 572 573static int 574calc_remaining(struct proc *p, int mode) 575{ 576 int remaining; 577 578 PROC_LOCK_ASSERT(p, MA_OWNED); 579 PROC_SLOCK_ASSERT(p, MA_OWNED); 580 if (mode == SINGLE_EXIT) 581 remaining = p->p_numthreads; 582 else if (mode == SINGLE_BOUNDARY) 583 remaining = p->p_numthreads - p->p_boundary_count; 584 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) 585 remaining = p->p_numthreads - p->p_suspcount; 586 else 587 panic("calc_remaining: wrong mode %d", mode); 588 return (remaining); 589} 590 591static int 592remain_for_mode(int mode) 593{ 594 595 return (mode == SINGLE_ALLPROC ? 0 : 1); 596} 597 598static int 599weed_inhib(int mode, struct thread *td2, struct proc *p) 600{ 601 int wakeup_swapper; 602 603 PROC_LOCK_ASSERT(p, MA_OWNED); 604 PROC_SLOCK_ASSERT(p, MA_OWNED); 605 THREAD_LOCK_ASSERT(td2, MA_OWNED); 606 607 wakeup_swapper = 0; 608 switch (mode) { 609 case SINGLE_EXIT: 610 if (TD_IS_SUSPENDED(td2)) 611 wakeup_swapper |= thread_unsuspend_one(td2, p, true); 612 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 613 wakeup_swapper |= sleepq_abort(td2, EINTR); 614 break; 615 case SINGLE_BOUNDARY: 616 case SINGLE_NO_EXIT: 617 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 618 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 619 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 620 wakeup_swapper |= sleepq_abort(td2, ERESTART); 621 break; 622 case SINGLE_ALLPROC: 623 /* 624 * ALLPROC suspend tries to avoid spurious EINTR for 625 * threads sleeping interruptable, by suspending the 626 * thread directly, similarly to sig_suspend_threads(). 627 * Since such sleep is not performed at the user 628 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP 629 * is used to avoid immediate un-suspend. 630 */ 631 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | 632 TDF_ALLPROCSUSP)) == 0) 633 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 634 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { 635 if ((td2->td_flags & TDF_SBDRY) == 0) { 636 thread_suspend_one(td2); 637 td2->td_flags |= TDF_ALLPROCSUSP; 638 } else { 639 wakeup_swapper |= sleepq_abort(td2, ERESTART); 640 } 641 } 642 break; 643 } 644 return (wakeup_swapper); 645} 646 647/* 648 * Enforce single-threading. 649 * 650 * Returns 1 if the caller must abort (another thread is waiting to 651 * exit the process or similar). Process is locked! 652 * Returns 0 when you are successfully the only thread running. 653 * A process has successfully single threaded in the suspend mode when 654 * There are no threads in user mode. Threads in the kernel must be 655 * allowed to continue until they get to the user boundary. They may even 656 * copy out their return values and data before suspending. They may however be 657 * accelerated in reaching the user boundary as we will wake up 658 * any sleeping threads that are interruptable. (PCATCH). 659 */ 660int 661thread_single(struct proc *p, int mode) 662{ 663 struct thread *td; 664 struct thread *td2; 665 int remaining, wakeup_swapper; 666 667 td = curthread; 668 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 669 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 670 ("invalid mode %d", mode)); 671 /* 672 * If allowing non-ALLPROC singlethreading for non-curproc 673 * callers, calc_remaining() and remain_for_mode() should be 674 * adjusted to also account for td->td_proc != p. For now 675 * this is not implemented because it is not used. 676 */ 677 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || 678 (mode != SINGLE_ALLPROC && td->td_proc == p), 679 ("mode %d proc %p curproc %p", mode, p, td->td_proc)); 680 mtx_assert(&Giant, MA_NOTOWNED); 681 PROC_LOCK_ASSERT(p, MA_OWNED); 682 683 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) 684 return (0); 685 686 /* Is someone already single threading? */ 687 if (p->p_singlethread != NULL && p->p_singlethread != td) 688 return (1); 689 690 if (mode == SINGLE_EXIT) { 691 p->p_flag |= P_SINGLE_EXIT; 692 p->p_flag &= ~P_SINGLE_BOUNDARY; 693 } else { 694 p->p_flag &= ~P_SINGLE_EXIT; 695 if (mode == SINGLE_BOUNDARY) 696 p->p_flag |= P_SINGLE_BOUNDARY; 697 else 698 p->p_flag &= ~P_SINGLE_BOUNDARY; 699 } 700 if (mode == SINGLE_ALLPROC) 701 p->p_flag |= P_TOTAL_STOP; 702 p->p_flag |= P_STOPPED_SINGLE; 703 PROC_SLOCK(p); 704 p->p_singlethread = td; 705 remaining = calc_remaining(p, mode); 706 while (remaining != remain_for_mode(mode)) { 707 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) 708 goto stopme; 709 wakeup_swapper = 0; 710 FOREACH_THREAD_IN_PROC(p, td2) { 711 if (td2 == td) 712 continue; 713 thread_lock(td2); 714 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 715 if (TD_IS_INHIBITED(td2)) { 716 wakeup_swapper |= weed_inhib(mode, td2, p); 717#ifdef SMP 718 } else if (TD_IS_RUNNING(td2) && td != td2) { 719 forward_signal(td2); 720#endif 721 } 722 thread_unlock(td2); 723 } 724 if (wakeup_swapper) 725 kick_proc0(); 726 remaining = calc_remaining(p, mode); 727 728 /* 729 * Maybe we suspended some threads.. was it enough? 730 */ 731 if (remaining == remain_for_mode(mode)) 732 break; 733 734stopme: 735 /* 736 * Wake us up when everyone else has suspended. 737 * In the mean time we suspend as well. 738 */ 739 thread_suspend_switch(td, p); 740 remaining = calc_remaining(p, mode); 741 } 742 if (mode == SINGLE_EXIT) { 743 /* 744 * Convert the process to an unthreaded process. The 745 * SINGLE_EXIT is called by exit1() or execve(), in 746 * both cases other threads must be retired. 747 */ 748 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); 749 p->p_singlethread = NULL; 750 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); 751 752 /* 753 * Wait for any remaining threads to exit cpu_throw(). 754 */ 755 while (p->p_exitthreads != 0) { 756 PROC_SUNLOCK(p); 757 PROC_UNLOCK(p); 758 sched_relinquish(td); 759 PROC_LOCK(p); 760 PROC_SLOCK(p); 761 } 762 } else if (mode == SINGLE_BOUNDARY) { 763 /* 764 * Wait until all suspended threads are removed from 765 * the processors. The thread_suspend_check() 766 * increments p_boundary_count while it is still 767 * running, which makes it possible for the execve() 768 * to destroy vmspace while our other threads are 769 * still using the address space. 770 * 771 * We lock the thread, which is only allowed to 772 * succeed after context switch code finished using 773 * the address space. 774 */ 775 FOREACH_THREAD_IN_PROC(p, td2) { 776 if (td2 == td) 777 continue; 778 thread_lock(td2); 779 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, 780 ("td %p not on boundary", td2)); 781 KASSERT(TD_IS_SUSPENDED(td2), 782 ("td %p is not suspended", td2)); 783 thread_unlock(td2); 784 } 785 } 786 PROC_SUNLOCK(p); 787 return (0); 788} 789 790bool 791thread_suspend_check_needed(void) 792{ 793 struct proc *p; 794 struct thread *td; 795 796 td = curthread; 797 p = td->td_proc; 798 PROC_LOCK_ASSERT(p, MA_OWNED); 799 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && 800 (td->td_dbgflags & TDB_SUSPEND) != 0)); 801} 802 803/* 804 * Called in from locations that can safely check to see 805 * whether we have to suspend or at least throttle for a 806 * single-thread event (e.g. fork). 807 * 808 * Such locations include userret(). 809 * If the "return_instead" argument is non zero, the thread must be able to 810 * accept 0 (caller may continue), or 1 (caller must abort) as a result. 811 * 812 * The 'return_instead' argument tells the function if it may do a 813 * thread_exit() or suspend, or whether the caller must abort and back 814 * out instead. 815 * 816 * If the thread that set the single_threading request has set the 817 * P_SINGLE_EXIT bit in the process flags then this call will never return 818 * if 'return_instead' is false, but will exit. 819 * 820 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 821 *---------------+--------------------+--------------------- 822 * 0 | returns 0 | returns 0 or 1 823 * | when ST ends | immediately 824 *---------------+--------------------+--------------------- 825 * 1 | thread exits | returns 1 826 * | | immediately 827 * 0 = thread_exit() or suspension ok, 828 * other = return error instead of stopping the thread. 829 * 830 * While a full suspension is under effect, even a single threading 831 * thread would be suspended if it made this call (but it shouldn't). 832 * This call should only be made from places where 833 * thread_exit() would be safe as that may be the outcome unless 834 * return_instead is set. 835 */ 836int 837thread_suspend_check(int return_instead) 838{ 839 struct thread *td; 840 struct proc *p; 841 int wakeup_swapper; 842 843 td = curthread; 844 p = td->td_proc; 845 mtx_assert(&Giant, MA_NOTOWNED); 846 PROC_LOCK_ASSERT(p, MA_OWNED); 847 while (thread_suspend_check_needed()) { 848 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 849 KASSERT(p->p_singlethread != NULL, 850 ("singlethread not set")); 851 /* 852 * The only suspension in action is a 853 * single-threading. Single threader need not stop. 854 * It is safe to access p->p_singlethread unlocked 855 * because it can only be set to our address by us. 856 */ 857 if (p->p_singlethread == td) 858 return (0); /* Exempt from stopping. */ 859 } 860 if ((p->p_flag & P_SINGLE_EXIT) && return_instead) 861 return (EINTR); 862 863 /* Should we goto user boundary if we didn't come from there? */ 864 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 865 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) 866 return (ERESTART); 867 868 /* 869 * Ignore suspend requests if they are deferred. 870 */ 871 if ((td->td_flags & TDF_SBDRY) != 0) { 872 KASSERT(return_instead, 873 ("TDF_SBDRY set for unsafe thread_suspend_check")); 874 return (0); 875 } 876 877 /* 878 * If the process is waiting for us to exit, 879 * this thread should just suicide. 880 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. 881 */ 882 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { 883 PROC_UNLOCK(p); 884 885 /* 886 * Allow Linux emulation layer to do some work 887 * before thread suicide. 888 */ 889 if (__predict_false(p->p_sysent->sv_thread_detach != NULL)) 890 (p->p_sysent->sv_thread_detach)(td); 891 kern_thr_exit(td); 892 panic("stopped thread did not exit"); 893 } 894 895 PROC_SLOCK(p); 896 thread_stopped(p); 897 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 898 if (p->p_numthreads == p->p_suspcount + 1) { 899 thread_lock(p->p_singlethread); 900 wakeup_swapper = thread_unsuspend_one( 901 p->p_singlethread, p, false); 902 thread_unlock(p->p_singlethread); 903 if (wakeup_swapper) 904 kick_proc0(); 905 } 906 } 907 PROC_UNLOCK(p); 908 thread_lock(td); 909 /* 910 * When a thread suspends, it just 911 * gets taken off all queues. 912 */ 913 thread_suspend_one(td); 914 if (return_instead == 0) { 915 p->p_boundary_count++; 916 td->td_flags |= TDF_BOUNDARY; 917 } 918 PROC_SUNLOCK(p); 919 mi_switch(SW_INVOL | SWT_SUSPEND, NULL); 920 thread_unlock(td); 921 PROC_LOCK(p); 922 } 923 return (0); 924} 925 926void 927thread_suspend_switch(struct thread *td, struct proc *p) 928{ 929 930 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 931 PROC_LOCK_ASSERT(p, MA_OWNED); 932 PROC_SLOCK_ASSERT(p, MA_OWNED); 933 /* 934 * We implement thread_suspend_one in stages here to avoid 935 * dropping the proc lock while the thread lock is owned. 936 */ 937 if (p == td->td_proc) { 938 thread_stopped(p); 939 p->p_suspcount++; 940 } 941 PROC_UNLOCK(p); 942 thread_lock(td); 943 td->td_flags &= ~TDF_NEEDSUSPCHK; 944 TD_SET_SUSPENDED(td); 945 sched_sleep(td, 0); 946 PROC_SUNLOCK(p); 947 DROP_GIANT(); 948 mi_switch(SW_VOL | SWT_SUSPEND, NULL); 949 thread_unlock(td); 950 PICKUP_GIANT(); 951 PROC_LOCK(p); 952 PROC_SLOCK(p); 953} 954 955void 956thread_suspend_one(struct thread *td) 957{ 958 struct proc *p; 959 960 p = td->td_proc; 961 PROC_SLOCK_ASSERT(p, MA_OWNED); 962 THREAD_LOCK_ASSERT(td, MA_OWNED); 963 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 964 p->p_suspcount++; 965 td->td_flags &= ~TDF_NEEDSUSPCHK; 966 TD_SET_SUSPENDED(td); 967 sched_sleep(td, 0); 968} 969 970static int 971thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) 972{ 973 974 THREAD_LOCK_ASSERT(td, MA_OWNED); 975 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); 976 TD_CLR_SUSPENDED(td); 977 td->td_flags &= ~TDF_ALLPROCSUSP; 978 if (td->td_proc == p) { 979 PROC_SLOCK_ASSERT(p, MA_OWNED); 980 p->p_suspcount--; 981 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { 982 td->td_flags &= ~TDF_BOUNDARY; 983 p->p_boundary_count--; 984 } 985 } 986 return (setrunnable(td)); 987} 988 989/* 990 * Allow all threads blocked by single threading to continue running. 991 */ 992void 993thread_unsuspend(struct proc *p) 994{ 995 struct thread *td; 996 int wakeup_swapper; 997 998 PROC_LOCK_ASSERT(p, MA_OWNED); 999 PROC_SLOCK_ASSERT(p, MA_OWNED); 1000 wakeup_swapper = 0; 1001 if (!P_SHOULDSTOP(p)) { 1002 FOREACH_THREAD_IN_PROC(p, td) { 1003 thread_lock(td); 1004 if (TD_IS_SUSPENDED(td)) { 1005 wakeup_swapper |= thread_unsuspend_one(td, p, 1006 true); 1007 } 1008 thread_unlock(td); 1009 } 1010 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 1011 p->p_numthreads == p->p_suspcount) { 1012 /* 1013 * Stopping everything also did the job for the single 1014 * threading request. Now we've downgraded to single-threaded, 1015 * let it continue. 1016 */ 1017 if (p->p_singlethread->td_proc == p) { 1018 thread_lock(p->p_singlethread); 1019 wakeup_swapper = thread_unsuspend_one( 1020 p->p_singlethread, p, false); 1021 thread_unlock(p->p_singlethread); 1022 } 1023 } 1024 if (wakeup_swapper) 1025 kick_proc0(); 1026} 1027 1028/* 1029 * End the single threading mode.. 1030 */ 1031void 1032thread_single_end(struct proc *p, int mode) 1033{ 1034 struct thread *td; 1035 int wakeup_swapper; 1036 1037 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 1038 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 1039 ("invalid mode %d", mode)); 1040 PROC_LOCK_ASSERT(p, MA_OWNED); 1041 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || 1042 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), 1043 ("mode %d does not match P_TOTAL_STOP", mode)); 1044 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, 1045 ("thread_single_end from other thread %p %p", 1046 curthread, p->p_singlethread)); 1047 KASSERT(mode != SINGLE_BOUNDARY || 1048 (p->p_flag & P_SINGLE_BOUNDARY) != 0, 1049 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); 1050 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | 1051 P_TOTAL_STOP); 1052 PROC_SLOCK(p); 1053 p->p_singlethread = NULL; 1054 wakeup_swapper = 0; 1055 /* 1056 * If there are other threads they may now run, 1057 * unless of course there is a blanket 'stop order' 1058 * on the process. The single threader must be allowed 1059 * to continue however as this is a bad place to stop. 1060 */ 1061 if (p->p_numthreads != remain_for_mode(mode) && !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 mode == SINGLE_BOUNDARY); 1067 } 1068 thread_unlock(td); 1069 } 1070 } 1071 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, 1072 ("inconsistent boundary count %d", p->p_boundary_count)); 1073 PROC_SUNLOCK(p); 1074 if (wakeup_swapper) 1075 kick_proc0(); 1076} 1077 1078struct thread * 1079thread_find(struct proc *p, lwpid_t tid) 1080{ 1081 struct thread *td; 1082 1083 PROC_LOCK_ASSERT(p, MA_OWNED); 1084 FOREACH_THREAD_IN_PROC(p, td) { 1085 if (td->td_tid == tid) 1086 break; 1087 } 1088 return (td); 1089} 1090 1091/* Locate a thread by number; return with proc lock held. */ 1092struct thread * 1093tdfind(lwpid_t tid, pid_t pid) 1094{ 1095#define RUN_THRESH 16 1096 struct thread *td; 1097 int run = 0; 1098 1099 rw_rlock(&tidhash_lock); 1100 LIST_FOREACH(td, TIDHASH(tid), td_hash) { 1101 if (td->td_tid == tid) { 1102 if (pid != -1 && td->td_proc->p_pid != pid) { 1103 td = NULL; 1104 break; 1105 } 1106 PROC_LOCK(td->td_proc); 1107 if (td->td_proc->p_state == PRS_NEW) { 1108 PROC_UNLOCK(td->td_proc); 1109 td = NULL; 1110 break; 1111 } 1112 if (run > RUN_THRESH) { 1113 if (rw_try_upgrade(&tidhash_lock)) { 1114 LIST_REMOVE(td, td_hash); 1115 LIST_INSERT_HEAD(TIDHASH(td->td_tid), 1116 td, td_hash); 1117 rw_wunlock(&tidhash_lock); 1118 return (td); 1119 } 1120 } 1121 break; 1122 } 1123 run++; 1124 } 1125 rw_runlock(&tidhash_lock); 1126 return (td); 1127} 1128 1129void 1130tidhash_add(struct thread *td) 1131{ 1132 rw_wlock(&tidhash_lock); 1133 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); 1134 rw_wunlock(&tidhash_lock); 1135} 1136 1137void 1138tidhash_remove(struct thread *td) 1139{ 1140 rw_wlock(&tidhash_lock); 1141 LIST_REMOVE(td, td_hash); 1142 rw_wunlock(&tidhash_lock); 1143} 1144